diff options
Diffstat (limited to 'drivers/gpu/drm/amd/amdgpu')
97 files changed, 10581 insertions, 10582 deletions
diff --git a/drivers/gpu/drm/amd/amdgpu/Makefile b/drivers/gpu/drm/amd/amdgpu/Makefile index 04c270757030..c7fcdcedaadb 100644 --- a/drivers/gpu/drm/amd/amdgpu/Makefile +++ b/drivers/gpu/drm/amd/amdgpu/Makefile @@ -2,10 +2,14 @@ # Makefile for the drm device driver. This driver provides support for the # Direct Rendering Infrastructure (DRI) in XFree86 4.1.0 and higher. -ccflags-y := -Iinclude/drm -Idrivers/gpu/drm/amd/include/asic_reg \ - -Idrivers/gpu/drm/amd/include \ - -Idrivers/gpu/drm/amd/amdgpu \ - -Idrivers/gpu/drm/amd/scheduler +FULL_AMD_PATH=$(src)/.. + +ccflags-y := -Iinclude/drm -I$(FULL_AMD_PATH)/include/asic_reg \ + -I$(FULL_AMD_PATH)/include \ + -I$(FULL_AMD_PATH)/amdgpu \ + -I$(FULL_AMD_PATH)/scheduler \ + -I$(FULL_AMD_PATH)/powerplay/inc \ + -I$(FULL_AMD_PATH)/acp/include amdgpu-y := amdgpu_drv.o @@ -17,12 +21,12 @@ amdgpu-y += amdgpu_device.o amdgpu_kms.o \ amdgpu_fb.o amdgpu_gem.o amdgpu_ring.o \ amdgpu_cs.o amdgpu_bios.o amdgpu_benchmark.o amdgpu_test.o \ amdgpu_pm.o atombios_dp.o amdgpu_afmt.o amdgpu_trace_points.o \ - atombios_encoders.o amdgpu_semaphore.o amdgpu_sa.o atombios_i2c.o \ + atombios_encoders.o amdgpu_sa.o atombios_i2c.o \ amdgpu_prime.o amdgpu_vm.o amdgpu_ib.o amdgpu_pll.o \ amdgpu_ucode.o amdgpu_bo_list.o amdgpu_ctx.o amdgpu_sync.o # add asic specific block -amdgpu-$(CONFIG_DRM_AMDGPU_CIK)+= cik.o gmc_v7_0.o cik_ih.o kv_smc.o kv_dpm.o \ +amdgpu-$(CONFIG_DRM_AMDGPU_CIK)+= cik.o cik_ih.o kv_smc.o kv_dpm.o \ ci_smc.o ci_dpm.o dce_v8_0.o gfx_v7_0.o cik_sdma.o uvd_v4_2.o vce_v2_0.o \ amdgpu_amdkfd_gfx_v7.o @@ -31,6 +35,7 @@ amdgpu-y += \ # add GMC block amdgpu-y += \ + gmc_v7_0.o \ gmc_v8_0.o # add IH block @@ -44,6 +49,7 @@ amdgpu-y += \ # add SMC block amdgpu-y += \ amdgpu_dpm.o \ + amdgpu_powerplay.o \ cz_smc.o cz_dpm.o \ tonga_smc.o tonga_dpm.o \ fiji_smc.o fiji_dpm.o \ @@ -87,13 +93,31 @@ amdgpu-y += amdgpu_cgs.o amdgpu-y += \ ../scheduler/gpu_scheduler.o \ ../scheduler/sched_fence.o \ - amdgpu_sched.o + amdgpu_job.o + +# ACP componet +ifneq ($(CONFIG_DRM_AMD_ACP),) +amdgpu-y += amdgpu_acp.o + +AMDACPPATH := ../acp +include $(FULL_AMD_PATH)/acp/Makefile + +amdgpu-y += $(AMD_ACP_FILES) +endif amdgpu-$(CONFIG_COMPAT) += amdgpu_ioc32.o amdgpu-$(CONFIG_VGA_SWITCHEROO) += amdgpu_atpx_handler.o amdgpu-$(CONFIG_ACPI) += amdgpu_acpi.o amdgpu-$(CONFIG_MMU_NOTIFIER) += amdgpu_mn.o +ifneq ($(CONFIG_DRM_AMD_POWERPLAY),) + +include $(FULL_AMD_PATH)/powerplay/Makefile + +amdgpu-y += $(AMD_POWERPLAY_FILES) + +endif + obj-$(CONFIG_DRM_AMDGPU)+= amdgpu.o CFLAGS_amdgpu_trace_points.o := -I$(src) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h index 0d13e6368b96..c4a21c6428f5 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h @@ -52,6 +52,8 @@ #include "amdgpu_irq.h" #include "amdgpu_ucode.h" #include "amdgpu_gds.h" +#include "amd_powerplay.h" +#include "amdgpu_acp.h" #include "gpu_scheduler.h" @@ -73,16 +75,18 @@ extern int amdgpu_dpm; extern int amdgpu_smc_load_fw; extern int amdgpu_aspm; extern int amdgpu_runtime_pm; -extern int amdgpu_hard_reset; extern unsigned amdgpu_ip_block_mask; extern int amdgpu_bapm; extern int amdgpu_deep_color; extern int amdgpu_vm_size; extern int amdgpu_vm_block_size; -extern int amdgpu_enable_scheduler; +extern int amdgpu_vm_fault_stop; +extern int amdgpu_vm_debug; extern int amdgpu_sched_jobs; extern int amdgpu_sched_hw_submission; -extern int amdgpu_enable_semaphores; +extern int amdgpu_powerplay; +extern unsigned amdgpu_pcie_gen_cap; +extern unsigned amdgpu_pcie_lane_cap; #define AMDGPU_WAIT_IDLE_TIMEOUT_IN_MS 3000 #define AMDGPU_MAX_USEC_TIMEOUT 100000 /* 100 ms */ @@ -102,9 +106,6 @@ extern int amdgpu_enable_semaphores; /* max number of IP instances */ #define AMDGPU_MAX_SDMA_INSTANCES 2 -/* number of hw syncs before falling back on blocking */ -#define AMDGPU_NUM_SYNCS 4 - /* hardcode that limit for now */ #define AMDGPU_VA_RESERVED_SIZE (8 << 20) @@ -128,47 +129,6 @@ extern int amdgpu_enable_semaphores; #define AMDGPU_RESET_VCE (1 << 13) #define AMDGPU_RESET_VCE1 (1 << 14) -/* CG block flags */ -#define AMDGPU_CG_BLOCK_GFX (1 << 0) -#define AMDGPU_CG_BLOCK_MC (1 << 1) -#define AMDGPU_CG_BLOCK_SDMA (1 << 2) -#define AMDGPU_CG_BLOCK_UVD (1 << 3) -#define AMDGPU_CG_BLOCK_VCE (1 << 4) -#define AMDGPU_CG_BLOCK_HDP (1 << 5) -#define AMDGPU_CG_BLOCK_BIF (1 << 6) - -/* CG flags */ -#define AMDGPU_CG_SUPPORT_GFX_MGCG (1 << 0) -#define AMDGPU_CG_SUPPORT_GFX_MGLS (1 << 1) -#define AMDGPU_CG_SUPPORT_GFX_CGCG (1 << 2) -#define AMDGPU_CG_SUPPORT_GFX_CGLS (1 << 3) -#define AMDGPU_CG_SUPPORT_GFX_CGTS (1 << 4) -#define AMDGPU_CG_SUPPORT_GFX_CGTS_LS (1 << 5) -#define AMDGPU_CG_SUPPORT_GFX_CP_LS (1 << 6) -#define AMDGPU_CG_SUPPORT_GFX_RLC_LS (1 << 7) -#define AMDGPU_CG_SUPPORT_MC_LS (1 << 8) -#define AMDGPU_CG_SUPPORT_MC_MGCG (1 << 9) -#define AMDGPU_CG_SUPPORT_SDMA_LS (1 << 10) -#define AMDGPU_CG_SUPPORT_SDMA_MGCG (1 << 11) -#define AMDGPU_CG_SUPPORT_BIF_LS (1 << 12) -#define AMDGPU_CG_SUPPORT_UVD_MGCG (1 << 13) -#define AMDGPU_CG_SUPPORT_VCE_MGCG (1 << 14) -#define AMDGPU_CG_SUPPORT_HDP_LS (1 << 15) -#define AMDGPU_CG_SUPPORT_HDP_MGCG (1 << 16) - -/* PG flags */ -#define AMDGPU_PG_SUPPORT_GFX_PG (1 << 0) -#define AMDGPU_PG_SUPPORT_GFX_SMG (1 << 1) -#define AMDGPU_PG_SUPPORT_GFX_DMG (1 << 2) -#define AMDGPU_PG_SUPPORT_UVD (1 << 3) -#define AMDGPU_PG_SUPPORT_VCE (1 << 4) -#define AMDGPU_PG_SUPPORT_CP (1 << 5) -#define AMDGPU_PG_SUPPORT_GDS (1 << 6) -#define AMDGPU_PG_SUPPORT_RLC_SMU_HS (1 << 7) -#define AMDGPU_PG_SUPPORT_SDMA (1 << 8) -#define AMDGPU_PG_SUPPORT_ACP (1 << 9) -#define AMDGPU_PG_SUPPORT_SAMU (1 << 10) - /* GFX current status */ #define AMDGPU_GFX_NORMAL_MODE 0x00000000L #define AMDGPU_GFX_SAFE_MODE 0x00000001L @@ -181,11 +141,9 @@ extern int amdgpu_enable_semaphores; #define CIK_CURSOR_HEIGHT 128 struct amdgpu_device; -struct amdgpu_fence; struct amdgpu_ib; struct amdgpu_vm; struct amdgpu_ring; -struct amdgpu_semaphore; struct amdgpu_cs_parser; struct amdgpu_job; struct amdgpu_irq_src; @@ -283,7 +241,7 @@ struct amdgpu_vm_pte_funcs { unsigned count); /* write pte one entry at a time with addr mapping */ void (*write_pte)(struct amdgpu_ib *ib, - uint64_t pe, + const dma_addr_t *pages_addr, uint64_t pe, uint64_t addr, unsigned count, uint32_t incr, uint32_t flags); /* for linear pte/pde updates without addr mapping */ @@ -291,8 +249,6 @@ struct amdgpu_vm_pte_funcs { uint64_t pe, uint64_t addr, unsigned count, uint32_t incr, uint32_t flags); - /* pad the indirect buffer to the necessary number of dw */ - void (*pad_ib)(struct amdgpu_ib *ib); }; /* provided by the gmc block */ @@ -330,12 +286,11 @@ struct amdgpu_ring_funcs { struct amdgpu_ib *ib); void (*emit_fence)(struct amdgpu_ring *ring, uint64_t addr, uint64_t seq, unsigned flags); - bool (*emit_semaphore)(struct amdgpu_ring *ring, - struct amdgpu_semaphore *semaphore, - bool emit_wait); + void (*emit_pipeline_sync)(struct amdgpu_ring *ring); void (*emit_vm_flush)(struct amdgpu_ring *ring, unsigned vm_id, uint64_t pd_addr); void (*emit_hdp_flush)(struct amdgpu_ring *ring); + void (*emit_hdp_invalidate)(struct amdgpu_ring *ring); void (*emit_gds_switch)(struct amdgpu_ring *ring, uint32_t vmid, uint32_t gds_base, uint32_t gds_size, uint32_t gws_base, uint32_t gws_size, @@ -343,9 +298,10 @@ struct amdgpu_ring_funcs { /* testing functions */ int (*test_ring)(struct amdgpu_ring *ring); int (*test_ib)(struct amdgpu_ring *ring); - bool (*is_lockup)(struct amdgpu_ring *ring); /* insert NOP packets */ void (*insert_nop)(struct amdgpu_ring *ring, uint32_t count); + /* pad the indirect buffer to the necessary number of dw */ + void (*pad_ib)(struct amdgpu_ring *ring, struct amdgpu_ib *ib); }; /* @@ -388,40 +344,27 @@ struct amdgpu_clock { * Fences. */ struct amdgpu_fence_driver { - struct amdgpu_ring *ring; uint64_t gpu_addr; volatile uint32_t *cpu_addr; /* sync_seq is protected by ring emission lock */ - uint64_t sync_seq[AMDGPU_MAX_RINGS]; - atomic64_t last_seq; + uint32_t sync_seq; + atomic_t last_seq; bool initialized; struct amdgpu_irq_src *irq_src; unsigned irq_type; - struct delayed_work lockup_work; - wait_queue_head_t fence_queue; + struct timer_list fallback_timer; + unsigned num_fences_mask; + spinlock_t lock; + struct fence **fences; }; /* some special values for the owner field */ #define AMDGPU_FENCE_OWNER_UNDEFINED ((void*)0ul) #define AMDGPU_FENCE_OWNER_VM ((void*)1ul) -#define AMDGPU_FENCE_OWNER_MOVE ((void*)2ul) #define AMDGPU_FENCE_FLAG_64BIT (1 << 0) #define AMDGPU_FENCE_FLAG_INT (1 << 1) -struct amdgpu_fence { - struct fence base; - - /* RB, DMA, etc. */ - struct amdgpu_ring *ring; - uint64_t seq; - - /* filp or special value for fence creator */ - void *owner; - - wait_queue_t fence_wake; -}; - struct amdgpu_user_fence { /* write-back bo */ struct amdgpu_bo *bo; @@ -433,71 +376,18 @@ int amdgpu_fence_driver_init(struct amdgpu_device *adev); void amdgpu_fence_driver_fini(struct amdgpu_device *adev); void amdgpu_fence_driver_force_completion(struct amdgpu_device *adev); -int amdgpu_fence_driver_init_ring(struct amdgpu_ring *ring); +int amdgpu_fence_driver_init_ring(struct amdgpu_ring *ring, + unsigned num_hw_submission); int amdgpu_fence_driver_start_ring(struct amdgpu_ring *ring, struct amdgpu_irq_src *irq_src, unsigned irq_type); void amdgpu_fence_driver_suspend(struct amdgpu_device *adev); void amdgpu_fence_driver_resume(struct amdgpu_device *adev); -int amdgpu_fence_emit(struct amdgpu_ring *ring, void *owner, - struct amdgpu_fence **fence); +int amdgpu_fence_emit(struct amdgpu_ring *ring, struct fence **fence); void amdgpu_fence_process(struct amdgpu_ring *ring); -int amdgpu_fence_wait_next(struct amdgpu_ring *ring); int amdgpu_fence_wait_empty(struct amdgpu_ring *ring); unsigned amdgpu_fence_count_emitted(struct amdgpu_ring *ring); -signed long amdgpu_fence_wait_any(struct amdgpu_device *adev, - struct fence **array, - uint32_t count, - bool intr, - signed long t); -struct amdgpu_fence *amdgpu_fence_ref(struct amdgpu_fence *fence); -void amdgpu_fence_unref(struct amdgpu_fence **fence); - -bool amdgpu_fence_need_sync(struct amdgpu_fence *fence, - struct amdgpu_ring *ring); -void amdgpu_fence_note_sync(struct amdgpu_fence *fence, - struct amdgpu_ring *ring); - -static inline struct amdgpu_fence *amdgpu_fence_later(struct amdgpu_fence *a, - struct amdgpu_fence *b) -{ - if (!a) { - return b; - } - - if (!b) { - return a; - } - - BUG_ON(a->ring != b->ring); - - if (a->seq > b->seq) { - return a; - } else { - return b; - } -} - -static inline bool amdgpu_fence_is_earlier(struct amdgpu_fence *a, - struct amdgpu_fence *b) -{ - if (!a) { - return false; - } - - if (!b) { - return true; - } - - BUG_ON(a->ring != b->ring); - - return a->seq < b->seq; -} - -int amdgpu_user_fence_emit(struct amdgpu_ring *ring, struct amdgpu_user_fence *user, - void *owner, struct amdgpu_fence **fence); - /* * TTM. */ @@ -516,6 +406,8 @@ struct amdgpu_mman { /* buffer handling */ const struct amdgpu_buffer_funcs *buffer_funcs; struct amdgpu_ring *buffer_funcs_ring; + /* Scheduler entity for buffer moves */ + struct amd_sched_entity entity; }; int amdgpu_copy_buffer(struct amdgpu_ring *ring, @@ -530,9 +422,9 @@ struct amdgpu_bo_list_entry { struct amdgpu_bo *robj; struct ttm_validate_buffer tv; struct amdgpu_bo_va *bo_va; - unsigned prefered_domains; - unsigned allowed_domains; uint32_t priority; + struct page **user_pages; + int user_invalidated; }; struct amdgpu_bo_va_mapping { @@ -567,7 +459,8 @@ struct amdgpu_bo { /* Protected by gem.mutex */ struct list_head list; /* Protected by tbo.reserved */ - u32 initial_domain; + u32 prefered_domains; + u32 allowed_domains; struct ttm_place placements[AMDGPU_GEM_DOMAIN_MAX + 1]; struct ttm_placement placement; struct ttm_buffer_object tbo; @@ -586,9 +479,9 @@ struct amdgpu_bo { /* Constant after initialization */ struct amdgpu_device *adev; struct drm_gem_object gem_base; + struct amdgpu_bo *parent; struct ttm_bo_kmap_obj dma_buf_vmap; - pid_t pid; struct amdgpu_mn *mn; struct list_head mn_list; }; @@ -637,11 +530,14 @@ int amdgpu_gem_debugfs_init(struct amdgpu_device *adev); * Assumption is that there won't be hole (all object on same * alignment). */ + +#define AMDGPU_SA_NUM_FENCE_LISTS 32 + struct amdgpu_sa_manager { wait_queue_head_t wq; struct amdgpu_bo *bo; struct list_head *hole; - struct list_head flist[AMDGPU_MAX_RINGS]; + struct list_head flist[AMDGPU_SA_NUM_FENCE_LISTS]; struct list_head olist; unsigned size; uint64_t gpu_addr; @@ -650,8 +546,6 @@ struct amdgpu_sa_manager { uint32_t align; }; -struct amdgpu_sa_bo; - /* sub-allocation buffer */ struct amdgpu_sa_bo { struct list_head olist; @@ -665,13 +559,7 @@ struct amdgpu_sa_bo { /* * GEM objects. */ -struct amdgpu_gem { - struct mutex mutex; - struct list_head objects; -}; - -int amdgpu_gem_init(struct amdgpu_device *adev); -void amdgpu_gem_fini(struct amdgpu_device *adev); +void amdgpu_gem_force_release(struct amdgpu_device *adev); int amdgpu_gem_object_create(struct amdgpu_device *adev, unsigned long size, int alignment, u32 initial_domain, u64 flags, bool kernel, @@ -683,32 +571,10 @@ int amdgpu_mode_dumb_create(struct drm_file *file_priv, int amdgpu_mode_dumb_mmap(struct drm_file *filp, struct drm_device *dev, uint32_t handle, uint64_t *offset_p); - -/* - * Semaphores. - */ -struct amdgpu_semaphore { - struct amdgpu_sa_bo *sa_bo; - signed waiters; - uint64_t gpu_addr; -}; - -int amdgpu_semaphore_create(struct amdgpu_device *adev, - struct amdgpu_semaphore **semaphore); -bool amdgpu_semaphore_emit_signal(struct amdgpu_ring *ring, - struct amdgpu_semaphore *semaphore); -bool amdgpu_semaphore_emit_wait(struct amdgpu_ring *ring, - struct amdgpu_semaphore *semaphore); -void amdgpu_semaphore_free(struct amdgpu_device *adev, - struct amdgpu_semaphore **semaphore, - struct fence *fence); - /* * Synchronization */ struct amdgpu_sync { - struct amdgpu_semaphore *semaphores[AMDGPU_NUM_SYNCS]; - struct amdgpu_fence *sync_to[AMDGPU_MAX_RINGS]; DECLARE_HASHTABLE(fences, 4); struct fence *last_vm_update; }; @@ -720,12 +586,11 @@ int amdgpu_sync_resv(struct amdgpu_device *adev, struct amdgpu_sync *sync, struct reservation_object *resv, void *owner); -int amdgpu_sync_rings(struct amdgpu_sync *sync, - struct amdgpu_ring *ring); struct fence *amdgpu_sync_get_fence(struct amdgpu_sync *sync); int amdgpu_sync_wait(struct amdgpu_sync *sync); -void amdgpu_sync_free(struct amdgpu_device *adev, struct amdgpu_sync *sync, - struct fence *fence); +void amdgpu_sync_free(struct amdgpu_sync *sync); +int amdgpu_sync_init(void); +void amdgpu_sync_fini(void); /* * GART structures, functions & helpers @@ -843,6 +708,7 @@ struct amdgpu_flip_work { struct fence *excl; unsigned shared_count; struct fence **shared; + struct fence_cb cb; }; @@ -855,12 +721,11 @@ struct amdgpu_ib { uint32_t length_dw; uint64_t gpu_addr; uint32_t *ptr; - struct amdgpu_ring *ring; - struct amdgpu_fence *fence; struct amdgpu_user_fence *user; struct amdgpu_vm *vm; + unsigned vm_id; + uint64_t vm_pd_addr; struct amdgpu_ctx *ctx; - struct amdgpu_sync sync; uint32_t gds_base, gds_size; uint32_t gws_base, gws_size; uint32_t oa_base, oa_size; @@ -879,13 +744,14 @@ enum amdgpu_ring_type { extern struct amd_sched_backend_ops amdgpu_sched_ops; -int amdgpu_sched_ib_submit_kernel_helper(struct amdgpu_device *adev, - struct amdgpu_ring *ring, - struct amdgpu_ib *ibs, - unsigned num_ibs, - int (*free_job)(struct amdgpu_job *), - void *owner, - struct fence **fence); +int amdgpu_job_alloc(struct amdgpu_device *adev, unsigned num_ibs, + struct amdgpu_job **job); +int amdgpu_job_alloc_with_ib(struct amdgpu_device *adev, unsigned size, + struct amdgpu_job **job); +void amdgpu_job_free(struct amdgpu_job *job); +int amdgpu_job_submit(struct amdgpu_job *job, struct amdgpu_ring *ring, + struct amd_sched_entity *entity, void *owner, + struct fence **f); struct amdgpu_ring { struct amdgpu_device *adev; @@ -894,7 +760,6 @@ struct amdgpu_ring { struct amd_gpu_scheduler sched; spinlock_t fence_lock; - struct mutex *ring_lock; struct amdgpu_bo *ring_obj; volatile uint32_t *ring; unsigned rptr_offs; @@ -903,18 +768,14 @@ struct amdgpu_ring { unsigned wptr; unsigned wptr_old; unsigned ring_size; - unsigned ring_free_dw; + unsigned max_dw; int count_dw; - atomic_t last_rptr; - atomic64_t last_activity; uint64_t gpu_addr; uint32_t align_mask; uint32_t ptr_mask; bool ready; u32 nop; u32 idx; - u64 last_semaphore_signal_addr; - u64 last_semaphore_wait_addr; u32 me; u32 pipe; u32 queue; @@ -927,7 +788,6 @@ struct amdgpu_ring { struct amdgpu_ctx *current_ctx; enum amdgpu_ring_type type; char name[16]; - bool is_pte_ring; }; /* @@ -960,23 +820,25 @@ struct amdgpu_ring { #define AMDGPU_PTE_FRAG_64KB (4 << 7) #define AMDGPU_LOG2_PAGES_PER_FRAG 4 +/* How to programm VM fault handling */ +#define AMDGPU_VM_FAULT_STOP_NEVER 0 +#define AMDGPU_VM_FAULT_STOP_FIRST 1 +#define AMDGPU_VM_FAULT_STOP_ALWAYS 2 + struct amdgpu_vm_pt { - struct amdgpu_bo *bo; + struct amdgpu_bo_list_entry entry; uint64_t addr; }; struct amdgpu_vm_id { - unsigned id; - uint64_t pd_gpu_addr; + struct amdgpu_vm_manager_id *mgr_id; + uint64_t pd_gpu_addr; /* last flushed PD/PT update */ - struct fence *flushed_updates; - /* last use of vmid */ - struct amdgpu_fence *last_id_use; + struct fence *flushed_updates; }; struct amdgpu_vm { - struct mutex mutex; - + /* tree of virtual addresses mapped */ struct rb_root va; /* protecting invalidated */ @@ -1001,33 +863,99 @@ struct amdgpu_vm { /* for id and flush management per ring */ struct amdgpu_vm_id ids[AMDGPU_MAX_RINGS]; + + /* protecting freed */ + spinlock_t freed_lock; + + /* Scheduler entity for page table updates */ + struct amd_sched_entity entity; +}; + +struct amdgpu_vm_manager_id { + struct list_head list; + struct fence *active; + atomic_long_t owner; + + uint32_t gds_base; + uint32_t gds_size; + uint32_t gws_base; + uint32_t gws_size; + uint32_t oa_base; + uint32_t oa_size; }; struct amdgpu_vm_manager { - struct amdgpu_fence *active[AMDGPU_NUM_VM]; - uint32_t max_pfn; - /* number of VMIDs */ - unsigned nvm; + /* Handling of VMIDs */ + struct mutex lock; + unsigned num_ids; + struct list_head ids_lru; + struct amdgpu_vm_manager_id ids[AMDGPU_NUM_VM]; + + uint32_t max_pfn; /* vram base address for page table entry */ - u64 vram_base_offset; + u64 vram_base_offset; /* is vm enabled? */ - bool enabled; - /* for hw to save the PD addr on suspend/resume */ - uint32_t saved_table_addr[AMDGPU_NUM_VM]; + bool enabled; /* vm pte handling */ const struct amdgpu_vm_pte_funcs *vm_pte_funcs; - struct amdgpu_ring *vm_pte_funcs_ring; + struct amdgpu_ring *vm_pte_rings[AMDGPU_MAX_RINGS]; + unsigned vm_pte_num_rings; + atomic_t vm_pte_next_ring; }; +void amdgpu_vm_manager_init(struct amdgpu_device *adev); +void amdgpu_vm_manager_fini(struct amdgpu_device *adev); +int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm); +void amdgpu_vm_fini(struct amdgpu_device *adev, struct amdgpu_vm *vm); +void amdgpu_vm_get_pd_bo(struct amdgpu_vm *vm, + struct list_head *validated, + struct amdgpu_bo_list_entry *entry); +void amdgpu_vm_get_pt_bos(struct amdgpu_vm *vm, struct list_head *duplicates); +void amdgpu_vm_move_pt_bos_in_lru(struct amdgpu_device *adev, + struct amdgpu_vm *vm); +int amdgpu_vm_grab_id(struct amdgpu_vm *vm, struct amdgpu_ring *ring, + struct amdgpu_sync *sync, struct fence *fence, + unsigned *vm_id, uint64_t *vm_pd_addr); +void amdgpu_vm_flush(struct amdgpu_ring *ring, + unsigned vm_id, uint64_t pd_addr, + uint32_t gds_base, uint32_t gds_size, + uint32_t gws_base, uint32_t gws_size, + uint32_t oa_base, uint32_t oa_size); +void amdgpu_vm_reset_id(struct amdgpu_device *adev, unsigned vm_id); +uint64_t amdgpu_vm_map_gart(const dma_addr_t *pages_addr, uint64_t addr); +int amdgpu_vm_update_page_directory(struct amdgpu_device *adev, + struct amdgpu_vm *vm); +int amdgpu_vm_clear_freed(struct amdgpu_device *adev, + struct amdgpu_vm *vm); +int amdgpu_vm_clear_invalids(struct amdgpu_device *adev, struct amdgpu_vm *vm, + struct amdgpu_sync *sync); +int amdgpu_vm_bo_update(struct amdgpu_device *adev, + struct amdgpu_bo_va *bo_va, + struct ttm_mem_reg *mem); +void amdgpu_vm_bo_invalidate(struct amdgpu_device *adev, + struct amdgpu_bo *bo); +struct amdgpu_bo_va *amdgpu_vm_bo_find(struct amdgpu_vm *vm, + struct amdgpu_bo *bo); +struct amdgpu_bo_va *amdgpu_vm_bo_add(struct amdgpu_device *adev, + struct amdgpu_vm *vm, + struct amdgpu_bo *bo); +int amdgpu_vm_bo_map(struct amdgpu_device *adev, + struct amdgpu_bo_va *bo_va, + uint64_t addr, uint64_t offset, + uint64_t size, uint32_t flags); +int amdgpu_vm_bo_unmap(struct amdgpu_device *adev, + struct amdgpu_bo_va *bo_va, + uint64_t addr); +void amdgpu_vm_bo_rmv(struct amdgpu_device *adev, + struct amdgpu_bo_va *bo_va); + /* * context related structures */ -#define AMDGPU_CTX_MAX_CS_PENDING 16 - struct amdgpu_ctx_ring { uint64_t sequence; - struct fence *fences[AMDGPU_CTX_MAX_CS_PENDING]; + struct fence **fences; struct amd_sched_entity entity; }; @@ -1036,6 +964,7 @@ struct amdgpu_ctx { struct amdgpu_device *adev; unsigned reset_counter; spinlock_t ring_lock; + struct fence **fences; struct amdgpu_ctx_ring rings[AMDGPU_MAX_RINGS]; }; @@ -1046,10 +975,6 @@ struct amdgpu_ctx_mgr { struct idr ctx_handles; }; -int amdgpu_ctx_init(struct amdgpu_device *adev, bool kernel, - struct amdgpu_ctx *ctx); -void amdgpu_ctx_fini(struct amdgpu_ctx *ctx); - struct amdgpu_ctx *amdgpu_ctx_get(struct amdgpu_fpriv *fpriv, uint32_t id); int amdgpu_ctx_put(struct amdgpu_ctx *ctx); @@ -1084,13 +1009,15 @@ struct amdgpu_bo_list { struct amdgpu_bo *gds_obj; struct amdgpu_bo *gws_obj; struct amdgpu_bo *oa_obj; - bool has_userptr; + unsigned first_userptr; unsigned num_entries; struct amdgpu_bo_list_entry *array; }; struct amdgpu_bo_list * amdgpu_bo_list_get(struct amdgpu_fpriv *fpriv, int id); +void amdgpu_bo_list_get_list(struct amdgpu_bo_list *list, + struct list_head *validated); void amdgpu_bo_list_put(struct amdgpu_bo_list *list); void amdgpu_bo_list_free(struct amdgpu_bo_list *list); @@ -1164,6 +1091,7 @@ struct amdgpu_gca_config { unsigned multi_gpu_tile_size; unsigned mc_arb_ramcfg; unsigned gb_addr_config; + unsigned num_rbs; uint32_t tile_mode_array[32]; uint32_t macrotile_mode_array[16]; @@ -1206,25 +1134,20 @@ struct amdgpu_gfx { unsigned ce_ram_size; }; -int amdgpu_ib_get(struct amdgpu_ring *ring, struct amdgpu_vm *vm, +int amdgpu_ib_get(struct amdgpu_device *adev, struct amdgpu_vm *vm, unsigned size, struct amdgpu_ib *ib); -void amdgpu_ib_free(struct amdgpu_device *adev, struct amdgpu_ib *ib); -int amdgpu_ib_schedule(struct amdgpu_device *adev, unsigned num_ibs, - struct amdgpu_ib *ib, void *owner); +void amdgpu_ib_free(struct amdgpu_device *adev, struct amdgpu_ib *ib, struct fence *f); +int amdgpu_ib_schedule(struct amdgpu_ring *ring, unsigned num_ibs, + struct amdgpu_ib *ib, struct fence *last_vm_update, + struct fence **f); int amdgpu_ib_pool_init(struct amdgpu_device *adev); void amdgpu_ib_pool_fini(struct amdgpu_device *adev); int amdgpu_ib_ring_tests(struct amdgpu_device *adev); -/* Ring access between begin & end cannot sleep */ -void amdgpu_ring_free_size(struct amdgpu_ring *ring); int amdgpu_ring_alloc(struct amdgpu_ring *ring, unsigned ndw); -int amdgpu_ring_lock(struct amdgpu_ring *ring, unsigned ndw); void amdgpu_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count); +void amdgpu_ring_generic_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib); void amdgpu_ring_commit(struct amdgpu_ring *ring); -void amdgpu_ring_unlock_commit(struct amdgpu_ring *ring); void amdgpu_ring_undo(struct amdgpu_ring *ring); -void amdgpu_ring_unlock_undo(struct amdgpu_ring *ring); -void amdgpu_ring_lockup_update(struct amdgpu_ring *ring); -bool amdgpu_ring_test_lockup(struct amdgpu_ring *ring); unsigned amdgpu_ring_backup(struct amdgpu_ring *ring, uint32_t **data); int amdgpu_ring_restore(struct amdgpu_ring *ring, @@ -1242,45 +1165,58 @@ struct amdgpu_cs_chunk { uint32_t chunk_id; uint32_t length_dw; uint32_t *kdata; - void __user *user_ptr; }; struct amdgpu_cs_parser { struct amdgpu_device *adev; struct drm_file *filp; struct amdgpu_ctx *ctx; - struct amdgpu_bo_list *bo_list; + /* chunks */ unsigned nchunks; struct amdgpu_cs_chunk *chunks; - /* relocations */ - struct amdgpu_bo_list_entry *vm_bos; - struct list_head validated; - struct amdgpu_ib *ibs; - uint32_t num_ibs; + /* scheduler job object */ + struct amdgpu_job *job; - struct ww_acquire_ctx ticket; + /* buffer objects */ + struct ww_acquire_ctx ticket; + struct amdgpu_bo_list *bo_list; + struct amdgpu_bo_list_entry vm_pd; + struct list_head validated; + struct fence *fence; + uint64_t bytes_moved_threshold; + uint64_t bytes_moved; /* user fence */ - struct amdgpu_user_fence uf; + struct amdgpu_bo_list_entry uf_entry; }; struct amdgpu_job { struct amd_sched_job base; struct amdgpu_device *adev; + struct amdgpu_ring *ring; + struct amdgpu_sync sync; struct amdgpu_ib *ibs; + struct fence *fence; /* the hw fence */ uint32_t num_ibs; - struct mutex job_lock; + void *owner; struct amdgpu_user_fence uf; - int (*free_job)(struct amdgpu_job *job); }; #define to_amdgpu_job(sched_job) \ container_of((sched_job), struct amdgpu_job, base) -static inline u32 amdgpu_get_ib_value(struct amdgpu_cs_parser *p, uint32_t ib_idx, int idx) +static inline u32 amdgpu_get_ib_value(struct amdgpu_cs_parser *p, + uint32_t ib_idx, int idx) { - return p->ibs[ib_idx].ptr[idx]; + return p->job->ibs[ib_idx].ptr[idx]; +} + +static inline void amdgpu_set_ib_value(struct amdgpu_cs_parser *p, + uint32_t ib_idx, int idx, + uint32_t value) +{ + p->job->ibs[ib_idx].ptr[idx] = value; } /* @@ -1299,31 +1235,7 @@ struct amdgpu_wb { int amdgpu_wb_get(struct amdgpu_device *adev, u32 *wb); void amdgpu_wb_free(struct amdgpu_device *adev, u32 wb); -/** - * struct amdgpu_pm - power management datas - * It keeps track of various data needed to take powermanagement decision. - */ -enum amdgpu_pm_state_type { - /* not used for dpm */ - POWER_STATE_TYPE_DEFAULT, - POWER_STATE_TYPE_POWERSAVE, - /* user selectable states */ - POWER_STATE_TYPE_BATTERY, - POWER_STATE_TYPE_BALANCED, - POWER_STATE_TYPE_PERFORMANCE, - /* internal states */ - POWER_STATE_TYPE_INTERNAL_UVD, - POWER_STATE_TYPE_INTERNAL_UVD_SD, - POWER_STATE_TYPE_INTERNAL_UVD_HD, - POWER_STATE_TYPE_INTERNAL_UVD_HD2, - POWER_STATE_TYPE_INTERNAL_UVD_MVC, - POWER_STATE_TYPE_INTERNAL_BOOT, - POWER_STATE_TYPE_INTERNAL_THERMAL, - POWER_STATE_TYPE_INTERNAL_ACPI, - POWER_STATE_TYPE_INTERNAL_ULV, - POWER_STATE_TYPE_INTERNAL_3DPERF, -}; enum amdgpu_int_thermal_type { THERMAL_TYPE_NONE, @@ -1556,6 +1468,7 @@ enum amdgpu_dpm_forced_level { AMDGPU_DPM_FORCED_LEVEL_AUTO = 0, AMDGPU_DPM_FORCED_LEVEL_LOW = 1, AMDGPU_DPM_FORCED_LEVEL_HIGH = 2, + AMDGPU_DPM_FORCED_LEVEL_MANUAL = 3, }; struct amdgpu_vce_state { @@ -1605,8 +1518,8 @@ struct amdgpu_dpm { /* vce requirements */ struct amdgpu_vce_state vce_states[AMDGPU_MAX_VCE_LEVELS]; enum amdgpu_vce_level vce_level; - enum amdgpu_pm_state_type state; - enum amdgpu_pm_state_type user_state; + enum amd_pm_state_type state; + enum amd_pm_state_type user_state; u32 platform_caps; u32 voltage_response_time; u32 backbias_response_time; @@ -1659,8 +1572,13 @@ struct amdgpu_pm { const struct firmware *fw; /* SMC firmware */ uint32_t fw_version; const struct amdgpu_dpm_funcs *funcs; + uint32_t pcie_gen_mask; + uint32_t pcie_mlw_mask; + struct amd_pp_display_configuration pm_display_cfg;/* set by DAL */ }; +void amdgpu_get_pcie_info(struct amdgpu_device *adev); + /* * UVD */ @@ -1680,6 +1598,7 @@ struct amdgpu_uvd { struct amdgpu_ring ring; struct amdgpu_irq_src irq; bool address_64_bit; + struct amd_sched_entity entity; }; /* @@ -1704,12 +1623,13 @@ struct amdgpu_vce { struct amdgpu_ring ring[AMDGPU_MAX_VCE_RINGS]; struct amdgpu_irq_src irq; unsigned harvest_config; + struct amd_sched_entity entity; }; /* * SDMA */ -struct amdgpu_sdma { +struct amdgpu_sdma_instance { /* SDMA firmware */ const struct firmware *fw; uint32_t fw_version; @@ -1719,6 +1639,13 @@ struct amdgpu_sdma { bool burst_nop; }; +struct amdgpu_sdma { + struct amdgpu_sdma_instance instance[AMDGPU_MAX_SDMA_INSTANCES]; + struct amdgpu_irq_src trap_irq; + struct amdgpu_irq_src illegal_inst_irq; + int num_instances; +}; + /* * Firmware */ @@ -1751,11 +1678,11 @@ void amdgpu_test_syncing(struct amdgpu_device *adev); int amdgpu_mn_register(struct amdgpu_bo *bo, unsigned long addr); void amdgpu_mn_unregister(struct amdgpu_bo *bo); #else -static int amdgpu_mn_register(struct amdgpu_bo *bo, unsigned long addr) +static inline int amdgpu_mn_register(struct amdgpu_bo *bo, unsigned long addr) { return -ENODEV; } -static void amdgpu_mn_unregister(struct amdgpu_bo *bo) {} +static inline void amdgpu_mn_unregister(struct amdgpu_bo *bo) {} #endif /* @@ -1821,6 +1748,8 @@ struct amdgpu_cu_info { */ struct amdgpu_asic_funcs { bool (*read_disabled_bios)(struct amdgpu_device *adev); + bool (*read_bios_from_rom)(struct amdgpu_device *adev, + u8 *bios, u32 length_bytes); int (*read_register)(struct amdgpu_device *adev, u32 se_num, u32 sh_num, u32 reg_offset, u32 *value); void (*set_vga_state)(struct amdgpu_device *adev, bool state); @@ -1929,6 +1858,18 @@ void amdgpu_cgs_destroy_device(void *cgs_device); /* + * CGS + */ +void *amdgpu_cgs_create_device(struct amdgpu_device *adev); +void amdgpu_cgs_destroy_device(void *cgs_device); + + +/* GPU virtualization */ +struct amdgpu_virtualization { + bool supports_sr_iov; +}; + +/* * Core structure, functions and helpers. */ typedef uint32_t (*amdgpu_rreg_t)(struct amdgpu_device*, uint32_t); @@ -1947,7 +1888,10 @@ struct amdgpu_device { struct device *dev; struct drm_device *ddev; struct pci_dev *pdev; - struct rw_semaphore exclusive_lock; + +#ifdef CONFIG_DRM_AMD_ACP + struct amdgpu_acp acp; +#endif /* ASIC */ enum amd_asic_type asic_type; @@ -1961,7 +1905,6 @@ struct amdgpu_device { bool suspend; bool need_dma32; bool accel_working; - bool needs_reset; struct work_struct reset_work; struct notifier_block acpi_nb; struct amdgpu_i2c_chan *i2c_bus[AMDGPU_MAX_I2C_BUS]; @@ -2026,7 +1969,6 @@ struct amdgpu_device { /* memory management */ struct amdgpu_mman mman; - struct amdgpu_gem gem; struct amdgpu_vram_scratch vram_scratch; struct amdgpu_wb wb; atomic64_t vram_usage; @@ -2044,7 +1986,6 @@ struct amdgpu_device { /* rings */ unsigned fence_context; - struct mutex ring_lock; unsigned num_rings; struct amdgpu_ring *rings[AMDGPU_MAX_RINGS]; bool ib_pool_ready; @@ -2053,6 +1994,11 @@ struct amdgpu_device { /* interrupts */ struct amdgpu_irq irq; + /* powerplay */ + struct amd_powerplay powerplay; + bool pp_enabled; + bool pp_force_state_enabled; + /* dpm */ struct amdgpu_pm pm; u32 cg_flags; @@ -2065,12 +2011,9 @@ struct amdgpu_device { struct amdgpu_gfx gfx; /* sdma */ - struct amdgpu_sdma sdma[AMDGPU_MAX_SDMA_INSTANCES]; - struct amdgpu_irq_src sdma_trap_irq; - struct amdgpu_irq_src sdma_illegal_inst_irq; + struct amdgpu_sdma sdma; /* uvd */ - bool has_uvd; struct amdgpu_uvd uvd; /* vce */ @@ -2095,8 +2038,7 @@ struct amdgpu_device { /* amdkfd interface */ struct kfd_dev *kfd; - /* kernel conext for IB submission */ - struct amdgpu_ctx kernel_ctx; + struct amdgpu_virtualization virtualization; }; bool amdgpu_device_is_px(struct drm_device *dev); @@ -2118,20 +2060,6 @@ u32 amdgpu_mm_rdoorbell(struct amdgpu_device *adev, u32 index); void amdgpu_mm_wdoorbell(struct amdgpu_device *adev, u32 index, u32 v); /* - * Cast helper - */ -extern const struct fence_ops amdgpu_fence_ops; -static inline struct amdgpu_fence *to_amdgpu_fence(struct fence *f) -{ - struct amdgpu_fence *__f = container_of(f, struct amdgpu_fence, base); - - if (__f->base.ops == &amdgpu_fence_ops) - return __f; - - return NULL; -} - -/* * Registers read & write functions. */ #define RREG32(reg) amdgpu_mm_rreg(adev, (reg), false) @@ -2201,20 +2129,20 @@ static inline void amdgpu_ring_write(struct amdgpu_ring *ring, uint32_t v) ring->ring[ring->wptr++] = v; ring->wptr &= ring->ptr_mask; ring->count_dw--; - ring->ring_free_dw--; } -static inline struct amdgpu_sdma * amdgpu_get_sdma_instance(struct amdgpu_ring *ring) +static inline struct amdgpu_sdma_instance * +amdgpu_get_sdma_instance(struct amdgpu_ring *ring) { struct amdgpu_device *adev = ring->adev; int i; - for (i = 0; i < AMDGPU_MAX_SDMA_INSTANCES; i++) - if (&adev->sdma[i].ring == ring) + for (i = 0; i < adev->sdma.num_instances; i++) + if (&adev->sdma.instance[i].ring == ring) break; if (i < AMDGPU_MAX_SDMA_INSTANCES) - return &adev->sdma[i]; + return &adev->sdma.instance[i]; else return NULL; } @@ -2230,27 +2158,28 @@ static inline struct amdgpu_sdma * amdgpu_get_sdma_instance(struct amdgpu_ring * #define amdgpu_asic_set_vce_clocks(adev, ev, ec) (adev)->asic_funcs->set_vce_clocks((adev), (ev), (ec)) #define amdgpu_asic_get_gpu_clock_counter(adev) (adev)->asic_funcs->get_gpu_clock_counter((adev)) #define amdgpu_asic_read_disabled_bios(adev) (adev)->asic_funcs->read_disabled_bios((adev)) +#define amdgpu_asic_read_bios_from_rom(adev, b, l) (adev)->asic_funcs->read_bios_from_rom((adev), (b), (l)) #define amdgpu_asic_read_register(adev, se, sh, offset, v)((adev)->asic_funcs->read_register((adev), (se), (sh), (offset), (v))) #define amdgpu_asic_get_cu_info(adev, info) (adev)->asic_funcs->get_cu_info((adev), (info)) #define amdgpu_gart_flush_gpu_tlb(adev, vmid) (adev)->gart.gart_funcs->flush_gpu_tlb((adev), (vmid)) #define amdgpu_gart_set_pte_pde(adev, pt, idx, addr, flags) (adev)->gart.gart_funcs->set_pte_pde((adev), (pt), (idx), (addr), (flags)) #define amdgpu_vm_copy_pte(adev, ib, pe, src, count) ((adev)->vm_manager.vm_pte_funcs->copy_pte((ib), (pe), (src), (count))) -#define amdgpu_vm_write_pte(adev, ib, pe, addr, count, incr, flags) ((adev)->vm_manager.vm_pte_funcs->write_pte((ib), (pe), (addr), (count), (incr), (flags))) +#define amdgpu_vm_write_pte(adev, ib, pa, pe, addr, count, incr, flags) ((adev)->vm_manager.vm_pte_funcs->write_pte((ib), (pa), (pe), (addr), (count), (incr), (flags))) #define amdgpu_vm_set_pte_pde(adev, ib, pe, addr, count, incr, flags) ((adev)->vm_manager.vm_pte_funcs->set_pte_pde((ib), (pe), (addr), (count), (incr), (flags))) -#define amdgpu_vm_pad_ib(adev, ib) ((adev)->vm_manager.vm_pte_funcs->pad_ib((ib))) #define amdgpu_ring_parse_cs(r, p, ib) ((r)->funcs->parse_cs((p), (ib))) #define amdgpu_ring_test_ring(r) (r)->funcs->test_ring((r)) #define amdgpu_ring_test_ib(r) (r)->funcs->test_ib((r)) -#define amdgpu_ring_is_lockup(r) (r)->funcs->is_lockup((r)) #define amdgpu_ring_get_rptr(r) (r)->funcs->get_rptr((r)) #define amdgpu_ring_get_wptr(r) (r)->funcs->get_wptr((r)) #define amdgpu_ring_set_wptr(r) (r)->funcs->set_wptr((r)) #define amdgpu_ring_emit_ib(r, ib) (r)->funcs->emit_ib((r), (ib)) +#define amdgpu_ring_emit_pipeline_sync(r) (r)->funcs->emit_pipeline_sync((r)) #define amdgpu_ring_emit_vm_flush(r, vmid, addr) (r)->funcs->emit_vm_flush((r), (vmid), (addr)) #define amdgpu_ring_emit_fence(r, addr, seq, flags) (r)->funcs->emit_fence((r), (addr), (seq), (flags)) -#define amdgpu_ring_emit_semaphore(r, semaphore, emit_wait) (r)->funcs->emit_semaphore((r), (semaphore), (emit_wait)) #define amdgpu_ring_emit_gds_switch(r, v, db, ds, wb, ws, ab, as) (r)->funcs->emit_gds_switch((r), (v), (db), (ds), (wb), (ws), (ab), (as)) #define amdgpu_ring_emit_hdp_flush(r) (r)->funcs->emit_hdp_flush((r)) +#define amdgpu_ring_emit_hdp_invalidate(r) (r)->funcs->emit_hdp_invalidate((r)) +#define amdgpu_ring_pad_ib(r, ib) ((r)->funcs->pad_ib((r), (ib))) #define amdgpu_ih_get_wptr(adev) (adev)->irq.ih_funcs->get_wptr((adev)) #define amdgpu_ih_decode_iv(adev, iv) (adev)->irq.ih_funcs->decode_iv((adev), (iv)) #define amdgpu_ih_set_rptr(adev) (adev)->irq.ih_funcs->set_rptr((adev)) @@ -2272,24 +2201,93 @@ static inline struct amdgpu_sdma * amdgpu_get_sdma_instance(struct amdgpu_ring * #define amdgpu_display_resume_mc_access(adev, s) (adev)->mode_info.funcs->resume_mc_access((adev), (s)) #define amdgpu_emit_copy_buffer(adev, ib, s, d, b) (adev)->mman.buffer_funcs->emit_copy_buffer((ib), (s), (d), (b)) #define amdgpu_emit_fill_buffer(adev, ib, s, d, b) (adev)->mman.buffer_funcs->emit_fill_buffer((ib), (s), (d), (b)) -#define amdgpu_dpm_get_temperature(adev) (adev)->pm.funcs->get_temperature((adev)) #define amdgpu_dpm_pre_set_power_state(adev) (adev)->pm.funcs->pre_set_power_state((adev)) #define amdgpu_dpm_set_power_state(adev) (adev)->pm.funcs->set_power_state((adev)) #define amdgpu_dpm_post_set_power_state(adev) (adev)->pm.funcs->post_set_power_state((adev)) #define amdgpu_dpm_display_configuration_changed(adev) (adev)->pm.funcs->display_configuration_changed((adev)) -#define amdgpu_dpm_get_sclk(adev, l) (adev)->pm.funcs->get_sclk((adev), (l)) -#define amdgpu_dpm_get_mclk(adev, l) (adev)->pm.funcs->get_mclk((adev), (l)) #define amdgpu_dpm_print_power_state(adev, ps) (adev)->pm.funcs->print_power_state((adev), (ps)) -#define amdgpu_dpm_debugfs_print_current_performance_level(adev, m) (adev)->pm.funcs->debugfs_print_current_performance_level((adev), (m)) -#define amdgpu_dpm_force_performance_level(adev, l) (adev)->pm.funcs->force_performance_level((adev), (l)) #define amdgpu_dpm_vblank_too_short(adev) (adev)->pm.funcs->vblank_too_short((adev)) -#define amdgpu_dpm_powergate_uvd(adev, g) (adev)->pm.funcs->powergate_uvd((adev), (g)) -#define amdgpu_dpm_powergate_vce(adev, g) (adev)->pm.funcs->powergate_vce((adev), (g)) #define amdgpu_dpm_enable_bapm(adev, e) (adev)->pm.funcs->enable_bapm((adev), (e)) -#define amdgpu_dpm_set_fan_control_mode(adev, m) (adev)->pm.funcs->set_fan_control_mode((adev), (m)) -#define amdgpu_dpm_get_fan_control_mode(adev) (adev)->pm.funcs->get_fan_control_mode((adev)) -#define amdgpu_dpm_set_fan_speed_percent(adev, s) (adev)->pm.funcs->set_fan_speed_percent((adev), (s)) -#define amdgpu_dpm_get_fan_speed_percent(adev, s) (adev)->pm.funcs->get_fan_speed_percent((adev), (s)) + +#define amdgpu_dpm_get_temperature(adev) \ + ((adev)->pp_enabled ? \ + (adev)->powerplay.pp_funcs->get_temperature((adev)->powerplay.pp_handle) : \ + (adev)->pm.funcs->get_temperature((adev))) + +#define amdgpu_dpm_set_fan_control_mode(adev, m) \ + ((adev)->pp_enabled ? \ + (adev)->powerplay.pp_funcs->set_fan_control_mode((adev)->powerplay.pp_handle, (m)) : \ + (adev)->pm.funcs->set_fan_control_mode((adev), (m))) + +#define amdgpu_dpm_get_fan_control_mode(adev) \ + ((adev)->pp_enabled ? \ + (adev)->powerplay.pp_funcs->get_fan_control_mode((adev)->powerplay.pp_handle) : \ + (adev)->pm.funcs->get_fan_control_mode((adev))) + +#define amdgpu_dpm_set_fan_speed_percent(adev, s) \ + ((adev)->pp_enabled ? \ + (adev)->powerplay.pp_funcs->set_fan_speed_percent((adev)->powerplay.pp_handle, (s)) : \ + (adev)->pm.funcs->set_fan_speed_percent((adev), (s))) + +#define amdgpu_dpm_get_fan_speed_percent(adev, s) \ + ((adev)->pp_enabled ? \ + (adev)->powerplay.pp_funcs->get_fan_speed_percent((adev)->powerplay.pp_handle, (s)) : \ + (adev)->pm.funcs->get_fan_speed_percent((adev), (s))) + +#define amdgpu_dpm_get_sclk(adev, l) \ + ((adev)->pp_enabled ? \ + (adev)->powerplay.pp_funcs->get_sclk((adev)->powerplay.pp_handle, (l)) : \ + (adev)->pm.funcs->get_sclk((adev), (l))) + +#define amdgpu_dpm_get_mclk(adev, l) \ + ((adev)->pp_enabled ? \ + (adev)->powerplay.pp_funcs->get_mclk((adev)->powerplay.pp_handle, (l)) : \ + (adev)->pm.funcs->get_mclk((adev), (l))) + + +#define amdgpu_dpm_force_performance_level(adev, l) \ + ((adev)->pp_enabled ? \ + (adev)->powerplay.pp_funcs->force_performance_level((adev)->powerplay.pp_handle, (l)) : \ + (adev)->pm.funcs->force_performance_level((adev), (l))) + +#define amdgpu_dpm_powergate_uvd(adev, g) \ + ((adev)->pp_enabled ? \ + (adev)->powerplay.pp_funcs->powergate_uvd((adev)->powerplay.pp_handle, (g)) : \ + (adev)->pm.funcs->powergate_uvd((adev), (g))) + +#define amdgpu_dpm_powergate_vce(adev, g) \ + ((adev)->pp_enabled ? \ + (adev)->powerplay.pp_funcs->powergate_vce((adev)->powerplay.pp_handle, (g)) : \ + (adev)->pm.funcs->powergate_vce((adev), (g))) + +#define amdgpu_dpm_debugfs_print_current_performance_level(adev, m) \ + ((adev)->pp_enabled ? \ + (adev)->powerplay.pp_funcs->print_current_performance_level((adev)->powerplay.pp_handle, (m)) : \ + (adev)->pm.funcs->debugfs_print_current_performance_level((adev), (m))) + +#define amdgpu_dpm_get_current_power_state(adev) \ + (adev)->powerplay.pp_funcs->get_current_power_state((adev)->powerplay.pp_handle) + +#define amdgpu_dpm_get_performance_level(adev) \ + (adev)->powerplay.pp_funcs->get_performance_level((adev)->powerplay.pp_handle) + +#define amdgpu_dpm_get_pp_num_states(adev, data) \ + (adev)->powerplay.pp_funcs->get_pp_num_states((adev)->powerplay.pp_handle, data) + +#define amdgpu_dpm_get_pp_table(adev, table) \ + (adev)->powerplay.pp_funcs->get_pp_table((adev)->powerplay.pp_handle, table) + +#define amdgpu_dpm_set_pp_table(adev, buf, size) \ + (adev)->powerplay.pp_funcs->set_pp_table((adev)->powerplay.pp_handle, buf, size) + +#define amdgpu_dpm_print_clock_levels(adev, type, buf) \ + (adev)->powerplay.pp_funcs->print_clock_levels((adev)->powerplay.pp_handle, type, buf) + +#define amdgpu_dpm_force_clock_level(adev, type, level) \ + (adev)->powerplay.pp_funcs->force_clock_level((adev)->powerplay.pp_handle, type, level) + +#define amdgpu_dpm_dispatch_task(adev, event_id, input, output) \ + (adev)->powerplay.pp_funcs->dispatch_tasks((adev)->powerplay.pp_handle, (event_id), (input), (output)) #define amdgpu_gds_switch(adev, r, v, d, w, a) (adev)->gds.funcs->patch_gds_switch((r), (v), (d), (w), (a)) @@ -2298,12 +2296,6 @@ int amdgpu_gpu_reset(struct amdgpu_device *adev); void amdgpu_pci_config_reset(struct amdgpu_device *adev); bool amdgpu_card_posted(struct amdgpu_device *adev); void amdgpu_update_display_priority(struct amdgpu_device *adev); -bool amdgpu_boot_test_post_card(struct amdgpu_device *adev); -struct amdgpu_cs_parser *amdgpu_cs_parser_create(struct amdgpu_device *adev, - struct drm_file *filp, - struct amdgpu_ctx *ctx, - struct amdgpu_ib *ibs, - uint32_t num_ibs); int amdgpu_cs_parser_init(struct amdgpu_cs_parser *p, void *data); int amdgpu_cs_get_ring(struct amdgpu_device *adev, u32 ip_type, @@ -2311,9 +2303,15 @@ int amdgpu_cs_get_ring(struct amdgpu_device *adev, u32 ip_type, struct amdgpu_ring **out_ring); void amdgpu_ttm_placement_from_domain(struct amdgpu_bo *rbo, u32 domain); bool amdgpu_ttm_bo_is_amdgpu_bo(struct ttm_buffer_object *bo); +int amdgpu_ttm_tt_get_user_pages(struct ttm_tt *ttm, struct page **pages); int amdgpu_ttm_tt_set_userptr(struct ttm_tt *ttm, uint64_t addr, uint32_t flags); bool amdgpu_ttm_tt_has_userptr(struct ttm_tt *ttm); +struct mm_struct *amdgpu_ttm_tt_get_usermm(struct ttm_tt *ttm); +bool amdgpu_ttm_tt_affect_userptr(struct ttm_tt *ttm, unsigned long start, + unsigned long end); +bool amdgpu_ttm_tt_userptr_invalidated(struct ttm_tt *ttm, + int *last_invalidated); bool amdgpu_ttm_tt_is_readonly(struct ttm_tt *ttm); uint32_t amdgpu_ttm_tt_pte_flags(struct amdgpu_device *adev, struct ttm_tt *ttm, struct ttm_mem_reg *mem); @@ -2350,10 +2348,10 @@ void amdgpu_driver_preclose_kms(struct drm_device *dev, struct drm_file *file_priv); int amdgpu_suspend_kms(struct drm_device *dev, bool suspend, bool fbcon); int amdgpu_resume_kms(struct drm_device *dev, bool resume, bool fbcon); -u32 amdgpu_get_vblank_counter_kms(struct drm_device *dev, int crtc); -int amdgpu_enable_vblank_kms(struct drm_device *dev, int crtc); -void amdgpu_disable_vblank_kms(struct drm_device *dev, int crtc); -int amdgpu_get_vblank_timestamp_kms(struct drm_device *dev, int crtc, +u32 amdgpu_get_vblank_counter_kms(struct drm_device *dev, unsigned int pipe); +int amdgpu_enable_vblank_kms(struct drm_device *dev, unsigned int pipe); +void amdgpu_disable_vblank_kms(struct drm_device *dev, unsigned int pipe); +int amdgpu_get_vblank_timestamp_kms(struct drm_device *dev, unsigned int pipe, int *max_error, struct timeval *vblank_time, unsigned flags); @@ -2361,49 +2359,6 @@ long amdgpu_kms_compat_ioctl(struct file *filp, unsigned int cmd, unsigned long arg); /* - * vm - */ -int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm); -void amdgpu_vm_fini(struct amdgpu_device *adev, struct amdgpu_vm *vm); -struct amdgpu_bo_list_entry *amdgpu_vm_get_bos(struct amdgpu_device *adev, - struct amdgpu_vm *vm, - struct list_head *head); -int amdgpu_vm_grab_id(struct amdgpu_vm *vm, struct amdgpu_ring *ring, - struct amdgpu_sync *sync); -void amdgpu_vm_flush(struct amdgpu_ring *ring, - struct amdgpu_vm *vm, - struct fence *updates); -void amdgpu_vm_fence(struct amdgpu_device *adev, - struct amdgpu_vm *vm, - struct amdgpu_fence *fence); -uint64_t amdgpu_vm_map_gart(struct amdgpu_device *adev, uint64_t addr); -int amdgpu_vm_update_page_directory(struct amdgpu_device *adev, - struct amdgpu_vm *vm); -int amdgpu_vm_clear_freed(struct amdgpu_device *adev, - struct amdgpu_vm *vm); -int amdgpu_vm_clear_invalids(struct amdgpu_device *adev, - struct amdgpu_vm *vm, struct amdgpu_sync *sync); -int amdgpu_vm_bo_update(struct amdgpu_device *adev, - struct amdgpu_bo_va *bo_va, - struct ttm_mem_reg *mem); -void amdgpu_vm_bo_invalidate(struct amdgpu_device *adev, - struct amdgpu_bo *bo); -struct amdgpu_bo_va *amdgpu_vm_bo_find(struct amdgpu_vm *vm, - struct amdgpu_bo *bo); -struct amdgpu_bo_va *amdgpu_vm_bo_add(struct amdgpu_device *adev, - struct amdgpu_vm *vm, - struct amdgpu_bo *bo); -int amdgpu_vm_bo_map(struct amdgpu_device *adev, - struct amdgpu_bo_va *bo_va, - uint64_t addr, uint64_t offset, - uint64_t size, uint32_t flags); -int amdgpu_vm_bo_unmap(struct amdgpu_device *adev, - struct amdgpu_bo_va *bo_va, - uint64_t addr); -void amdgpu_vm_bo_rmv(struct amdgpu_device *adev, - struct amdgpu_bo_va *bo_va); -int amdgpu_vm_free_job(struct amdgpu_job *job); -/* * functions used by amdgpu_encoder.c */ struct amdgpu_afmt_acr { diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_acp.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_acp.c new file mode 100644 index 000000000000..d6b0bff510aa --- /dev/null +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_acp.c @@ -0,0 +1,500 @@ +/* + * Copyright 2015 Advanced Micro Devices, Inc. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice shall be included in + * all copies or substantial portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR + * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, + * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR + * OTHER DEALINGS IN THE SOFTWARE. + * + * Authors: AMD + * + */ + +#include <linux/irqdomain.h> +#include <linux/pm_domain.h> +#include <linux/platform_device.h> +#include <sound/designware_i2s.h> +#include <sound/pcm.h> + +#include "amdgpu.h" +#include "atom.h" +#include "amdgpu_acp.h" + +#include "acp_gfx_if.h" + +#define ACP_TILE_ON_MASK 0x03 +#define ACP_TILE_OFF_MASK 0x02 +#define ACP_TILE_ON_RETAIN_REG_MASK 0x1f +#define ACP_TILE_OFF_RETAIN_REG_MASK 0x20 + +#define ACP_TILE_P1_MASK 0x3e +#define ACP_TILE_P2_MASK 0x3d +#define ACP_TILE_DSP0_MASK 0x3b +#define ACP_TILE_DSP1_MASK 0x37 + +#define ACP_TILE_DSP2_MASK 0x2f + +#define ACP_DMA_REGS_END 0x146c0 +#define ACP_I2S_PLAY_REGS_START 0x14840 +#define ACP_I2S_PLAY_REGS_END 0x148b4 +#define ACP_I2S_CAP_REGS_START 0x148b8 +#define ACP_I2S_CAP_REGS_END 0x1496c + +#define ACP_I2S_COMP1_CAP_REG_OFFSET 0xac +#define ACP_I2S_COMP2_CAP_REG_OFFSET 0xa8 +#define ACP_I2S_COMP1_PLAY_REG_OFFSET 0x6c +#define ACP_I2S_COMP2_PLAY_REG_OFFSET 0x68 + +#define mmACP_PGFSM_RETAIN_REG 0x51c9 +#define mmACP_PGFSM_CONFIG_REG 0x51ca +#define mmACP_PGFSM_READ_REG_0 0x51cc + +#define mmACP_MEM_SHUT_DOWN_REQ_LO 0x51f8 +#define mmACP_MEM_SHUT_DOWN_REQ_HI 0x51f9 +#define mmACP_MEM_SHUT_DOWN_STS_LO 0x51fa +#define mmACP_MEM_SHUT_DOWN_STS_HI 0x51fb + +#define ACP_TIMEOUT_LOOP 0x000000FF +#define ACP_DEVS 3 +#define ACP_SRC_ID 162 + +enum { + ACP_TILE_P1 = 0, + ACP_TILE_P2, + ACP_TILE_DSP0, + ACP_TILE_DSP1, + ACP_TILE_DSP2, +}; + +static int acp_sw_init(void *handle) +{ + struct amdgpu_device *adev = (struct amdgpu_device *)handle; + + adev->acp.parent = adev->dev; + + adev->acp.cgs_device = + amdgpu_cgs_create_device(adev); + if (!adev->acp.cgs_device) + return -EINVAL; + + return 0; +} + +static int acp_sw_fini(void *handle) +{ + struct amdgpu_device *adev = (struct amdgpu_device *)handle; + + if (adev->acp.cgs_device) + amdgpu_cgs_destroy_device(adev->acp.cgs_device); + + return 0; +} + +/* power off a tile/block within ACP */ +static int acp_suspend_tile(void *cgs_dev, int tile) +{ + u32 val = 0; + u32 count = 0; + + if ((tile < ACP_TILE_P1) || (tile > ACP_TILE_DSP2)) { + pr_err("Invalid ACP tile : %d to suspend\n", tile); + return -1; + } + + val = cgs_read_register(cgs_dev, mmACP_PGFSM_READ_REG_0 + tile); + val &= ACP_TILE_ON_MASK; + + if (val == 0x0) { + val = cgs_read_register(cgs_dev, mmACP_PGFSM_RETAIN_REG); + val = val | (1 << tile); + cgs_write_register(cgs_dev, mmACP_PGFSM_RETAIN_REG, val); + cgs_write_register(cgs_dev, mmACP_PGFSM_CONFIG_REG, + 0x500 + tile); + + count = ACP_TIMEOUT_LOOP; + while (true) { + val = cgs_read_register(cgs_dev, mmACP_PGFSM_READ_REG_0 + + tile); + val = val & ACP_TILE_ON_MASK; + if (val == ACP_TILE_OFF_MASK) + break; + if (--count == 0) { + pr_err("Timeout reading ACP PGFSM status\n"); + return -ETIMEDOUT; + } + udelay(100); + } + + val = cgs_read_register(cgs_dev, mmACP_PGFSM_RETAIN_REG); + + val |= ACP_TILE_OFF_RETAIN_REG_MASK; + cgs_write_register(cgs_dev, mmACP_PGFSM_RETAIN_REG, val); + } + return 0; +} + +/* power on a tile/block within ACP */ +static int acp_resume_tile(void *cgs_dev, int tile) +{ + u32 val = 0; + u32 count = 0; + + if ((tile < ACP_TILE_P1) || (tile > ACP_TILE_DSP2)) { + pr_err("Invalid ACP tile to resume\n"); + return -1; + } + + val = cgs_read_register(cgs_dev, mmACP_PGFSM_READ_REG_0 + tile); + val = val & ACP_TILE_ON_MASK; + + if (val != 0x0) { + cgs_write_register(cgs_dev, mmACP_PGFSM_CONFIG_REG, + 0x600 + tile); + count = ACP_TIMEOUT_LOOP; + while (true) { + val = cgs_read_register(cgs_dev, mmACP_PGFSM_READ_REG_0 + + tile); + val = val & ACP_TILE_ON_MASK; + if (val == 0x0) + break; + if (--count == 0) { + pr_err("Timeout reading ACP PGFSM status\n"); + return -ETIMEDOUT; + } + udelay(100); + } + val = cgs_read_register(cgs_dev, mmACP_PGFSM_RETAIN_REG); + if (tile == ACP_TILE_P1) + val = val & (ACP_TILE_P1_MASK); + else if (tile == ACP_TILE_P2) + val = val & (ACP_TILE_P2_MASK); + + cgs_write_register(cgs_dev, mmACP_PGFSM_RETAIN_REG, val); + } + return 0; +} + +struct acp_pm_domain { + void *cgs_dev; + struct generic_pm_domain gpd; +}; + +static int acp_poweroff(struct generic_pm_domain *genpd) +{ + int i, ret; + struct acp_pm_domain *apd; + + apd = container_of(genpd, struct acp_pm_domain, gpd); + if (apd != NULL) { + /* Donot return abruptly if any of power tile fails to suspend. + * Log it and continue powering off other tile + */ + for (i = 4; i >= 0 ; i--) { + ret = acp_suspend_tile(apd->cgs_dev, ACP_TILE_P1 + i); + if (ret) + pr_err("ACP tile %d tile suspend failed\n", i); + } + } + return 0; +} + +static int acp_poweron(struct generic_pm_domain *genpd) +{ + int i, ret; + struct acp_pm_domain *apd; + + apd = container_of(genpd, struct acp_pm_domain, gpd); + if (apd != NULL) { + for (i = 0; i < 2; i++) { + ret = acp_resume_tile(apd->cgs_dev, ACP_TILE_P1 + i); + if (ret) { + pr_err("ACP tile %d resume failed\n", i); + break; + } + } + + /* Disable DSPs which are not going to be used */ + for (i = 0; i < 3; i++) { + ret = acp_suspend_tile(apd->cgs_dev, ACP_TILE_DSP0 + i); + /* Continue suspending other DSP, even if one fails */ + if (ret) + pr_err("ACP DSP %d suspend failed\n", i); + } + } + return 0; +} + +static struct device *get_mfd_cell_dev(const char *device_name, int r) +{ + char auto_dev_name[25]; + struct device *dev; + + snprintf(auto_dev_name, sizeof(auto_dev_name), + "%s.%d.auto", device_name, r); + dev = bus_find_device_by_name(&platform_bus_type, NULL, auto_dev_name); + dev_info(dev, "device %s added to pm domain\n", auto_dev_name); + + return dev; +} + +/** + * acp_hw_init - start and test ACP block + * + * @adev: amdgpu_device pointer + * + */ +static int acp_hw_init(void *handle) +{ + int r, i; + uint64_t acp_base; + struct device *dev; + struct i2s_platform_data *i2s_pdata; + + struct amdgpu_device *adev = (struct amdgpu_device *)handle; + + const struct amdgpu_ip_block_version *ip_version = + amdgpu_get_ip_block(adev, AMD_IP_BLOCK_TYPE_ACP); + + if (!ip_version) + return -EINVAL; + + r = amd_acp_hw_init(adev->acp.cgs_device, + ip_version->major, ip_version->minor); + /* -ENODEV means board uses AZ rather than ACP */ + if (r == -ENODEV) + return 0; + else if (r) + return r; + + r = cgs_get_pci_resource(adev->acp.cgs_device, CGS_RESOURCE_TYPE_MMIO, + 0x5289, 0, &acp_base); + if (r == -ENODEV) + return 0; + else if (r) + return r; + + adev->acp.acp_genpd = kzalloc(sizeof(struct acp_pm_domain), GFP_KERNEL); + if (adev->acp.acp_genpd == NULL) + return -ENOMEM; + + adev->acp.acp_genpd->gpd.name = "ACP_AUDIO"; + adev->acp.acp_genpd->gpd.power_off = acp_poweroff; + adev->acp.acp_genpd->gpd.power_on = acp_poweron; + + + adev->acp.acp_genpd->cgs_dev = adev->acp.cgs_device; + + pm_genpd_init(&adev->acp.acp_genpd->gpd, NULL, false); + + adev->acp.acp_cell = kzalloc(sizeof(struct mfd_cell) * ACP_DEVS, + GFP_KERNEL); + + if (adev->acp.acp_cell == NULL) + return -ENOMEM; + + adev->acp.acp_res = kzalloc(sizeof(struct resource) * 4, GFP_KERNEL); + + if (adev->acp.acp_res == NULL) { + kfree(adev->acp.acp_cell); + return -ENOMEM; + } + + i2s_pdata = kzalloc(sizeof(struct i2s_platform_data) * 2, GFP_KERNEL); + if (i2s_pdata == NULL) { + kfree(adev->acp.acp_res); + kfree(adev->acp.acp_cell); + return -ENOMEM; + } + + i2s_pdata[0].quirks = DW_I2S_QUIRK_COMP_REG_OFFSET; + i2s_pdata[0].cap = DWC_I2S_PLAY; + i2s_pdata[0].snd_rates = SNDRV_PCM_RATE_8000_96000; + i2s_pdata[0].i2s_reg_comp1 = ACP_I2S_COMP1_PLAY_REG_OFFSET; + i2s_pdata[0].i2s_reg_comp2 = ACP_I2S_COMP2_PLAY_REG_OFFSET; + + i2s_pdata[1].quirks = DW_I2S_QUIRK_COMP_REG_OFFSET | + DW_I2S_QUIRK_COMP_PARAM1; + i2s_pdata[1].cap = DWC_I2S_RECORD; + i2s_pdata[1].snd_rates = SNDRV_PCM_RATE_8000_96000; + i2s_pdata[1].i2s_reg_comp1 = ACP_I2S_COMP1_CAP_REG_OFFSET; + i2s_pdata[1].i2s_reg_comp2 = ACP_I2S_COMP2_CAP_REG_OFFSET; + + adev->acp.acp_res[0].name = "acp2x_dma"; + adev->acp.acp_res[0].flags = IORESOURCE_MEM; + adev->acp.acp_res[0].start = acp_base; + adev->acp.acp_res[0].end = acp_base + ACP_DMA_REGS_END; + + adev->acp.acp_res[1].name = "acp2x_dw_i2s_play"; + adev->acp.acp_res[1].flags = IORESOURCE_MEM; + adev->acp.acp_res[1].start = acp_base + ACP_I2S_PLAY_REGS_START; + adev->acp.acp_res[1].end = acp_base + ACP_I2S_PLAY_REGS_END; + + adev->acp.acp_res[2].name = "acp2x_dw_i2s_cap"; + adev->acp.acp_res[2].flags = IORESOURCE_MEM; + adev->acp.acp_res[2].start = acp_base + ACP_I2S_CAP_REGS_START; + adev->acp.acp_res[2].end = acp_base + ACP_I2S_CAP_REGS_END; + + adev->acp.acp_res[3].name = "acp2x_dma_irq"; + adev->acp.acp_res[3].flags = IORESOURCE_IRQ; + adev->acp.acp_res[3].start = amdgpu_irq_create_mapping(adev, 162); + adev->acp.acp_res[3].end = adev->acp.acp_res[3].start; + + adev->acp.acp_cell[0].name = "acp_audio_dma"; + adev->acp.acp_cell[0].num_resources = 4; + adev->acp.acp_cell[0].resources = &adev->acp.acp_res[0]; + + adev->acp.acp_cell[1].name = "designware-i2s"; + adev->acp.acp_cell[1].num_resources = 1; + adev->acp.acp_cell[1].resources = &adev->acp.acp_res[1]; + adev->acp.acp_cell[1].platform_data = &i2s_pdata[0]; + adev->acp.acp_cell[1].pdata_size = sizeof(struct i2s_platform_data); + + adev->acp.acp_cell[2].name = "designware-i2s"; + adev->acp.acp_cell[2].num_resources = 1; + adev->acp.acp_cell[2].resources = &adev->acp.acp_res[2]; + adev->acp.acp_cell[2].platform_data = &i2s_pdata[1]; + adev->acp.acp_cell[2].pdata_size = sizeof(struct i2s_platform_data); + + r = mfd_add_hotplug_devices(adev->acp.parent, adev->acp.acp_cell, + ACP_DEVS); + if (r) + return r; + + for (i = 0; i < ACP_DEVS ; i++) { + dev = get_mfd_cell_dev(adev->acp.acp_cell[i].name, i); + r = pm_genpd_add_device(&adev->acp.acp_genpd->gpd, dev); + if (r) { + dev_err(dev, "Failed to add dev to genpd\n"); + return r; + } + } + + return 0; +} + +/** + * acp_hw_fini - stop the hardware block + * + * @adev: amdgpu_device pointer + * + */ +static int acp_hw_fini(void *handle) +{ + int i, ret; + struct device *dev; + + struct amdgpu_device *adev = (struct amdgpu_device *)handle; + + for (i = 0; i < ACP_DEVS ; i++) { + dev = get_mfd_cell_dev(adev->acp.acp_cell[i].name, i); + ret = pm_genpd_remove_device(&adev->acp.acp_genpd->gpd, dev); + /* If removal fails, dont giveup and try rest */ + if (ret) + dev_err(dev, "remove dev from genpd failed\n"); + } + + mfd_remove_devices(adev->acp.parent); + kfree(adev->acp.acp_res); + kfree(adev->acp.acp_genpd); + kfree(adev->acp.acp_cell); + + return 0; +} + +static int acp_suspend(void *handle) +{ + return 0; +} + +static int acp_resume(void *handle) +{ + int i, ret; + struct acp_pm_domain *apd; + struct amdgpu_device *adev = (struct amdgpu_device *)handle; + + /* SMU block will power on ACP irrespective of ACP runtime status. + * Power off explicitly based on genpd ACP runtime status so that ACP + * hw and ACP-genpd status are in sync. + * 'suspend_power_off' represents "Power status before system suspend" + */ + if (adev->acp.acp_genpd->gpd.suspend_power_off == true) { + apd = container_of(&adev->acp.acp_genpd->gpd, + struct acp_pm_domain, gpd); + + for (i = 4; i >= 0 ; i--) { + ret = acp_suspend_tile(apd->cgs_dev, ACP_TILE_P1 + i); + if (ret) + pr_err("ACP tile %d tile suspend failed\n", i); + } + } + return 0; +} + +static int acp_early_init(void *handle) +{ + return 0; +} + +static bool acp_is_idle(void *handle) +{ + return true; +} + +static int acp_wait_for_idle(void *handle) +{ + return 0; +} + +static int acp_soft_reset(void *handle) +{ + return 0; +} + +static void acp_print_status(void *handle) +{ + struct amdgpu_device *adev = (struct amdgpu_device *)handle; + + dev_info(adev->dev, "ACP STATUS\n"); +} + +static int acp_set_clockgating_state(void *handle, + enum amd_clockgating_state state) +{ + return 0; +} + +static int acp_set_powergating_state(void *handle, + enum amd_powergating_state state) +{ + return 0; +} + +const struct amd_ip_funcs acp_ip_funcs = { + .early_init = acp_early_init, + .late_init = NULL, + .sw_init = acp_sw_init, + .sw_fini = acp_sw_fini, + .hw_init = acp_hw_init, + .hw_fini = acp_hw_fini, + .suspend = acp_suspend, + .resume = acp_resume, + .is_idle = acp_is_idle, + .wait_for_idle = acp_wait_for_idle, + .soft_reset = acp_soft_reset, + .print_status = acp_print_status, + .set_clockgating_state = acp_set_clockgating_state, + .set_powergating_state = acp_set_powergating_state, +}; diff --git a/drivers/gpu/drm/amd/amdgpu/smu8.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_acp.h index d758d07b6a31..f6e32a639107 100644 --- a/drivers/gpu/drm/amd/amdgpu/smu8.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_acp.h @@ -1,5 +1,5 @@ /* - * Copyright 2014 Advanced Micro Devices, Inc. + * Copyright 2015 Advanced Micro Devices, Inc. * * Permission is hereby granted, free of charge, to any person obtaining a * copy of this software and associated documentation files (the "Software"), @@ -19,54 +19,24 @@ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR * OTHER DEALINGS IN THE SOFTWARE. * + * Authors: AMD + * */ -#ifndef SMU8_H -#define SMU8_H - -#pragma pack(push, 1) +#ifndef __AMDGPU_ACP_H__ +#define __AMDGPU_ACP_H__ -#define ENABLE_DEBUG_FEATURES +#include <linux/mfd/core.h> -struct SMU8_Firmware_Header { - uint32_t Version; - uint32_t ImageSize; - uint32_t CodeSize; - uint32_t HeaderSize; - uint32_t EntryPoint; - uint32_t Rtos; - uint32_t UcodeLoadStatus; - uint32_t DpmTable; - uint32_t FanTable; - uint32_t PmFuseTable; - uint32_t Globals; - uint32_t Reserved[20]; - uint32_t Signature; +struct amdgpu_acp { + struct device *parent; + void *cgs_device; + struct amd_acp_private *private; + struct mfd_cell *acp_cell; + struct resource *acp_res; + struct acp_pm_domain *acp_genpd; }; -struct SMU8_MultimediaPowerLogData { - uint32_t avgTotalPower; - uint32_t avgGpuPower; - uint32_t avgUvdPower; - uint32_t avgVcePower; - - uint32_t avgSclk; - uint32_t avgDclk; - uint32_t avgVclk; - uint32_t avgEclk; - - uint32_t startTimeHi; - uint32_t startTimeLo; - - uint32_t endTimeHi; - uint32_t endTimeLo; -}; - -#define SMU8_FIRMWARE_HEADER_LOCATION 0x1FF80 -#define SMU8_UNBCSR_START_ADDR 0xC0100000 - -#define SMN_MP1_SRAM_START_ADDR 0x10000000 - -#pragma pack(pop) +extern const struct amd_ip_funcs acp_ip_funcs; -#endif +#endif /* __AMDGPU_ACP_H__ */ diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_acpi.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_acpi.c index aef4a7aac0f7..5cd7b736a9de 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_acpi.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_acpi.c @@ -25,71 +25,14 @@ #include <linux/acpi.h> #include <linux/slab.h> #include <linux/power_supply.h> -#include <linux/vga_switcheroo.h> #include <acpi/video.h> #include <drm/drmP.h> #include <drm/drm_crtc_helper.h> #include "amdgpu.h" -#include "amdgpu_acpi.h" +#include "amd_acpi.h" #include "atom.h" -#define ACPI_AC_CLASS "ac_adapter" - extern void amdgpu_pm_acpi_event_handler(struct amdgpu_device *adev); - -struct atif_verify_interface { - u16 size; /* structure size in bytes (includes size field) */ - u16 version; /* version */ - u32 notification_mask; /* supported notifications mask */ - u32 function_bits; /* supported functions bit vector */ -} __packed; - -struct atif_system_params { - u16 size; /* structure size in bytes (includes size field) */ - u32 valid_mask; /* valid flags mask */ - u32 flags; /* flags */ - u8 command_code; /* notify command code */ -} __packed; - -struct atif_sbios_requests { - u16 size; /* structure size in bytes (includes size field) */ - u32 pending; /* pending sbios requests */ - u8 panel_exp_mode; /* panel expansion mode */ - u8 thermal_gfx; /* thermal state: target gfx controller */ - u8 thermal_state; /* thermal state: state id (0: exit state, non-0: state) */ - u8 forced_power_gfx; /* forced power state: target gfx controller */ - u8 forced_power_state; /* forced power state: state id */ - u8 system_power_src; /* system power source */ - u8 backlight_level; /* panel backlight level (0-255) */ -} __packed; - -#define ATIF_NOTIFY_MASK 0x3 -#define ATIF_NOTIFY_NONE 0 -#define ATIF_NOTIFY_81 1 -#define ATIF_NOTIFY_N 2 - -struct atcs_verify_interface { - u16 size; /* structure size in bytes (includes size field) */ - u16 version; /* version */ - u32 function_bits; /* supported functions bit vector */ -} __packed; - -#define ATCS_VALID_FLAGS_MASK 0x3 - -struct atcs_pref_req_input { - u16 size; /* structure size in bytes (includes size field) */ - u16 client_id; /* client id (bit 2-0: func num, 7-3: dev num, 15-8: bus num) */ - u16 valid_flags_mask; /* valid flags mask */ - u16 flags; /* flags */ - u8 req_type; /* request type */ - u8 perf_req; /* performance request */ -} __packed; - -struct atcs_pref_req_output { - u16 size; /* structure size in bytes (includes size field) */ - u8 ret_val; /* return value */ -} __packed; - /* Call the ATIF method */ /** diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_acpi.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_acpi.h deleted file mode 100644 index 01a29c3d7011..000000000000 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_acpi.h +++ /dev/null @@ -1,445 +0,0 @@ -/* - * Copyright 2012 Advanced Micro Devices, Inc. - * - * Permission is hereby granted, free of charge, to any person obtaining a - * copy of this software and associated documentation files (the "Software"), - * to deal in the Software without restriction, including without limitation - * the rights to use, copy, modify, merge, publish, distribute, sublicense, - * and/or sell copies of the Software, and to permit persons to whom the - * Software is furnished to do so, subject to the following conditions: - * - * The above copyright notice and this permission notice shall be included in - * all copies or substantial portions of the Software. - * - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR - * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, - * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL - * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR - * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, - * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR - * OTHER DEALINGS IN THE SOFTWARE. - * - */ - -#ifndef AMDGPU_ACPI_H -#define AMDGPU_ACPI_H - -struct amdgpu_device; -struct acpi_bus_event; - -int amdgpu_atif_handler(struct amdgpu_device *adev, - struct acpi_bus_event *event); - -/* AMD hw uses four ACPI control methods: - * 1. ATIF - * ARG0: (ACPI_INTEGER) function code - * ARG1: (ACPI_BUFFER) parameter buffer, 256 bytes - * OUTPUT: (ACPI_BUFFER) output buffer, 256 bytes - * ATIF provides an entry point for the gfx driver to interact with the sbios. - * The AMD ACPI notification mechanism uses Notify (VGA, 0x81) or a custom - * notification. Which notification is used as indicated by the ATIF Control - * Method GET_SYSTEM_PARAMETERS. When the driver receives Notify (VGA, 0x81) or - * a custom notification it invokes ATIF Control Method GET_SYSTEM_BIOS_REQUESTS - * to identify pending System BIOS requests and associated parameters. For - * example, if one of the pending requests is DISPLAY_SWITCH_REQUEST, the driver - * will perform display device detection and invoke ATIF Control Method - * SELECT_ACTIVE_DISPLAYS. - * - * 2. ATPX - * ARG0: (ACPI_INTEGER) function code - * ARG1: (ACPI_BUFFER) parameter buffer, 256 bytes - * OUTPUT: (ACPI_BUFFER) output buffer, 256 bytes - * ATPX methods are used on PowerXpress systems to handle mux switching and - * discrete GPU power control. - * - * 3. ATRM - * ARG0: (ACPI_INTEGER) offset of vbios rom data - * ARG1: (ACPI_BUFFER) size of the buffer to fill (up to 4K). - * OUTPUT: (ACPI_BUFFER) output buffer - * ATRM provides an interfacess to access the discrete GPU vbios image on - * PowerXpress systems with multiple GPUs. - * - * 4. ATCS - * ARG0: (ACPI_INTEGER) function code - * ARG1: (ACPI_BUFFER) parameter buffer, 256 bytes - * OUTPUT: (ACPI_BUFFER) output buffer, 256 bytes - * ATCS provides an interface to AMD chipset specific functionality. - * - */ -/* ATIF */ -#define ATIF_FUNCTION_VERIFY_INTERFACE 0x0 -/* ARG0: ATIF_FUNCTION_VERIFY_INTERFACE - * ARG1: none - * OUTPUT: - * WORD - structure size in bytes (includes size field) - * WORD - version - * DWORD - supported notifications mask - * DWORD - supported functions bit vector - */ -/* Notifications mask */ -# define ATIF_DISPLAY_SWITCH_REQUEST_SUPPORTED (1 << 0) -# define ATIF_EXPANSION_MODE_CHANGE_REQUEST_SUPPORTED (1 << 1) -# define ATIF_THERMAL_STATE_CHANGE_REQUEST_SUPPORTED (1 << 2) -# define ATIF_FORCED_POWER_STATE_CHANGE_REQUEST_SUPPORTED (1 << 3) -# define ATIF_SYSTEM_POWER_SOURCE_CHANGE_REQUEST_SUPPORTED (1 << 4) -# define ATIF_DISPLAY_CONF_CHANGE_REQUEST_SUPPORTED (1 << 5) -# define ATIF_PX_GFX_SWITCH_REQUEST_SUPPORTED (1 << 6) -# define ATIF_PANEL_BRIGHTNESS_CHANGE_REQUEST_SUPPORTED (1 << 7) -# define ATIF_DGPU_DISPLAY_EVENT_SUPPORTED (1 << 8) -/* supported functions vector */ -# define ATIF_GET_SYSTEM_PARAMETERS_SUPPORTED (1 << 0) -# define ATIF_GET_SYSTEM_BIOS_REQUESTS_SUPPORTED (1 << 1) -# define ATIF_SELECT_ACTIVE_DISPLAYS_SUPPORTED (1 << 2) -# define ATIF_GET_LID_STATE_SUPPORTED (1 << 3) -# define ATIF_GET_TV_STANDARD_FROM_CMOS_SUPPORTED (1 << 4) -# define ATIF_SET_TV_STANDARD_IN_CMOS_SUPPORTED (1 << 5) -# define ATIF_GET_PANEL_EXPANSION_MODE_FROM_CMOS_SUPPORTED (1 << 6) -# define ATIF_SET_PANEL_EXPANSION_MODE_IN_CMOS_SUPPORTED (1 << 7) -# define ATIF_TEMPERATURE_CHANGE_NOTIFICATION_SUPPORTED (1 << 12) -# define ATIF_GET_GRAPHICS_DEVICE_TYPES_SUPPORTED (1 << 14) -#define ATIF_FUNCTION_GET_SYSTEM_PARAMETERS 0x1 -/* ARG0: ATIF_FUNCTION_GET_SYSTEM_PARAMETERS - * ARG1: none - * OUTPUT: - * WORD - structure size in bytes (includes size field) - * DWORD - valid flags mask - * DWORD - flags - * - * OR - * - * WORD - structure size in bytes (includes size field) - * DWORD - valid flags mask - * DWORD - flags - * BYTE - notify command code - * - * flags - * bits 1:0: - * 0 - Notify(VGA, 0x81) is not used for notification - * 1 - Notify(VGA, 0x81) is used for notification - * 2 - Notify(VGA, n) is used for notification where - * n (0xd0-0xd9) is specified in notify command code. - * bit 2: - * 1 - lid changes not reported though int10 - */ -#define ATIF_FUNCTION_GET_SYSTEM_BIOS_REQUESTS 0x2 -/* ARG0: ATIF_FUNCTION_GET_SYSTEM_BIOS_REQUESTS - * ARG1: none - * OUTPUT: - * WORD - structure size in bytes (includes size field) - * DWORD - pending sbios requests - * BYTE - panel expansion mode - * BYTE - thermal state: target gfx controller - * BYTE - thermal state: state id (0: exit state, non-0: state) - * BYTE - forced power state: target gfx controller - * BYTE - forced power state: state id - * BYTE - system power source - * BYTE - panel backlight level (0-255) - */ -/* pending sbios requests */ -# define ATIF_DISPLAY_SWITCH_REQUEST (1 << 0) -# define ATIF_EXPANSION_MODE_CHANGE_REQUEST (1 << 1) -# define ATIF_THERMAL_STATE_CHANGE_REQUEST (1 << 2) -# define ATIF_FORCED_POWER_STATE_CHANGE_REQUEST (1 << 3) -# define ATIF_SYSTEM_POWER_SOURCE_CHANGE_REQUEST (1 << 4) -# define ATIF_DISPLAY_CONF_CHANGE_REQUEST (1 << 5) -# define ATIF_PX_GFX_SWITCH_REQUEST (1 << 6) -# define ATIF_PANEL_BRIGHTNESS_CHANGE_REQUEST (1 << 7) -# define ATIF_DGPU_DISPLAY_EVENT (1 << 8) -/* panel expansion mode */ -# define ATIF_PANEL_EXPANSION_DISABLE 0 -# define ATIF_PANEL_EXPANSION_FULL 1 -# define ATIF_PANEL_EXPANSION_ASPECT 2 -/* target gfx controller */ -# define ATIF_TARGET_GFX_SINGLE 0 -# define ATIF_TARGET_GFX_PX_IGPU 1 -# define ATIF_TARGET_GFX_PX_DGPU 2 -/* system power source */ -# define ATIF_POWER_SOURCE_AC 1 -# define ATIF_POWER_SOURCE_DC 2 -# define ATIF_POWER_SOURCE_RESTRICTED_AC_1 3 -# define ATIF_POWER_SOURCE_RESTRICTED_AC_2 4 -#define ATIF_FUNCTION_SELECT_ACTIVE_DISPLAYS 0x3 -/* ARG0: ATIF_FUNCTION_SELECT_ACTIVE_DISPLAYS - * ARG1: - * WORD - structure size in bytes (includes size field) - * WORD - selected displays - * WORD - connected displays - * OUTPUT: - * WORD - structure size in bytes (includes size field) - * WORD - selected displays - */ -# define ATIF_LCD1 (1 << 0) -# define ATIF_CRT1 (1 << 1) -# define ATIF_TV (1 << 2) -# define ATIF_DFP1 (1 << 3) -# define ATIF_CRT2 (1 << 4) -# define ATIF_LCD2 (1 << 5) -# define ATIF_DFP2 (1 << 7) -# define ATIF_CV (1 << 8) -# define ATIF_DFP3 (1 << 9) -# define ATIF_DFP4 (1 << 10) -# define ATIF_DFP5 (1 << 11) -# define ATIF_DFP6 (1 << 12) -#define ATIF_FUNCTION_GET_LID_STATE 0x4 -/* ARG0: ATIF_FUNCTION_GET_LID_STATE - * ARG1: none - * OUTPUT: - * WORD - structure size in bytes (includes size field) - * BYTE - lid state (0: open, 1: closed) - * - * GET_LID_STATE only works at boot and resume, for general lid - * status, use the kernel provided status - */ -#define ATIF_FUNCTION_GET_TV_STANDARD_FROM_CMOS 0x5 -/* ARG0: ATIF_FUNCTION_GET_TV_STANDARD_FROM_CMOS - * ARG1: none - * OUTPUT: - * WORD - structure size in bytes (includes size field) - * BYTE - 0 - * BYTE - TV standard - */ -# define ATIF_TV_STD_NTSC 0 -# define ATIF_TV_STD_PAL 1 -# define ATIF_TV_STD_PALM 2 -# define ATIF_TV_STD_PAL60 3 -# define ATIF_TV_STD_NTSCJ 4 -# define ATIF_TV_STD_PALCN 5 -# define ATIF_TV_STD_PALN 6 -# define ATIF_TV_STD_SCART_RGB 9 -#define ATIF_FUNCTION_SET_TV_STANDARD_IN_CMOS 0x6 -/* ARG0: ATIF_FUNCTION_SET_TV_STANDARD_IN_CMOS - * ARG1: - * WORD - structure size in bytes (includes size field) - * BYTE - 0 - * BYTE - TV standard - * OUTPUT: none - */ -#define ATIF_FUNCTION_GET_PANEL_EXPANSION_MODE_FROM_CMOS 0x7 -/* ARG0: ATIF_FUNCTION_GET_PANEL_EXPANSION_MODE_FROM_CMOS - * ARG1: none - * OUTPUT: - * WORD - structure size in bytes (includes size field) - * BYTE - panel expansion mode - */ -#define ATIF_FUNCTION_SET_PANEL_EXPANSION_MODE_IN_CMOS 0x8 -/* ARG0: ATIF_FUNCTION_SET_PANEL_EXPANSION_MODE_IN_CMOS - * ARG1: - * WORD - structure size in bytes (includes size field) - * BYTE - panel expansion mode - * OUTPUT: none - */ -#define ATIF_FUNCTION_TEMPERATURE_CHANGE_NOTIFICATION 0xD -/* ARG0: ATIF_FUNCTION_TEMPERATURE_CHANGE_NOTIFICATION - * ARG1: - * WORD - structure size in bytes (includes size field) - * WORD - gfx controller id - * BYTE - current temperature (degress Celsius) - * OUTPUT: none - */ -#define ATIF_FUNCTION_GET_GRAPHICS_DEVICE_TYPES 0xF -/* ARG0: ATIF_FUNCTION_GET_GRAPHICS_DEVICE_TYPES - * ARG1: none - * OUTPUT: - * WORD - number of gfx devices - * WORD - device structure size in bytes (excludes device size field) - * DWORD - flags \ - * WORD - bus number } repeated structure - * WORD - device number / - */ -/* flags */ -# define ATIF_PX_REMOVABLE_GRAPHICS_DEVICE (1 << 0) -# define ATIF_XGP_PORT (1 << 1) -# define ATIF_VGA_ENABLED_GRAPHICS_DEVICE (1 << 2) -# define ATIF_XGP_PORT_IN_DOCK (1 << 3) - -/* ATPX */ -#define ATPX_FUNCTION_VERIFY_INTERFACE 0x0 -/* ARG0: ATPX_FUNCTION_VERIFY_INTERFACE - * ARG1: none - * OUTPUT: - * WORD - structure size in bytes (includes size field) - * WORD - version - * DWORD - supported functions bit vector - */ -/* supported functions vector */ -# define ATPX_GET_PX_PARAMETERS_SUPPORTED (1 << 0) -# define ATPX_POWER_CONTROL_SUPPORTED (1 << 1) -# define ATPX_DISPLAY_MUX_CONTROL_SUPPORTED (1 << 2) -# define ATPX_I2C_MUX_CONTROL_SUPPORTED (1 << 3) -# define ATPX_GRAPHICS_DEVICE_SWITCH_START_NOTIFICATION_SUPPORTED (1 << 4) -# define ATPX_GRAPHICS_DEVICE_SWITCH_END_NOTIFICATION_SUPPORTED (1 << 5) -# define ATPX_GET_DISPLAY_CONNECTORS_MAPPING_SUPPORTED (1 << 7) -# define ATPX_GET_DISPLAY_DETECTION_PORTS_SUPPORTED (1 << 8) -#define ATPX_FUNCTION_GET_PX_PARAMETERS 0x1 -/* ARG0: ATPX_FUNCTION_GET_PX_PARAMETERS - * ARG1: none - * OUTPUT: - * WORD - structure size in bytes (includes size field) - * DWORD - valid flags mask - * DWORD - flags - */ -/* flags */ -# define ATPX_LVDS_I2C_AVAILABLE_TO_BOTH_GPUS (1 << 0) -# define ATPX_CRT1_I2C_AVAILABLE_TO_BOTH_GPUS (1 << 1) -# define ATPX_DVI1_I2C_AVAILABLE_TO_BOTH_GPUS (1 << 2) -# define ATPX_CRT1_RGB_SIGNAL_MUXED (1 << 3) -# define ATPX_TV_SIGNAL_MUXED (1 << 4) -# define ATPX_DFP_SIGNAL_MUXED (1 << 5) -# define ATPX_SEPARATE_MUX_FOR_I2C (1 << 6) -# define ATPX_DYNAMIC_PX_SUPPORTED (1 << 7) -# define ATPX_ACF_NOT_SUPPORTED (1 << 8) -# define ATPX_FIXED_NOT_SUPPORTED (1 << 9) -# define ATPX_DYNAMIC_DGPU_POWER_OFF_SUPPORTED (1 << 10) -# define ATPX_DGPU_REQ_POWER_FOR_DISPLAYS (1 << 11) -#define ATPX_FUNCTION_POWER_CONTROL 0x2 -/* ARG0: ATPX_FUNCTION_POWER_CONTROL - * ARG1: - * WORD - structure size in bytes (includes size field) - * BYTE - dGPU power state (0: power off, 1: power on) - * OUTPUT: none - */ -#define ATPX_FUNCTION_DISPLAY_MUX_CONTROL 0x3 -/* ARG0: ATPX_FUNCTION_DISPLAY_MUX_CONTROL - * ARG1: - * WORD - structure size in bytes (includes size field) - * WORD - display mux control (0: iGPU, 1: dGPU) - * OUTPUT: none - */ -# define ATPX_INTEGRATED_GPU 0 -# define ATPX_DISCRETE_GPU 1 -#define ATPX_FUNCTION_I2C_MUX_CONTROL 0x4 -/* ARG0: ATPX_FUNCTION_I2C_MUX_CONTROL - * ARG1: - * WORD - structure size in bytes (includes size field) - * WORD - i2c/aux/hpd mux control (0: iGPU, 1: dGPU) - * OUTPUT: none - */ -#define ATPX_FUNCTION_GRAPHICS_DEVICE_SWITCH_START_NOTIFICATION 0x5 -/* ARG0: ATPX_FUNCTION_GRAPHICS_DEVICE_SWITCH_START_NOTIFICATION - * ARG1: - * WORD - structure size in bytes (includes size field) - * WORD - target gpu (0: iGPU, 1: dGPU) - * OUTPUT: none - */ -#define ATPX_FUNCTION_GRAPHICS_DEVICE_SWITCH_END_NOTIFICATION 0x6 -/* ARG0: ATPX_FUNCTION_GRAPHICS_DEVICE_SWITCH_END_NOTIFICATION - * ARG1: - * WORD - structure size in bytes (includes size field) - * WORD - target gpu (0: iGPU, 1: dGPU) - * OUTPUT: none - */ -#define ATPX_FUNCTION_GET_DISPLAY_CONNECTORS_MAPPING 0x8 -/* ARG0: ATPX_FUNCTION_GET_DISPLAY_CONNECTORS_MAPPING - * ARG1: none - * OUTPUT: - * WORD - number of display connectors - * WORD - connector structure size in bytes (excludes connector size field) - * BYTE - flags \ - * BYTE - ATIF display vector bit position } repeated - * BYTE - adapter id (0: iGPU, 1-n: dGPU ordered by pcie bus number) } structure - * WORD - connector ACPI id / - */ -/* flags */ -# define ATPX_DISPLAY_OUTPUT_SUPPORTED_BY_ADAPTER_ID_DEVICE (1 << 0) -# define ATPX_DISPLAY_HPD_SUPPORTED_BY_ADAPTER_ID_DEVICE (1 << 1) -# define ATPX_DISPLAY_I2C_SUPPORTED_BY_ADAPTER_ID_DEVICE (1 << 2) -#define ATPX_FUNCTION_GET_DISPLAY_DETECTION_PORTS 0x9 -/* ARG0: ATPX_FUNCTION_GET_DISPLAY_DETECTION_PORTS - * ARG1: none - * OUTPUT: - * WORD - number of HPD/DDC ports - * WORD - port structure size in bytes (excludes port size field) - * BYTE - ATIF display vector bit position \ - * BYTE - hpd id } reapeated structure - * BYTE - ddc id / - * - * available on A+A systems only - */ -/* hpd id */ -# define ATPX_HPD_NONE 0 -# define ATPX_HPD1 1 -# define ATPX_HPD2 2 -# define ATPX_HPD3 3 -# define ATPX_HPD4 4 -# define ATPX_HPD5 5 -# define ATPX_HPD6 6 -/* ddc id */ -# define ATPX_DDC_NONE 0 -# define ATPX_DDC1 1 -# define ATPX_DDC2 2 -# define ATPX_DDC3 3 -# define ATPX_DDC4 4 -# define ATPX_DDC5 5 -# define ATPX_DDC6 6 -# define ATPX_DDC7 7 -# define ATPX_DDC8 8 - -/* ATCS */ -#define ATCS_FUNCTION_VERIFY_INTERFACE 0x0 -/* ARG0: ATCS_FUNCTION_VERIFY_INTERFACE - * ARG1: none - * OUTPUT: - * WORD - structure size in bytes (includes size field) - * WORD - version - * DWORD - supported functions bit vector - */ -/* supported functions vector */ -# define ATCS_GET_EXTERNAL_STATE_SUPPORTED (1 << 0) -# define ATCS_PCIE_PERFORMANCE_REQUEST_SUPPORTED (1 << 1) -# define ATCS_PCIE_DEVICE_READY_NOTIFICATION_SUPPORTED (1 << 2) -# define ATCS_SET_PCIE_BUS_WIDTH_SUPPORTED (1 << 3) -#define ATCS_FUNCTION_GET_EXTERNAL_STATE 0x1 -/* ARG0: ATCS_FUNCTION_GET_EXTERNAL_STATE - * ARG1: none - * OUTPUT: - * WORD - structure size in bytes (includes size field) - * DWORD - valid flags mask - * DWORD - flags (0: undocked, 1: docked) - */ -/* flags */ -# define ATCS_DOCKED (1 << 0) -#define ATCS_FUNCTION_PCIE_PERFORMANCE_REQUEST 0x2 -/* ARG0: ATCS_FUNCTION_PCIE_PERFORMANCE_REQUEST - * ARG1: - * WORD - structure size in bytes (includes size field) - * WORD - client id (bit 2-0: func num, 7-3: dev num, 15-8: bus num) - * WORD - valid flags mask - * WORD - flags - * BYTE - request type - * BYTE - performance request - * OUTPUT: - * WORD - structure size in bytes (includes size field) - * BYTE - return value - */ -/* flags */ -# define ATCS_ADVERTISE_CAPS (1 << 0) -# define ATCS_WAIT_FOR_COMPLETION (1 << 1) -/* request type */ -# define ATCS_PCIE_LINK_SPEED 1 -/* performance request */ -# define ATCS_REMOVE 0 -# define ATCS_FORCE_LOW_POWER 1 -# define ATCS_PERF_LEVEL_1 2 /* PCIE Gen 1 */ -# define ATCS_PERF_LEVEL_2 3 /* PCIE Gen 2 */ -# define ATCS_PERF_LEVEL_3 4 /* PCIE Gen 3 */ -/* return value */ -# define ATCS_REQUEST_REFUSED 1 -# define ATCS_REQUEST_COMPLETE 2 -# define ATCS_REQUEST_IN_PROGRESS 3 -#define ATCS_FUNCTION_PCIE_DEVICE_READY_NOTIFICATION 0x3 -/* ARG0: ATCS_FUNCTION_PCIE_DEVICE_READY_NOTIFICATION - * ARG1: none - * OUTPUT: none - */ -#define ATCS_FUNCTION_SET_PCIE_BUS_WIDTH 0x4 -/* ARG0: ATCS_FUNCTION_SET_PCIE_BUS_WIDTH - * ARG1: - * WORD - structure size in bytes (includes size field) - * WORD - client id (bit 2-0: func num, 7-3: dev num, 15-8: bus num) - * BYTE - number of active lanes - * OUTPUT: - * WORD - structure size in bytes (includes size field) - * BYTE - number of active lanes - */ - -#endif diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c index 84d68d658f8a..32809f749903 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c @@ -30,25 +30,38 @@ const struct kfd2kgd_calls *kfd2kgd; const struct kgd2kfd_calls *kgd2kfd; bool (*kgd2kfd_init_p)(unsigned, const struct kgd2kfd_calls**); -bool amdgpu_amdkfd_init(void) +int amdgpu_amdkfd_init(void) { + int ret; + #if defined(CONFIG_HSA_AMD_MODULE) - bool (*kgd2kfd_init_p)(unsigned, const struct kgd2kfd_calls**); + int (*kgd2kfd_init_p)(unsigned, const struct kgd2kfd_calls**); kgd2kfd_init_p = symbol_request(kgd2kfd_init); if (kgd2kfd_init_p == NULL) - return false; + return -ENOENT; + + ret = kgd2kfd_init_p(KFD_INTERFACE_VERSION, &kgd2kfd); + if (ret) { + symbol_put(kgd2kfd_init); + kgd2kfd = NULL; + } + +#elif defined(CONFIG_HSA_AMD) + ret = kgd2kfd_init(KFD_INTERFACE_VERSION, &kgd2kfd); + if (ret) + kgd2kfd = NULL; + +#else + ret = -ENOENT; #endif - return true; + + return ret; } bool amdgpu_amdkfd_load_interface(struct amdgpu_device *rdev) { -#if defined(CONFIG_HSA_AMD_MODULE) - bool (*kgd2kfd_init_p)(unsigned, const struct kgd2kfd_calls**); -#endif - switch (rdev->asic_type) { #ifdef CONFIG_DRM_AMDGPU_CIK case CHIP_KAVERI: @@ -62,35 +75,7 @@ bool amdgpu_amdkfd_load_interface(struct amdgpu_device *rdev) return false; } -#if defined(CONFIG_HSA_AMD_MODULE) - kgd2kfd_init_p = symbol_request(kgd2kfd_init); - - if (kgd2kfd_init_p == NULL) { - kfd2kgd = NULL; - return false; - } - - if (!kgd2kfd_init_p(KFD_INTERFACE_VERSION, &kgd2kfd)) { - symbol_put(kgd2kfd_init); - kfd2kgd = NULL; - kgd2kfd = NULL; - - return false; - } - return true; -#elif defined(CONFIG_HSA_AMD) - if (!kgd2kfd_init(KFD_INTERFACE_VERSION, &kgd2kfd)) { - kfd2kgd = NULL; - kgd2kfd = NULL; - return false; - } - - return true; -#else - kfd2kgd = NULL; - return false; -#endif } void amdgpu_amdkfd_fini(void) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h index a8be765542e6..de530f68d4e3 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h @@ -36,7 +36,7 @@ struct kgd_mem { void *cpu_ptr; }; -bool amdgpu_amdkfd_init(void); +int amdgpu_amdkfd_init(void); void amdgpu_amdkfd_fini(void); bool amdgpu_amdkfd_load_interface(struct amdgpu_device *rdev); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c index dd2037bc0b4a..362bedc9e507 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c @@ -154,7 +154,7 @@ static const struct kfd2kgd_calls kfd2kgd = { .get_fw_version = get_fw_version }; -struct kfd2kgd_calls *amdgpu_amdkfd_gfx_7_get_functions() +struct kfd2kgd_calls *amdgpu_amdkfd_gfx_7_get_functions(void) { return (struct kfd2kgd_calls *)&kfd2kgd; } @@ -649,12 +649,12 @@ static uint16_t get_fw_version(struct kgd_dev *kgd, enum kgd_engine_type type) case KGD_ENGINE_SDMA1: hdr = (const union amdgpu_firmware_header *) - adev->sdma[0].fw->data; + adev->sdma.instance[0].fw->data; break; case KGD_ENGINE_SDMA2: hdr = (const union amdgpu_firmware_header *) - adev->sdma[1].fw->data; + adev->sdma.instance[1].fw->data; break; default: diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c index dfd1d503bccf..04b744d64b57 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c @@ -115,7 +115,7 @@ static const struct kfd2kgd_calls kfd2kgd = { .get_fw_version = get_fw_version }; -struct kfd2kgd_calls *amdgpu_amdkfd_gfx_8_0_get_functions() +struct kfd2kgd_calls *amdgpu_amdkfd_gfx_8_0_get_functions(void) { return (struct kfd2kgd_calls *)&kfd2kgd; } @@ -523,12 +523,12 @@ static uint16_t get_fw_version(struct kgd_dev *kgd, enum kgd_engine_type type) case KGD_ENGINE_SDMA1: hdr = (const union amdgpu_firmware_header *) - adev->sdma[0].fw->data; + adev->sdma.instance[0].fw->data; break; case KGD_ENGINE_SDMA2: hdr = (const union amdgpu_firmware_header *) - adev->sdma[1].fw->data; + adev->sdma.instance[1].fw->data; break; default: diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_atombios.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_atombios.c index 9416e0f5c1db..84b0ce39ee14 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_atombios.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_atombios.c @@ -1514,6 +1514,19 @@ int amdgpu_atombios_init_mc_reg_table(struct amdgpu_device *adev, return -EINVAL; } +bool amdgpu_atombios_has_gpu_virtualization_table(struct amdgpu_device *adev) +{ + int index = GetIndexIntoMasterTable(DATA, GPUVirtualizationInfo); + u8 frev, crev; + u16 data_offset, size; + + if (amdgpu_atom_parse_data_header(adev->mode_info.atom_context, index, &size, + &frev, &crev, &data_offset)) + return true; + + return false; +} + void amdgpu_atombios_scratch_regs_lock(struct amdgpu_device *adev, bool lock) { uint32_t bios_6_scratch; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_atombios.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_atombios.h index 0ebb959ea435..9e1442053fe4 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_atombios.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_atombios.h @@ -196,6 +196,8 @@ int amdgpu_atombios_init_mc_reg_table(struct amdgpu_device *adev, u8 module_index, struct atom_mc_reg_table *reg_table); +bool amdgpu_atombios_has_gpu_virtualization_table(struct amdgpu_device *adev); + void amdgpu_atombios_scratch_regs_lock(struct amdgpu_device *adev, bool lock); void amdgpu_atombios_scratch_regs_init(struct amdgpu_device *adev); void amdgpu_atombios_scratch_regs_save(struct amdgpu_device *adev); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_atpx_handler.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_atpx_handler.c index 3f7aaa45bf8e..0020a0ea43ff 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_atpx_handler.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_atpx_handler.c @@ -11,7 +11,7 @@ #include <linux/acpi.h> #include <linux/pci.h> -#include "amdgpu_acpi.h" +#include "amd_acpi.h" struct amdgpu_atpx_functions { bool px_params; @@ -63,6 +63,10 @@ bool amdgpu_has_atpx(void) { return amdgpu_atpx_priv.atpx_detected; } +bool amdgpu_has_atpx_dgpu_power_cntl(void) { + return amdgpu_atpx_priv.atpx.functions.power_cntl; +} + /** * amdgpu_atpx_call - call an ATPX method * @@ -142,10 +146,6 @@ static void amdgpu_atpx_parse_functions(struct amdgpu_atpx_functions *f, u32 mas */ static int amdgpu_atpx_validate(struct amdgpu_atpx *atpx) { - /* make sure required functions are enabled */ - /* dGPU power control is required */ - atpx->functions.power_cntl = true; - if (atpx->functions.px_params) { union acpi_object *info; struct atpx_px_params output; @@ -501,7 +501,7 @@ static int amdgpu_atpx_get_client_id(struct pci_dev *pdev) return VGA_SWITCHEROO_DIS; } -static struct vga_switcheroo_handler amdgpu_atpx_handler = { +static const struct vga_switcheroo_handler amdgpu_atpx_handler = { .switchto = amdgpu_atpx_switchto, .power_state = amdgpu_atpx_power_state, .init = amdgpu_atpx_init, @@ -536,7 +536,7 @@ static bool amdgpu_atpx_detect(void) if (has_atpx && vga_count == 2) { acpi_get_name(amdgpu_atpx_priv.atpx.handle, ACPI_FULL_PATHNAME, &buffer); - printk(KERN_INFO "VGA switcheroo: detected switching method %s handle\n", + printk(KERN_INFO "vga_switcheroo: detected switching method %s handle\n", acpi_method_name); amdgpu_atpx_priv.atpx_detected = true; return true; @@ -552,13 +552,14 @@ static bool amdgpu_atpx_detect(void) void amdgpu_register_atpx_handler(void) { bool r; + enum vga_switcheroo_handler_flags_t handler_flags = 0; /* detect if we have any ATPX + 2 VGA in the system */ r = amdgpu_atpx_detect(); if (!r) return; - vga_switcheroo_register_handler(&amdgpu_atpx_handler); + vga_switcheroo_register_handler(&amdgpu_atpx_handler, handler_flags); } /** diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_bios.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_bios.c index 02add0a508cb..80add22375ee 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_bios.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_bios.c @@ -29,13 +29,19 @@ #include "amdgpu.h" #include "atom.h" -#include <linux/vga_switcheroo.h> #include <linux/slab.h> #include <linux/acpi.h> /* * BIOS. */ +#define AMD_VBIOS_SIGNATURE " 761295520" +#define AMD_VBIOS_SIGNATURE_OFFSET 0x30 +#define AMD_VBIOS_SIGNATURE_SIZE sizeof(AMD_VBIOS_SIGNATURE) +#define AMD_VBIOS_SIGNATURE_END (AMD_VBIOS_SIGNATURE_OFFSET + AMD_VBIOS_SIGNATURE_SIZE) +#define AMD_IS_VALID_VBIOS(p) ((p)[0] == 0x55 && (p)[1] == 0xAA) +#define AMD_VBIOS_LENGTH(p) ((p)[2] << 9) + /* If you boot an IGP board with a discrete card as the primary, * the IGP rom is not accessible via the rom bar as the IGP rom is * part of the system bios. On boot, the system bios puts a @@ -59,7 +65,7 @@ static bool igp_read_bios_from_vram(struct amdgpu_device *adev) return false; } - if (size == 0 || bios[0] != 0x55 || bios[1] != 0xaa) { + if (size == 0 || !AMD_IS_VALID_VBIOS(bios)) { iounmap(bios); return false; } @@ -75,7 +81,7 @@ static bool igp_read_bios_from_vram(struct amdgpu_device *adev) bool amdgpu_read_bios(struct amdgpu_device *adev) { - uint8_t __iomem *bios, val1, val2; + uint8_t __iomem *bios, val[2]; size_t size; adev->bios = NULL; @@ -85,10 +91,10 @@ bool amdgpu_read_bios(struct amdgpu_device *adev) return false; } - val1 = readb(&bios[0]); - val2 = readb(&bios[1]); + val[0] = readb(&bios[0]); + val[1] = readb(&bios[1]); - if (size == 0 || val1 != 0x55 || val2 != 0xaa) { + if (size == 0 || !AMD_IS_VALID_VBIOS(val)) { pci_unmap_rom(adev->pdev, bios); return false; } @@ -102,6 +108,38 @@ bool amdgpu_read_bios(struct amdgpu_device *adev) return true; } +static bool amdgpu_read_bios_from_rom(struct amdgpu_device *adev) +{ + u8 header[AMD_VBIOS_SIGNATURE_END+1] = {0}; + int len; + + if (!adev->asic_funcs->read_bios_from_rom) + return false; + + /* validate VBIOS signature */ + if (amdgpu_asic_read_bios_from_rom(adev, &header[0], sizeof(header)) == false) + return false; + header[AMD_VBIOS_SIGNATURE_END] = 0; + + if ((!AMD_IS_VALID_VBIOS(header)) || + 0 != memcmp((char *)&header[AMD_VBIOS_SIGNATURE_OFFSET], + AMD_VBIOS_SIGNATURE, + strlen(AMD_VBIOS_SIGNATURE))) + return false; + + /* valid vbios, go on */ + len = AMD_VBIOS_LENGTH(header); + len = ALIGN(len, 4); + adev->bios = kmalloc(len, GFP_KERNEL); + if (!adev->bios) { + DRM_ERROR("no memory to allocate for BIOS\n"); + return false; + } + + /* read complete BIOS */ + return amdgpu_asic_read_bios_from_rom(adev, adev->bios, len); +} + static bool amdgpu_read_platform_bios(struct amdgpu_device *adev) { uint8_t __iomem *bios; @@ -114,7 +152,7 @@ static bool amdgpu_read_platform_bios(struct amdgpu_device *adev) return false; } - if (size == 0 || bios[0] != 0x55 || bios[1] != 0xaa) { + if (size == 0 || !AMD_IS_VALID_VBIOS(bios)) { return false; } adev->bios = kmemdup(bios, size, GFP_KERNEL); @@ -231,7 +269,7 @@ static bool amdgpu_atrm_get_bios(struct amdgpu_device *adev) break; } - if (i == 0 || adev->bios[0] != 0x55 || adev->bios[1] != 0xaa) { + if (i == 0 || !AMD_IS_VALID_VBIOS(adev->bios)) { kfree(adev->bios); return false; } @@ -321,6 +359,9 @@ bool amdgpu_get_bios(struct amdgpu_device *adev) if (r == false) r = amdgpu_read_bios(adev); if (r == false) { + r = amdgpu_read_bios_from_rom(adev); + } + if (r == false) { r = amdgpu_read_disabled_bios(adev); } if (r == false) { @@ -331,7 +372,7 @@ bool amdgpu_get_bios(struct amdgpu_device *adev) adev->bios = NULL; return false; } - if (adev->bios[0] != 0x55 || adev->bios[1] != 0xaa) { + if (!AMD_IS_VALID_VBIOS(adev->bios)) { printk("BIOS signature incorrect %x %x\n", adev->bios[0], adev->bios[1]); goto free_bios; } diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c index f82a2dd83874..eacd810fc09b 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c @@ -32,6 +32,9 @@ #include "amdgpu.h" #include "amdgpu_trace.h" +#define AMDGPU_BO_LIST_MAX_PRIORITY 32u +#define AMDGPU_BO_LIST_NUM_BUCKETS (AMDGPU_BO_LIST_MAX_PRIORITY + 1) + static int amdgpu_bo_list_create(struct amdgpu_fpriv *fpriv, struct amdgpu_bo_list **result, int *id) @@ -88,8 +91,9 @@ static int amdgpu_bo_list_set(struct amdgpu_device *adev, struct amdgpu_bo *gws_obj = adev->gds.gws_gfx_bo; struct amdgpu_bo *oa_obj = adev->gds.oa_gfx_bo; - bool has_userptr = false; + unsigned last_entry = 0, first_userptr = num_entries; unsigned i; + int r; array = drm_malloc_ab(num_entries, sizeof(struct amdgpu_bo_list_entry)); if (!array) @@ -97,33 +101,43 @@ static int amdgpu_bo_list_set(struct amdgpu_device *adev, memset(array, 0, num_entries * sizeof(struct amdgpu_bo_list_entry)); for (i = 0; i < num_entries; ++i) { - struct amdgpu_bo_list_entry *entry = &array[i]; + struct amdgpu_bo_list_entry *entry; struct drm_gem_object *gobj; + struct amdgpu_bo *bo; + struct mm_struct *usermm; gobj = drm_gem_object_lookup(adev->ddev, filp, info[i].bo_handle); - if (!gobj) + if (!gobj) { + r = -ENOENT; goto error_free; + } - entry->robj = amdgpu_bo_ref(gem_to_amdgpu_bo(gobj)); + bo = amdgpu_bo_ref(gem_to_amdgpu_bo(gobj)); drm_gem_object_unreference_unlocked(gobj); - entry->priority = info[i].bo_priority; - entry->prefered_domains = entry->robj->initial_domain; - entry->allowed_domains = entry->prefered_domains; - if (entry->allowed_domains == AMDGPU_GEM_DOMAIN_VRAM) - entry->allowed_domains |= AMDGPU_GEM_DOMAIN_GTT; - if (amdgpu_ttm_tt_has_userptr(entry->robj->tbo.ttm)) { - has_userptr = true; - entry->prefered_domains = AMDGPU_GEM_DOMAIN_GTT; - entry->allowed_domains = AMDGPU_GEM_DOMAIN_GTT; + + usermm = amdgpu_ttm_tt_get_usermm(bo->tbo.ttm); + if (usermm) { + if (usermm != current->mm) { + amdgpu_bo_unref(&bo); + r = -EPERM; + goto error_free; + } + entry = &array[--first_userptr]; + } else { + entry = &array[last_entry++]; } + + entry->robj = bo; + entry->priority = min(info[i].bo_priority, + AMDGPU_BO_LIST_MAX_PRIORITY); entry->tv.bo = &entry->robj->tbo; entry->tv.shared = true; - if (entry->prefered_domains == AMDGPU_GEM_DOMAIN_GDS) + if (entry->robj->prefered_domains == AMDGPU_GEM_DOMAIN_GDS) gds_obj = entry->robj; - if (entry->prefered_domains == AMDGPU_GEM_DOMAIN_GWS) + if (entry->robj->prefered_domains == AMDGPU_GEM_DOMAIN_GWS) gws_obj = entry->robj; - if (entry->prefered_domains == AMDGPU_GEM_DOMAIN_OA) + if (entry->robj->prefered_domains == AMDGPU_GEM_DOMAIN_OA) oa_obj = entry->robj; trace_amdgpu_bo_list_set(list, entry->robj); @@ -137,15 +151,17 @@ static int amdgpu_bo_list_set(struct amdgpu_device *adev, list->gds_obj = gds_obj; list->gws_obj = gws_obj; list->oa_obj = oa_obj; - list->has_userptr = has_userptr; + list->first_userptr = first_userptr; list->array = array; list->num_entries = num_entries; return 0; error_free: + while (i--) + amdgpu_bo_unref(&array[i].robj); drm_free_large(array); - return -ENOENT; + return r; } struct amdgpu_bo_list * @@ -161,6 +177,37 @@ amdgpu_bo_list_get(struct amdgpu_fpriv *fpriv, int id) return result; } +void amdgpu_bo_list_get_list(struct amdgpu_bo_list *list, + struct list_head *validated) +{ + /* This is based on the bucket sort with O(n) time complexity. + * An item with priority "i" is added to bucket[i]. The lists are then + * concatenated in descending order. + */ + struct list_head bucket[AMDGPU_BO_LIST_NUM_BUCKETS]; + unsigned i; + + for (i = 0; i < AMDGPU_BO_LIST_NUM_BUCKETS; i++) + INIT_LIST_HEAD(&bucket[i]); + + /* Since buffers which appear sooner in the relocation list are + * likely to be used more often than buffers which appear later + * in the list, the sort mustn't change the ordering of buffers + * with the same priority, i.e. it must be stable. + */ + for (i = 0; i < list->num_entries; i++) { + unsigned priority = list->array[i].priority; + + list_add_tail(&list->array[i].tv.head, + &bucket[priority]); + list->array[i].user_pages = NULL; + } + + /* Connect the sorted buckets in the output list. */ + for (i = 0; i < AMDGPU_BO_LIST_NUM_BUCKETS; i++) + list_splice(&bucket[i], validated); +} + void amdgpu_bo_list_put(struct amdgpu_bo_list *list) { mutex_unlock(&list->lock); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cgs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cgs.c index 8e995148f56e..7a4b101e10c6 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cgs.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cgs.c @@ -24,6 +24,7 @@ #include <linux/list.h> #include <linux/slab.h> #include <linux/pci.h> +#include <linux/acpi.h> #include <drm/drmP.h> #include <linux/firmware.h> #include <drm/amdgpu_drm.h> @@ -32,7 +33,6 @@ #include "atom.h" #include "amdgpu_ucode.h" - struct amdgpu_cgs_device { struct cgs_device base; struct amdgpu_device *adev; @@ -398,6 +398,41 @@ static void amdgpu_cgs_write_pci_config_dword(void *cgs_device, unsigned addr, WARN(ret, "pci_write_config_dword error"); } + +static int amdgpu_cgs_get_pci_resource(void *cgs_device, + enum cgs_resource_type resource_type, + uint64_t size, + uint64_t offset, + uint64_t *resource_base) +{ + CGS_FUNC_ADEV; + + if (resource_base == NULL) + return -EINVAL; + + switch (resource_type) { + case CGS_RESOURCE_TYPE_MMIO: + if (adev->rmmio_size == 0) + return -ENOENT; + if ((offset + size) > adev->rmmio_size) + return -EINVAL; + *resource_base = adev->rmmio_base; + return 0; + case CGS_RESOURCE_TYPE_DOORBELL: + if (adev->doorbell.size == 0) + return -ENOENT; + if ((offset + size) > adev->doorbell.size) + return -EINVAL; + *resource_base = adev->doorbell.base; + return 0; + case CGS_RESOURCE_TYPE_FB: + case CGS_RESOURCE_TYPE_IO: + case CGS_RESOURCE_TYPE_ROM: + default: + return -EINVAL; + } +} + static const void *amdgpu_cgs_atom_get_data_table(void *cgs_device, unsigned table, uint16_t *size, uint8_t *frev, uint8_t *crev) @@ -703,6 +738,9 @@ static int amdgpu_cgs_get_firmware_info(void *cgs_device, case CHIP_TONGA: strcpy(fw_name, "amdgpu/tonga_smc.bin"); break; + case CHIP_FIJI: + strcpy(fw_name, "amdgpu/fiji_smc.bin"); + break; default: DRM_ERROR("SMC firmware not supported\n"); return -EINVAL; @@ -736,6 +774,294 @@ static int amdgpu_cgs_get_firmware_info(void *cgs_device, return 0; } +static int amdgpu_cgs_query_system_info(void *cgs_device, + struct cgs_system_info *sys_info) +{ + CGS_FUNC_ADEV; + + if (NULL == sys_info) + return -ENODEV; + + if (sizeof(struct cgs_system_info) != sys_info->size) + return -ENODEV; + + switch (sys_info->info_id) { + case CGS_SYSTEM_INFO_ADAPTER_BDF_ID: + sys_info->value = adev->pdev->devfn | (adev->pdev->bus->number << 8); + break; + case CGS_SYSTEM_INFO_PCIE_GEN_INFO: + sys_info->value = adev->pm.pcie_gen_mask; + break; + case CGS_SYSTEM_INFO_PCIE_MLW: + sys_info->value = adev->pm.pcie_mlw_mask; + break; + case CGS_SYSTEM_INFO_CG_FLAGS: + sys_info->value = adev->cg_flags; + break; + case CGS_SYSTEM_INFO_PG_FLAGS: + sys_info->value = adev->pg_flags; + break; + default: + return -ENODEV; + } + + return 0; +} + +static int amdgpu_cgs_get_active_displays_info(void *cgs_device, + struct cgs_display_info *info) +{ + CGS_FUNC_ADEV; + struct amdgpu_crtc *amdgpu_crtc; + struct drm_device *ddev = adev->ddev; + struct drm_crtc *crtc; + uint32_t line_time_us, vblank_lines; + + if (info == NULL) + return -EINVAL; + + if (adev->mode_info.num_crtc && adev->mode_info.mode_config_initialized) { + list_for_each_entry(crtc, + &ddev->mode_config.crtc_list, head) { + amdgpu_crtc = to_amdgpu_crtc(crtc); + if (crtc->enabled) { + info->active_display_mask |= (1 << amdgpu_crtc->crtc_id); + info->display_count++; + } + if (info->mode_info != NULL && + crtc->enabled && amdgpu_crtc->enabled && + amdgpu_crtc->hw_mode.clock) { + line_time_us = (amdgpu_crtc->hw_mode.crtc_htotal * 1000) / + amdgpu_crtc->hw_mode.clock; + vblank_lines = amdgpu_crtc->hw_mode.crtc_vblank_end - + amdgpu_crtc->hw_mode.crtc_vdisplay + + (amdgpu_crtc->v_border * 2); + info->mode_info->vblank_time_us = vblank_lines * line_time_us; + info->mode_info->refresh_rate = drm_mode_vrefresh(&amdgpu_crtc->hw_mode); + info->mode_info->ref_clock = adev->clock.spll.reference_freq; + info->mode_info++; + } + } + } + + return 0; +} + +/** \brief evaluate acpi namespace object, handle or pathname must be valid + * \param cgs_device + * \param info input/output arguments for the control method + * \return status + */ + +#if defined(CONFIG_ACPI) +static int amdgpu_cgs_acpi_eval_object(void *cgs_device, + struct cgs_acpi_method_info *info) +{ + CGS_FUNC_ADEV; + acpi_handle handle; + struct acpi_object_list input; + struct acpi_buffer output = { ACPI_ALLOCATE_BUFFER, NULL }; + union acpi_object *params = NULL; + union acpi_object *obj = NULL; + uint8_t name[5] = {'\0'}; + struct cgs_acpi_method_argument *argument = NULL; + uint32_t i, count; + acpi_status status; + int result; + uint32_t func_no = 0xFFFFFFFF; + + handle = ACPI_HANDLE(&adev->pdev->dev); + if (!handle) + return -ENODEV; + + memset(&input, 0, sizeof(struct acpi_object_list)); + + /* validate input info */ + if (info->size != sizeof(struct cgs_acpi_method_info)) + return -EINVAL; + + input.count = info->input_count; + if (info->input_count > 0) { + if (info->pinput_argument == NULL) + return -EINVAL; + argument = info->pinput_argument; + func_no = argument->value; + for (i = 0; i < info->input_count; i++) { + if (((argument->type == ACPI_TYPE_STRING) || + (argument->type == ACPI_TYPE_BUFFER)) && + (argument->pointer == NULL)) + return -EINVAL; + argument++; + } + } + + if (info->output_count > 0) { + if (info->poutput_argument == NULL) + return -EINVAL; + argument = info->poutput_argument; + for (i = 0; i < info->output_count; i++) { + if (((argument->type == ACPI_TYPE_STRING) || + (argument->type == ACPI_TYPE_BUFFER)) + && (argument->pointer == NULL)) + return -EINVAL; + argument++; + } + } + + /* The path name passed to acpi_evaluate_object should be null terminated */ + if ((info->field & CGS_ACPI_FIELD_METHOD_NAME) != 0) { + strncpy(name, (char *)&(info->name), sizeof(uint32_t)); + name[4] = '\0'; + } + + /* parse input parameters */ + if (input.count > 0) { + input.pointer = params = + kzalloc(sizeof(union acpi_object) * input.count, GFP_KERNEL); + if (params == NULL) + return -EINVAL; + + argument = info->pinput_argument; + + for (i = 0; i < input.count; i++) { + params->type = argument->type; + switch (params->type) { + case ACPI_TYPE_INTEGER: + params->integer.value = argument->value; + break; + case ACPI_TYPE_STRING: + params->string.length = argument->method_length; + params->string.pointer = argument->pointer; + break; + case ACPI_TYPE_BUFFER: + params->buffer.length = argument->method_length; + params->buffer.pointer = argument->pointer; + break; + default: + break; + } + params++; + argument++; + } + } + + /* parse output info */ + count = info->output_count; + argument = info->poutput_argument; + + /* evaluate the acpi method */ + status = acpi_evaluate_object(handle, name, &input, &output); + + if (ACPI_FAILURE(status)) { + result = -EIO; + goto error; + } + + /* return the output info */ + obj = output.pointer; + + if (count > 1) { + if ((obj->type != ACPI_TYPE_PACKAGE) || + (obj->package.count != count)) { + result = -EIO; + goto error; + } + params = obj->package.elements; + } else + params = obj; + + if (params == NULL) { + result = -EIO; + goto error; + } + + for (i = 0; i < count; i++) { + if (argument->type != params->type) { + result = -EIO; + goto error; + } + switch (params->type) { + case ACPI_TYPE_INTEGER: + argument->value = params->integer.value; + break; + case ACPI_TYPE_STRING: + if ((params->string.length != argument->data_length) || + (params->string.pointer == NULL)) { + result = -EIO; + goto error; + } + strncpy(argument->pointer, + params->string.pointer, + params->string.length); + break; + case ACPI_TYPE_BUFFER: + if (params->buffer.pointer == NULL) { + result = -EIO; + goto error; + } + memcpy(argument->pointer, + params->buffer.pointer, + argument->data_length); + break; + default: + break; + } + argument++; + params++; + } + +error: + if (obj != NULL) + kfree(obj); + kfree((void *)input.pointer); + return result; +} +#else +static int amdgpu_cgs_acpi_eval_object(void *cgs_device, + struct cgs_acpi_method_info *info) +{ + return -EIO; +} +#endif + +int amdgpu_cgs_call_acpi_method(void *cgs_device, + uint32_t acpi_method, + uint32_t acpi_function, + void *pinput, void *poutput, + uint32_t output_count, + uint32_t input_size, + uint32_t output_size) +{ + struct cgs_acpi_method_argument acpi_input[2] = { {0}, {0} }; + struct cgs_acpi_method_argument acpi_output = {0}; + struct cgs_acpi_method_info info = {0}; + + acpi_input[0].type = CGS_ACPI_TYPE_INTEGER; + acpi_input[0].method_length = sizeof(uint32_t); + acpi_input[0].data_length = sizeof(uint32_t); + acpi_input[0].value = acpi_function; + + acpi_input[1].type = CGS_ACPI_TYPE_BUFFER; + acpi_input[1].method_length = CGS_ACPI_MAX_BUFFER_SIZE; + acpi_input[1].data_length = input_size; + acpi_input[1].pointer = pinput; + + acpi_output.type = CGS_ACPI_TYPE_BUFFER; + acpi_output.method_length = CGS_ACPI_MAX_BUFFER_SIZE; + acpi_output.data_length = output_size; + acpi_output.pointer = poutput; + + info.size = sizeof(struct cgs_acpi_method_info); + info.field = CGS_ACPI_FIELD_METHOD_NAME | CGS_ACPI_FIELD_INPUT_ARGUMENT_COUNT; + info.input_count = 2; + info.name = acpi_method; + info.pinput_argument = acpi_input; + info.output_count = output_count; + info.poutput_argument = &acpi_output; + + return amdgpu_cgs_acpi_eval_object(cgs_device, &info); +} + static const struct cgs_ops amdgpu_cgs_ops = { amdgpu_cgs_gpu_mem_info, amdgpu_cgs_gmap_kmem, @@ -756,6 +1082,7 @@ static const struct cgs_ops amdgpu_cgs_ops = { amdgpu_cgs_write_pci_config_byte, amdgpu_cgs_write_pci_config_word, amdgpu_cgs_write_pci_config_dword, + amdgpu_cgs_get_pci_resource, amdgpu_cgs_atom_get_data_table, amdgpu_cgs_atom_get_cmd_table_revs, amdgpu_cgs_atom_exec_cmd_table, @@ -768,7 +1095,10 @@ static const struct cgs_ops amdgpu_cgs_ops = { amdgpu_cgs_set_camera_voltages, amdgpu_cgs_get_firmware_info, amdgpu_cgs_set_powergating_state, - amdgpu_cgs_set_clockgating_state + amdgpu_cgs_set_clockgating_state, + amdgpu_cgs_get_active_displays_info, + amdgpu_cgs_call_acpi_method, + amdgpu_cgs_query_system_info, }; static const struct cgs_os_ops amdgpu_cgs_os_ops = { diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_connectors.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_connectors.c index 89c3dd62ba21..119cdc2c43e7 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_connectors.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_connectors.c @@ -77,7 +77,7 @@ void amdgpu_connector_hotplug(struct drm_connector *connector) } else if (amdgpu_atombios_dp_needs_link_train(amdgpu_connector)) { /* Don't try to start link training before we * have the dpcd */ - if (!amdgpu_atombios_dp_get_dpcd(amdgpu_connector)) + if (amdgpu_atombios_dp_get_dpcd(amdgpu_connector)) return; /* set it to OFF so that drm_helper_connector_dpms() diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c index fd16652aa277..9392e50a7ba4 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c @@ -25,52 +25,12 @@ * Jerome Glisse <glisse@freedesktop.org> */ #include <linux/list_sort.h> +#include <linux/pagemap.h> #include <drm/drmP.h> #include <drm/amdgpu_drm.h> #include "amdgpu.h" #include "amdgpu_trace.h" -#define AMDGPU_CS_MAX_PRIORITY 32u -#define AMDGPU_CS_NUM_BUCKETS (AMDGPU_CS_MAX_PRIORITY + 1) - -/* This is based on the bucket sort with O(n) time complexity. - * An item with priority "i" is added to bucket[i]. The lists are then - * concatenated in descending order. - */ -struct amdgpu_cs_buckets { - struct list_head bucket[AMDGPU_CS_NUM_BUCKETS]; -}; - -static void amdgpu_cs_buckets_init(struct amdgpu_cs_buckets *b) -{ - unsigned i; - - for (i = 0; i < AMDGPU_CS_NUM_BUCKETS; i++) - INIT_LIST_HEAD(&b->bucket[i]); -} - -static void amdgpu_cs_buckets_add(struct amdgpu_cs_buckets *b, - struct list_head *item, unsigned priority) -{ - /* Since buffers which appear sooner in the relocation list are - * likely to be used more often than buffers which appear later - * in the list, the sort mustn't change the ordering of buffers - * with the same priority, i.e. it must be stable. - */ - list_add_tail(item, &b->bucket[min(priority, AMDGPU_CS_MAX_PRIORITY)]); -} - -static void amdgpu_cs_buckets_get_list(struct amdgpu_cs_buckets *b, - struct list_head *out_list) -{ - unsigned i; - - /* Connect the sorted buckets in the output list. */ - for (i = 0; i < AMDGPU_CS_NUM_BUCKETS; i++) { - list_splice(&b->bucket[i], out_list); - } -} - int amdgpu_cs_get_ring(struct amdgpu_device *adev, u32 ip_type, u32 ip_instance, u32 ring, struct amdgpu_ring **out_ring) @@ -104,10 +64,11 @@ int amdgpu_cs_get_ring(struct amdgpu_device *adev, u32 ip_type, } break; case AMDGPU_HW_IP_DMA: - if (ring < 2) { - *out_ring = &adev->sdma[ring].ring; + if (ring < adev->sdma.num_instances) { + *out_ring = &adev->sdma.instance[ring].ring; } else { - DRM_ERROR("only two SDMA rings are supported\n"); + DRM_ERROR("only %d SDMA rings are supported\n", + adev->sdma.num_instances); return -EINVAL; } break; @@ -126,37 +87,45 @@ int amdgpu_cs_get_ring(struct amdgpu_device *adev, u32 ip_type, return 0; } -struct amdgpu_cs_parser *amdgpu_cs_parser_create(struct amdgpu_device *adev, - struct drm_file *filp, - struct amdgpu_ctx *ctx, - struct amdgpu_ib *ibs, - uint32_t num_ibs) +static int amdgpu_cs_user_fence_chunk(struct amdgpu_cs_parser *p, + struct amdgpu_user_fence *uf, + struct drm_amdgpu_cs_chunk_fence *fence_data) { - struct amdgpu_cs_parser *parser; - int i; + struct drm_gem_object *gobj; + uint32_t handle; - parser = kzalloc(sizeof(struct amdgpu_cs_parser), GFP_KERNEL); - if (!parser) - return NULL; + handle = fence_data->handle; + gobj = drm_gem_object_lookup(p->adev->ddev, p->filp, + fence_data->handle); + if (gobj == NULL) + return -EINVAL; + + uf->bo = amdgpu_bo_ref(gem_to_amdgpu_bo(gobj)); + uf->offset = fence_data->offset; + + if (amdgpu_ttm_tt_get_usermm(uf->bo->tbo.ttm)) { + drm_gem_object_unreference_unlocked(gobj); + return -EINVAL; + } - parser->adev = adev; - parser->filp = filp; - parser->ctx = ctx; - parser->ibs = ibs; - parser->num_ibs = num_ibs; - for (i = 0; i < num_ibs; i++) - ibs[i].ctx = ctx; + p->uf_entry.robj = amdgpu_bo_ref(uf->bo); + p->uf_entry.priority = 0; + p->uf_entry.tv.bo = &p->uf_entry.robj->tbo; + p->uf_entry.tv.shared = true; + p->uf_entry.user_pages = NULL; - return parser; + drm_gem_object_unreference_unlocked(gobj); + return 0; } int amdgpu_cs_parser_init(struct amdgpu_cs_parser *p, void *data) { + struct amdgpu_fpriv *fpriv = p->filp->driver_priv; union drm_amdgpu_cs *cs = data; uint64_t *chunk_array_user; uint64_t *chunk_array; - struct amdgpu_fpriv *fpriv = p->filp->driver_priv; - unsigned size; + struct amdgpu_user_fence uf = {}; + unsigned size, num_ibs = 0; int i; int ret; @@ -173,15 +142,12 @@ int amdgpu_cs_parser_init(struct amdgpu_cs_parser *p, void *data) goto free_chunk; } - p->bo_list = amdgpu_bo_list_get(fpriv, cs->in.bo_list_handle); - /* get chunks */ - INIT_LIST_HEAD(&p->validated); chunk_array_user = (uint64_t __user *)(unsigned long)(cs->in.chunks); if (copy_from_user(chunk_array, chunk_array_user, sizeof(uint64_t)*cs->in.num_chunks)) { ret = -EFAULT; - goto put_bo_list; + goto put_ctx; } p->nchunks = cs->in.num_chunks; @@ -189,7 +155,7 @@ int amdgpu_cs_parser_init(struct amdgpu_cs_parser *p, void *data) GFP_KERNEL); if (!p->chunks) { ret = -ENOMEM; - goto put_bo_list; + goto put_ctx; } for (i = 0; i < p->nchunks; i++) { @@ -209,7 +175,6 @@ int amdgpu_cs_parser_init(struct amdgpu_cs_parser *p, void *data) size = p->chunks[i].length_dw; cdata = (void __user *)(unsigned long)user_chunk.chunk_data; - p->chunks[i].user_ptr = cdata; p->chunks[i].kdata = drm_malloc_ab(size, sizeof(uint32_t)); if (p->chunks[i].kdata == NULL) { @@ -225,31 +190,20 @@ int amdgpu_cs_parser_init(struct amdgpu_cs_parser *p, void *data) switch (p->chunks[i].chunk_id) { case AMDGPU_CHUNK_ID_IB: - p->num_ibs++; + ++num_ibs; break; case AMDGPU_CHUNK_ID_FENCE: size = sizeof(struct drm_amdgpu_cs_chunk_fence); - if (p->chunks[i].length_dw * sizeof(uint32_t) >= size) { - uint32_t handle; - struct drm_gem_object *gobj; - struct drm_amdgpu_cs_chunk_fence *fence_data; - - fence_data = (void *)p->chunks[i].kdata; - handle = fence_data->handle; - gobj = drm_gem_object_lookup(p->adev->ddev, - p->filp, handle); - if (gobj == NULL) { - ret = -EINVAL; - goto free_partial_kdata; - } - - p->uf.bo = gem_to_amdgpu_bo(gobj); - p->uf.offset = fence_data->offset; - } else { + if (p->chunks[i].length_dw * sizeof(uint32_t) < size) { ret = -EINVAL; goto free_partial_kdata; } + + ret = amdgpu_cs_user_fence_chunk(p, &uf, (void *)p->chunks[i].kdata); + if (ret) + goto free_partial_kdata; + break; case AMDGPU_CHUNK_ID_DEPENDENCIES: @@ -261,12 +215,11 @@ int amdgpu_cs_parser_init(struct amdgpu_cs_parser *p, void *data) } } - - p->ibs = kcalloc(p->num_ibs, sizeof(struct amdgpu_ib), GFP_KERNEL); - if (!p->ibs) { - ret = -ENOMEM; + ret = amdgpu_job_alloc(p->adev, num_ibs, &p->job); + if (ret) goto free_all_kdata; - } + + p->job->uf = uf; kfree(chunk_array); return 0; @@ -277,9 +230,7 @@ free_partial_kdata: for (; i >= 0; i--) drm_free_large(p->chunks[i].kdata); kfree(p->chunks); -put_bo_list: - if (p->bo_list) - amdgpu_bo_list_put(p->bo_list); +put_ctx: amdgpu_ctx_put(p->ctx); free_chunk: kfree(chunk_array); @@ -339,101 +290,225 @@ static u64 amdgpu_cs_get_threshold_for_moves(struct amdgpu_device *adev) return max(bytes_moved_threshold, 1024*1024ull); } -int amdgpu_cs_list_validate(struct amdgpu_device *adev, - struct amdgpu_vm *vm, +int amdgpu_cs_list_validate(struct amdgpu_cs_parser *p, struct list_head *validated) { struct amdgpu_bo_list_entry *lobj; - struct amdgpu_bo *bo; - u64 bytes_moved = 0, initial_bytes_moved; - u64 bytes_moved_threshold = amdgpu_cs_get_threshold_for_moves(adev); + u64 initial_bytes_moved; int r; list_for_each_entry(lobj, validated, tv.head) { - bo = lobj->robj; - if (!bo->pin_count) { - u32 domain = lobj->prefered_domains; - u32 current_domain = - amdgpu_mem_type_to_domain(bo->tbo.mem.mem_type); - - /* Check if this buffer will be moved and don't move it - * if we have moved too many buffers for this IB already. - * - * Note that this allows moving at least one buffer of - * any size, because it doesn't take the current "bo" - * into account. We don't want to disallow buffer moves - * completely. - */ - if ((lobj->allowed_domains & current_domain) != 0 && - (domain & current_domain) == 0 && /* will be moved */ - bytes_moved > bytes_moved_threshold) { - /* don't move it */ - domain = current_domain; - } + struct amdgpu_bo *bo = lobj->robj; + bool binding_userptr = false; + struct mm_struct *usermm; + uint32_t domain; + + usermm = amdgpu_ttm_tt_get_usermm(bo->tbo.ttm); + if (usermm && usermm != current->mm) + return -EPERM; + + /* Check if we have user pages and nobody bound the BO already */ + if (lobj->user_pages && bo->tbo.ttm->state != tt_bound) { + size_t size = sizeof(struct page *); + + size *= bo->tbo.ttm->num_pages; + memcpy(bo->tbo.ttm->pages, lobj->user_pages, size); + binding_userptr = true; + } - retry: - amdgpu_ttm_placement_from_domain(bo, domain); - initial_bytes_moved = atomic64_read(&adev->num_bytes_moved); - r = ttm_bo_validate(&bo->tbo, &bo->placement, true, false); - bytes_moved += atomic64_read(&adev->num_bytes_moved) - - initial_bytes_moved; - - if (unlikely(r)) { - if (r != -ERESTARTSYS && domain != lobj->allowed_domains) { - domain = lobj->allowed_domains; - goto retry; - } - return r; + if (bo->pin_count) + continue; + + /* Avoid moving this one if we have moved too many buffers + * for this IB already. + * + * Note that this allows moving at least one buffer of + * any size, because it doesn't take the current "bo" + * into account. We don't want to disallow buffer moves + * completely. + */ + if (p->bytes_moved <= p->bytes_moved_threshold) + domain = bo->prefered_domains; + else + domain = bo->allowed_domains; + + retry: + amdgpu_ttm_placement_from_domain(bo, domain); + initial_bytes_moved = atomic64_read(&bo->adev->num_bytes_moved); + r = ttm_bo_validate(&bo->tbo, &bo->placement, true, false); + p->bytes_moved += atomic64_read(&bo->adev->num_bytes_moved) - + initial_bytes_moved; + + if (unlikely(r)) { + if (r != -ERESTARTSYS && domain != bo->allowed_domains) { + domain = bo->allowed_domains; + goto retry; } + return r; + } + + if (binding_userptr) { + drm_free_large(lobj->user_pages); + lobj->user_pages = NULL; } - lobj->bo_va = amdgpu_vm_bo_find(vm, bo); } return 0; } -static int amdgpu_cs_parser_relocs(struct amdgpu_cs_parser *p) +static int amdgpu_cs_parser_bos(struct amdgpu_cs_parser *p, + union drm_amdgpu_cs *cs) { struct amdgpu_fpriv *fpriv = p->filp->driver_priv; - struct amdgpu_cs_buckets buckets; + struct amdgpu_bo_list_entry *e; struct list_head duplicates; bool need_mmap_lock = false; - int i, r; + unsigned i, tries = 10; + int r; - if (p->bo_list) { - need_mmap_lock = p->bo_list->has_userptr; - amdgpu_cs_buckets_init(&buckets); - for (i = 0; i < p->bo_list->num_entries; i++) - amdgpu_cs_buckets_add(&buckets, &p->bo_list->array[i].tv.head, - p->bo_list->array[i].priority); + INIT_LIST_HEAD(&p->validated); - amdgpu_cs_buckets_get_list(&buckets, &p->validated); + p->bo_list = amdgpu_bo_list_get(fpriv, cs->in.bo_list_handle); + if (p->bo_list) { + need_mmap_lock = p->bo_list->first_userptr != + p->bo_list->num_entries; + amdgpu_bo_list_get_list(p->bo_list, &p->validated); } - p->vm_bos = amdgpu_vm_get_bos(p->adev, &fpriv->vm, - &p->validated); + INIT_LIST_HEAD(&duplicates); + amdgpu_vm_get_pd_bo(&fpriv->vm, &p->validated, &p->vm_pd); + + if (p->job->uf.bo) + list_add(&p->uf_entry.tv.head, &p->validated); if (need_mmap_lock) down_read(¤t->mm->mmap_sem); - INIT_LIST_HEAD(&duplicates); - r = ttm_eu_reserve_buffers(&p->ticket, &p->validated, true, &duplicates); - if (unlikely(r != 0)) - goto error_reserve; + while (1) { + struct list_head need_pages; + unsigned i; + + r = ttm_eu_reserve_buffers(&p->ticket, &p->validated, true, + &duplicates); + if (unlikely(r != 0)) + goto error_free_pages; + + /* Without a BO list we don't have userptr BOs */ + if (!p->bo_list) + break; + + INIT_LIST_HEAD(&need_pages); + for (i = p->bo_list->first_userptr; + i < p->bo_list->num_entries; ++i) { + + e = &p->bo_list->array[i]; + + if (amdgpu_ttm_tt_userptr_invalidated(e->robj->tbo.ttm, + &e->user_invalidated) && e->user_pages) { + + /* We acquired a page array, but somebody + * invalidated it. Free it an try again + */ + release_pages(e->user_pages, + e->robj->tbo.ttm->num_pages, + false); + drm_free_large(e->user_pages); + e->user_pages = NULL; + } + + if (e->robj->tbo.ttm->state != tt_bound && + !e->user_pages) { + list_del(&e->tv.head); + list_add(&e->tv.head, &need_pages); + + amdgpu_bo_unreserve(e->robj); + } + } + + if (list_empty(&need_pages)) + break; - r = amdgpu_cs_list_validate(p->adev, &fpriv->vm, &p->validated); + /* Unreserve everything again. */ + ttm_eu_backoff_reservation(&p->ticket, &p->validated); + + /* We tried to often, just abort */ + if (!--tries) { + r = -EDEADLK; + goto error_free_pages; + } + + /* Fill the page arrays for all useptrs. */ + list_for_each_entry(e, &need_pages, tv.head) { + struct ttm_tt *ttm = e->robj->tbo.ttm; + + e->user_pages = drm_calloc_large(ttm->num_pages, + sizeof(struct page*)); + if (!e->user_pages) { + r = -ENOMEM; + goto error_free_pages; + } + + r = amdgpu_ttm_tt_get_user_pages(ttm, e->user_pages); + if (r) { + drm_free_large(e->user_pages); + e->user_pages = NULL; + goto error_free_pages; + } + } + + /* And try again. */ + list_splice(&need_pages, &p->validated); + } + + amdgpu_vm_get_pt_bos(&fpriv->vm, &duplicates); + + p->bytes_moved_threshold = amdgpu_cs_get_threshold_for_moves(p->adev); + p->bytes_moved = 0; + + r = amdgpu_cs_list_validate(p, &duplicates); + if (r) + goto error_validate; + + r = amdgpu_cs_list_validate(p, &p->validated); if (r) goto error_validate; - r = amdgpu_cs_list_validate(p->adev, &fpriv->vm, &duplicates); + if (p->bo_list) { + struct amdgpu_vm *vm = &fpriv->vm; + unsigned i; + + for (i = 0; i < p->bo_list->num_entries; i++) { + struct amdgpu_bo *bo = p->bo_list->array[i].robj; + + p->bo_list->array[i].bo_va = amdgpu_vm_bo_find(vm, bo); + } + } error_validate: - if (r) + if (r) { + amdgpu_vm_move_pt_bos_in_lru(p->adev, &fpriv->vm); ttm_eu_backoff_reservation(&p->ticket, &p->validated); + } + +error_free_pages: -error_reserve: if (need_mmap_lock) up_read(¤t->mm->mmap_sem); + if (p->bo_list) { + for (i = p->bo_list->first_userptr; + i < p->bo_list->num_entries; ++i) { + e = &p->bo_list->array[i]; + + if (!e->user_pages) + continue; + + release_pages(e->user_pages, + e->robj->tbo.ttm->num_pages, + false); + drm_free_large(e->user_pages); + } + } + return r; } @@ -444,7 +519,7 @@ static int amdgpu_cs_sync_rings(struct amdgpu_cs_parser *p) list_for_each_entry(e, &p->validated, tv.head) { struct reservation_object *resv = e->robj->tbo.resv; - r = amdgpu_sync_resv(p->adev, &p->ibs[0].sync, resv, p->filp); + r = amdgpu_sync_resv(p->adev, &p->job->sync, resv, p->filp); if (r) return r; @@ -462,9 +537,22 @@ static int cmp_size_smaller_first(void *priv, struct list_head *a, return (int)la->robj->tbo.num_pages - (int)lb->robj->tbo.num_pages; } -static void amdgpu_cs_parser_fini_early(struct amdgpu_cs_parser *parser, int error, bool backoff) +/** + * cs_parser_fini() - clean parser states + * @parser: parser structure holding parsing context. + * @error: error number + * + * If error is set than unvalidate buffer, otherwise just free memory + * used by parsing context. + **/ +static void amdgpu_cs_parser_fini(struct amdgpu_cs_parser *parser, int error, bool backoff) { + struct amdgpu_fpriv *fpriv = parser->filp->driver_priv; + unsigned i; + if (!error) { + amdgpu_vm_move_pt_bos_in_lru(parser->adev, &fpriv->vm); + /* Sort the buffer list from the smallest to largest buffer, * which affects the order of buffers in the LRU list. * This assures that the smallest buffers are added first @@ -478,51 +566,25 @@ static void amdgpu_cs_parser_fini_early(struct amdgpu_cs_parser *parser, int err list_sort(NULL, &parser->validated, cmp_size_smaller_first); ttm_eu_fence_buffer_objects(&parser->ticket, - &parser->validated, - &parser->ibs[parser->num_ibs-1].fence->base); + &parser->validated, + parser->fence); } else if (backoff) { ttm_eu_backoff_reservation(&parser->ticket, &parser->validated); } -} + fence_put(parser->fence); -static void amdgpu_cs_parser_fini_late(struct amdgpu_cs_parser *parser) -{ - unsigned i; if (parser->ctx) amdgpu_ctx_put(parser->ctx); if (parser->bo_list) amdgpu_bo_list_put(parser->bo_list); - drm_free_large(parser->vm_bos); for (i = 0; i < parser->nchunks; i++) drm_free_large(parser->chunks[i].kdata); kfree(parser->chunks); - if (!amdgpu_enable_scheduler) - { - if (parser->ibs) - for (i = 0; i < parser->num_ibs; i++) - amdgpu_ib_free(parser->adev, &parser->ibs[i]); - kfree(parser->ibs); - if (parser->uf.bo) - drm_gem_object_unreference_unlocked(&parser->uf.bo->gem_base); - } - - kfree(parser); -} - -/** - * cs_parser_fini() - clean parser states - * @parser: parser structure holding parsing context. - * @error: error number - * - * If error is set than unvalidate buffer, otherwise just free memory - * used by parsing context. - **/ -static void amdgpu_cs_parser_fini(struct amdgpu_cs_parser *parser, int error, bool backoff) -{ - amdgpu_cs_parser_fini_early(parser, error, backoff); - amdgpu_cs_parser_fini_late(parser); + if (parser->job) + amdgpu_job_free(parser->job); + amdgpu_bo_unref(&parser->uf_entry.robj); } static int amdgpu_bo_vm_update_pte(struct amdgpu_cs_parser *p, @@ -537,7 +599,7 @@ static int amdgpu_bo_vm_update_pte(struct amdgpu_cs_parser *p, if (r) return r; - r = amdgpu_sync_fence(adev, &p->ibs[0].sync, vm->page_directory_fence); + r = amdgpu_sync_fence(adev, &p->job->sync, vm->page_directory_fence); if (r) return r; @@ -563,48 +625,51 @@ static int amdgpu_bo_vm_update_pte(struct amdgpu_cs_parser *p, return r; f = bo_va->last_pt_update; - r = amdgpu_sync_fence(adev, &p->ibs[0].sync, f); + r = amdgpu_sync_fence(adev, &p->job->sync, f); if (r) return r; } + + } + + r = amdgpu_vm_clear_invalids(adev, vm, &p->job->sync); + + if (amdgpu_vm_debug && p->bo_list) { + /* Invalidate all BOs to test for userspace bugs */ + for (i = 0; i < p->bo_list->num_entries; i++) { + /* ignore duplicates */ + bo = p->bo_list->array[i].robj; + if (!bo) + continue; + + amdgpu_vm_bo_invalidate(adev, bo); + } } - return amdgpu_vm_clear_invalids(adev, vm, &p->ibs[0].sync); + return r; } static int amdgpu_cs_ib_vm_chunk(struct amdgpu_device *adev, - struct amdgpu_cs_parser *parser) + struct amdgpu_cs_parser *p) { - struct amdgpu_fpriv *fpriv = parser->filp->driver_priv; + struct amdgpu_fpriv *fpriv = p->filp->driver_priv; struct amdgpu_vm *vm = &fpriv->vm; - struct amdgpu_ring *ring; + struct amdgpu_ring *ring = p->job->ring; int i, r; - if (parser->num_ibs == 0) - return 0; - /* Only for UVD/VCE VM emulation */ - for (i = 0; i < parser->num_ibs; i++) { - ring = parser->ibs[i].ring; - if (ring->funcs->parse_cs) { - r = amdgpu_ring_parse_cs(ring, parser, i); + if (ring->funcs->parse_cs) { + for (i = 0; i < p->job->num_ibs; i++) { + r = amdgpu_ring_parse_cs(ring, p, i); if (r) return r; } } - mutex_lock(&vm->mutex); - r = amdgpu_bo_vm_update_pte(parser, vm); - if (r) { - goto out; - } - amdgpu_cs_sync_rings(parser); - if (!amdgpu_enable_scheduler) - r = amdgpu_ib_schedule(adev, parser->num_ibs, parser->ibs, - parser->filp); + r = amdgpu_bo_vm_update_pte(p, vm); + if (!r) + amdgpu_cs_sync_rings(p); -out: - mutex_unlock(&vm->mutex); return r; } @@ -626,14 +691,14 @@ static int amdgpu_cs_ib_fill(struct amdgpu_device *adev, int i, j; int r; - for (i = 0, j = 0; i < parser->nchunks && j < parser->num_ibs; i++) { + for (i = 0, j = 0; i < parser->nchunks && j < parser->job->num_ibs; i++) { struct amdgpu_cs_chunk *chunk; struct amdgpu_ib *ib; struct drm_amdgpu_cs_chunk_ib *chunk_ib; struct amdgpu_ring *ring; chunk = &parser->chunks[i]; - ib = &parser->ibs[j]; + ib = &parser->job->ibs[j]; chunk_ib = (struct drm_amdgpu_cs_chunk_ib *)chunk->kdata; if (chunk->chunk_id != AMDGPU_CHUNK_ID_IB) @@ -645,6 +710,11 @@ static int amdgpu_cs_ib_fill(struct amdgpu_device *adev, if (r) return r; + if (parser->job->ring && parser->job->ring != ring) + return -EINVAL; + + parser->job->ring = ring; + if (ring->funcs->parse_cs) { struct amdgpu_bo_va_mapping *m; struct amdgpu_bo *aobj = NULL; @@ -673,7 +743,7 @@ static int amdgpu_cs_ib_fill(struct amdgpu_device *adev, offset = ((uint64_t)m->it.start) * AMDGPU_GPU_PAGE_SIZE; kptr += chunk_ib->va_start - offset; - r = amdgpu_ib_get(ring, NULL, chunk_ib->ib_bytes, ib); + r = amdgpu_ib_get(adev, NULL, chunk_ib->ib_bytes, ib); if (r) { DRM_ERROR("Failed to get ib !\n"); return r; @@ -682,7 +752,7 @@ static int amdgpu_cs_ib_fill(struct amdgpu_device *adev, memcpy(ib->ptr, kptr, chunk_ib->ib_bytes); amdgpu_bo_kunmap(aobj); } else { - r = amdgpu_ib_get(ring, vm, 0, ib); + r = amdgpu_ib_get(adev, vm, 0, ib); if (r) { DRM_ERROR("Failed to get ib !\n"); return r; @@ -697,15 +767,12 @@ static int amdgpu_cs_ib_fill(struct amdgpu_device *adev, j++; } - if (!parser->num_ibs) - return 0; - /* add GDS resources to first IB */ if (parser->bo_list) { struct amdgpu_bo *gds = parser->bo_list->gds_obj; struct amdgpu_bo *gws = parser->bo_list->gws_obj; struct amdgpu_bo *oa = parser->bo_list->oa_obj; - struct amdgpu_ib *ib = &parser->ibs[0]; + struct amdgpu_ib *ib = &parser->job->ibs[0]; if (gds) { ib->gds_base = amdgpu_bo_gpu_offset(gds); @@ -721,15 +788,15 @@ static int amdgpu_cs_ib_fill(struct amdgpu_device *adev, } } /* wrap the last IB with user fence */ - if (parser->uf.bo) { - struct amdgpu_ib *ib = &parser->ibs[parser->num_ibs - 1]; + if (parser->job->uf.bo) { + struct amdgpu_ib *ib = &parser->job->ibs[parser->job->num_ibs - 1]; /* UVD & VCE fw doesn't support user fences */ - if (ib->ring->type == AMDGPU_RING_TYPE_UVD || - ib->ring->type == AMDGPU_RING_TYPE_VCE) + if (parser->job->ring->type == AMDGPU_RING_TYPE_UVD || + parser->job->ring->type == AMDGPU_RING_TYPE_VCE) return -EINVAL; - ib->user = &parser->uf; + ib->user = &parser->job->uf; } return 0; @@ -739,14 +806,8 @@ static int amdgpu_cs_dependencies(struct amdgpu_device *adev, struct amdgpu_cs_parser *p) { struct amdgpu_fpriv *fpriv = p->filp->driver_priv; - struct amdgpu_ib *ib; int i, j, r; - if (!p->num_ibs) - return 0; - - /* Add dependencies to first IB */ - ib = &p->ibs[0]; for (i = 0; i < p->nchunks; ++i) { struct drm_amdgpu_cs_chunk_dep *deps; struct amdgpu_cs_chunk *chunk; @@ -784,7 +845,8 @@ static int amdgpu_cs_dependencies(struct amdgpu_device *adev, return r; } else if (fence) { - r = amdgpu_sync_fence(adev, &ib->sync, fence); + r = amdgpu_sync_fence(adev, &p->job->sync, + fence); fence_put(fence); amdgpu_ctx_put(ctx); if (r) @@ -796,15 +858,36 @@ static int amdgpu_cs_dependencies(struct amdgpu_device *adev, return 0; } -static int amdgpu_cs_free_job(struct amdgpu_job *job) +static int amdgpu_cs_submit(struct amdgpu_cs_parser *p, + union drm_amdgpu_cs *cs) { - int i; - if (job->ibs) - for (i = 0; i < job->num_ibs; i++) - amdgpu_ib_free(job->adev, &job->ibs[i]); - kfree(job->ibs); - if (job->uf.bo) - drm_gem_object_unreference_unlocked(&job->uf.bo->gem_base); + struct amdgpu_ring *ring = p->job->ring; + struct amd_sched_fence *fence; + struct amdgpu_job *job; + + job = p->job; + p->job = NULL; + + job->base.sched = &ring->sched; + job->base.s_entity = &p->ctx->rings[ring->idx].entity; + job->owner = p->filp; + + fence = amd_sched_fence_create(job->base.s_entity, p->filp); + if (!fence) { + amdgpu_job_free(job); + return -ENOMEM; + } + + job->base.s_fence = fence; + p->fence = fence_get(&fence->base); + + cs->out.handle = amdgpu_ctx_add_fence(p->ctx, ring, + &fence->base); + job->ibs[job->num_ibs - 1].sequence = cs->out.handle; + + trace_amdgpu_cs_ioctl(job); + amd_sched_entity_push_job(&job->base); + return 0; } @@ -812,40 +895,35 @@ int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp) { struct amdgpu_device *adev = dev->dev_private; union drm_amdgpu_cs *cs = data; - struct amdgpu_cs_parser *parser; + struct amdgpu_cs_parser parser = {}; bool reserved_buffers = false; int i, r; - down_read(&adev->exclusive_lock); - if (!adev->accel_working) { - up_read(&adev->exclusive_lock); + if (!adev->accel_working) return -EBUSY; - } - parser = amdgpu_cs_parser_create(adev, filp, NULL, NULL, 0); - if (!parser) - return -ENOMEM; - r = amdgpu_cs_parser_init(parser, data); + parser.adev = adev; + parser.filp = filp; + + r = amdgpu_cs_parser_init(&parser, data); if (r) { DRM_ERROR("Failed to initialize parser !\n"); - kfree(parser); - up_read(&adev->exclusive_lock); + amdgpu_cs_parser_fini(&parser, r, false); r = amdgpu_cs_handle_lockup(adev, r); return r; } - - r = amdgpu_cs_parser_relocs(parser); + r = amdgpu_cs_parser_bos(&parser, data); if (r == -ENOMEM) DRM_ERROR("Not enough memory for command submission!\n"); else if (r && r != -ERESTARTSYS) DRM_ERROR("Failed to process the buffer list %d!\n", r); else if (!r) { reserved_buffers = true; - r = amdgpu_cs_ib_fill(adev, parser); + r = amdgpu_cs_ib_fill(adev, &parser); } if (!r) { - r = amdgpu_cs_dependencies(adev, parser); + r = amdgpu_cs_dependencies(adev, &parser); if (r) DRM_ERROR("Failed in the dependencies handling %d!\n", r); } @@ -853,61 +931,17 @@ int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp) if (r) goto out; - for (i = 0; i < parser->num_ibs; i++) - trace_amdgpu_cs(parser, i); + for (i = 0; i < parser.job->num_ibs; i++) + trace_amdgpu_cs(&parser, i); - r = amdgpu_cs_ib_vm_chunk(adev, parser); + r = amdgpu_cs_ib_vm_chunk(adev, &parser); if (r) goto out; - if (amdgpu_enable_scheduler && parser->num_ibs) { - struct amdgpu_job *job; - struct amdgpu_ring * ring = parser->ibs->ring; - job = kzalloc(sizeof(struct amdgpu_job), GFP_KERNEL); - if (!job) - return -ENOMEM; - job->base.sched = &ring->sched; - job->base.s_entity = &parser->ctx->rings[ring->idx].entity; - job->adev = parser->adev; - job->ibs = parser->ibs; - job->num_ibs = parser->num_ibs; - job->base.owner = parser->filp; - mutex_init(&job->job_lock); - if (job->ibs[job->num_ibs - 1].user) { - memcpy(&job->uf, &parser->uf, - sizeof(struct amdgpu_user_fence)); - job->ibs[job->num_ibs - 1].user = &job->uf; - } - - job->free_job = amdgpu_cs_free_job; - mutex_lock(&job->job_lock); - r = amd_sched_entity_push_job(&job->base); - if (r) { - mutex_unlock(&job->job_lock); - amdgpu_cs_free_job(job); - kfree(job); - goto out; - } - cs->out.handle = - amdgpu_ctx_add_fence(parser->ctx, ring, - &job->base.s_fence->base); - parser->ibs[parser->num_ibs - 1].sequence = cs->out.handle; - - list_sort(NULL, &parser->validated, cmp_size_smaller_first); - ttm_eu_fence_buffer_objects(&parser->ticket, - &parser->validated, - &job->base.s_fence->base); - - mutex_unlock(&job->job_lock); - amdgpu_cs_parser_fini_late(parser); - up_read(&adev->exclusive_lock); - return 0; - } + r = amdgpu_cs_submit(&parser, cs); - cs->out.handle = parser->ibs[parser->num_ibs - 1].sequence; out: - amdgpu_cs_parser_fini(parser, r, reserved_buffers); - up_read(&adev->exclusive_lock); + amdgpu_cs_parser_fini(&parser, r, reserved_buffers); r = amdgpu_cs_handle_lockup(adev, r); return r; } @@ -975,30 +1009,36 @@ struct amdgpu_bo_va_mapping * amdgpu_cs_find_mapping(struct amdgpu_cs_parser *parser, uint64_t addr, struct amdgpu_bo **bo) { - struct amdgpu_bo_list_entry *reloc; struct amdgpu_bo_va_mapping *mapping; + unsigned i; + + if (!parser->bo_list) + return NULL; addr /= AMDGPU_GPU_PAGE_SIZE; - list_for_each_entry(reloc, &parser->validated, tv.head) { - if (!reloc->bo_va) + for (i = 0; i < parser->bo_list->num_entries; i++) { + struct amdgpu_bo_list_entry *lobj; + + lobj = &parser->bo_list->array[i]; + if (!lobj->bo_va) continue; - list_for_each_entry(mapping, &reloc->bo_va->valids, list) { + list_for_each_entry(mapping, &lobj->bo_va->valids, list) { if (mapping->it.start > addr || addr > mapping->it.last) continue; - *bo = reloc->bo_va->bo; + *bo = lobj->bo_va->bo; return mapping; } - list_for_each_entry(mapping, &reloc->bo_va->invalids, list) { + list_for_each_entry(mapping, &lobj->bo_va->invalids, list) { if (mapping->it.start > addr || addr > mapping->it.last) continue; - *bo = reloc->bo_va->bo; + *bo = lobj->bo_va->bo; return mapping; } } diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c index e0b80ccdfe8a..17e13621fae9 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c @@ -25,8 +25,7 @@ #include <drm/drmP.h> #include "amdgpu.h" -int amdgpu_ctx_init(struct amdgpu_device *adev, bool kernel, - struct amdgpu_ctx *ctx) +static int amdgpu_ctx_init(struct amdgpu_device *adev, struct amdgpu_ctx *ctx) { unsigned i, j; int r; @@ -35,49 +34,53 @@ int amdgpu_ctx_init(struct amdgpu_device *adev, bool kernel, ctx->adev = adev; kref_init(&ctx->refcount); spin_lock_init(&ctx->ring_lock); - for (i = 0; i < AMDGPU_MAX_RINGS; ++i) + ctx->fences = kcalloc(amdgpu_sched_jobs * AMDGPU_MAX_RINGS, + sizeof(struct fence*), GFP_KERNEL); + if (!ctx->fences) + return -ENOMEM; + + for (i = 0; i < AMDGPU_MAX_RINGS; ++i) { ctx->rings[i].sequence = 1; + ctx->rings[i].fences = &ctx->fences[amdgpu_sched_jobs * i]; + } + /* create context entity for each ring */ + for (i = 0; i < adev->num_rings; i++) { + struct amdgpu_ring *ring = adev->rings[i]; + struct amd_sched_rq *rq; + + rq = &ring->sched.sched_rq[AMD_SCHED_PRIORITY_NORMAL]; + r = amd_sched_entity_init(&ring->sched, &ctx->rings[i].entity, + rq, amdgpu_sched_jobs); + if (r) + break; + } - if (amdgpu_enable_scheduler) { - /* create context entity for each ring */ - for (i = 0; i < adev->num_rings; i++) { - struct amd_sched_rq *rq; - if (kernel) - rq = &adev->rings[i]->sched.kernel_rq; - else - rq = &adev->rings[i]->sched.sched_rq; - r = amd_sched_entity_init(&adev->rings[i]->sched, - &ctx->rings[i].entity, - rq, amdgpu_sched_jobs); - if (r) - break; - } - - if (i < adev->num_rings) { - for (j = 0; j < i; j++) - amd_sched_entity_fini(&adev->rings[j]->sched, - &ctx->rings[j].entity); - kfree(ctx); - return r; - } + if (i < adev->num_rings) { + for (j = 0; j < i; j++) + amd_sched_entity_fini(&adev->rings[j]->sched, + &ctx->rings[j].entity); + kfree(ctx->fences); + return r; } return 0; } -void amdgpu_ctx_fini(struct amdgpu_ctx *ctx) +static void amdgpu_ctx_fini(struct amdgpu_ctx *ctx) { struct amdgpu_device *adev = ctx->adev; unsigned i, j; + if (!adev) + return; + for (i = 0; i < AMDGPU_MAX_RINGS; ++i) - for (j = 0; j < AMDGPU_CTX_MAX_CS_PENDING; ++j) + for (j = 0; j < amdgpu_sched_jobs; ++j) fence_put(ctx->rings[i].fences[j]); + kfree(ctx->fences); - if (amdgpu_enable_scheduler) { - for (i = 0; i < adev->num_rings; i++) - amd_sched_entity_fini(&adev->rings[i]->sched, - &ctx->rings[i].entity); - } + for (i = 0; i < adev->num_rings; i++) + amd_sched_entity_fini(&adev->rings[i]->sched, + &ctx->rings[i].entity); } static int amdgpu_ctx_alloc(struct amdgpu_device *adev, @@ -100,9 +103,13 @@ static int amdgpu_ctx_alloc(struct amdgpu_device *adev, return r; } *id = (uint32_t)r; - r = amdgpu_ctx_init(adev, false, ctx); + r = amdgpu_ctx_init(adev, ctx); + if (r) { + idr_remove(&mgr->ctx_handles, *id); + *id = 0; + kfree(ctx); + } mutex_unlock(&mgr->lock); - return r; } @@ -184,18 +191,18 @@ int amdgpu_ctx_ioctl(struct drm_device *dev, void *data, id = args->in.ctx_id; switch (args->in.op) { - case AMDGPU_CTX_OP_ALLOC_CTX: - r = amdgpu_ctx_alloc(adev, fpriv, &id); - args->out.alloc.ctx_id = id; - break; - case AMDGPU_CTX_OP_FREE_CTX: - r = amdgpu_ctx_free(fpriv, id); - break; - case AMDGPU_CTX_OP_QUERY_STATE: - r = amdgpu_ctx_query(adev, fpriv, id, &args->out); - break; - default: - return -EINVAL; + case AMDGPU_CTX_OP_ALLOC_CTX: + r = amdgpu_ctx_alloc(adev, fpriv, &id); + args->out.alloc.ctx_id = id; + break; + case AMDGPU_CTX_OP_FREE_CTX: + r = amdgpu_ctx_free(fpriv, id); + break; + case AMDGPU_CTX_OP_QUERY_STATE: + r = amdgpu_ctx_query(adev, fpriv, id, &args->out); + break; + default: + return -EINVAL; } return r; @@ -236,7 +243,7 @@ uint64_t amdgpu_ctx_add_fence(struct amdgpu_ctx *ctx, struct amdgpu_ring *ring, unsigned idx = 0; struct fence *other = NULL; - idx = seq % AMDGPU_CTX_MAX_CS_PENDING; + idx = seq & (amdgpu_sched_jobs - 1); other = cring->fences[idx]; if (other) { signed long r; @@ -271,12 +278,12 @@ struct fence *amdgpu_ctx_get_fence(struct amdgpu_ctx *ctx, } - if (seq + AMDGPU_CTX_MAX_CS_PENDING < cring->sequence) { + if (seq + amdgpu_sched_jobs < cring->sequence) { spin_unlock(&ctx->ring_lock); return NULL; } - fence = fence_get(cring->fences[seq % AMDGPU_CTX_MAX_CS_PENDING]); + fence = fence_get(cring->fences[seq & (amdgpu_sched_jobs - 1)]); spin_unlock(&ctx->ring_lock); return fence; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c index 6068d8207d10..612117478b57 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c @@ -38,6 +38,7 @@ #include "amdgpu_i2c.h" #include "atom.h" #include "amdgpu_atombios.h" +#include "amd_pcie.h" #ifdef CONFIG_DRM_AMDGPU_CIK #include "cik.h" #endif @@ -57,9 +58,16 @@ static const char *amdgpu_asic_name[] = { "TONGA", "FIJI", "CARRIZO", + "STONEY", "LAST", }; +#if defined(CONFIG_VGA_SWITCHEROO) +bool amdgpu_has_atpx_dgpu_power_cntl(void); +#else +static inline bool amdgpu_has_atpx_dgpu_power_cntl(void) { return false; } +#endif + bool amdgpu_device_is_px(struct drm_device *dev) { struct amdgpu_device *adev = dev->dev_private; @@ -634,31 +642,6 @@ bool amdgpu_card_posted(struct amdgpu_device *adev) } /** - * amdgpu_boot_test_post_card - check and possibly initialize the hw - * - * @adev: amdgpu_device pointer - * - * Check if the asic is initialized and if not, attempt to initialize - * it (all asics). - * Returns true if initialized or false if not. - */ -bool amdgpu_boot_test_post_card(struct amdgpu_device *adev) -{ - if (amdgpu_card_posted(adev)) - return true; - - if (adev->bios) { - DRM_INFO("GPU not posted. posting now...\n"); - if (adev->is_atom_bios) - amdgpu_atom_asic_init(adev->mode_info.atom_context); - return true; - } else { - dev_err(adev->dev, "Card not posted and no BIOS - ignoring\n"); - return false; - } -} - -/** * amdgpu_dummy_page_init - init dummy page used by the driver * * @adev: amdgpu_device pointer @@ -948,11 +931,14 @@ static bool amdgpu_check_pot_argument(int arg) */ static void amdgpu_check_arguments(struct amdgpu_device *adev) { - /* vramlimit must be a power of two */ - if (!amdgpu_check_pot_argument(amdgpu_vram_limit)) { - dev_warn(adev->dev, "vram limit (%d) must be a power of 2\n", - amdgpu_vram_limit); - amdgpu_vram_limit = 0; + if (amdgpu_sched_jobs < 4) { + dev_warn(adev->dev, "sched jobs (%d) must be at least 4\n", + amdgpu_sched_jobs); + amdgpu_sched_jobs = 4; + } else if (!amdgpu_check_pot_argument(amdgpu_sched_jobs)){ + dev_warn(adev->dev, "sched jobs (%d) must be a power of 2\n", + amdgpu_sched_jobs); + amdgpu_sched_jobs = roundup_pow_of_two(amdgpu_sched_jobs); } if (amdgpu_gart_size != -1) { @@ -1022,7 +1008,7 @@ static void amdgpu_check_arguments(struct amdgpu_device *adev) * amdgpu_switcheroo_set_state - set switcheroo state * * @pdev: pci dev pointer - * @state: vga switcheroo state + * @state: vga_switcheroo state * * Callback for the switcheroo driver. Suspends or resumes the * the asics before or after it is powered up using ACPI methods. @@ -1165,7 +1151,8 @@ static int amdgpu_early_init(struct amdgpu_device *adev) case CHIP_TONGA: case CHIP_FIJI: case CHIP_CARRIZO: - if (adev->asic_type == CHIP_CARRIZO) + case CHIP_STONEY: + if (adev->asic_type == CHIP_CARRIZO || adev->asic_type == CHIP_STONEY) adev->family = AMDGPU_FAMILY_CZ; else adev->family = AMDGPU_FAMILY_VI; @@ -1212,12 +1199,14 @@ static int amdgpu_early_init(struct amdgpu_device *adev) } else { if (adev->ip_blocks[i].funcs->early_init) { r = adev->ip_blocks[i].funcs->early_init((void *)adev); - if (r == -ENOENT) + if (r == -ENOENT) { adev->ip_block_status[i].valid = false; - else if (r) + } else if (r) { + DRM_ERROR("early_init %d failed %d\n", i, r); return r; - else + } else { adev->ip_block_status[i].valid = true; + } } else { adev->ip_block_status[i].valid = true; } @@ -1235,20 +1224,28 @@ static int amdgpu_init(struct amdgpu_device *adev) if (!adev->ip_block_status[i].valid) continue; r = adev->ip_blocks[i].funcs->sw_init((void *)adev); - if (r) + if (r) { + DRM_ERROR("sw_init %d failed %d\n", i, r); return r; + } adev->ip_block_status[i].sw = true; /* need to do gmc hw init early so we can allocate gpu mem */ if (adev->ip_blocks[i].type == AMD_IP_BLOCK_TYPE_GMC) { r = amdgpu_vram_scratch_init(adev); - if (r) + if (r) { + DRM_ERROR("amdgpu_vram_scratch_init failed %d\n", r); return r; + } r = adev->ip_blocks[i].funcs->hw_init((void *)adev); - if (r) + if (r) { + DRM_ERROR("hw_init %d failed %d\n", i, r); return r; + } r = amdgpu_wb_init(adev); - if (r) + if (r) { + DRM_ERROR("amdgpu_wb_init failed %d\n", r); return r; + } adev->ip_block_status[i].hw = true; } } @@ -1260,8 +1257,10 @@ static int amdgpu_init(struct amdgpu_device *adev) if (adev->ip_blocks[i].type == AMD_IP_BLOCK_TYPE_GMC) continue; r = adev->ip_blocks[i].funcs->hw_init((void *)adev); - if (r) + if (r) { + DRM_ERROR("hw_init %d failed %d\n", i, r); return r; + } adev->ip_block_status[i].hw = true; } @@ -1278,12 +1277,16 @@ static int amdgpu_late_init(struct amdgpu_device *adev) /* enable clockgating to save power */ r = adev->ip_blocks[i].funcs->set_clockgating_state((void *)adev, AMD_CG_STATE_GATE); - if (r) + if (r) { + DRM_ERROR("set_clockgating_state(gate) %d failed %d\n", i, r); return r; + } if (adev->ip_blocks[i].funcs->late_init) { r = adev->ip_blocks[i].funcs->late_init((void *)adev); - if (r) + if (r) { + DRM_ERROR("late_init %d failed %d\n", i, r); return r; + } } } @@ -1304,10 +1307,15 @@ static int amdgpu_fini(struct amdgpu_device *adev) /* ungate blocks before hw fini so that we can shutdown the blocks safely */ r = adev->ip_blocks[i].funcs->set_clockgating_state((void *)adev, AMD_CG_STATE_UNGATE); - if (r) + if (r) { + DRM_ERROR("set_clockgating_state(ungate) %d failed %d\n", i, r); return r; + } r = adev->ip_blocks[i].funcs->hw_fini((void *)adev); /* XXX handle errors */ + if (r) { + DRM_DEBUG("hw_fini %d failed %d\n", i, r); + } adev->ip_block_status[i].hw = false; } @@ -1316,6 +1324,9 @@ static int amdgpu_fini(struct amdgpu_device *adev) continue; r = adev->ip_blocks[i].funcs->sw_fini((void *)adev); /* XXX handle errors */ + if (r) { + DRM_DEBUG("sw_fini %d failed %d\n", i, r); + } adev->ip_block_status[i].sw = false; adev->ip_block_status[i].valid = false; } @@ -1333,9 +1344,15 @@ static int amdgpu_suspend(struct amdgpu_device *adev) /* ungate blocks so that suspend can properly shut them down */ r = adev->ip_blocks[i].funcs->set_clockgating_state((void *)adev, AMD_CG_STATE_UNGATE); + if (r) { + DRM_ERROR("set_clockgating_state(ungate) %d failed %d\n", i, r); + } /* XXX handle errors */ r = adev->ip_blocks[i].funcs->suspend(adev); /* XXX handle errors */ + if (r) { + DRM_ERROR("suspend %d failed %d\n", i, r); + } } return 0; @@ -1349,8 +1366,10 @@ static int amdgpu_resume(struct amdgpu_device *adev) if (!adev->ip_block_status[i].valid) continue; r = adev->ip_blocks[i].funcs->resume(adev); - if (r) + if (r) { + DRM_ERROR("resume %d failed %d\n", i, r); return r; + } } return 0; @@ -1390,7 +1409,7 @@ int amdgpu_device_init(struct amdgpu_device *adev, adev->mman.buffer_funcs = NULL; adev->mman.buffer_funcs_ring = NULL; adev->vm_manager.vm_pte_funcs = NULL; - adev->vm_manager.vm_pte_funcs_ring = NULL; + adev->vm_manager.vm_pte_num_rings = 0; adev->gart.gart_funcs = NULL; adev->fence_context = fence_context_alloc(AMDGPU_MAX_RINGS); @@ -1411,14 +1430,12 @@ int amdgpu_device_init(struct amdgpu_device *adev, /* mutex initialization are all done here so we * can recall function without having locking issues */ - mutex_init(&adev->ring_lock); + mutex_init(&adev->vm_manager.lock); atomic_set(&adev->irq.ih.lock, 0); - mutex_init(&adev->gem.mutex); mutex_init(&adev->pm.mutex); mutex_init(&adev->gfx.gpu_clock_mutex); mutex_init(&adev->srbm_mutex); mutex_init(&adev->grbm_idx_mutex); - init_rwsem(&adev->exclusive_lock); mutex_init(&adev->mn_lock); hash_init(adev->mn_hash); @@ -1468,7 +1485,7 @@ int amdgpu_device_init(struct amdgpu_device *adev, if (amdgpu_runtime_pm == 1) runtime = true; - if (amdgpu_device_is_px(ddev)) + if (amdgpu_device_is_px(ddev) && amdgpu_has_atpx_dgpu_power_cntl()) runtime = true; vga_switcheroo_register_client(adev->pdev, &amdgpu_switcheroo_ops, runtime); if (runtime) @@ -1483,11 +1500,18 @@ int amdgpu_device_init(struct amdgpu_device *adev, return -EINVAL; } r = amdgpu_atombios_init(adev); - if (r) + if (r) { + dev_err(adev->dev, "amdgpu_atombios_init failed\n"); return r; + } + + /* See if the asic supports SR-IOV */ + adev->virtualization.supports_sr_iov = + amdgpu_atombios_has_gpu_virtualization_table(adev); /* Post card if necessary */ - if (!amdgpu_card_posted(adev)) { + if (!amdgpu_card_posted(adev) || + adev->virtualization.supports_sr_iov) { if (!adev->bios) { dev_err(adev->dev, "Card not posted and no BIOS - ignoring\n"); return -EINVAL; @@ -1498,21 +1522,26 @@ int amdgpu_device_init(struct amdgpu_device *adev, /* Initialize clocks */ r = amdgpu_atombios_get_clock_info(adev); - if (r) + if (r) { + dev_err(adev->dev, "amdgpu_atombios_get_clock_info failed\n"); return r; + } /* init i2c buses */ amdgpu_atombios_i2c_init(adev); /* Fence driver */ r = amdgpu_fence_driver_init(adev); - if (r) + if (r) { + dev_err(adev->dev, "amdgpu_fence_driver_init failed\n"); return r; + } /* init the mode config */ drm_mode_config_init(adev->ddev); r = amdgpu_init(adev); if (r) { + dev_err(adev->dev, "amdgpu_init failed\n"); amdgpu_fini(adev); return r; } @@ -1527,11 +1556,6 @@ int amdgpu_device_init(struct amdgpu_device *adev, return r; } - r = amdgpu_ctx_init(adev, true, &adev->kernel_ctx); - if (r) { - dev_err(adev->dev, "failed to create kernel context (%d).\n", r); - return r; - } r = amdgpu_ib_ring_tests(adev); if (r) DRM_ERROR("ib ring test failed (%d).\n", r); @@ -1569,8 +1593,10 @@ int amdgpu_device_init(struct amdgpu_device *adev, * explicit gating rather than handling it automatically. */ r = amdgpu_late_init(adev); - if (r) + if (r) { + dev_err(adev->dev, "amdgpu_late_init failed\n"); return r; + } return 0; } @@ -1593,7 +1619,6 @@ void amdgpu_device_fini(struct amdgpu_device *adev) adev->shutdown = true; /* evict vram memory */ amdgpu_bo_evict_vram(adev); - amdgpu_ctx_fini(&adev->kernel_ctx); amdgpu_ib_pool_fini(adev); amdgpu_fence_driver_fini(adev); amdgpu_fbdev_fini(adev); @@ -1657,11 +1682,21 @@ int amdgpu_suspend_kms(struct drm_device *dev, bool suspend, bool fbcon) } drm_modeset_unlock_all(dev); - /* unpin the front buffers */ + /* unpin the front buffers and cursors */ list_for_each_entry(crtc, &dev->mode_config.crtc_list, head) { + struct amdgpu_crtc *amdgpu_crtc = to_amdgpu_crtc(crtc); struct amdgpu_framebuffer *rfb = to_amdgpu_framebuffer(crtc->primary->fb); struct amdgpu_bo *robj; + if (amdgpu_crtc->cursor_bo) { + struct amdgpu_bo *aobj = gem_to_amdgpu_bo(amdgpu_crtc->cursor_bo); + r = amdgpu_bo_reserve(aobj, false); + if (r == 0) { + amdgpu_bo_unpin(aobj); + amdgpu_bo_unreserve(aobj); + } + } + if (rfb == NULL || rfb->obj == NULL) { continue; } @@ -1713,6 +1748,7 @@ int amdgpu_resume_kms(struct drm_device *dev, bool resume, bool fbcon) { struct drm_connector *connector; struct amdgpu_device *adev = dev->dev_private; + struct drm_crtc *crtc; int r; if (dev->switch_power_state == DRM_SWITCH_POWER_OFF) @@ -1732,20 +1768,43 @@ int amdgpu_resume_kms(struct drm_device *dev, bool resume, bool fbcon) } /* post card */ - amdgpu_atom_asic_init(adev->mode_info.atom_context); + if (!amdgpu_card_posted(adev)) + amdgpu_atom_asic_init(adev->mode_info.atom_context); r = amdgpu_resume(adev); + if (r) + DRM_ERROR("amdgpu_resume failed (%d).\n", r); amdgpu_fence_driver_resume(adev); - r = amdgpu_ib_ring_tests(adev); - if (r) - DRM_ERROR("ib ring test failed (%d).\n", r); + if (resume) { + r = amdgpu_ib_ring_tests(adev); + if (r) + DRM_ERROR("ib ring test failed (%d).\n", r); + } r = amdgpu_late_init(adev); if (r) return r; + /* pin cursors */ + list_for_each_entry(crtc, &dev->mode_config.crtc_list, head) { + struct amdgpu_crtc *amdgpu_crtc = to_amdgpu_crtc(crtc); + + if (amdgpu_crtc->cursor_bo) { + struct amdgpu_bo *aobj = gem_to_amdgpu_bo(amdgpu_crtc->cursor_bo); + r = amdgpu_bo_reserve(aobj, false); + if (r == 0) { + r = amdgpu_bo_pin(aobj, + AMDGPU_GEM_DOMAIN_VRAM, + &amdgpu_crtc->cursor_addr); + if (r != 0) + DRM_ERROR("Failed to pin cursor BO (%d)\n", r); + amdgpu_bo_unreserve(aobj); + } + } + } + /* blat the mode back in */ if (fbcon) { drm_helper_resume_force_mode(dev); @@ -1758,6 +1817,7 @@ int amdgpu_resume_kms(struct drm_device *dev, bool resume, bool fbcon) } drm_kms_helper_poll_enable(dev); + drm_helper_hpd_irq_event(dev); if (fbcon) { amdgpu_fbdev_set_suspend(adev, 0); @@ -1785,14 +1845,6 @@ int amdgpu_gpu_reset(struct amdgpu_device *adev) int i, r; int resched; - down_write(&adev->exclusive_lock); - - if (!adev->needs_reset) { - up_write(&adev->exclusive_lock); - return 0; - } - - adev->needs_reset = false; atomic_inc(&adev->gpu_reset_counter); /* block TTM */ @@ -1815,6 +1867,9 @@ int amdgpu_gpu_reset(struct amdgpu_device *adev) retry: r = amdgpu_asic_reset(adev); + /* post card */ + amdgpu_atom_asic_init(adev->mode_info.atom_context); + if (!r) { dev_info(adev->dev, "GPU reset succeeded, trying to resume\n"); r = amdgpu_resume(adev); @@ -1856,10 +1911,103 @@ retry: dev_info(adev->dev, "GPU reset failed\n"); } - up_write(&adev->exclusive_lock); return r; } +#define AMDGPU_DEFAULT_PCIE_GEN_MASK 0x30007 /* gen: chipset 1/2, asic 1/2/3 */ +#define AMDGPU_DEFAULT_PCIE_MLW_MASK 0x2f0000 /* 1/2/4/8/16 lanes */ + +void amdgpu_get_pcie_info(struct amdgpu_device *adev) +{ + u32 mask; + int ret; + + if (amdgpu_pcie_gen_cap) + adev->pm.pcie_gen_mask = amdgpu_pcie_gen_cap; + + if (amdgpu_pcie_lane_cap) + adev->pm.pcie_mlw_mask = amdgpu_pcie_lane_cap; + + /* covers APUs as well */ + if (pci_is_root_bus(adev->pdev->bus)) { + if (adev->pm.pcie_gen_mask == 0) + adev->pm.pcie_gen_mask = AMDGPU_DEFAULT_PCIE_GEN_MASK; + if (adev->pm.pcie_mlw_mask == 0) + adev->pm.pcie_mlw_mask = AMDGPU_DEFAULT_PCIE_MLW_MASK; + return; + } + + if (adev->pm.pcie_gen_mask == 0) { + ret = drm_pcie_get_speed_cap_mask(adev->ddev, &mask); + if (!ret) { + adev->pm.pcie_gen_mask = (CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN1 | + CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN2 | + CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN3); + + if (mask & DRM_PCIE_SPEED_25) + adev->pm.pcie_gen_mask |= CAIL_PCIE_LINK_SPEED_SUPPORT_GEN1; + if (mask & DRM_PCIE_SPEED_50) + adev->pm.pcie_gen_mask |= CAIL_PCIE_LINK_SPEED_SUPPORT_GEN2; + if (mask & DRM_PCIE_SPEED_80) + adev->pm.pcie_gen_mask |= CAIL_PCIE_LINK_SPEED_SUPPORT_GEN3; + } else { + adev->pm.pcie_gen_mask = AMDGPU_DEFAULT_PCIE_GEN_MASK; + } + } + if (adev->pm.pcie_mlw_mask == 0) { + ret = drm_pcie_get_max_link_width(adev->ddev, &mask); + if (!ret) { + switch (mask) { + case 32: + adev->pm.pcie_mlw_mask = (CAIL_PCIE_LINK_WIDTH_SUPPORT_X32 | + CAIL_PCIE_LINK_WIDTH_SUPPORT_X16 | + CAIL_PCIE_LINK_WIDTH_SUPPORT_X12 | + CAIL_PCIE_LINK_WIDTH_SUPPORT_X8 | + CAIL_PCIE_LINK_WIDTH_SUPPORT_X4 | + CAIL_PCIE_LINK_WIDTH_SUPPORT_X2 | + CAIL_PCIE_LINK_WIDTH_SUPPORT_X1); + break; + case 16: + adev->pm.pcie_mlw_mask = (CAIL_PCIE_LINK_WIDTH_SUPPORT_X16 | + CAIL_PCIE_LINK_WIDTH_SUPPORT_X12 | + CAIL_PCIE_LINK_WIDTH_SUPPORT_X8 | + CAIL_PCIE_LINK_WIDTH_SUPPORT_X4 | + CAIL_PCIE_LINK_WIDTH_SUPPORT_X2 | + CAIL_PCIE_LINK_WIDTH_SUPPORT_X1); + break; + case 12: + adev->pm.pcie_mlw_mask = (CAIL_PCIE_LINK_WIDTH_SUPPORT_X12 | + CAIL_PCIE_LINK_WIDTH_SUPPORT_X8 | + CAIL_PCIE_LINK_WIDTH_SUPPORT_X4 | + CAIL_PCIE_LINK_WIDTH_SUPPORT_X2 | + CAIL_PCIE_LINK_WIDTH_SUPPORT_X1); + break; + case 8: + adev->pm.pcie_mlw_mask = (CAIL_PCIE_LINK_WIDTH_SUPPORT_X8 | + CAIL_PCIE_LINK_WIDTH_SUPPORT_X4 | + CAIL_PCIE_LINK_WIDTH_SUPPORT_X2 | + CAIL_PCIE_LINK_WIDTH_SUPPORT_X1); + break; + case 4: + adev->pm.pcie_mlw_mask = (CAIL_PCIE_LINK_WIDTH_SUPPORT_X4 | + CAIL_PCIE_LINK_WIDTH_SUPPORT_X2 | + CAIL_PCIE_LINK_WIDTH_SUPPORT_X1); + break; + case 2: + adev->pm.pcie_mlw_mask = (CAIL_PCIE_LINK_WIDTH_SUPPORT_X2 | + CAIL_PCIE_LINK_WIDTH_SUPPORT_X1); + break; + case 1: + adev->pm.pcie_mlw_mask = CAIL_PCIE_LINK_WIDTH_SUPPORT_X1; + break; + default: + break; + } + } else { + adev->pm.pcie_mlw_mask = AMDGPU_DEFAULT_PCIE_MLW_MASK; + } + } +} /* * Debugfs diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_display.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_display.c index 6c9e0902a414..f0ed974bd4e0 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_display.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_display.c @@ -35,35 +35,30 @@ #include <drm/drm_crtc_helper.h> #include <drm/drm_edid.h> -static void amdgpu_flip_wait_fence(struct amdgpu_device *adev, - struct fence **f) +static void amdgpu_flip_callback(struct fence *f, struct fence_cb *cb) { - struct amdgpu_fence *fence; - long r; + struct amdgpu_flip_work *work = + container_of(cb, struct amdgpu_flip_work, cb); - if (*f == NULL) - return; + fence_put(f); + schedule_work(&work->flip_work); +} - fence = to_amdgpu_fence(*f); - if (fence) { - r = fence_wait(&fence->base, false); - if (r == -EDEADLK) { - up_read(&adev->exclusive_lock); - r = amdgpu_gpu_reset(adev); - down_read(&adev->exclusive_lock); - } - } else - r = fence_wait(*f, false); +static bool amdgpu_flip_handle_fence(struct amdgpu_flip_work *work, + struct fence **f) +{ + struct fence *fence= *f; - if (r) - DRM_ERROR("failed to wait on page flip fence (%ld)!\n", r); + if (fence == NULL) + return false; - /* We continue with the page flip even if we failed to wait on - * the fence, otherwise the DRM core and userspace will be - * confused about which BO the CRTC is scanning out - */ - fence_put(*f); *f = NULL; + + if (!fence_add_callback(fence, &work->cb, amdgpu_flip_callback)) + return true; + + fence_put(*f); + return false; } static void amdgpu_flip_work_func(struct work_struct *__work) @@ -75,23 +70,73 @@ static void amdgpu_flip_work_func(struct work_struct *__work) struct drm_crtc *crtc = &amdgpuCrtc->base; unsigned long flags; - unsigned i; + unsigned i, repcnt = 4; + int vpos, hpos, stat, min_udelay = 0; + struct drm_vblank_crtc *vblank = &crtc->dev->vblank[work->crtc_id]; + + if (amdgpu_flip_handle_fence(work, &work->excl)) + return; - down_read(&adev->exclusive_lock); - amdgpu_flip_wait_fence(adev, &work->excl); for (i = 0; i < work->shared_count; ++i) - amdgpu_flip_wait_fence(adev, &work->shared[i]); + if (amdgpu_flip_handle_fence(work, &work->shared[i])) + return; /* We borrow the event spin lock for protecting flip_status */ spin_lock_irqsave(&crtc->dev->event_lock, flags); - /* do the flip (mmio) */ - adev->mode_info.funcs->page_flip(adev, work->crtc_id, work->base); + /* If this happens to execute within the "virtually extended" vblank + * interval before the start of the real vblank interval then it needs + * to delay programming the mmio flip until the real vblank is entered. + * This prevents completing a flip too early due to the way we fudge + * our vblank counter and vblank timestamps in order to work around the + * problem that the hw fires vblank interrupts before actual start of + * vblank (when line buffer refilling is done for a frame). It + * complements the fudging logic in amdgpu_get_crtc_scanoutpos() for + * timestamping and amdgpu_get_vblank_counter_kms() for vblank counts. + * + * In practice this won't execute very often unless on very fast + * machines because the time window for this to happen is very small. + */ + while (amdgpuCrtc->enabled && --repcnt) { + /* GET_DISTANCE_TO_VBLANKSTART returns distance to real vblank + * start in hpos, and to the "fudged earlier" vblank start in + * vpos. + */ + stat = amdgpu_get_crtc_scanoutpos(adev->ddev, work->crtc_id, + GET_DISTANCE_TO_VBLANKSTART, + &vpos, &hpos, NULL, NULL, + &crtc->hwmode); + + if ((stat & (DRM_SCANOUTPOS_VALID | DRM_SCANOUTPOS_ACCURATE)) != + (DRM_SCANOUTPOS_VALID | DRM_SCANOUTPOS_ACCURATE) || + !(vpos >= 0 && hpos <= 0)) + break; + + /* Sleep at least until estimated real start of hw vblank */ + min_udelay = (-hpos + 1) * max(vblank->linedur_ns / 1000, 5); + if (min_udelay > vblank->framedur_ns / 2000) { + /* Don't wait ridiculously long - something is wrong */ + repcnt = 0; + break; + } + spin_unlock_irqrestore(&crtc->dev->event_lock, flags); + usleep_range(min_udelay, 2 * min_udelay); + spin_lock_irqsave(&crtc->dev->event_lock, flags); + }; + + if (!repcnt) + DRM_DEBUG_DRIVER("Delay problem on crtc %d: min_udelay %d, " + "framedur %d, linedur %d, stat %d, vpos %d, " + "hpos %d\n", work->crtc_id, min_udelay, + vblank->framedur_ns / 1000, + vblank->linedur_ns / 1000, stat, vpos, hpos); + /* set the flip status */ amdgpuCrtc->pflip_status = AMDGPU_FLIP_SUBMITTED; - spin_unlock_irqrestore(&crtc->dev->event_lock, flags); - up_read(&adev->exclusive_lock); + + /* Do the flip (mmio) */ + adev->mode_info.funcs->page_flip(adev, work->crtc_id, work->base); } /* @@ -114,7 +159,7 @@ static void amdgpu_unpin_work_func(struct work_struct *__work) } else DRM_ERROR("failed to reserve buffer after flip\n"); - drm_gem_object_unreference_unlocked(&work->old_rbo->gem_base); + amdgpu_bo_unref(&work->old_rbo); kfree(work->shared); kfree(work); } @@ -153,8 +198,8 @@ int amdgpu_crtc_page_flip(struct drm_crtc *crtc, obj = old_amdgpu_fb->obj; /* take a reference to the old object */ - drm_gem_object_reference(obj); work->old_rbo = gem_to_amdgpu_bo(obj); + amdgpu_bo_ref(work->old_rbo); new_amdgpu_fb = to_amdgpu_framebuffer(fb); obj = new_amdgpu_fb->obj; @@ -210,7 +255,7 @@ int amdgpu_crtc_page_flip(struct drm_crtc *crtc, /* update crtc fb */ crtc->primary->fb = fb; spin_unlock_irqrestore(&crtc->dev->event_lock, flags); - queue_work(amdgpu_crtc->pflip_queue, &work->flip_work); + amdgpu_flip_work_func(&work->flip_work); return 0; vblank_cleanup: @@ -227,7 +272,7 @@ pflip_cleanup: amdgpu_bo_unreserve(new_rbo); cleanup: - drm_gem_object_unreference_unlocked(&work->old_rbo->gem_base); + amdgpu_bo_unref(&work->old_rbo); fence_put(work->excl); for (i = 0; i < work->shared_count; ++i) fence_put(work->shared[i]); @@ -486,7 +531,7 @@ static const struct drm_framebuffer_funcs amdgpu_fb_funcs = { int amdgpu_framebuffer_init(struct drm_device *dev, struct amdgpu_framebuffer *rfb, - struct drm_mode_fb_cmd2 *mode_cmd, + const struct drm_mode_fb_cmd2 *mode_cmd, struct drm_gem_object *obj) { int ret; @@ -503,7 +548,7 @@ amdgpu_framebuffer_init(struct drm_device *dev, static struct drm_framebuffer * amdgpu_user_framebuffer_create(struct drm_device *dev, struct drm_file *file_priv, - struct drm_mode_fb_cmd2 *mode_cmd) + const struct drm_mode_fb_cmd2 *mode_cmd) { struct drm_gem_object *obj; struct amdgpu_framebuffer *amdgpu_fb; @@ -715,8 +760,17 @@ bool amdgpu_crtc_scaling_mode_fixup(struct drm_crtc *crtc, * an optional accurate timestamp of when query happened. * * \param dev Device to query. - * \param crtc Crtc to query. + * \param pipe Crtc to query. * \param flags Flags from caller (DRM_CALLED_FROM_VBLIRQ or 0). + * For driver internal use only also supports these flags: + * + * USE_REAL_VBLANKSTART to use the real start of vblank instead + * of a fudged earlier start of vblank. + * + * GET_DISTANCE_TO_VBLANKSTART to return distance to the + * fudged earlier start of vblank in *vpos and the distance + * to true start of vblank in *hpos. + * * \param *vpos Location where vertical scanout position should be stored. * \param *hpos Location where horizontal scanout position should go. * \param *stime Target location for timestamp taken immediately before @@ -738,8 +792,10 @@ bool amdgpu_crtc_scaling_mode_fixup(struct drm_crtc *crtc, * unknown small number of scanlines wrt. real scanout position. * */ -int amdgpu_get_crtc_scanoutpos(struct drm_device *dev, int crtc, unsigned int flags, - int *vpos, int *hpos, ktime_t *stime, ktime_t *etime) +int amdgpu_get_crtc_scanoutpos(struct drm_device *dev, unsigned int pipe, + unsigned int flags, int *vpos, int *hpos, + ktime_t *stime, ktime_t *etime, + const struct drm_display_mode *mode) { u32 vbl = 0, position = 0; int vbl_start, vbl_end, vtotal, ret = 0; @@ -753,7 +809,7 @@ int amdgpu_get_crtc_scanoutpos(struct drm_device *dev, int crtc, unsigned int fl if (stime) *stime = ktime_get(); - if (amdgpu_display_page_flip_get_scanoutpos(adev, crtc, &vbl, &position) == 0) + if (amdgpu_display_page_flip_get_scanoutpos(adev, pipe, &vbl, &position) == 0) ret |= DRM_SCANOUTPOS_VALID; /* Get optional system timestamp after query. */ @@ -775,14 +831,44 @@ int amdgpu_get_crtc_scanoutpos(struct drm_device *dev, int crtc, unsigned int fl } else { /* No: Fake something reasonable which gives at least ok results. */ - vbl_start = adev->mode_info.crtcs[crtc]->base.hwmode.crtc_vdisplay; + vbl_start = mode->crtc_vdisplay; vbl_end = 0; } + /* Called from driver internal vblank counter query code? */ + if (flags & GET_DISTANCE_TO_VBLANKSTART) { + /* Caller wants distance from real vbl_start in *hpos */ + *hpos = *vpos - vbl_start; + } + + /* Fudge vblank to start a few scanlines earlier to handle the + * problem that vblank irqs fire a few scanlines before start + * of vblank. Some driver internal callers need the true vblank + * start to be used and signal this via the USE_REAL_VBLANKSTART flag. + * + * The cause of the "early" vblank irq is that the irq is triggered + * by the line buffer logic when the line buffer read position enters + * the vblank, whereas our crtc scanout position naturally lags the + * line buffer read position. + */ + if (!(flags & USE_REAL_VBLANKSTART)) + vbl_start -= adev->mode_info.crtcs[pipe]->lb_vblank_lead_lines; + /* Test scanout position against vblank region. */ if ((*vpos < vbl_start) && (*vpos >= vbl_end)) in_vbl = false; + /* In vblank? */ + if (in_vbl) + ret |= DRM_SCANOUTPOS_IN_VBLANK; + + /* Called from driver internal vblank counter query code? */ + if (flags & GET_DISTANCE_TO_VBLANKSTART) { + /* Caller wants distance from fudged earlier vbl_start */ + *vpos -= vbl_start; + return ret; + } + /* Check if inside vblank area and apply corrective offsets: * vpos will then be >=0 in video scanout area, but negative * within vblank area, counting down the number of lines until @@ -791,39 +877,13 @@ int amdgpu_get_crtc_scanoutpos(struct drm_device *dev, int crtc, unsigned int fl /* Inside "upper part" of vblank area? Apply corrective offset if so: */ if (in_vbl && (*vpos >= vbl_start)) { - vtotal = adev->mode_info.crtcs[crtc]->base.hwmode.crtc_vtotal; + vtotal = mode->crtc_vtotal; *vpos = *vpos - vtotal; } /* Correct for shifted end of vbl at vbl_end. */ *vpos = *vpos - vbl_end; - /* In vblank? */ - if (in_vbl) - ret |= DRM_SCANOUTPOS_IN_VBLANK; - - /* Is vpos outside nominal vblank area, but less than - * 1/100 of a frame height away from start of vblank? - * If so, assume this isn't a massively delayed vblank - * interrupt, but a vblank interrupt that fired a few - * microseconds before true start of vblank. Compensate - * by adding a full frame duration to the final timestamp. - * Happens, e.g., on ATI R500, R600. - * - * We only do this if DRM_CALLED_FROM_VBLIRQ. - */ - if ((flags & DRM_CALLED_FROM_VBLIRQ) && !in_vbl) { - vbl_start = adev->mode_info.crtcs[crtc]->base.hwmode.crtc_vdisplay; - vtotal = adev->mode_info.crtcs[crtc]->base.hwmode.crtc_vtotal; - - if (vbl_start - *vpos < vtotal / 100) { - *vpos -= vtotal; - - /* Signal this correction as "applied". */ - ret |= 0x8; - } - } - return ret; } diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c index b190c2a83680..f1e17d60055a 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c @@ -69,17 +69,19 @@ int amdgpu_dpm = -1; int amdgpu_smc_load_fw = 1; int amdgpu_aspm = -1; int amdgpu_runtime_pm = -1; -int amdgpu_hard_reset = 0; unsigned amdgpu_ip_block_mask = 0xffffffff; int amdgpu_bapm = -1; int amdgpu_deep_color = 0; -int amdgpu_vm_size = 8; +int amdgpu_vm_size = 64; int amdgpu_vm_block_size = -1; +int amdgpu_vm_fault_stop = 0; +int amdgpu_vm_debug = 0; int amdgpu_exp_hw_support = 0; -int amdgpu_enable_scheduler = 0; -int amdgpu_sched_jobs = 16; +int amdgpu_sched_jobs = 32; int amdgpu_sched_hw_submission = 2; -int amdgpu_enable_semaphores = 1; +int amdgpu_powerplay = -1; +unsigned amdgpu_pcie_gen_cap = 0; +unsigned amdgpu_pcie_lane_cap = 0; MODULE_PARM_DESC(vramlimit, "Restrict VRAM for testing, in megabytes"); module_param_named(vramlimit, amdgpu_vram_limit, int, 0600); @@ -123,9 +125,6 @@ module_param_named(aspm, amdgpu_aspm, int, 0444); MODULE_PARM_DESC(runpm, "PX runtime pm (1 = force enable, 0 = disable, -1 = PX only default)"); module_param_named(runpm, amdgpu_runtime_pm, int, 0444); -MODULE_PARM_DESC(hard_reset, "PCI config reset (1 = force enable, 0 = disable (default))"); -module_param_named(hard_reset, amdgpu_hard_reset, int, 0444); - MODULE_PARM_DESC(ip_block_mask, "IP Block Mask (all blocks enabled (default))"); module_param_named(ip_block_mask, amdgpu_ip_block_mask, uint, 0444); @@ -135,26 +134,37 @@ module_param_named(bapm, amdgpu_bapm, int, 0444); MODULE_PARM_DESC(deep_color, "Deep Color support (1 = enable, 0 = disable (default))"); module_param_named(deep_color, amdgpu_deep_color, int, 0444); -MODULE_PARM_DESC(vm_size, "VM address space size in gigabytes (default 8GB)"); +MODULE_PARM_DESC(vm_size, "VM address space size in gigabytes (default 64GB)"); module_param_named(vm_size, amdgpu_vm_size, int, 0444); MODULE_PARM_DESC(vm_block_size, "VM page table size in bits (default depending on vm_size)"); module_param_named(vm_block_size, amdgpu_vm_block_size, int, 0444); +MODULE_PARM_DESC(vm_fault_stop, "Stop on VM fault (0 = never (default), 1 = print first, 2 = always)"); +module_param_named(vm_fault_stop, amdgpu_vm_fault_stop, int, 0444); + +MODULE_PARM_DESC(vm_debug, "Debug VM handling (0 = disabled (default), 1 = enabled)"); +module_param_named(vm_debug, amdgpu_vm_debug, int, 0644); + MODULE_PARM_DESC(exp_hw_support, "experimental hw support (1 = enable, 0 = disable (default))"); module_param_named(exp_hw_support, amdgpu_exp_hw_support, int, 0444); -MODULE_PARM_DESC(enable_scheduler, "enable SW GPU scheduler (1 = enable, 0 = disable ((default))"); -module_param_named(enable_scheduler, amdgpu_enable_scheduler, int, 0444); - -MODULE_PARM_DESC(sched_jobs, "the max number of jobs supported in the sw queue (default 16)"); +MODULE_PARM_DESC(sched_jobs, "the max number of jobs supported in the sw queue (default 32)"); module_param_named(sched_jobs, amdgpu_sched_jobs, int, 0444); MODULE_PARM_DESC(sched_hw_submission, "the max number of HW submissions (default 2)"); module_param_named(sched_hw_submission, amdgpu_sched_hw_submission, int, 0444); -MODULE_PARM_DESC(enable_semaphores, "Enable semaphores (1 = enable (default), 0 = disable)"); -module_param_named(enable_semaphores, amdgpu_enable_semaphores, int, 0644); +#ifdef CONFIG_DRM_AMD_POWERPLAY +MODULE_PARM_DESC(powerplay, "Powerplay component (1 = enable, 0 = disable, -1 = auto (default))"); +module_param_named(powerplay, amdgpu_powerplay, int, 0444); +#endif + +MODULE_PARM_DESC(pcie_gen_cap, "PCIE Gen Caps (0: autodetect (default))"); +module_param_named(pcie_gen_cap, amdgpu_pcie_gen_cap, uint, 0444); + +MODULE_PARM_DESC(pcie_lane_cap, "PCIE Lane Caps (0: autodetect (default))"); +module_param_named(pcie_lane_cap, amdgpu_pcie_lane_cap, uint, 0444); static struct pci_device_id pciidlist[] = { #ifdef CONFIG_DRM_AMDGPU_CIK @@ -242,11 +252,11 @@ static struct pci_device_id pciidlist[] = { {0x1002, 0x985F, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_MULLINS|AMD_IS_MOBILITY|AMD_IS_APU}, #endif /* topaz */ - {0x1002, 0x6900, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_TOPAZ|AMD_EXP_HW_SUPPORT}, - {0x1002, 0x6901, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_TOPAZ|AMD_EXP_HW_SUPPORT}, - {0x1002, 0x6902, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_TOPAZ|AMD_EXP_HW_SUPPORT}, - {0x1002, 0x6903, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_TOPAZ|AMD_EXP_HW_SUPPORT}, - {0x1002, 0x6907, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_TOPAZ|AMD_EXP_HW_SUPPORT}, + {0x1002, 0x6900, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_TOPAZ}, + {0x1002, 0x6901, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_TOPAZ}, + {0x1002, 0x6902, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_TOPAZ}, + {0x1002, 0x6903, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_TOPAZ}, + {0x1002, 0x6907, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_TOPAZ}, /* tonga */ {0x1002, 0x6920, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_TONGA}, {0x1002, 0x6921, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_TONGA}, @@ -265,6 +275,8 @@ static struct pci_device_id pciidlist[] = { {0x1002, 0x9875, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_CARRIZO|AMD_IS_APU}, {0x1002, 0x9876, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_CARRIZO|AMD_IS_APU}, {0x1002, 0x9877, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_CARRIZO|AMD_IS_APU}, + /* stoney */ + {0x1002, 0x98E4, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_STONEY|AMD_IS_APU}, {0, 0, 0} }; @@ -306,6 +318,14 @@ static int amdgpu_pci_probe(struct pci_dev *pdev, return -ENODEV; } + /* + * Initialize amdkfd before starting radeon. If it was not loaded yet, + * defer radeon probing + */ + ret = amdgpu_amdkfd_init(); + if (ret == -EPROBE_DEFER) + return ret; + /* Get rid of things like offb */ ret = amdgpu_kick_out_firmware_fb(pdev); if (ret) @@ -535,6 +555,7 @@ static struct pci_driver amdgpu_kms_pci_driver = { static int __init amdgpu_init(void) { + amdgpu_sync_init(); #ifdef CONFIG_VGA_CONSOLE if (vgacon_text_force()) { DRM_ERROR("VGACON disables amdgpu kernel modesetting.\n"); @@ -548,8 +569,6 @@ static int __init amdgpu_init(void) driver->num_ioctls = amdgpu_max_kms_ioctl; amdgpu_register_atpx_handler(); - amdgpu_amdkfd_init(); - /* let modprobe override vga console setting */ return drm_pci_init(driver, pdriver); } @@ -559,6 +578,7 @@ static void __exit amdgpu_exit(void) amdgpu_amdkfd_fini(); drm_pci_exit(driver, pdriver); amdgpu_unregister_atpx_handler(); + amdgpu_sync_fini(); } module_init(amdgpu_init); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_fb.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_fb.c index 96290d9cddca..919146780a15 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_fb.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_fb.c @@ -45,7 +45,6 @@ struct amdgpu_fbdev { struct drm_fb_helper helper; struct amdgpu_framebuffer rfb; - struct list_head fbdev_list; struct amdgpu_device *adev; }; @@ -207,6 +206,7 @@ static int amdgpufb_create(struct drm_fb_helper *helper, } info->par = rfbdev; + info->skip_vt_switch = true; ret = amdgpu_framebuffer_init(adev->ddev, &rfbdev->rfb, &mode_cmd, gobj); if (ret) { @@ -263,7 +263,7 @@ out_unref: } if (fb && ret) { - drm_gem_object_unreference(gobj); + drm_gem_object_unreference_unlocked(gobj); drm_framebuffer_unregister_private(fb); drm_framebuffer_cleanup(fb); kfree(fb); @@ -333,6 +333,10 @@ int amdgpu_fbdev_init(struct amdgpu_device *adev) if (!adev->mode_info.mode_config_initialized) return 0; + /* don't init fbdev if there are no connectors */ + if (list_empty(&adev->ddev->mode_config.connector_list)) + return 0; + /* select 8 bpp console on low vram cards */ if (adev->mc.real_vram_size <= (32*1024*1024)) bpp_sel = 8; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c index b3fc26c59787..4303b447efe8 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c @@ -47,6 +47,30 @@ * that the the relevant GPU caches have been flushed. */ +struct amdgpu_fence { + struct fence base; + + /* RB, DMA, etc. */ + struct amdgpu_ring *ring; +}; + +static struct kmem_cache *amdgpu_fence_slab; +static atomic_t amdgpu_fence_slab_ref = ATOMIC_INIT(0); + +/* + * Cast helper + */ +static const struct fence_ops amdgpu_fence_ops; +static inline struct amdgpu_fence *to_amdgpu_fence(struct fence *f) +{ + struct amdgpu_fence *__f = container_of(f, struct amdgpu_fence, base); + + if (__f->base.ops == &amdgpu_fence_ops) + return __f; + + return NULL; +} + /** * amdgpu_fence_write - write a fence value * @@ -79,345 +103,123 @@ static u32 amdgpu_fence_read(struct amdgpu_ring *ring) if (drv->cpu_addr) seq = le32_to_cpu(*drv->cpu_addr); else - seq = lower_32_bits(atomic64_read(&drv->last_seq)); + seq = atomic_read(&drv->last_seq); return seq; } /** - * amdgpu_fence_schedule_check - schedule lockup check - * - * @ring: pointer to struct amdgpu_ring - * - * Queues a delayed work item to check for lockups. - */ -static void amdgpu_fence_schedule_check(struct amdgpu_ring *ring) -{ - /* - * Do not reset the timer here with mod_delayed_work, - * this can livelock in an interaction with TTM delayed destroy. - */ - queue_delayed_work(system_power_efficient_wq, - &ring->fence_drv.lockup_work, - AMDGPU_FENCE_JIFFIES_TIMEOUT); -} - -/** * amdgpu_fence_emit - emit a fence on the requested ring * * @ring: ring the fence is associated with - * @owner: creator of the fence - * @fence: amdgpu fence object + * @f: resulting fence object * * Emits a fence command on the requested ring (all asics). * Returns 0 on success, -ENOMEM on failure. */ -int amdgpu_fence_emit(struct amdgpu_ring *ring, void *owner, - struct amdgpu_fence **fence) +int amdgpu_fence_emit(struct amdgpu_ring *ring, struct fence **f) { struct amdgpu_device *adev = ring->adev; + struct amdgpu_fence *fence; + struct fence **ptr; + uint32_t seq; - /* we are protected by the ring emission mutex */ - *fence = kmalloc(sizeof(struct amdgpu_fence), GFP_KERNEL); - if ((*fence) == NULL) { + fence = kmem_cache_alloc(amdgpu_fence_slab, GFP_KERNEL); + if (fence == NULL) return -ENOMEM; - } - (*fence)->seq = ++ring->fence_drv.sync_seq[ring->idx]; - (*fence)->ring = ring; - (*fence)->owner = owner; - fence_init(&(*fence)->base, &amdgpu_fence_ops, - &ring->fence_drv.fence_queue.lock, - adev->fence_context + ring->idx, - (*fence)->seq); + + seq = ++ring->fence_drv.sync_seq; + fence->ring = ring; + fence_init(&fence->base, &amdgpu_fence_ops, + &ring->fence_drv.lock, + adev->fence_context + ring->idx, + seq); amdgpu_ring_emit_fence(ring, ring->fence_drv.gpu_addr, - (*fence)->seq, - AMDGPU_FENCE_FLAG_INT); - trace_amdgpu_fence_emit(ring->adev->ddev, ring->idx, (*fence)->seq); - return 0; -} + seq, AMDGPU_FENCE_FLAG_INT); -/** - * amdgpu_fence_check_signaled - callback from fence_queue - * - * this function is called with fence_queue lock held, which is also used - * for the fence locking itself, so unlocked variants are used for - * fence_signal, and remove_wait_queue. - */ -static int amdgpu_fence_check_signaled(wait_queue_t *wait, unsigned mode, int flags, void *key) -{ - struct amdgpu_fence *fence; - struct amdgpu_device *adev; - u64 seq; - int ret; + ptr = &ring->fence_drv.fences[seq & ring->fence_drv.num_fences_mask]; + /* This function can't be called concurrently anyway, otherwise + * emitting the fence would mess up the hardware ring buffer. + */ + BUG_ON(rcu_dereference_protected(*ptr, 1)); - fence = container_of(wait, struct amdgpu_fence, fence_wake); - adev = fence->ring->adev; + rcu_assign_pointer(*ptr, fence_get(&fence->base)); - /* - * We cannot use amdgpu_fence_process here because we're already - * in the waitqueue, in a call from wake_up_all. - */ - seq = atomic64_read(&fence->ring->fence_drv.last_seq); - if (seq >= fence->seq) { - ret = fence_signal_locked(&fence->base); - if (!ret) - FENCE_TRACE(&fence->base, "signaled from irq context\n"); - else - FENCE_TRACE(&fence->base, "was already signaled\n"); + *f = &fence->base; - __remove_wait_queue(&fence->ring->fence_drv.fence_queue, &fence->fence_wake); - fence_put(&fence->base); - } else - FENCE_TRACE(&fence->base, "pending\n"); return 0; } /** - * amdgpu_fence_activity - check for fence activity + * amdgpu_fence_schedule_fallback - schedule fallback check * * @ring: pointer to struct amdgpu_ring * - * Checks the current fence value and calculates the last - * signalled fence value. Returns true if activity occured - * on the ring, and the fence_queue should be waken up. + * Start a timer as fallback to our interrupts. */ -static bool amdgpu_fence_activity(struct amdgpu_ring *ring) +static void amdgpu_fence_schedule_fallback(struct amdgpu_ring *ring) { - uint64_t seq, last_seq, last_emitted; - unsigned count_loop = 0; - bool wake = false; - - /* Note there is a scenario here for an infinite loop but it's - * very unlikely to happen. For it to happen, the current polling - * process need to be interrupted by another process and another - * process needs to update the last_seq btw the atomic read and - * xchg of the current process. - * - * More over for this to go in infinite loop there need to be - * continuously new fence signaled ie amdgpu_fence_read needs - * to return a different value each time for both the currently - * polling process and the other process that xchg the last_seq - * btw atomic read and xchg of the current process. And the - * value the other process set as last seq must be higher than - * the seq value we just read. Which means that current process - * need to be interrupted after amdgpu_fence_read and before - * atomic xchg. - * - * To be even more safe we count the number of time we loop and - * we bail after 10 loop just accepting the fact that we might - * have temporarly set the last_seq not to the true real last - * seq but to an older one. - */ - last_seq = atomic64_read(&ring->fence_drv.last_seq); - do { - last_emitted = ring->fence_drv.sync_seq[ring->idx]; - seq = amdgpu_fence_read(ring); - seq |= last_seq & 0xffffffff00000000LL; - if (seq < last_seq) { - seq &= 0xffffffff; - seq |= last_emitted & 0xffffffff00000000LL; - } - - if (seq <= last_seq || seq > last_emitted) { - break; - } - /* If we loop over we don't want to return without - * checking if a fence is signaled as it means that the - * seq we just read is different from the previous on. - */ - wake = true; - last_seq = seq; - if ((count_loop++) > 10) { - /* We looped over too many time leave with the - * fact that we might have set an older fence - * seq then the current real last seq as signaled - * by the hw. - */ - break; - } - } while (atomic64_xchg(&ring->fence_drv.last_seq, seq) > seq); - - if (seq < last_emitted) - amdgpu_fence_schedule_check(ring); - - return wake; -} - -/** - * amdgpu_fence_check_lockup - check for hardware lockup - * - * @work: delayed work item - * - * Checks for fence activity and if there is none probe - * the hardware if a lockup occured. - */ -static void amdgpu_fence_check_lockup(struct work_struct *work) -{ - struct amdgpu_fence_driver *fence_drv; - struct amdgpu_ring *ring; - - fence_drv = container_of(work, struct amdgpu_fence_driver, - lockup_work.work); - ring = fence_drv->ring; - - if (!down_read_trylock(&ring->adev->exclusive_lock)) { - /* just reschedule the check if a reset is going on */ - amdgpu_fence_schedule_check(ring); - return; - } - - if (amdgpu_fence_activity(ring)) { - wake_up_all(&ring->fence_drv.fence_queue); - } - else if (amdgpu_ring_is_lockup(ring)) { - /* good news we believe it's a lockup */ - dev_warn(ring->adev->dev, "GPU lockup (current fence id " - "0x%016llx last fence id 0x%016llx on ring %d)\n", - (uint64_t)atomic64_read(&fence_drv->last_seq), - fence_drv->sync_seq[ring->idx], ring->idx); - - /* remember that we need an reset */ - ring->adev->needs_reset = true; - wake_up_all(&ring->fence_drv.fence_queue); - } - up_read(&ring->adev->exclusive_lock); + mod_timer(&ring->fence_drv.fallback_timer, + jiffies + AMDGPU_FENCE_JIFFIES_TIMEOUT); } /** - * amdgpu_fence_process - process a fence + * amdgpu_fence_process - check for fence activity * - * @adev: amdgpu_device pointer - * @ring: ring index the fence is associated with + * @ring: pointer to struct amdgpu_ring * - * Checks the current fence value and wakes the fence queue - * if the sequence number has increased (all asics). + * Checks the current fence value and calculates the last + * signalled fence value. Wakes the fence queue if the + * sequence number has increased. */ void amdgpu_fence_process(struct amdgpu_ring *ring) { - if (amdgpu_fence_activity(ring)) - wake_up_all(&ring->fence_drv.fence_queue); -} - -/** - * amdgpu_fence_seq_signaled - check if a fence sequence number has signaled - * - * @ring: ring the fence is associated with - * @seq: sequence number - * - * Check if the last signaled fence sequnce number is >= the requested - * sequence number (all asics). - * Returns true if the fence has signaled (current fence value - * is >= requested value) or false if it has not (current fence - * value is < the requested value. Helper function for - * amdgpu_fence_signaled(). - */ -static bool amdgpu_fence_seq_signaled(struct amdgpu_ring *ring, u64 seq) -{ - if (atomic64_read(&ring->fence_drv.last_seq) >= seq) - return true; - - /* poll new last sequence at least once */ - amdgpu_fence_process(ring); - if (atomic64_read(&ring->fence_drv.last_seq) >= seq) - return true; - - return false; -} - -static bool amdgpu_fence_is_signaled(struct fence *f) -{ - struct amdgpu_fence *fence = to_amdgpu_fence(f); - struct amdgpu_ring *ring = fence->ring; - struct amdgpu_device *adev = ring->adev; - - if (atomic64_read(&ring->fence_drv.last_seq) >= fence->seq) - return true; - - if (down_read_trylock(&adev->exclusive_lock)) { - amdgpu_fence_process(ring); - up_read(&adev->exclusive_lock); + struct amdgpu_fence_driver *drv = &ring->fence_drv; + uint32_t seq, last_seq; + int r; - if (atomic64_read(&ring->fence_drv.last_seq) >= fence->seq) - return true; - } - return false; -} + do { + last_seq = atomic_read(&ring->fence_drv.last_seq); + seq = amdgpu_fence_read(ring); -/** - * amdgpu_fence_enable_signaling - enable signalling on fence - * @fence: fence - * - * This function is called with fence_queue lock held, and adds a callback - * to fence_queue that checks if this fence is signaled, and if so it - * signals the fence and removes itself. - */ -static bool amdgpu_fence_enable_signaling(struct fence *f) -{ - struct amdgpu_fence *fence = to_amdgpu_fence(f); - struct amdgpu_ring *ring = fence->ring; + } while (atomic_cmpxchg(&drv->last_seq, last_seq, seq) != last_seq); - if (atomic64_read(&ring->fence_drv.last_seq) >= fence->seq) - return false; + if (seq != ring->fence_drv.sync_seq) + amdgpu_fence_schedule_fallback(ring); - fence->fence_wake.flags = 0; - fence->fence_wake.private = NULL; - fence->fence_wake.func = amdgpu_fence_check_signaled; - __add_wait_queue(&ring->fence_drv.fence_queue, &fence->fence_wake); - fence_get(f); - FENCE_TRACE(&fence->base, "armed on ring %i!\n", ring->idx); - return true; -} + while (last_seq != seq) { + struct fence *fence, **ptr; -/* - * amdgpu_ring_wait_seq_timeout - wait for seq of the specific ring to signal - * @ring: ring to wait on for the seq number - * @seq: seq number wait for - * - * return value: - * 0: seq signaled, and gpu not hang - * -EDEADL: GPU hang detected - * -EINVAL: some paramter is not valid - */ -static int amdgpu_fence_ring_wait_seq(struct amdgpu_ring *ring, uint64_t seq) -{ - struct amdgpu_device *adev = ring->adev; - bool signaled = false; + ptr = &drv->fences[++last_seq & drv->num_fences_mask]; - BUG_ON(!ring); - if (seq > ring->fence_drv.sync_seq[ring->idx]) - return -EINVAL; + /* There is always exactly one thread signaling this fence slot */ + fence = rcu_dereference_protected(*ptr, 1); + rcu_assign_pointer(*ptr, NULL); - if (atomic64_read(&ring->fence_drv.last_seq) >= seq) - return 0; + BUG_ON(!fence); - wait_event(ring->fence_drv.fence_queue, ( - (signaled = amdgpu_fence_seq_signaled(ring, seq)) - || adev->needs_reset)); + r = fence_signal(fence); + if (!r) + FENCE_TRACE(fence, "signaled from irq context\n"); + else + BUG(); - if (signaled) - return 0; - else - return -EDEADLK; + fence_put(fence); + } } /** - * amdgpu_fence_wait_next - wait for the next fence to signal + * amdgpu_fence_fallback - fallback for hardware interrupts * - * @adev: amdgpu device pointer - * @ring: ring index the fence is associated with + * @work: delayed work item * - * Wait for the next fence on the requested ring to signal (all asics). - * Returns 0 if the next fence has passed, error for all other cases. - * Caller must hold ring lock. + * Checks for fence activity. */ -int amdgpu_fence_wait_next(struct amdgpu_ring *ring) +static void amdgpu_fence_fallback(unsigned long arg) { - uint64_t seq = atomic64_read(&ring->fence_drv.last_seq) + 1ULL; - - if (seq >= ring->fence_drv.sync_seq[ring->idx]) - return -ENOENT; + struct amdgpu_ring *ring = (void *)arg; - return amdgpu_fence_ring_wait_seq(ring, seq); + amdgpu_fence_process(ring); } /** @@ -428,46 +230,28 @@ int amdgpu_fence_wait_next(struct amdgpu_ring *ring) * * Wait for all fences on the requested ring to signal (all asics). * Returns 0 if the fences have passed, error for all other cases. - * Caller must hold ring lock. */ int amdgpu_fence_wait_empty(struct amdgpu_ring *ring) { - uint64_t seq = ring->fence_drv.sync_seq[ring->idx]; + uint64_t seq = ACCESS_ONCE(ring->fence_drv.sync_seq); + struct fence *fence, **ptr; + int r; if (!seq) return 0; - return amdgpu_fence_ring_wait_seq(ring, seq); -} - -/** - * amdgpu_fence_ref - take a ref on a fence - * - * @fence: amdgpu fence object - * - * Take a reference on a fence (all asics). - * Returns the fence. - */ -struct amdgpu_fence *amdgpu_fence_ref(struct amdgpu_fence *fence) -{ - fence_get(&fence->base); - return fence; -} - -/** - * amdgpu_fence_unref - remove a ref on a fence - * - * @fence: amdgpu fence object - * - * Remove a reference on a fence (all asics). - */ -void amdgpu_fence_unref(struct amdgpu_fence **fence) -{ - struct amdgpu_fence *tmp = *fence; + ptr = &ring->fence_drv.fences[seq & ring->fence_drv.num_fences_mask]; + rcu_read_lock(); + fence = rcu_dereference(*ptr); + if (!fence || !fence_get_rcu(fence)) { + rcu_read_unlock(); + return 0; + } + rcu_read_unlock(); - *fence = NULL; - if (tmp) - fence_put(&tmp->base); + r = fence_wait(fence, false); + fence_put(fence); + return r; } /** @@ -487,75 +271,10 @@ unsigned amdgpu_fence_count_emitted(struct amdgpu_ring *ring) * but it's ok to report slightly wrong fence count here. */ amdgpu_fence_process(ring); - emitted = ring->fence_drv.sync_seq[ring->idx] - - atomic64_read(&ring->fence_drv.last_seq); - /* to avoid 32bits warp around */ - if (emitted > 0x10000000) - emitted = 0x10000000; - - return (unsigned)emitted; -} - -/** - * amdgpu_fence_need_sync - do we need a semaphore - * - * @fence: amdgpu fence object - * @dst_ring: which ring to check against - * - * Check if the fence needs to be synced against another ring - * (all asics). If so, we need to emit a semaphore. - * Returns true if we need to sync with another ring, false if - * not. - */ -bool amdgpu_fence_need_sync(struct amdgpu_fence *fence, - struct amdgpu_ring *dst_ring) -{ - struct amdgpu_fence_driver *fdrv; - - if (!fence) - return false; - - if (fence->ring == dst_ring) - return false; - - /* we are protected by the ring mutex */ - fdrv = &dst_ring->fence_drv; - if (fence->seq <= fdrv->sync_seq[fence->ring->idx]) - return false; - - return true; -} - -/** - * amdgpu_fence_note_sync - record the sync point - * - * @fence: amdgpu fence object - * @dst_ring: which ring to check against - * - * Note the sequence number at which point the fence will - * be synced with the requested ring (all asics). - */ -void amdgpu_fence_note_sync(struct amdgpu_fence *fence, - struct amdgpu_ring *dst_ring) -{ - struct amdgpu_fence_driver *dst, *src; - unsigned i; - - if (!fence) - return; - - if (fence->ring == dst_ring) - return; - - /* we are protected by the ring mutex */ - src = &fence->ring->fence_drv; - dst = &dst_ring->fence_drv; - for (i = 0; i < AMDGPU_MAX_RINGS; ++i) { - if (i == dst_ring->idx) - continue; - - dst->sync_seq[i] = max(dst->sync_seq[i], src->sync_seq[i]); - } + emitted = 0x100000000ull; + emitted -= atomic_read(&ring->fence_drv.last_seq); + emitted += ACCESS_ONCE(ring->fence_drv.sync_seq); + return lower_32_bits(emitted); } /** @@ -587,7 +306,7 @@ int amdgpu_fence_driver_start_ring(struct amdgpu_ring *ring, ring->fence_drv.cpu_addr = adev->uvd.cpu_addr + index; ring->fence_drv.gpu_addr = adev->uvd.gpu_addr + index; } - amdgpu_fence_write(ring, atomic64_read(&ring->fence_drv.last_seq)); + amdgpu_fence_write(ring, atomic_read(&ring->fence_drv.last_seq)); amdgpu_irq_get(adev, irq_src, irq_type); ring->fence_drv.irq_src = irq_src; @@ -605,36 +324,55 @@ int amdgpu_fence_driver_start_ring(struct amdgpu_ring *ring, * for the requested ring. * * @ring: ring to init the fence driver on + * @num_hw_submission: number of entries on the hardware queue * * Init the fence driver for the requested ring (all asics). * Helper function for amdgpu_fence_driver_init(). */ -int amdgpu_fence_driver_init_ring(struct amdgpu_ring *ring) +int amdgpu_fence_driver_init_ring(struct amdgpu_ring *ring, + unsigned num_hw_submission) { - int i, r; + long timeout; + int r; + + /* Check that num_hw_submission is a power of two */ + if ((num_hw_submission & (num_hw_submission - 1)) != 0) + return -EINVAL; ring->fence_drv.cpu_addr = NULL; ring->fence_drv.gpu_addr = 0; - for (i = 0; i < AMDGPU_MAX_RINGS; ++i) - ring->fence_drv.sync_seq[i] = 0; - - atomic64_set(&ring->fence_drv.last_seq, 0); + ring->fence_drv.sync_seq = 0; + atomic_set(&ring->fence_drv.last_seq, 0); ring->fence_drv.initialized = false; - INIT_DELAYED_WORK(&ring->fence_drv.lockup_work, - amdgpu_fence_check_lockup); - ring->fence_drv.ring = ring; + setup_timer(&ring->fence_drv.fallback_timer, amdgpu_fence_fallback, + (unsigned long)ring); - init_waitqueue_head(&ring->fence_drv.fence_queue); + ring->fence_drv.num_fences_mask = num_hw_submission - 1; + spin_lock_init(&ring->fence_drv.lock); + ring->fence_drv.fences = kcalloc(num_hw_submission, sizeof(void *), + GFP_KERNEL); + if (!ring->fence_drv.fences) + return -ENOMEM; - if (amdgpu_enable_scheduler) { - r = amd_sched_init(&ring->sched, &amdgpu_sched_ops, - amdgpu_sched_hw_submission, ring->name); - if (r) { - DRM_ERROR("Failed to create scheduler on ring %s.\n", - ring->name); - return r; - } + timeout = msecs_to_jiffies(amdgpu_lockup_timeout); + if (timeout == 0) { + /* + * FIXME: + * Delayed workqueue cannot use it directly, + * so the scheduler will not use delayed workqueue if + * MAX_SCHEDULE_TIMEOUT is set. + * Currently keep it simple and silly. + */ + timeout = MAX_SCHEDULE_TIMEOUT; + } + r = amd_sched_init(&ring->sched, &amdgpu_sched_ops, + num_hw_submission, + timeout, ring->name); + if (r) { + DRM_ERROR("Failed to create scheduler on ring %s.\n", + ring->name); + return r; } return 0; @@ -654,6 +392,13 @@ int amdgpu_fence_driver_init_ring(struct amdgpu_ring *ring) */ int amdgpu_fence_driver_init(struct amdgpu_device *adev) { + if (atomic_inc_return(&amdgpu_fence_slab_ref) == 1) { + amdgpu_fence_slab = kmem_cache_create( + "amdgpu_fence", sizeof(struct amdgpu_fence), 0, + SLAB_HWCACHE_ALIGN, NULL); + if (!amdgpu_fence_slab) + return -ENOMEM; + } if (amdgpu_debugfs_fence_init(adev)) dev_err(adev->dev, "fence debugfs file creation failed\n"); @@ -670,11 +415,12 @@ int amdgpu_fence_driver_init(struct amdgpu_device *adev) */ void amdgpu_fence_driver_fini(struct amdgpu_device *adev) { - int i, r; + unsigned i, j; + int r; - mutex_lock(&adev->ring_lock); for (i = 0; i < AMDGPU_MAX_RINGS; i++) { struct amdgpu_ring *ring = adev->rings[i]; + if (!ring || !ring->fence_drv.initialized) continue; r = amdgpu_fence_wait_empty(ring); @@ -682,13 +428,18 @@ void amdgpu_fence_driver_fini(struct amdgpu_device *adev) /* no need to trigger GPU reset as we are unloading */ amdgpu_fence_driver_force_completion(adev); } - wake_up_all(&ring->fence_drv.fence_queue); amdgpu_irq_put(adev, ring->fence_drv.irq_src, ring->fence_drv.irq_type); amd_sched_fini(&ring->sched); + del_timer_sync(&ring->fence_drv.fallback_timer); + for (j = 0; j <= ring->fence_drv.num_fences_mask; ++j) + fence_put(ring->fence_drv.fences[i]); + kfree(ring->fence_drv.fences); ring->fence_drv.initialized = false; } - mutex_unlock(&adev->ring_lock); + + if (atomic_dec_and_test(&amdgpu_fence_slab_ref)) + kmem_cache_destroy(amdgpu_fence_slab); } /** @@ -703,7 +454,6 @@ void amdgpu_fence_driver_suspend(struct amdgpu_device *adev) { int i, r; - mutex_lock(&adev->ring_lock); for (i = 0; i < AMDGPU_MAX_RINGS; i++) { struct amdgpu_ring *ring = adev->rings[i]; if (!ring || !ring->fence_drv.initialized) @@ -720,7 +470,6 @@ void amdgpu_fence_driver_suspend(struct amdgpu_device *adev) amdgpu_irq_put(adev, ring->fence_drv.irq_src, ring->fence_drv.irq_type); } - mutex_unlock(&adev->ring_lock); } /** @@ -739,7 +488,6 @@ void amdgpu_fence_driver_resume(struct amdgpu_device *adev) { int i; - mutex_lock(&adev->ring_lock); for (i = 0; i < AMDGPU_MAX_RINGS; i++) { struct amdgpu_ring *ring = adev->rings[i]; if (!ring || !ring->fence_drv.initialized) @@ -749,7 +497,6 @@ void amdgpu_fence_driver_resume(struct amdgpu_device *adev) amdgpu_irq_get(adev, ring->fence_drv.irq_src, ring->fence_drv.irq_type); } - mutex_unlock(&adev->ring_lock); } /** @@ -769,59 +516,13 @@ void amdgpu_fence_driver_force_completion(struct amdgpu_device *adev) if (!ring || !ring->fence_drv.initialized) continue; - amdgpu_fence_write(ring, ring->fence_drv.sync_seq[i]); + amdgpu_fence_write(ring, ring->fence_drv.sync_seq); } } - /* - * Fence debugfs + * Common fence implementation */ -#if defined(CONFIG_DEBUG_FS) -static int amdgpu_debugfs_fence_info(struct seq_file *m, void *data) -{ - struct drm_info_node *node = (struct drm_info_node *)m->private; - struct drm_device *dev = node->minor->dev; - struct amdgpu_device *adev = dev->dev_private; - int i, j; - - for (i = 0; i < AMDGPU_MAX_RINGS; ++i) { - struct amdgpu_ring *ring = adev->rings[i]; - if (!ring || !ring->fence_drv.initialized) - continue; - - amdgpu_fence_process(ring); - - seq_printf(m, "--- ring %d (%s) ---\n", i, ring->name); - seq_printf(m, "Last signaled fence 0x%016llx\n", - (unsigned long long)atomic64_read(&ring->fence_drv.last_seq)); - seq_printf(m, "Last emitted 0x%016llx\n", - ring->fence_drv.sync_seq[i]); - - for (j = 0; j < AMDGPU_MAX_RINGS; ++j) { - struct amdgpu_ring *other = adev->rings[j]; - if (i != j && other && other->fence_drv.initialized && - ring->fence_drv.sync_seq[j]) - seq_printf(m, "Last sync to ring %d 0x%016llx\n", - j, ring->fence_drv.sync_seq[j]); - } - } - return 0; -} - -static struct drm_info_list amdgpu_debugfs_fence_list[] = { - {"amdgpu_fence_info", &amdgpu_debugfs_fence_info, 0, NULL}, -}; -#endif - -int amdgpu_debugfs_fence_init(struct amdgpu_device *adev) -{ -#if defined(CONFIG_DEBUG_FS) - return amdgpu_debugfs_add_files(adev, amdgpu_debugfs_fence_list, 1); -#else - return 0; -#endif -} static const char *amdgpu_fence_get_driver_name(struct fence *fence) { @@ -834,130 +535,118 @@ static const char *amdgpu_fence_get_timeline_name(struct fence *f) return (const char *)fence->ring->name; } -static inline bool amdgpu_test_signaled(struct amdgpu_fence *fence) +/** + * amdgpu_fence_enable_signaling - enable signalling on fence + * @fence: fence + * + * This function is called with fence_queue lock held, and adds a callback + * to fence_queue that checks if this fence is signaled, and if so it + * signals the fence and removes itself. + */ +static bool amdgpu_fence_enable_signaling(struct fence *f) { - return test_bit(FENCE_FLAG_SIGNALED_BIT, &fence->base.flags); -} + struct amdgpu_fence *fence = to_amdgpu_fence(f); + struct amdgpu_ring *ring = fence->ring; -static bool amdgpu_test_signaled_any(struct fence **fences, uint32_t count) -{ - int idx; - struct fence *fence; - - for (idx = 0; idx < count; ++idx) { - fence = fences[idx]; - if (fence) { - if (test_bit(FENCE_FLAG_SIGNALED_BIT, &fence->flags)) - return true; - } - } - return false; -} + if (!timer_pending(&ring->fence_drv.fallback_timer)) + amdgpu_fence_schedule_fallback(ring); -struct amdgpu_wait_cb { - struct fence_cb base; - struct task_struct *task; -}; + FENCE_TRACE(&fence->base, "armed on ring %i!\n", ring->idx); -static void amdgpu_fence_wait_cb(struct fence *fence, struct fence_cb *cb) -{ - struct amdgpu_wait_cb *wait = - container_of(cb, struct amdgpu_wait_cb, base); - wake_up_process(wait->task); + return true; } -static signed long amdgpu_fence_default_wait(struct fence *f, bool intr, - signed long t) +/** + * amdgpu_fence_free - free up the fence memory + * + * @rcu: RCU callback head + * + * Free up the fence memory after the RCU grace period. + */ +static void amdgpu_fence_free(struct rcu_head *rcu) { + struct fence *f = container_of(rcu, struct fence, rcu); struct amdgpu_fence *fence = to_amdgpu_fence(f); - struct amdgpu_device *adev = fence->ring->adev; - - return amdgpu_fence_wait_any(adev, &f, 1, intr, t); + kmem_cache_free(amdgpu_fence_slab, fence); } /** - * Wait the fence array with timeout + * amdgpu_fence_release - callback that fence can be freed * - * @adev: amdgpu device - * @array: the fence array with amdgpu fence pointer - * @count: the number of the fence array - * @intr: when sleep, set the current task interruptable or not - * @t: timeout to wait + * @fence: fence * - * It will return when any fence is signaled or timeout. + * This function is called when the reference count becomes zero. + * It just RCU schedules freeing up the fence. */ -signed long amdgpu_fence_wait_any(struct amdgpu_device *adev, - struct fence **array, uint32_t count, - bool intr, signed long t) +static void amdgpu_fence_release(struct fence *f) { - struct amdgpu_wait_cb *cb; - struct fence *fence; - unsigned idx; - - BUG_ON(!array); - - cb = kcalloc(count, sizeof(struct amdgpu_wait_cb), GFP_KERNEL); - if (cb == NULL) { - t = -ENOMEM; - goto err_free_cb; - } - - for (idx = 0; idx < count; ++idx) { - fence = array[idx]; - if (fence) { - cb[idx].task = current; - if (fence_add_callback(fence, - &cb[idx].base, amdgpu_fence_wait_cb)) { - /* The fence is already signaled */ - goto fence_rm_cb; - } - } - } + call_rcu(&f->rcu, amdgpu_fence_free); +} - while (t > 0) { - if (intr) - set_current_state(TASK_INTERRUPTIBLE); - else - set_current_state(TASK_UNINTERRUPTIBLE); +static const struct fence_ops amdgpu_fence_ops = { + .get_driver_name = amdgpu_fence_get_driver_name, + .get_timeline_name = amdgpu_fence_get_timeline_name, + .enable_signaling = amdgpu_fence_enable_signaling, + .wait = fence_default_wait, + .release = amdgpu_fence_release, +}; - /* - * amdgpu_test_signaled_any must be called after - * set_current_state to prevent a race with wake_up_process - */ - if (amdgpu_test_signaled_any(array, count)) - break; +/* + * Fence debugfs + */ +#if defined(CONFIG_DEBUG_FS) +static int amdgpu_debugfs_fence_info(struct seq_file *m, void *data) +{ + struct drm_info_node *node = (struct drm_info_node *)m->private; + struct drm_device *dev = node->minor->dev; + struct amdgpu_device *adev = dev->dev_private; + int i; - if (adev->needs_reset) { - t = -EDEADLK; - break; - } + for (i = 0; i < AMDGPU_MAX_RINGS; ++i) { + struct amdgpu_ring *ring = adev->rings[i]; + if (!ring || !ring->fence_drv.initialized) + continue; - t = schedule_timeout(t); + amdgpu_fence_process(ring); - if (t > 0 && intr && signal_pending(current)) - t = -ERESTARTSYS; + seq_printf(m, "--- ring %d (%s) ---\n", i, ring->name); + seq_printf(m, "Last signaled fence 0x%08x\n", + atomic_read(&ring->fence_drv.last_seq)); + seq_printf(m, "Last emitted 0x%08x\n", + ring->fence_drv.sync_seq); } + return 0; +} - __set_current_state(TASK_RUNNING); - -fence_rm_cb: - for (idx = 0; idx < count; ++idx) { - fence = array[idx]; - if (fence && cb[idx].base.func) - fence_remove_callback(fence, &cb[idx].base); - } +/** + * amdgpu_debugfs_gpu_reset - manually trigger a gpu reset + * + * Manually trigger a gpu reset at the next fence wait. + */ +static int amdgpu_debugfs_gpu_reset(struct seq_file *m, void *data) +{ + struct drm_info_node *node = (struct drm_info_node *) m->private; + struct drm_device *dev = node->minor->dev; + struct amdgpu_device *adev = dev->dev_private; -err_free_cb: - kfree(cb); + seq_printf(m, "gpu reset\n"); + amdgpu_gpu_reset(adev); - return t; + return 0; } -const struct fence_ops amdgpu_fence_ops = { - .get_driver_name = amdgpu_fence_get_driver_name, - .get_timeline_name = amdgpu_fence_get_timeline_name, - .enable_signaling = amdgpu_fence_enable_signaling, - .signaled = amdgpu_fence_is_signaled, - .wait = amdgpu_fence_default_wait, - .release = NULL, +static struct drm_info_list amdgpu_debugfs_fence_list[] = { + {"amdgpu_fence_info", &amdgpu_debugfs_fence_info, 0, NULL}, + {"amdgpu_gpu_reset", &amdgpu_debugfs_gpu_reset, 0, NULL} }; +#endif + +int amdgpu_debugfs_fence_init(struct amdgpu_device *adev) +{ +#if defined(CONFIG_DEBUG_FS) + return amdgpu_debugfs_add_files(adev, amdgpu_debugfs_fence_list, 2); +#else + return 0; +#endif +} + diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c index 7297ca3a0ba7..fa6a27bff298 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c @@ -26,6 +26,7 @@ * Jerome Glisse */ #include <linux/ktime.h> +#include <linux/pagemap.h> #include <drm/drmP.h> #include <drm/amdgpu_drm.h> #include "amdgpu.h" @@ -83,24 +84,32 @@ retry: return r; } *obj = &robj->gem_base; - robj->pid = task_pid_nr(current); - - mutex_lock(&adev->gem.mutex); - list_add_tail(&robj->list, &adev->gem.objects); - mutex_unlock(&adev->gem.mutex); return 0; } -int amdgpu_gem_init(struct amdgpu_device *adev) +void amdgpu_gem_force_release(struct amdgpu_device *adev) { - INIT_LIST_HEAD(&adev->gem.objects); - return 0; -} + struct drm_device *ddev = adev->ddev; + struct drm_file *file; -void amdgpu_gem_fini(struct amdgpu_device *adev) -{ - amdgpu_bo_force_delete(adev); + mutex_lock(&ddev->struct_mutex); + + list_for_each_entry(file, &ddev->filelist, lhead) { + struct drm_gem_object *gobj; + int handle; + + WARN_ONCE(1, "Still active user space clients!\n"); + spin_lock(&file->table_lock); + idr_for_each_entry(&file->object_idr, gobj, handle) { + WARN_ONCE(1, "And also active allocations!\n"); + drm_gem_object_unreference(gobj); + } + idr_destroy(&file->object_idr); + spin_unlock(&file->table_lock); + } + + mutex_unlock(&ddev->struct_mutex); } /* @@ -115,11 +124,9 @@ int amdgpu_gem_object_open(struct drm_gem_object *obj, struct drm_file *file_pri struct amdgpu_vm *vm = &fpriv->vm; struct amdgpu_bo_va *bo_va; int r; - r = amdgpu_bo_reserve(rbo, false); - if (r) { + if (r) return r; - } bo_va = amdgpu_vm_bo_find(vm, rbo); if (!bo_va) { @@ -128,33 +135,46 @@ int amdgpu_gem_object_open(struct drm_gem_object *obj, struct drm_file *file_pri ++bo_va->ref_count; } amdgpu_bo_unreserve(rbo); - return 0; } void amdgpu_gem_object_close(struct drm_gem_object *obj, struct drm_file *file_priv) { - struct amdgpu_bo *rbo = gem_to_amdgpu_bo(obj); - struct amdgpu_device *adev = rbo->adev; + struct amdgpu_bo *bo = gem_to_amdgpu_bo(obj); + struct amdgpu_device *adev = bo->adev; struct amdgpu_fpriv *fpriv = file_priv->driver_priv; struct amdgpu_vm *vm = &fpriv->vm; + + struct amdgpu_bo_list_entry vm_pd; + struct list_head list, duplicates; + struct ttm_validate_buffer tv; + struct ww_acquire_ctx ticket; struct amdgpu_bo_va *bo_va; int r; - r = amdgpu_bo_reserve(rbo, true); + INIT_LIST_HEAD(&list); + INIT_LIST_HEAD(&duplicates); + + tv.bo = &bo->tbo; + tv.shared = true; + list_add(&tv.head, &list); + + amdgpu_vm_get_pd_bo(vm, &list, &vm_pd); + + r = ttm_eu_reserve_buffers(&ticket, &list, false, &duplicates); if (r) { dev_err(adev->dev, "leaking bo va because " "we fail to reserve bo (%d)\n", r); return; } - bo_va = amdgpu_vm_bo_find(vm, rbo); + bo_va = amdgpu_vm_bo_find(vm, bo); if (bo_va) { if (--bo_va->ref_count == 0) { amdgpu_vm_bo_rmv(adev, bo_va); } } - amdgpu_bo_unreserve(rbo); + ttm_eu_backoff_reservation(&ticket, &list); } static int amdgpu_gem_handle_lockup(struct amdgpu_device *adev, int r) @@ -181,7 +201,6 @@ int amdgpu_gem_create_ioctl(struct drm_device *dev, void *data, bool kernel = false; int r; - down_read(&adev->exclusive_lock); /* create a gem object to contain this object in */ if (args->in.domains & (AMDGPU_GEM_DOMAIN_GDS | AMDGPU_GEM_DOMAIN_GWS | AMDGPU_GEM_DOMAIN_OA)) { @@ -214,11 +233,9 @@ int amdgpu_gem_create_ioctl(struct drm_device *dev, void *data, memset(args, 0, sizeof(*args)); args->out.handle = handle; - up_read(&adev->exclusive_lock); return 0; error_unlock: - up_read(&adev->exclusive_lock); r = amdgpu_gem_handle_lockup(adev, r); return r; } @@ -242,16 +259,13 @@ int amdgpu_gem_userptr_ioctl(struct drm_device *dev, void *data, AMDGPU_GEM_USERPTR_REGISTER)) return -EINVAL; - if (!(args->flags & AMDGPU_GEM_USERPTR_ANONONLY) || - !(args->flags & AMDGPU_GEM_USERPTR_REGISTER)) { + if (!(args->flags & AMDGPU_GEM_USERPTR_READONLY) && + !(args->flags & AMDGPU_GEM_USERPTR_REGISTER)) { - /* if we want to write to it we must require anonymous - memory and install a MMU notifier */ + /* if we want to write to it we must install a MMU notifier */ return -EACCES; } - down_read(&adev->exclusive_lock); - /* create a gem object to contain this object in */ r = amdgpu_gem_object_create(adev, args->size, 0, AMDGPU_GEM_DOMAIN_CPU, 0, @@ -260,6 +274,8 @@ int amdgpu_gem_userptr_ioctl(struct drm_device *dev, void *data, goto handle_lockup; bo = gem_to_amdgpu_bo(gobj); + bo->prefered_domains = AMDGPU_GEM_DOMAIN_GTT; + bo->allowed_domains = AMDGPU_GEM_DOMAIN_GTT; r = amdgpu_ttm_tt_set_userptr(bo->tbo.ttm, args->addr, args->flags); if (r) goto release_object; @@ -272,18 +288,23 @@ int amdgpu_gem_userptr_ioctl(struct drm_device *dev, void *data, if (args->flags & AMDGPU_GEM_USERPTR_VALIDATE) { down_read(¤t->mm->mmap_sem); + + r = amdgpu_ttm_tt_get_user_pages(bo->tbo.ttm, + bo->tbo.ttm->pages); + if (r) + goto unlock_mmap_sem; + r = amdgpu_bo_reserve(bo, true); - if (r) { - up_read(¤t->mm->mmap_sem); - goto release_object; - } + if (r) + goto free_pages; amdgpu_ttm_placement_from_domain(bo, AMDGPU_GEM_DOMAIN_GTT); r = ttm_bo_validate(&bo->tbo, &bo->placement, true, false); amdgpu_bo_unreserve(bo); - up_read(¤t->mm->mmap_sem); if (r) - goto release_object; + goto free_pages; + + up_read(¤t->mm->mmap_sem); } r = drm_gem_handle_create(filp, gobj, &handle); @@ -293,14 +314,18 @@ int amdgpu_gem_userptr_ioctl(struct drm_device *dev, void *data, goto handle_lockup; args->handle = handle; - up_read(&adev->exclusive_lock); return 0; +free_pages: + release_pages(bo->tbo.ttm->pages, bo->tbo.ttm->num_pages, false); + +unlock_mmap_sem: + up_read(¤t->mm->mmap_sem); + release_object: drm_gem_object_unreference_unlocked(gobj); handle_lockup: - up_read(&adev->exclusive_lock); r = amdgpu_gem_handle_lockup(adev, r); return r; @@ -318,7 +343,7 @@ int amdgpu_mode_dumb_mmap(struct drm_file *filp, return -ENOENT; } robj = gem_to_amdgpu_bo(gobj); - if (amdgpu_ttm_tt_has_userptr(robj->tbo.ttm) || + if (amdgpu_ttm_tt_get_usermm(robj->tbo.ttm) || (robj->flags & AMDGPU_GEM_CREATE_NO_CPU_ACCESS)) { drm_gem_object_unreference_unlocked(gobj); return -EPERM; @@ -458,7 +483,7 @@ static void amdgpu_gem_va_update_vm(struct amdgpu_device *adev, struct amdgpu_bo_va *bo_va, uint32_t operation) { struct ttm_validate_buffer tv, *entry; - struct amdgpu_bo_list_entry *vm_bos; + struct amdgpu_bo_list_entry vm_pd; struct ww_acquire_ctx ticket; struct list_head list, duplicates; unsigned domain; @@ -471,15 +496,14 @@ static void amdgpu_gem_va_update_vm(struct amdgpu_device *adev, tv.shared = true; list_add(&tv.head, &list); - vm_bos = amdgpu_vm_get_bos(adev, bo_va->vm, &list); - if (!vm_bos) - return; + amdgpu_vm_get_pd_bo(bo_va->vm, &list, &vm_pd); /* Provide duplicates to avoid -EALREADY */ r = ttm_eu_reserve_buffers(&ticket, &list, true, &duplicates); if (r) - goto error_free; + goto error_print; + amdgpu_vm_get_pt_bos(bo_va->vm, &duplicates); list_for_each_entry(entry, &list, head) { domain = amdgpu_mem_type_to_domain(entry->bo->mem.mem_type); /* if anything is swapped out don't swap it in here, @@ -487,25 +511,29 @@ static void amdgpu_gem_va_update_vm(struct amdgpu_device *adev, if (domain == AMDGPU_GEM_DOMAIN_CPU) goto error_unreserve; } + list_for_each_entry(entry, &duplicates, head) { + domain = amdgpu_mem_type_to_domain(entry->bo->mem.mem_type); + /* if anything is swapped out don't swap it in here, + just abort and wait for the next CS */ + if (domain == AMDGPU_GEM_DOMAIN_CPU) + goto error_unreserve; + } - mutex_lock(&bo_va->vm->mutex); - r = amdgpu_vm_clear_freed(adev, bo_va->vm); + r = amdgpu_vm_update_page_directory(adev, bo_va->vm); if (r) - goto error_unlock; + goto error_unreserve; + r = amdgpu_vm_clear_freed(adev, bo_va->vm); + if (r) + goto error_unreserve; if (operation == AMDGPU_VA_OP_MAP) r = amdgpu_vm_bo_update(adev, bo_va, &bo_va->bo->tbo.mem); -error_unlock: - mutex_unlock(&bo_va->vm->mutex); - error_unreserve: ttm_eu_backoff_reservation(&ticket, &list); -error_free: - drm_free_large(vm_bos); - +error_print: if (r && r != -ERESTARTSYS) DRM_ERROR("Couldn't update BO_VA (%d)\n", r); } @@ -521,6 +549,9 @@ int amdgpu_gem_va_ioctl(struct drm_device *dev, void *data, struct amdgpu_fpriv *fpriv = filp->driver_priv; struct amdgpu_bo *rbo; struct amdgpu_bo_va *bo_va; + struct ttm_validate_buffer tv, tv_pd; + struct ww_acquire_ctx ticket; + struct list_head list, duplicates; uint32_t invalid_flags, va_flags = 0; int r = 0; @@ -556,9 +587,18 @@ int amdgpu_gem_va_ioctl(struct drm_device *dev, void *data, gobj = drm_gem_object_lookup(dev, filp, args->handle); if (gobj == NULL) return -ENOENT; - rbo = gem_to_amdgpu_bo(gobj); - r = amdgpu_bo_reserve(rbo, false); + INIT_LIST_HEAD(&list); + INIT_LIST_HEAD(&duplicates); + tv.bo = &rbo->tbo; + tv.shared = true; + list_add(&tv.head, &list); + + tv_pd.bo = &fpriv->vm.page_directory->tbo; + tv_pd.shared = true; + list_add(&tv_pd.head, &list); + + r = ttm_eu_reserve_buffers(&ticket, &list, true, &duplicates); if (r) { drm_gem_object_unreference_unlocked(gobj); return r; @@ -566,7 +606,8 @@ int amdgpu_gem_va_ioctl(struct drm_device *dev, void *data, bo_va = amdgpu_vm_bo_find(&fpriv->vm, rbo); if (!bo_va) { - amdgpu_bo_unreserve(rbo); + ttm_eu_backoff_reservation(&ticket, &list); + drm_gem_object_unreference_unlocked(gobj); return -ENOENT; } @@ -588,8 +629,9 @@ int amdgpu_gem_va_ioctl(struct drm_device *dev, void *data, default: break; } - - if (!r && !(args->flags & AMDGPU_VM_DELAY_UPDATE)) + ttm_eu_backoff_reservation(&ticket, &list); + if (!r && !(args->flags & AMDGPU_VM_DELAY_UPDATE) && + !amdgpu_vm_debug) amdgpu_gem_va_update_vm(adev, bo_va, args->operation); drm_gem_object_unreference_unlocked(gobj); @@ -621,7 +663,7 @@ int amdgpu_gem_op_ioctl(struct drm_device *dev, void *data, info.bo_size = robj->gem_base.size; info.alignment = robj->tbo.mem.page_alignment << PAGE_SHIFT; - info.domains = robj->initial_domain; + info.domains = robj->prefered_domains; info.domain_flags = robj->flags; amdgpu_bo_unreserve(robj); if (copy_to_user(out, &info, sizeof(info))) @@ -629,14 +671,18 @@ int amdgpu_gem_op_ioctl(struct drm_device *dev, void *data, break; } case AMDGPU_GEM_OP_SET_PLACEMENT: - if (amdgpu_ttm_tt_has_userptr(robj->tbo.ttm)) { + if (amdgpu_ttm_tt_get_usermm(robj->tbo.ttm)) { r = -EPERM; amdgpu_bo_unreserve(robj); break; } - robj->initial_domain = args->value & (AMDGPU_GEM_DOMAIN_VRAM | - AMDGPU_GEM_DOMAIN_GTT | - AMDGPU_GEM_DOMAIN_CPU); + robj->prefered_domains = args->value & (AMDGPU_GEM_DOMAIN_VRAM | + AMDGPU_GEM_DOMAIN_GTT | + AMDGPU_GEM_DOMAIN_CPU); + robj->allowed_domains = robj->prefered_domains; + if (robj->allowed_domains == AMDGPU_GEM_DOMAIN_VRAM) + robj->allowed_domains |= AMDGPU_GEM_DOMAIN_GTT; + amdgpu_bo_unreserve(robj); break; default: @@ -681,38 +727,73 @@ int amdgpu_mode_dumb_create(struct drm_file *file_priv, } #if defined(CONFIG_DEBUG_FS) +static int amdgpu_debugfs_gem_bo_info(int id, void *ptr, void *data) +{ + struct drm_gem_object *gobj = ptr; + struct amdgpu_bo *bo = gem_to_amdgpu_bo(gobj); + struct seq_file *m = data; + + unsigned domain; + const char *placement; + unsigned pin_count; + + domain = amdgpu_mem_type_to_domain(bo->tbo.mem.mem_type); + switch (domain) { + case AMDGPU_GEM_DOMAIN_VRAM: + placement = "VRAM"; + break; + case AMDGPU_GEM_DOMAIN_GTT: + placement = " GTT"; + break; + case AMDGPU_GEM_DOMAIN_CPU: + default: + placement = " CPU"; + break; + } + seq_printf(m, "\t0x%08x: %12ld byte %s @ 0x%010Lx", + id, amdgpu_bo_size(bo), placement, + amdgpu_bo_gpu_offset(bo)); + + pin_count = ACCESS_ONCE(bo->pin_count); + if (pin_count) + seq_printf(m, " pin count %d", pin_count); + seq_printf(m, "\n"); + + return 0; +} + static int amdgpu_debugfs_gem_info(struct seq_file *m, void *data) { struct drm_info_node *node = (struct drm_info_node *)m->private; struct drm_device *dev = node->minor->dev; - struct amdgpu_device *adev = dev->dev_private; - struct amdgpu_bo *rbo; - unsigned i = 0; + struct drm_file *file; + int r; - mutex_lock(&adev->gem.mutex); - list_for_each_entry(rbo, &adev->gem.objects, list) { - unsigned domain; - const char *placement; + r = mutex_lock_interruptible(&dev->struct_mutex); + if (r) + return r; - domain = amdgpu_mem_type_to_domain(rbo->tbo.mem.mem_type); - switch (domain) { - case AMDGPU_GEM_DOMAIN_VRAM: - placement = "VRAM"; - break; - case AMDGPU_GEM_DOMAIN_GTT: - placement = " GTT"; - break; - case AMDGPU_GEM_DOMAIN_CPU: - default: - placement = " CPU"; - break; - } - seq_printf(m, "bo[0x%08x] %8ldkB %8ldMB %s pid %8ld\n", - i, amdgpu_bo_size(rbo) >> 10, amdgpu_bo_size(rbo) >> 20, - placement, (unsigned long)rbo->pid); - i++; + list_for_each_entry(file, &dev->filelist, lhead) { + struct task_struct *task; + + /* + * Although we have a valid reference on file->pid, that does + * not guarantee that the task_struct who called get_pid() is + * still alive (e.g. get_pid(current) => fork() => exit()). + * Therefore, we need to protect this ->comm access using RCU. + */ + rcu_read_lock(); + task = pid_task(file->pid, PIDTYPE_PID); + seq_printf(m, "pid %8d command %s:\n", pid_nr(file->pid), + task ? task->comm : "<unknown>"); + rcu_read_unlock(); + + spin_lock(&file->table_lock); + idr_for_each(&file->object_idr, amdgpu_debugfs_gem_bo_info, m); + spin_unlock(&file->table_lock); } - mutex_unlock(&adev->gem.mutex); + + mutex_unlock(&dev->struct_mutex); return 0; } diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c index c439735ee670..8443cea6821a 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c @@ -55,14 +55,13 @@ static int amdgpu_debugfs_sa_init(struct amdgpu_device *adev); * suballocator. * Returns 0 on success, error on failure. */ -int amdgpu_ib_get(struct amdgpu_ring *ring, struct amdgpu_vm *vm, +int amdgpu_ib_get(struct amdgpu_device *adev, struct amdgpu_vm *vm, unsigned size, struct amdgpu_ib *ib) { - struct amdgpu_device *adev = ring->adev; int r; if (size) { - r = amdgpu_sa_bo_new(adev, &adev->ring_tmp_bo, + r = amdgpu_sa_bo_new(&adev->ring_tmp_bo, &ib->sa_bo, size, 256); if (r) { dev_err(adev->dev, "failed to get a new IB (%d)\n", r); @@ -75,10 +74,8 @@ int amdgpu_ib_get(struct amdgpu_ring *ring, struct amdgpu_vm *vm, ib->gpu_addr = amdgpu_sa_bo_gpu_addr(ib->sa_bo); } - amdgpu_sync_create(&ib->sync); - - ib->ring = ring; ib->vm = vm; + ib->vm_id = 0; return 0; } @@ -88,14 +85,13 @@ int amdgpu_ib_get(struct amdgpu_ring *ring, struct amdgpu_vm *vm, * * @adev: amdgpu_device pointer * @ib: IB object to free + * @f: the fence SA bo need wait on for the ib alloation * * Free an IB (all asics). */ -void amdgpu_ib_free(struct amdgpu_device *adev, struct amdgpu_ib *ib) +void amdgpu_ib_free(struct amdgpu_device *adev, struct amdgpu_ib *ib, struct fence *f) { - amdgpu_sync_free(adev, &ib->sync, &ib->fence->base); - amdgpu_sa_bo_free(adev, &ib->sa_bo, &ib->fence->base); - amdgpu_fence_unref(&ib->fence); + amdgpu_sa_bo_free(adev, &ib->sa_bo, f); } /** @@ -104,7 +100,7 @@ void amdgpu_ib_free(struct amdgpu_device *adev, struct amdgpu_ib *ib) * @adev: amdgpu_device pointer * @num_ibs: number of IBs to schedule * @ibs: IB objects to schedule - * @owner: owner for creating the fences + * @f: fence created during this submission * * Schedule an IB on the associated ring (all asics). * Returns 0 on success, error on failure. @@ -119,20 +115,21 @@ void amdgpu_ib_free(struct amdgpu_device *adev, struct amdgpu_ib *ib) * a CONST_IB), it will be put on the ring prior to the DE IB. Prior * to SI there was just a DE IB. */ -int amdgpu_ib_schedule(struct amdgpu_device *adev, unsigned num_ibs, - struct amdgpu_ib *ibs, void *owner) +int amdgpu_ib_schedule(struct amdgpu_ring *ring, unsigned num_ibs, + struct amdgpu_ib *ibs, struct fence *last_vm_update, + struct fence **f) { + struct amdgpu_device *adev = ring->adev; struct amdgpu_ib *ib = &ibs[0]; - struct amdgpu_ring *ring; struct amdgpu_ctx *ctx, *old_ctx; struct amdgpu_vm *vm; + struct fence *hwf; unsigned i; int r = 0; if (num_ibs == 0) return -EINVAL; - ring = ibs->ring; ctx = ibs->ctx; vm = ibs->vm; @@ -140,42 +137,24 @@ int amdgpu_ib_schedule(struct amdgpu_device *adev, unsigned num_ibs, dev_err(adev->dev, "couldn't schedule ib\n"); return -EINVAL; } - r = amdgpu_sync_wait(&ibs->sync); - if (r) { - dev_err(adev->dev, "IB sync failed (%d).\n", r); - return r; - } - r = amdgpu_ring_lock(ring, (256 + AMDGPU_NUM_SYNCS * 8) * num_ibs); - if (r) { - dev_err(adev->dev, "scheduling IB failed (%d).\n", r); - return r; - } - if (vm) { - /* grab a vm id if necessary */ - r = amdgpu_vm_grab_id(ibs->vm, ibs->ring, &ibs->sync); - if (r) { - amdgpu_ring_unlock_undo(ring); - return r; - } + if (vm && !ibs->vm_id) { + dev_err(adev->dev, "VM IB without ID\n"); + return -EINVAL; } - r = amdgpu_sync_rings(&ibs->sync, ring); + r = amdgpu_ring_alloc(ring, 256 * num_ibs); if (r) { - amdgpu_ring_unlock_undo(ring); - dev_err(adev->dev, "failed to sync rings (%d)\n", r); + dev_err(adev->dev, "scheduling IB failed (%d).\n", r); return r; } if (vm) { /* do context switch */ - amdgpu_vm_flush(ring, vm, ib->sync.last_vm_update); - - if (ring->funcs->emit_gds_switch) - amdgpu_ring_emit_gds_switch(ring, ib->vm->ids[ring->idx].id, - ib->gds_base, ib->gds_size, - ib->gws_base, ib->gws_size, - ib->oa_base, ib->oa_size); + amdgpu_vm_flush(ring, ib->vm_id, ib->vm_pd_addr, + ib->gds_base, ib->gds_size, + ib->gws_base, ib->gws_size, + ib->oa_base, ib->oa_size); if (ring->funcs->emit_hdp_flush) amdgpu_ring_emit_hdp_flush(ring); @@ -185,27 +164,32 @@ int amdgpu_ib_schedule(struct amdgpu_device *adev, unsigned num_ibs, for (i = 0; i < num_ibs; ++i) { ib = &ibs[i]; - if (ib->ring != ring || ib->ctx != ctx || ib->vm != vm) { + if (ib->ctx != ctx || ib->vm != vm) { ring->current_ctx = old_ctx; - amdgpu_ring_unlock_undo(ring); + if (ib->vm_id) + amdgpu_vm_reset_id(adev, ib->vm_id); + amdgpu_ring_undo(ring); return -EINVAL; } amdgpu_ring_emit_ib(ring, ib); ring->current_ctx = ctx; } - r = amdgpu_fence_emit(ring, owner, &ib->fence); + if (vm) { + if (ring->funcs->emit_hdp_invalidate) + amdgpu_ring_emit_hdp_invalidate(ring); + } + + r = amdgpu_fence_emit(ring, &hwf); if (r) { dev_err(adev->dev, "failed to emit fence (%d)\n", r); ring->current_ctx = old_ctx; - amdgpu_ring_unlock_undo(ring); + if (ib->vm_id) + amdgpu_vm_reset_id(adev, ib->vm_id); + amdgpu_ring_undo(ring); return r; } - if (!amdgpu_enable_scheduler && ib->ctx) - ib->sequence = amdgpu_ctx_add_fence(ib->ctx, ring, - &ib->fence->base); - /* wrap the last IB with fence */ if (ib->user) { uint64_t addr = amdgpu_bo_gpu_offset(ib->user->bo); @@ -214,10 +198,10 @@ int amdgpu_ib_schedule(struct amdgpu_device *adev, unsigned num_ibs, AMDGPU_FENCE_FLAG_64BIT); } - if (ib->vm) - amdgpu_vm_fence(adev, ib->vm, ib->fence); + if (f) + *f = fence_get(hwf); - amdgpu_ring_unlock_commit(ring); + amdgpu_ring_commit(ring); return 0; } @@ -298,7 +282,6 @@ int amdgpu_ib_ring_tests(struct amdgpu_device *adev) r = amdgpu_ring_test_ib(ring); if (r) { ring->ready = false; - adev->needs_reset = false; if (ring == &adev->gfx.gfx_ring[0]) { /* oh, oh, that's really bad */ diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_irq.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_irq.c index 7c42ff670080..f594cfaa97e5 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_irq.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_irq.c @@ -25,6 +25,7 @@ * Alex Deucher * Jerome Glisse */ +#include <linux/irq.h> #include <drm/drmP.h> #include <drm/drm_crtc_helper.h> #include <drm/amdgpu_drm.h> @@ -312,6 +313,7 @@ int amdgpu_irq_add_id(struct amdgpu_device *adev, unsigned src_id, } adev->irq.sources[src_id] = source; + return 0; } @@ -335,15 +337,19 @@ void amdgpu_irq_dispatch(struct amdgpu_device *adev, return; } - src = adev->irq.sources[src_id]; - if (!src) { - DRM_DEBUG("Unhandled interrupt src_id: %d\n", src_id); - return; - } + if (adev->irq.virq[src_id]) { + generic_handle_irq(irq_find_mapping(adev->irq.domain, src_id)); + } else { + src = adev->irq.sources[src_id]; + if (!src) { + DRM_DEBUG("Unhandled interrupt src_id: %d\n", src_id); + return; + } - r = src->funcs->process(adev, src, entry); - if (r) - DRM_ERROR("error processing interrupt (%d)\n", r); + r = src->funcs->process(adev, src, entry); + if (r) + DRM_ERROR("error processing interrupt (%d)\n", r); + } } /** @@ -461,3 +467,90 @@ bool amdgpu_irq_enabled(struct amdgpu_device *adev, struct amdgpu_irq_src *src, return !!atomic_read(&src->enabled_types[type]); } + +/* gen irq */ +static void amdgpu_irq_mask(struct irq_data *irqd) +{ + /* XXX */ +} + +static void amdgpu_irq_unmask(struct irq_data *irqd) +{ + /* XXX */ +} + +static struct irq_chip amdgpu_irq_chip = { + .name = "amdgpu-ih", + .irq_mask = amdgpu_irq_mask, + .irq_unmask = amdgpu_irq_unmask, +}; + +static int amdgpu_irqdomain_map(struct irq_domain *d, + unsigned int irq, irq_hw_number_t hwirq) +{ + if (hwirq >= AMDGPU_MAX_IRQ_SRC_ID) + return -EPERM; + + irq_set_chip_and_handler(irq, + &amdgpu_irq_chip, handle_simple_irq); + return 0; +} + +static struct irq_domain_ops amdgpu_hw_irqdomain_ops = { + .map = amdgpu_irqdomain_map, +}; + +/** + * amdgpu_irq_add_domain - create a linear irq domain + * + * @adev: amdgpu device pointer + * + * Create an irq domain for GPU interrupt sources + * that may be driven by another driver (e.g., ACP). + */ +int amdgpu_irq_add_domain(struct amdgpu_device *adev) +{ + adev->irq.domain = irq_domain_add_linear(NULL, AMDGPU_MAX_IRQ_SRC_ID, + &amdgpu_hw_irqdomain_ops, adev); + if (!adev->irq.domain) { + DRM_ERROR("GPU irq add domain failed\n"); + return -ENODEV; + } + + return 0; +} + +/** + * amdgpu_irq_remove_domain - remove the irq domain + * + * @adev: amdgpu device pointer + * + * Remove the irq domain for GPU interrupt sources + * that may be driven by another driver (e.g., ACP). + */ +void amdgpu_irq_remove_domain(struct amdgpu_device *adev) +{ + if (adev->irq.domain) { + irq_domain_remove(adev->irq.domain); + adev->irq.domain = NULL; + } +} + +/** + * amdgpu_irq_create_mapping - create a mapping between a domain irq and a + * Linux irq + * + * @adev: amdgpu device pointer + * @src_id: IH source id + * + * Create a mapping between a domain irq (GPU IH src id) and a Linux irq + * Use this for components that generate a GPU interrupt, but are driven + * by a different driver (e.g., ACP). + * Returns the Linux irq. + */ +unsigned amdgpu_irq_create_mapping(struct amdgpu_device *adev, unsigned src_id) +{ + adev->irq.virq[src_id] = irq_create_mapping(adev->irq.domain, src_id); + + return adev->irq.virq[src_id]; +} diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_irq.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_irq.h index 17b01aef4278..e124b59f39c1 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_irq.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_irq.h @@ -24,6 +24,7 @@ #ifndef __AMDGPU_IRQ_H__ #define __AMDGPU_IRQ_H__ +#include <linux/irqdomain.h> #include "amdgpu_ih.h" #define AMDGPU_MAX_IRQ_SRC_ID 0x100 @@ -65,6 +66,10 @@ struct amdgpu_irq { /* interrupt ring */ struct amdgpu_ih_ring ih; const struct amdgpu_ih_funcs *ih_funcs; + + /* gen irq stuff */ + struct irq_domain *domain; /* GPU irq controller domain */ + unsigned virq[AMDGPU_MAX_IRQ_SRC_ID]; }; void amdgpu_irq_preinstall(struct drm_device *dev); @@ -90,4 +95,8 @@ int amdgpu_irq_put(struct amdgpu_device *adev, struct amdgpu_irq_src *src, bool amdgpu_irq_enabled(struct amdgpu_device *adev, struct amdgpu_irq_src *src, unsigned type); +int amdgpu_irq_add_domain(struct amdgpu_device *adev); +void amdgpu_irq_remove_domain(struct amdgpu_device *adev); +unsigned amdgpu_irq_create_mapping(struct amdgpu_device *adev, unsigned src_id); + #endif diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c new file mode 100644 index 000000000000..9c9b19e2f353 --- /dev/null +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c @@ -0,0 +1,171 @@ +/* + * Copyright 2015 Advanced Micro Devices, Inc. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice shall be included in + * all copies or substantial portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR + * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, + * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR + * OTHER DEALINGS IN THE SOFTWARE. + * + * + */ +#include <linux/kthread.h> +#include <linux/wait.h> +#include <linux/sched.h> +#include <drm/drmP.h> +#include "amdgpu.h" +#include "amdgpu_trace.h" + +int amdgpu_job_alloc(struct amdgpu_device *adev, unsigned num_ibs, + struct amdgpu_job **job) +{ + size_t size = sizeof(struct amdgpu_job); + + if (num_ibs == 0) + return -EINVAL; + + size += sizeof(struct amdgpu_ib) * num_ibs; + + *job = kzalloc(size, GFP_KERNEL); + if (!*job) + return -ENOMEM; + + (*job)->adev = adev; + (*job)->ibs = (void *)&(*job)[1]; + (*job)->num_ibs = num_ibs; + + amdgpu_sync_create(&(*job)->sync); + + return 0; +} + +int amdgpu_job_alloc_with_ib(struct amdgpu_device *adev, unsigned size, + struct amdgpu_job **job) +{ + int r; + + r = amdgpu_job_alloc(adev, 1, job); + if (r) + return r; + + r = amdgpu_ib_get(adev, NULL, size, &(*job)->ibs[0]); + if (r) + kfree(*job); + + return r; +} + +void amdgpu_job_free(struct amdgpu_job *job) +{ + unsigned i; + struct fence *f; + /* use sched fence if available */ + f = (job->base.s_fence)? &job->base.s_fence->base : job->fence; + + for (i = 0; i < job->num_ibs; ++i) + amdgpu_sa_bo_free(job->adev, &job->ibs[i].sa_bo, f); + fence_put(job->fence); + + amdgpu_bo_unref(&job->uf.bo); + amdgpu_sync_free(&job->sync); + kfree(job); +} + +int amdgpu_job_submit(struct amdgpu_job *job, struct amdgpu_ring *ring, + struct amd_sched_entity *entity, void *owner, + struct fence **f) +{ + job->ring = ring; + job->base.sched = &ring->sched; + job->base.s_entity = entity; + job->base.s_fence = amd_sched_fence_create(job->base.s_entity, owner); + if (!job->base.s_fence) + return -ENOMEM; + + *f = fence_get(&job->base.s_fence->base); + + job->owner = owner; + amd_sched_entity_push_job(&job->base); + + return 0; +} + +static struct fence *amdgpu_job_dependency(struct amd_sched_job *sched_job) +{ + struct amdgpu_job *job = to_amdgpu_job(sched_job); + struct amdgpu_vm *vm = job->ibs->vm; + + struct fence *fence = amdgpu_sync_get_fence(&job->sync); + + if (fence == NULL && vm && !job->ibs->vm_id) { + struct amdgpu_ring *ring = job->ring; + unsigned i, vm_id; + uint64_t vm_pd_addr; + int r; + + r = amdgpu_vm_grab_id(vm, ring, &job->sync, + &job->base.s_fence->base, + &vm_id, &vm_pd_addr); + if (r) + DRM_ERROR("Error getting VM ID (%d)\n", r); + else { + for (i = 0; i < job->num_ibs; ++i) { + job->ibs[i].vm_id = vm_id; + job->ibs[i].vm_pd_addr = vm_pd_addr; + } + } + + fence = amdgpu_sync_get_fence(&job->sync); + } + + return fence; +} + +static struct fence *amdgpu_job_run(struct amd_sched_job *sched_job) +{ + struct fence *fence = NULL; + struct amdgpu_job *job; + int r; + + if (!sched_job) { + DRM_ERROR("job is null\n"); + return NULL; + } + job = to_amdgpu_job(sched_job); + + r = amdgpu_sync_wait(&job->sync); + if (r) { + DRM_ERROR("failed to sync wait (%d)\n", r); + return NULL; + } + + trace_amdgpu_sched_run_job(job); + r = amdgpu_ib_schedule(job->ring, job->num_ibs, job->ibs, + job->sync.last_vm_update, &fence); + if (r) { + DRM_ERROR("Error scheduling IBs (%d)\n", r); + goto err; + } + +err: + job->fence = fence; + amdgpu_job_free(job); + return fence; +} + +struct amd_sched_backend_ops amdgpu_sched_ops = { + .dependency = amdgpu_job_dependency, + .run_job = amdgpu_job_run, +}; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c index 5d11e798230c..7805a8706af7 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c @@ -218,8 +218,8 @@ static int amdgpu_info_ioctl(struct drm_device *dev, void *data, struct drm_file break; case AMDGPU_HW_IP_DMA: type = AMD_IP_BLOCK_TYPE_SDMA; - ring_mask = adev->sdma[0].ring.ready ? 1 : 0; - ring_mask |= ((adev->sdma[1].ring.ready ? 1 : 0) << 1); + for (i = 0; i < adev->sdma.num_instances; i++) + ring_mask |= ((adev->sdma.instance[i].ring.ready ? 1 : 0) << i); ib_start_alignment = AMDGPU_GPU_PAGE_SIZE; ib_size_alignment = 1; break; @@ -341,10 +341,10 @@ static int amdgpu_info_ioctl(struct drm_device *dev, void *data, struct drm_file fw_info.feature = 0; break; case AMDGPU_INFO_FW_SDMA: - if (info->query_fw.index >= 2) + if (info->query_fw.index >= adev->sdma.num_instances) return -EINVAL; - fw_info.ver = adev->sdma[info->query_fw.index].fw_version; - fw_info.feature = adev->sdma[info->query_fw.index].feature_version; + fw_info.ver = adev->sdma.instance[info->query_fw.index].fw_version; + fw_info.feature = adev->sdma.instance[info->query_fw.index].feature_version; break; default: return -EINVAL; @@ -447,8 +447,7 @@ static int amdgpu_info_ioctl(struct drm_device *dev, void *data, struct drm_file dev_info.max_memory_clock = adev->pm.default_mclk * 10; } dev_info.enabled_rb_pipes_mask = adev->gfx.config.backend_enable_mask; - dev_info.num_rb_pipes = adev->gfx.config.max_backends_per_se * - adev->gfx.config.max_shader_engines; + dev_info.num_rb_pipes = adev->gfx.config.num_rbs; dev_info.num_hw_gfx_contexts = adev->gfx.config.max_hw_contexts; dev_info._pad = 0; dev_info.ids_flags = 0; @@ -489,7 +488,7 @@ static int amdgpu_info_ioctl(struct drm_device *dev, void *data, struct drm_file * * @dev: drm dev pointer * - * Switch vga switcheroo state after last close (all asics). + * Switch vga_switcheroo state after last close (all asics). */ void amdgpu_driver_lastclose_kms(struct drm_device *dev) { @@ -603,36 +602,82 @@ void amdgpu_driver_preclose_kms(struct drm_device *dev, * amdgpu_get_vblank_counter_kms - get frame count * * @dev: drm dev pointer - * @crtc: crtc to get the frame count from + * @pipe: crtc to get the frame count from * * Gets the frame count on the requested crtc (all asics). * Returns frame count on success, -EINVAL on failure. */ -u32 amdgpu_get_vblank_counter_kms(struct drm_device *dev, int crtc) +u32 amdgpu_get_vblank_counter_kms(struct drm_device *dev, unsigned int pipe) { struct amdgpu_device *adev = dev->dev_private; + int vpos, hpos, stat; + u32 count; - if (crtc < 0 || crtc >= adev->mode_info.num_crtc) { - DRM_ERROR("Invalid crtc %d\n", crtc); + if (pipe >= adev->mode_info.num_crtc) { + DRM_ERROR("Invalid crtc %u\n", pipe); return -EINVAL; } - return amdgpu_display_vblank_get_counter(adev, crtc); + /* The hw increments its frame counter at start of vsync, not at start + * of vblank, as is required by DRM core vblank counter handling. + * Cook the hw count here to make it appear to the caller as if it + * incremented at start of vblank. We measure distance to start of + * vblank in vpos. vpos therefore will be >= 0 between start of vblank + * and start of vsync, so vpos >= 0 means to bump the hw frame counter + * result by 1 to give the proper appearance to caller. + */ + if (adev->mode_info.crtcs[pipe]) { + /* Repeat readout if needed to provide stable result if + * we cross start of vsync during the queries. + */ + do { + count = amdgpu_display_vblank_get_counter(adev, pipe); + /* Ask amdgpu_get_crtc_scanoutpos to return vpos as + * distance to start of vblank, instead of regular + * vertical scanout pos. + */ + stat = amdgpu_get_crtc_scanoutpos( + dev, pipe, GET_DISTANCE_TO_VBLANKSTART, + &vpos, &hpos, NULL, NULL, + &adev->mode_info.crtcs[pipe]->base.hwmode); + } while (count != amdgpu_display_vblank_get_counter(adev, pipe)); + + if (((stat & (DRM_SCANOUTPOS_VALID | DRM_SCANOUTPOS_ACCURATE)) != + (DRM_SCANOUTPOS_VALID | DRM_SCANOUTPOS_ACCURATE))) { + DRM_DEBUG_VBL("Query failed! stat %d\n", stat); + } else { + DRM_DEBUG_VBL("crtc %d: dist from vblank start %d\n", + pipe, vpos); + + /* Bump counter if we are at >= leading edge of vblank, + * but before vsync where vpos would turn negative and + * the hw counter really increments. + */ + if (vpos >= 0) + count++; + } + } else { + /* Fallback to use value as is. */ + count = amdgpu_display_vblank_get_counter(adev, pipe); + DRM_DEBUG_VBL("NULL mode info! Returned count may be wrong.\n"); + } + + return count; } /** * amdgpu_enable_vblank_kms - enable vblank interrupt * * @dev: drm dev pointer - * @crtc: crtc to enable vblank interrupt for + * @pipe: crtc to enable vblank interrupt for * * Enable the interrupt on the requested crtc (all asics). * Returns 0 on success, -EINVAL on failure. */ -int amdgpu_enable_vblank_kms(struct drm_device *dev, int crtc) +int amdgpu_enable_vblank_kms(struct drm_device *dev, unsigned int pipe) { struct amdgpu_device *adev = dev->dev_private; - int idx = amdgpu_crtc_idx_to_irq_type(adev, crtc); + int idx = amdgpu_crtc_idx_to_irq_type(adev, pipe); return amdgpu_irq_get(adev, &adev->crtc_irq, idx); } @@ -641,14 +686,14 @@ int amdgpu_enable_vblank_kms(struct drm_device *dev, int crtc) * amdgpu_disable_vblank_kms - disable vblank interrupt * * @dev: drm dev pointer - * @crtc: crtc to disable vblank interrupt for + * @pipe: crtc to disable vblank interrupt for * * Disable the interrupt on the requested crtc (all asics). */ -void amdgpu_disable_vblank_kms(struct drm_device *dev, int crtc) +void amdgpu_disable_vblank_kms(struct drm_device *dev, unsigned int pipe) { struct amdgpu_device *adev = dev->dev_private; - int idx = amdgpu_crtc_idx_to_irq_type(adev, crtc); + int idx = amdgpu_crtc_idx_to_irq_type(adev, pipe); amdgpu_irq_put(adev, &adev->crtc_irq, idx); } @@ -666,41 +711,47 @@ void amdgpu_disable_vblank_kms(struct drm_device *dev, int crtc) * scanout position. (all asics). * Returns postive status flags on success, negative error on failure. */ -int amdgpu_get_vblank_timestamp_kms(struct drm_device *dev, int crtc, +int amdgpu_get_vblank_timestamp_kms(struct drm_device *dev, unsigned int pipe, int *max_error, struct timeval *vblank_time, unsigned flags) { - struct drm_crtc *drmcrtc; + struct drm_crtc *crtc; struct amdgpu_device *adev = dev->dev_private; - if (crtc < 0 || crtc >= dev->num_crtcs) { - DRM_ERROR("Invalid crtc %d\n", crtc); + if (pipe >= dev->num_crtcs) { + DRM_ERROR("Invalid crtc %u\n", pipe); return -EINVAL; } /* Get associated drm_crtc: */ - drmcrtc = &adev->mode_info.crtcs[crtc]->base; + crtc = &adev->mode_info.crtcs[pipe]->base; + if (!crtc) { + /* This can occur on driver load if some component fails to + * initialize completely and driver is unloaded */ + DRM_ERROR("Uninitialized crtc %d\n", pipe); + return -EINVAL; + } /* Helper routine in DRM core does all the work: */ - return drm_calc_vbltimestamp_from_scanoutpos(dev, crtc, max_error, + return drm_calc_vbltimestamp_from_scanoutpos(dev, pipe, max_error, vblank_time, flags, - drmcrtc, &drmcrtc->hwmode); + &crtc->hwmode); } const struct drm_ioctl_desc amdgpu_ioctls_kms[] = { - DRM_IOCTL_DEF_DRV(AMDGPU_GEM_CREATE, amdgpu_gem_create_ioctl, DRM_AUTH|DRM_UNLOCKED|DRM_RENDER_ALLOW), - DRM_IOCTL_DEF_DRV(AMDGPU_CTX, amdgpu_ctx_ioctl, DRM_AUTH|DRM_UNLOCKED|DRM_RENDER_ALLOW), - DRM_IOCTL_DEF_DRV(AMDGPU_BO_LIST, amdgpu_bo_list_ioctl, DRM_AUTH|DRM_UNLOCKED|DRM_RENDER_ALLOW), + DRM_IOCTL_DEF_DRV(AMDGPU_GEM_CREATE, amdgpu_gem_create_ioctl, DRM_AUTH|DRM_RENDER_ALLOW), + DRM_IOCTL_DEF_DRV(AMDGPU_CTX, amdgpu_ctx_ioctl, DRM_AUTH|DRM_RENDER_ALLOW), + DRM_IOCTL_DEF_DRV(AMDGPU_BO_LIST, amdgpu_bo_list_ioctl, DRM_AUTH|DRM_RENDER_ALLOW), /* KMS */ - DRM_IOCTL_DEF_DRV(AMDGPU_GEM_MMAP, amdgpu_gem_mmap_ioctl, DRM_AUTH|DRM_UNLOCKED|DRM_RENDER_ALLOW), - DRM_IOCTL_DEF_DRV(AMDGPU_GEM_WAIT_IDLE, amdgpu_gem_wait_idle_ioctl, DRM_AUTH|DRM_UNLOCKED|DRM_RENDER_ALLOW), - DRM_IOCTL_DEF_DRV(AMDGPU_CS, amdgpu_cs_ioctl, DRM_AUTH|DRM_UNLOCKED|DRM_RENDER_ALLOW), - DRM_IOCTL_DEF_DRV(AMDGPU_INFO, amdgpu_info_ioctl, DRM_AUTH|DRM_UNLOCKED|DRM_RENDER_ALLOW), - DRM_IOCTL_DEF_DRV(AMDGPU_WAIT_CS, amdgpu_cs_wait_ioctl, DRM_AUTH|DRM_UNLOCKED|DRM_RENDER_ALLOW), - DRM_IOCTL_DEF_DRV(AMDGPU_GEM_METADATA, amdgpu_gem_metadata_ioctl, DRM_AUTH|DRM_UNLOCKED|DRM_RENDER_ALLOW), - DRM_IOCTL_DEF_DRV(AMDGPU_GEM_VA, amdgpu_gem_va_ioctl, DRM_AUTH|DRM_UNLOCKED|DRM_RENDER_ALLOW), - DRM_IOCTL_DEF_DRV(AMDGPU_GEM_OP, amdgpu_gem_op_ioctl, DRM_AUTH|DRM_UNLOCKED|DRM_RENDER_ALLOW), - DRM_IOCTL_DEF_DRV(AMDGPU_GEM_USERPTR, amdgpu_gem_userptr_ioctl, DRM_AUTH|DRM_UNLOCKED|DRM_RENDER_ALLOW), + DRM_IOCTL_DEF_DRV(AMDGPU_GEM_MMAP, amdgpu_gem_mmap_ioctl, DRM_AUTH|DRM_RENDER_ALLOW), + DRM_IOCTL_DEF_DRV(AMDGPU_GEM_WAIT_IDLE, amdgpu_gem_wait_idle_ioctl, DRM_AUTH|DRM_RENDER_ALLOW), + DRM_IOCTL_DEF_DRV(AMDGPU_CS, amdgpu_cs_ioctl, DRM_AUTH|DRM_RENDER_ALLOW), + DRM_IOCTL_DEF_DRV(AMDGPU_INFO, amdgpu_info_ioctl, DRM_AUTH|DRM_RENDER_ALLOW), + DRM_IOCTL_DEF_DRV(AMDGPU_WAIT_CS, amdgpu_cs_wait_ioctl, DRM_AUTH|DRM_RENDER_ALLOW), + DRM_IOCTL_DEF_DRV(AMDGPU_GEM_METADATA, amdgpu_gem_metadata_ioctl, DRM_AUTH|DRM_RENDER_ALLOW), + DRM_IOCTL_DEF_DRV(AMDGPU_GEM_VA, amdgpu_gem_va_ioctl, DRM_AUTH|DRM_RENDER_ALLOW), + DRM_IOCTL_DEF_DRV(AMDGPU_GEM_OP, amdgpu_gem_op_ioctl, DRM_AUTH|DRM_RENDER_ALLOW), + DRM_IOCTL_DEF_DRV(AMDGPU_GEM_USERPTR, amdgpu_gem_userptr_ioctl, DRM_AUTH|DRM_RENDER_ALLOW), }; int amdgpu_max_kms_ioctl = ARRAY_SIZE(amdgpu_ioctls_kms); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_mn.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_mn.c index b1969f2b2038..9f4a45cd2aab 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_mn.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_mn.c @@ -77,8 +77,6 @@ static void amdgpu_mn_destroy(struct work_struct *work) hash_del(&rmn->node); rbtree_postorder_for_each_entry_safe(node, next_node, &rmn->objects, it.rb) { - - interval_tree_remove(&node->it, &rmn->objects); list_for_each_entry_safe(bo, next_bo, &node->bos, mn_list) { bo->mn = NULL; list_del_init(&bo->mn_list); @@ -87,7 +85,7 @@ static void amdgpu_mn_destroy(struct work_struct *work) } mutex_unlock(&rmn->lock); mutex_unlock(&adev->mn_lock); - mmu_notifier_unregister(&rmn->mn, rmn->mm); + mmu_notifier_unregister_no_release(&rmn->mn, rmn->mm); kfree(rmn); } @@ -108,6 +106,76 @@ static void amdgpu_mn_release(struct mmu_notifier *mn, } /** + * amdgpu_mn_invalidate_node - unmap all BOs of a node + * + * @node: the node with the BOs to unmap + * + * We block for all BOs and unmap them by move them + * into system domain again. + */ +static void amdgpu_mn_invalidate_node(struct amdgpu_mn_node *node, + unsigned long start, + unsigned long end) +{ + struct amdgpu_bo *bo; + long r; + + list_for_each_entry(bo, &node->bos, mn_list) { + + if (!amdgpu_ttm_tt_affect_userptr(bo->tbo.ttm, start, end)) + continue; + + r = amdgpu_bo_reserve(bo, true); + if (r) { + DRM_ERROR("(%ld) failed to reserve user bo\n", r); + continue; + } + + r = reservation_object_wait_timeout_rcu(bo->tbo.resv, + true, false, MAX_SCHEDULE_TIMEOUT); + if (r <= 0) + DRM_ERROR("(%ld) failed to wait for user bo\n", r); + + amdgpu_ttm_placement_from_domain(bo, AMDGPU_GEM_DOMAIN_CPU); + r = ttm_bo_validate(&bo->tbo, &bo->placement, false, false); + if (r) + DRM_ERROR("(%ld) failed to validate user bo\n", r); + + amdgpu_bo_unreserve(bo); + } +} + +/** + * amdgpu_mn_invalidate_page - callback to notify about mm change + * + * @mn: our notifier + * @mn: the mm this callback is about + * @address: address of invalidate page + * + * Invalidation of a single page. Blocks for all BOs mapping it + * and unmap them by move them into system domain again. + */ +static void amdgpu_mn_invalidate_page(struct mmu_notifier *mn, + struct mm_struct *mm, + unsigned long address) +{ + struct amdgpu_mn *rmn = container_of(mn, struct amdgpu_mn, mn); + struct interval_tree_node *it; + + mutex_lock(&rmn->lock); + + it = interval_tree_iter_first(&rmn->objects, address, address); + if (it) { + struct amdgpu_mn_node *node; + + node = container_of(it, struct amdgpu_mn_node, it); + amdgpu_mn_invalidate_node(node, address, address); + } + + mutex_unlock(&rmn->lock); +} + +/** * amdgpu_mn_invalidate_range_start - callback to notify about mm change * * @mn: our notifier @@ -134,35 +202,11 @@ static void amdgpu_mn_invalidate_range_start(struct mmu_notifier *mn, it = interval_tree_iter_first(&rmn->objects, start, end); while (it) { struct amdgpu_mn_node *node; - struct amdgpu_bo *bo; - long r; node = container_of(it, struct amdgpu_mn_node, it); it = interval_tree_iter_next(it, start, end); - list_for_each_entry(bo, &node->bos, mn_list) { - - if (!bo->tbo.ttm || bo->tbo.ttm->state != tt_bound) - continue; - - r = amdgpu_bo_reserve(bo, true); - if (r) { - DRM_ERROR("(%ld) failed to reserve user bo\n", r); - continue; - } - - r = reservation_object_wait_timeout_rcu(bo->tbo.resv, - true, false, MAX_SCHEDULE_TIMEOUT); - if (r <= 0) - DRM_ERROR("(%ld) failed to wait for user bo\n", r); - - amdgpu_ttm_placement_from_domain(bo, AMDGPU_GEM_DOMAIN_CPU); - r = ttm_bo_validate(&bo->tbo, &bo->placement, false, false); - if (r) - DRM_ERROR("(%ld) failed to validate user bo\n", r); - - amdgpu_bo_unreserve(bo); - } + amdgpu_mn_invalidate_node(node, start, end); } mutex_unlock(&rmn->lock); @@ -170,6 +214,7 @@ static void amdgpu_mn_invalidate_range_start(struct mmu_notifier *mn, static const struct mmu_notifier_ops amdgpu_mn_ops = { .release = amdgpu_mn_release, + .invalidate_page = amdgpu_mn_invalidate_page, .invalidate_range_start = amdgpu_mn_invalidate_range_start, }; @@ -186,8 +231,8 @@ static struct amdgpu_mn *amdgpu_mn_get(struct amdgpu_device *adev) struct amdgpu_mn *rmn; int r; - down_write(&mm->mmap_sem); mutex_lock(&adev->mn_lock); + down_write(&mm->mmap_sem); hash_for_each_possible(adev->mn_hash, rmn, node, (unsigned long)mm) if (rmn->mm == mm) @@ -212,14 +257,14 @@ static struct amdgpu_mn *amdgpu_mn_get(struct amdgpu_device *adev) hash_add(adev->mn_hash, &rmn->node, (unsigned long)mm); release_locks: - mutex_unlock(&adev->mn_lock); up_write(&mm->mmap_sem); + mutex_unlock(&adev->mn_lock); return rmn; free_rmn: - mutex_unlock(&adev->mn_lock); up_write(&mm->mmap_sem); + mutex_unlock(&adev->mn_lock); kfree(rmn); return ERR_PTR(r); @@ -297,6 +342,7 @@ void amdgpu_mn_unregister(struct amdgpu_bo *bo) struct list_head *head; mutex_lock(&adev->mn_lock); + rmn = bo->mn; if (rmn == NULL) { mutex_unlock(&adev->mn_lock); @@ -304,6 +350,7 @@ void amdgpu_mn_unregister(struct amdgpu_bo *bo) } mutex_lock(&rmn->lock); + /* save the next list entry for later */ head = bo->mn_list.next; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_mode.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_mode.h index 7bd470d9ac30..8d432e6901af 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_mode.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_mode.h @@ -35,6 +35,7 @@ #include <drm/drm_dp_helper.h> #include <drm/drm_fixed.h> #include <drm/drm_crtc_helper.h> +#include <drm/drm_fb_helper.h> #include <drm/drm_plane_helper.h> #include <linux/i2c.h> #include <linux/i2c-algo-bit.h> @@ -373,6 +374,10 @@ struct amdgpu_crtc { uint32_t crtc_offset; struct drm_gem_object *cursor_bo; uint64_t cursor_addr; + int cursor_x; + int cursor_y; + int cursor_hot_x; + int cursor_hot_y; int cursor_width; int cursor_height; int max_cursor_width; @@ -385,7 +390,6 @@ struct amdgpu_crtc { struct drm_display_mode native_mode; u32 pll_id; /* page flipping */ - struct workqueue_struct *pflip_queue; struct amdgpu_flip_work *pflip_works; enum amdgpu_flip_status pflip_status; int deferred_flip_completion; @@ -403,6 +407,7 @@ struct amdgpu_crtc { u32 line_time; u32 wm_low; u32 wm_high; + u32 lb_vblank_lead_lines; struct drm_display_mode hw_mode; }; @@ -524,6 +529,10 @@ struct amdgpu_framebuffer { #define ENCODER_MODE_IS_DP(em) (((em) == ATOM_ENCODER_MODE_DP) || \ ((em) == ATOM_ENCODER_MODE_DP_MST)) +/* Driver internal use only flags of amdgpu_get_crtc_scanoutpos() */ +#define USE_REAL_VBLANKSTART (1 << 30) +#define GET_DISTANCE_TO_VBLANKSTART (1 << 31) + void amdgpu_link_encoder_connector(struct drm_device *dev); struct drm_connector * @@ -540,14 +549,14 @@ bool amdgpu_ddc_probe(struct amdgpu_connector *amdgpu_connector, bool use_aux); void amdgpu_encoder_set_active_device(struct drm_encoder *encoder); -int amdgpu_get_crtc_scanoutpos(struct drm_device *dev, int crtc, - unsigned int flags, - int *vpos, int *hpos, ktime_t *stime, - ktime_t *etime); +int amdgpu_get_crtc_scanoutpos(struct drm_device *dev, unsigned int pipe, + unsigned int flags, int *vpos, int *hpos, + ktime_t *stime, ktime_t *etime, + const struct drm_display_mode *mode); int amdgpu_framebuffer_init(struct drm_device *dev, struct amdgpu_framebuffer *rfb, - struct drm_mode_fb_cmd2 *mode_cmd, + const struct drm_mode_fb_cmd2 *mode_cmd, struct drm_gem_object *obj); int amdgpufb_remove(struct drm_device *dev, struct drm_framebuffer *fb); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c index 1a7708f365f3..151a2d42c639 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c @@ -33,6 +33,7 @@ #include <linux/slab.h> #include <drm/drmP.h> #include <drm/amdgpu_drm.h> +#include <drm/drm_cache.h> #include "amdgpu.h" #include "amdgpu_trace.h" @@ -96,10 +97,8 @@ static void amdgpu_ttm_bo_destroy(struct ttm_buffer_object *tbo) amdgpu_update_memory_usage(bo->adev, &bo->tbo.mem, NULL); - mutex_lock(&bo->adev->gem.mutex); - list_del_init(&bo->list); - mutex_unlock(&bo->adev->gem.mutex); drm_gem_object_release(&bo->gem_base); + amdgpu_bo_unref(&bo->parent); kfree(bo->metadata); kfree(bo); } @@ -132,6 +131,8 @@ static void amdgpu_ttm_placement_init(struct amdgpu_device *adev, placements[c].fpfn = 0; placements[c++].flags = TTM_PL_FLAG_WC | TTM_PL_FLAG_UNCACHED | TTM_PL_FLAG_VRAM; + if (!(flags & AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED)) + placements[c - 1].flags |= TTM_PL_FLAG_TOPDOWN; } if (domain & AMDGPU_GEM_DOMAIN_GTT) { @@ -250,14 +251,24 @@ int amdgpu_bo_create_restricted(struct amdgpu_device *adev, bo->adev = adev; INIT_LIST_HEAD(&bo->list); INIT_LIST_HEAD(&bo->va); - bo->initial_domain = domain & (AMDGPU_GEM_DOMAIN_VRAM | - AMDGPU_GEM_DOMAIN_GTT | - AMDGPU_GEM_DOMAIN_CPU | - AMDGPU_GEM_DOMAIN_GDS | - AMDGPU_GEM_DOMAIN_GWS | - AMDGPU_GEM_DOMAIN_OA); + bo->prefered_domains = domain & (AMDGPU_GEM_DOMAIN_VRAM | + AMDGPU_GEM_DOMAIN_GTT | + AMDGPU_GEM_DOMAIN_CPU | + AMDGPU_GEM_DOMAIN_GDS | + AMDGPU_GEM_DOMAIN_GWS | + AMDGPU_GEM_DOMAIN_OA); + bo->allowed_domains = bo->prefered_domains; + if (!kernel && bo->allowed_domains == AMDGPU_GEM_DOMAIN_VRAM) + bo->allowed_domains |= AMDGPU_GEM_DOMAIN_GTT; bo->flags = flags; + + /* For architectures that don't support WC memory, + * mask out the WC flag from the BO + */ + if (!drm_arch_can_wc_memory()) + bo->flags &= ~AMDGPU_GEM_CREATE_CPU_GTT_USWC; + amdgpu_fill_placement_to_bo(bo, placement); /* Kernel allocation are uninterruptible */ r = ttm_bo_init(&adev->mman.bdev, &bo->tbo, size, type, @@ -297,7 +308,7 @@ int amdgpu_bo_create(struct amdgpu_device *adev, int amdgpu_bo_kmap(struct amdgpu_bo *bo, void **ptr) { bool is_iomem; - int r; + long r; if (bo->flags & AMDGPU_GEM_CREATE_NO_CPU_ACCESS) return -EPERM; @@ -308,14 +319,20 @@ int amdgpu_bo_kmap(struct amdgpu_bo *bo, void **ptr) } return 0; } + + r = reservation_object_wait_timeout_rcu(bo->tbo.resv, false, false, + MAX_SCHEDULE_TIMEOUT); + if (r < 0) + return r; + r = ttm_bo_kmap(&bo->tbo, 0, bo->tbo.num_pages, &bo->kmap); - if (r) { + if (r) return r; - } + bo->kptr = ttm_kmap_obj_virtual(&bo->kmap, &is_iomem); - if (ptr) { + if (ptr) *ptr = bo->kptr; - } + return 0; } @@ -356,7 +373,7 @@ int amdgpu_bo_pin_restricted(struct amdgpu_bo *bo, u32 domain, int r, i; unsigned fpfn, lpfn; - if (amdgpu_ttm_tt_has_userptr(bo->tbo.ttm)) + if (amdgpu_ttm_tt_get_usermm(bo->tbo.ttm)) return -EPERM; if (WARN_ON_ONCE(min_offset > max_offset)) @@ -396,7 +413,8 @@ int amdgpu_bo_pin_restricted(struct amdgpu_bo *bo, u32 domain, } if (fpfn > bo->placements[i].fpfn) bo->placements[i].fpfn = fpfn; - if (lpfn && lpfn < bo->placements[i].lpfn) + if (!bo->placements[i].lpfn || + (lpfn && lpfn < bo->placements[i].lpfn)) bo->placements[i].lpfn = lpfn; bo->placements[i].flags |= TTM_PL_FLAG_NO_EVICT; } @@ -458,26 +476,6 @@ int amdgpu_bo_evict_vram(struct amdgpu_device *adev) return ttm_bo_evict_mm(&adev->mman.bdev, TTM_PL_VRAM); } -void amdgpu_bo_force_delete(struct amdgpu_device *adev) -{ - struct amdgpu_bo *bo, *n; - - if (list_empty(&adev->gem.objects)) { - return; - } - dev_err(adev->dev, "Userspace still has active objects !\n"); - list_for_each_entry_safe(bo, n, &adev->gem.objects, list) { - dev_err(adev->dev, "%p %p %lu %lu force free\n", - &bo->gem_base, bo, (unsigned long)bo->gem_base.size, - *((unsigned long *)&bo->gem_base.refcount)); - mutex_lock(&bo->adev->gem.mutex); - list_del_init(&bo->list); - mutex_unlock(&bo->adev->gem.mutex); - /* this should unref the ttm bo */ - drm_gem_object_unreference_unlocked(&bo->gem_base); - } -} - int amdgpu_bo_init(struct amdgpu_device *adev) { /* Add an MTRR for the VRAM */ diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h index 3c2ff4567798..acc08018c6cc 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h @@ -96,6 +96,7 @@ static inline void amdgpu_bo_unreserve(struct amdgpu_bo *bo) */ static inline u64 amdgpu_bo_gpu_offset(struct amdgpu_bo *bo) { + WARN_ON_ONCE(bo->tbo.mem.mem_type == TTM_PL_SYSTEM); return bo->tbo.offset; } @@ -148,7 +149,6 @@ int amdgpu_bo_pin_restricted(struct amdgpu_bo *bo, u32 domain, u64 *gpu_addr); int amdgpu_bo_unpin(struct amdgpu_bo *bo); int amdgpu_bo_evict_vram(struct amdgpu_device *adev); -void amdgpu_bo_force_delete(struct amdgpu_device *adev); int amdgpu_bo_init(struct amdgpu_device *adev); void amdgpu_bo_fini(struct amdgpu_device *adev); int amdgpu_bo_fbdev_mmap(struct amdgpu_bo *bo, @@ -189,10 +189,9 @@ int amdgpu_sa_bo_manager_start(struct amdgpu_device *adev, struct amdgpu_sa_manager *sa_manager); int amdgpu_sa_bo_manager_suspend(struct amdgpu_device *adev, struct amdgpu_sa_manager *sa_manager); -int amdgpu_sa_bo_new(struct amdgpu_device *adev, - struct amdgpu_sa_manager *sa_manager, - struct amdgpu_sa_bo **sa_bo, - unsigned size, unsigned align); +int amdgpu_sa_bo_new(struct amdgpu_sa_manager *sa_manager, + struct amdgpu_sa_bo **sa_bo, + unsigned size, unsigned align); void amdgpu_sa_bo_free(struct amdgpu_device *adev, struct amdgpu_sa_bo **sa_bo, struct fence *fence); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_pm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_pm.c index 22a8c7d3a3ab..ff9597ce268c 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_pm.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_pm.c @@ -30,10 +30,16 @@ #include <linux/hwmon.h> #include <linux/hwmon-sysfs.h> +#include "amd_powerplay.h" + static int amdgpu_debugfs_pm_init(struct amdgpu_device *adev); void amdgpu_pm_acpi_event_handler(struct amdgpu_device *adev) { + if (adev->pp_enabled) + /* TODO */ + return; + if (adev->pm.dpm_enabled) { mutex_lock(&adev->pm.mutex); if (power_supply_is_system_supplied() > 0) @@ -52,7 +58,12 @@ static ssize_t amdgpu_get_dpm_state(struct device *dev, { struct drm_device *ddev = dev_get_drvdata(dev); struct amdgpu_device *adev = ddev->dev_private; - enum amdgpu_pm_state_type pm = adev->pm.dpm.user_state; + enum amd_pm_state_type pm; + + if (adev->pp_enabled) { + pm = amdgpu_dpm_get_current_power_state(adev); + } else + pm = adev->pm.dpm.user_state; return snprintf(buf, PAGE_SIZE, "%s\n", (pm == POWER_STATE_TYPE_BATTERY) ? "battery" : @@ -66,40 +77,63 @@ static ssize_t amdgpu_set_dpm_state(struct device *dev, { struct drm_device *ddev = dev_get_drvdata(dev); struct amdgpu_device *adev = ddev->dev_private; + enum amd_pm_state_type state; - mutex_lock(&adev->pm.mutex); if (strncmp("battery", buf, strlen("battery")) == 0) - adev->pm.dpm.user_state = POWER_STATE_TYPE_BATTERY; + state = POWER_STATE_TYPE_BATTERY; else if (strncmp("balanced", buf, strlen("balanced")) == 0) - adev->pm.dpm.user_state = POWER_STATE_TYPE_BALANCED; + state = POWER_STATE_TYPE_BALANCED; else if (strncmp("performance", buf, strlen("performance")) == 0) - adev->pm.dpm.user_state = POWER_STATE_TYPE_PERFORMANCE; + state = POWER_STATE_TYPE_PERFORMANCE; else { - mutex_unlock(&adev->pm.mutex); count = -EINVAL; goto fail; } - mutex_unlock(&adev->pm.mutex); - /* Can't set dpm state when the card is off */ - if (!(adev->flags & AMD_IS_PX) || - (ddev->switch_power_state == DRM_SWITCH_POWER_ON)) - amdgpu_pm_compute_clocks(adev); + if (adev->pp_enabled) { + amdgpu_dpm_dispatch_task(adev, AMD_PP_EVENT_ENABLE_USER_STATE, &state, NULL); + } else { + mutex_lock(&adev->pm.mutex); + adev->pm.dpm.user_state = state; + mutex_unlock(&adev->pm.mutex); + + /* Can't set dpm state when the card is off */ + if (!(adev->flags & AMD_IS_PX) || + (ddev->switch_power_state == DRM_SWITCH_POWER_ON)) + amdgpu_pm_compute_clocks(adev); + } fail: return count; } static ssize_t amdgpu_get_dpm_forced_performance_level(struct device *dev, - struct device_attribute *attr, - char *buf) + struct device_attribute *attr, + char *buf) { struct drm_device *ddev = dev_get_drvdata(dev); struct amdgpu_device *adev = ddev->dev_private; - enum amdgpu_dpm_forced_level level = adev->pm.dpm.forced_level; - return snprintf(buf, PAGE_SIZE, "%s\n", - (level == AMDGPU_DPM_FORCED_LEVEL_AUTO) ? "auto" : - (level == AMDGPU_DPM_FORCED_LEVEL_LOW) ? "low" : "high"); + if ((adev->flags & AMD_IS_PX) && + (ddev->switch_power_state != DRM_SWITCH_POWER_ON)) + return snprintf(buf, PAGE_SIZE, "off\n"); + + if (adev->pp_enabled) { + enum amd_dpm_forced_level level; + + level = amdgpu_dpm_get_performance_level(adev); + return snprintf(buf, PAGE_SIZE, "%s\n", + (level == AMD_DPM_FORCED_LEVEL_AUTO) ? "auto" : + (level == AMD_DPM_FORCED_LEVEL_LOW) ? "low" : + (level == AMD_DPM_FORCED_LEVEL_HIGH) ? "high" : + (level == AMD_DPM_FORCED_LEVEL_MANUAL) ? "manual" : "unknown"); + } else { + enum amdgpu_dpm_forced_level level; + + level = adev->pm.dpm.forced_level; + return snprintf(buf, PAGE_SIZE, "%s\n", + (level == AMDGPU_DPM_FORCED_LEVEL_AUTO) ? "auto" : + (level == AMDGPU_DPM_FORCED_LEVEL_LOW) ? "low" : "high"); + } } static ssize_t amdgpu_set_dpm_forced_performance_level(struct device *dev, @@ -112,29 +146,307 @@ static ssize_t amdgpu_set_dpm_forced_performance_level(struct device *dev, enum amdgpu_dpm_forced_level level; int ret = 0; - mutex_lock(&adev->pm.mutex); + /* Can't force performance level when the card is off */ + if ((adev->flags & AMD_IS_PX) && + (ddev->switch_power_state != DRM_SWITCH_POWER_ON)) + return -EINVAL; + if (strncmp("low", buf, strlen("low")) == 0) { level = AMDGPU_DPM_FORCED_LEVEL_LOW; } else if (strncmp("high", buf, strlen("high")) == 0) { level = AMDGPU_DPM_FORCED_LEVEL_HIGH; } else if (strncmp("auto", buf, strlen("auto")) == 0) { level = AMDGPU_DPM_FORCED_LEVEL_AUTO; + } else if (strncmp("manual", buf, strlen("manual")) == 0) { + level = AMDGPU_DPM_FORCED_LEVEL_MANUAL; } else { count = -EINVAL; goto fail; } - if (adev->pm.funcs->force_performance_level) { + + if (adev->pp_enabled) + amdgpu_dpm_force_performance_level(adev, level); + else { + mutex_lock(&adev->pm.mutex); if (adev->pm.dpm.thermal_active) { count = -EINVAL; + mutex_unlock(&adev->pm.mutex); goto fail; } ret = amdgpu_dpm_force_performance_level(adev, level); if (ret) count = -EINVAL; + else + adev->pm.dpm.forced_level = level; + mutex_unlock(&adev->pm.mutex); } fail: - mutex_unlock(&adev->pm.mutex); + return count; +} + +static ssize_t amdgpu_get_pp_num_states(struct device *dev, + struct device_attribute *attr, + char *buf) +{ + struct drm_device *ddev = dev_get_drvdata(dev); + struct amdgpu_device *adev = ddev->dev_private; + struct pp_states_info data; + int i, buf_len; + + if (adev->pp_enabled) + amdgpu_dpm_get_pp_num_states(adev, &data); + + buf_len = snprintf(buf, PAGE_SIZE, "states: %d\n", data.nums); + for (i = 0; i < data.nums; i++) + buf_len += snprintf(buf + buf_len, PAGE_SIZE, "%d %s\n", i, + (data.states[i] == POWER_STATE_TYPE_INTERNAL_BOOT) ? "boot" : + (data.states[i] == POWER_STATE_TYPE_BATTERY) ? "battery" : + (data.states[i] == POWER_STATE_TYPE_BALANCED) ? "balanced" : + (data.states[i] == POWER_STATE_TYPE_PERFORMANCE) ? "performance" : "default"); + + return buf_len; +} + +static ssize_t amdgpu_get_pp_cur_state(struct device *dev, + struct device_attribute *attr, + char *buf) +{ + struct drm_device *ddev = dev_get_drvdata(dev); + struct amdgpu_device *adev = ddev->dev_private; + struct pp_states_info data; + enum amd_pm_state_type pm = 0; + int i = 0; + + if (adev->pp_enabled) { + + pm = amdgpu_dpm_get_current_power_state(adev); + amdgpu_dpm_get_pp_num_states(adev, &data); + + for (i = 0; i < data.nums; i++) { + if (pm == data.states[i]) + break; + } + + if (i == data.nums) + i = -EINVAL; + } + + return snprintf(buf, PAGE_SIZE, "%d\n", i); +} + +static ssize_t amdgpu_get_pp_force_state(struct device *dev, + struct device_attribute *attr, + char *buf) +{ + struct drm_device *ddev = dev_get_drvdata(dev); + struct amdgpu_device *adev = ddev->dev_private; + struct pp_states_info data; + enum amd_pm_state_type pm = 0; + int i; + + if (adev->pp_force_state_enabled && adev->pp_enabled) { + pm = amdgpu_dpm_get_current_power_state(adev); + amdgpu_dpm_get_pp_num_states(adev, &data); + + for (i = 0; i < data.nums; i++) { + if (pm == data.states[i]) + break; + } + + if (i == data.nums) + i = -EINVAL; + + return snprintf(buf, PAGE_SIZE, "%d\n", i); + + } else + return snprintf(buf, PAGE_SIZE, "\n"); +} + +static ssize_t amdgpu_set_pp_force_state(struct device *dev, + struct device_attribute *attr, + const char *buf, + size_t count) +{ + struct drm_device *ddev = dev_get_drvdata(dev); + struct amdgpu_device *adev = ddev->dev_private; + enum amd_pm_state_type state = 0; + long idx; + int ret; + + if (strlen(buf) == 1) + adev->pp_force_state_enabled = false; + else { + ret = kstrtol(buf, 0, &idx); + + if (ret) { + count = -EINVAL; + goto fail; + } + + if (adev->pp_enabled) { + struct pp_states_info data; + amdgpu_dpm_get_pp_num_states(adev, &data); + state = data.states[idx]; + /* only set user selected power states */ + if (state != POWER_STATE_TYPE_INTERNAL_BOOT && + state != POWER_STATE_TYPE_DEFAULT) { + amdgpu_dpm_dispatch_task(adev, + AMD_PP_EVENT_ENABLE_USER_STATE, &state, NULL); + adev->pp_force_state_enabled = true; + } + } + } +fail: + return count; +} + +static ssize_t amdgpu_get_pp_table(struct device *dev, + struct device_attribute *attr, + char *buf) +{ + struct drm_device *ddev = dev_get_drvdata(dev); + struct amdgpu_device *adev = ddev->dev_private; + char *table = NULL; + int size, i; + + if (adev->pp_enabled) + size = amdgpu_dpm_get_pp_table(adev, &table); + else + return 0; + + if (size >= PAGE_SIZE) + size = PAGE_SIZE - 1; + for (i = 0; i < size; i++) { + sprintf(buf + i, "%02x", table[i]); + } + sprintf(buf + i, "\n"); + + return size; +} + +static ssize_t amdgpu_set_pp_table(struct device *dev, + struct device_attribute *attr, + const char *buf, + size_t count) +{ + struct drm_device *ddev = dev_get_drvdata(dev); + struct amdgpu_device *adev = ddev->dev_private; + + if (adev->pp_enabled) + amdgpu_dpm_set_pp_table(adev, buf, count); + + return count; +} + +static ssize_t amdgpu_get_pp_dpm_sclk(struct device *dev, + struct device_attribute *attr, + char *buf) +{ + struct drm_device *ddev = dev_get_drvdata(dev); + struct amdgpu_device *adev = ddev->dev_private; + ssize_t size = 0; + + if (adev->pp_enabled) + size = amdgpu_dpm_print_clock_levels(adev, PP_SCLK, buf); + + return size; +} + +static ssize_t amdgpu_set_pp_dpm_sclk(struct device *dev, + struct device_attribute *attr, + const char *buf, + size_t count) +{ + struct drm_device *ddev = dev_get_drvdata(dev); + struct amdgpu_device *adev = ddev->dev_private; + int ret; + long level; + + ret = kstrtol(buf, 0, &level); + + if (ret) { + count = -EINVAL; + goto fail; + } + + if (adev->pp_enabled) + amdgpu_dpm_force_clock_level(adev, PP_SCLK, level); +fail: + return count; +} + +static ssize_t amdgpu_get_pp_dpm_mclk(struct device *dev, + struct device_attribute *attr, + char *buf) +{ + struct drm_device *ddev = dev_get_drvdata(dev); + struct amdgpu_device *adev = ddev->dev_private; + ssize_t size = 0; + + if (adev->pp_enabled) + size = amdgpu_dpm_print_clock_levels(adev, PP_MCLK, buf); + + return size; +} + +static ssize_t amdgpu_set_pp_dpm_mclk(struct device *dev, + struct device_attribute *attr, + const char *buf, + size_t count) +{ + struct drm_device *ddev = dev_get_drvdata(dev); + struct amdgpu_device *adev = ddev->dev_private; + int ret; + long level; + + ret = kstrtol(buf, 0, &level); + + if (ret) { + count = -EINVAL; + goto fail; + } + + if (adev->pp_enabled) + amdgpu_dpm_force_clock_level(adev, PP_MCLK, level); +fail: + return count; +} + +static ssize_t amdgpu_get_pp_dpm_pcie(struct device *dev, + struct device_attribute *attr, + char *buf) +{ + struct drm_device *ddev = dev_get_drvdata(dev); + struct amdgpu_device *adev = ddev->dev_private; + ssize_t size = 0; + + if (adev->pp_enabled) + size = amdgpu_dpm_print_clock_levels(adev, PP_PCIE, buf); + + return size; +} + +static ssize_t amdgpu_set_pp_dpm_pcie(struct device *dev, + struct device_attribute *attr, + const char *buf, + size_t count) +{ + struct drm_device *ddev = dev_get_drvdata(dev); + struct amdgpu_device *adev = ddev->dev_private; + int ret; + long level; + + ret = kstrtol(buf, 0, &level); + + if (ret) { + count = -EINVAL; + goto fail; + } + + if (adev->pp_enabled) + amdgpu_dpm_force_clock_level(adev, PP_PCIE, level); +fail: return count; } @@ -142,18 +454,41 @@ static DEVICE_ATTR(power_dpm_state, S_IRUGO | S_IWUSR, amdgpu_get_dpm_state, amd static DEVICE_ATTR(power_dpm_force_performance_level, S_IRUGO | S_IWUSR, amdgpu_get_dpm_forced_performance_level, amdgpu_set_dpm_forced_performance_level); +static DEVICE_ATTR(pp_num_states, S_IRUGO, amdgpu_get_pp_num_states, NULL); +static DEVICE_ATTR(pp_cur_state, S_IRUGO, amdgpu_get_pp_cur_state, NULL); +static DEVICE_ATTR(pp_force_state, S_IRUGO | S_IWUSR, + amdgpu_get_pp_force_state, + amdgpu_set_pp_force_state); +static DEVICE_ATTR(pp_table, S_IRUGO | S_IWUSR, + amdgpu_get_pp_table, + amdgpu_set_pp_table); +static DEVICE_ATTR(pp_dpm_sclk, S_IRUGO | S_IWUSR, + amdgpu_get_pp_dpm_sclk, + amdgpu_set_pp_dpm_sclk); +static DEVICE_ATTR(pp_dpm_mclk, S_IRUGO | S_IWUSR, + amdgpu_get_pp_dpm_mclk, + amdgpu_set_pp_dpm_mclk); +static DEVICE_ATTR(pp_dpm_pcie, S_IRUGO | S_IWUSR, + amdgpu_get_pp_dpm_pcie, + amdgpu_set_pp_dpm_pcie); static ssize_t amdgpu_hwmon_show_temp(struct device *dev, struct device_attribute *attr, char *buf) { struct amdgpu_device *adev = dev_get_drvdata(dev); + struct drm_device *ddev = adev->ddev; int temp; - if (adev->pm.funcs->get_temperature) - temp = amdgpu_dpm_get_temperature(adev); - else + /* Can't get temperature when the card is off */ + if ((adev->flags & AMD_IS_PX) && + (ddev->switch_power_state != DRM_SWITCH_POWER_ON)) + return -EINVAL; + + if (!adev->pp_enabled && !adev->pm.funcs->get_temperature) temp = 0; + else + temp = amdgpu_dpm_get_temperature(adev); return snprintf(buf, PAGE_SIZE, "%d\n", temp); } @@ -181,8 +516,10 @@ static ssize_t amdgpu_hwmon_get_pwm1_enable(struct device *dev, struct amdgpu_device *adev = dev_get_drvdata(dev); u32 pwm_mode = 0; - if (adev->pm.funcs->get_fan_control_mode) - pwm_mode = amdgpu_dpm_get_fan_control_mode(adev); + if (!adev->pp_enabled && !adev->pm.funcs->get_fan_control_mode) + return -EINVAL; + + pwm_mode = amdgpu_dpm_get_fan_control_mode(adev); /* never 0 (full-speed), fuse or smc-controlled always */ return sprintf(buf, "%i\n", pwm_mode == FDO_PWM_MODE_STATIC ? 1 : 2); @@ -197,7 +534,7 @@ static ssize_t amdgpu_hwmon_set_pwm1_enable(struct device *dev, int err; int value; - if(!adev->pm.funcs->set_fan_control_mode) + if (!adev->pp_enabled && !adev->pm.funcs->set_fan_control_mode) return -EINVAL; err = kstrtoint(buf, 10, &value); @@ -290,11 +627,11 @@ static struct attribute *hwmon_attributes[] = { static umode_t hwmon_attributes_visible(struct kobject *kobj, struct attribute *attr, int index) { - struct device *dev = container_of(kobj, struct device, kobj); + struct device *dev = kobj_to_dev(kobj); struct amdgpu_device *adev = dev_get_drvdata(dev); umode_t effective_mode = attr->mode; - /* Skip attributes if DPM is not enabled */ + /* Skip limit attributes if DPM is not enabled */ if (!adev->pm.dpm_enabled && (attr == &sensor_dev_attr_temp1_crit.dev_attr.attr || attr == &sensor_dev_attr_temp1_crit_hyst.dev_attr.attr || @@ -304,6 +641,9 @@ static umode_t hwmon_attributes_visible(struct kobject *kobj, attr == &sensor_dev_attr_pwm1_min.dev_attr.attr)) return 0; + if (adev->pp_enabled) + return effective_mode; + /* Skip fan attributes if fan is not present */ if (adev->pm.no_fan && (attr == &sensor_dev_attr_pwm1.dev_attr.attr || @@ -351,7 +691,7 @@ void amdgpu_dpm_thermal_work_handler(struct work_struct *work) container_of(work, struct amdgpu_device, pm.dpm.thermal.work); /* switch to the thermal state */ - enum amdgpu_pm_state_type dpm_state = POWER_STATE_TYPE_INTERNAL_THERMAL; + enum amd_pm_state_type dpm_state = POWER_STATE_TYPE_INTERNAL_THERMAL; if (!adev->pm.dpm_enabled) return; @@ -379,7 +719,7 @@ void amdgpu_dpm_thermal_work_handler(struct work_struct *work) } static struct amdgpu_ps *amdgpu_dpm_pick_power_state(struct amdgpu_device *adev, - enum amdgpu_pm_state_type dpm_state) + enum amd_pm_state_type dpm_state) { int i; struct amdgpu_ps *ps; @@ -516,7 +856,7 @@ static void amdgpu_dpm_change_power_state_locked(struct amdgpu_device *adev) { int i; struct amdgpu_ps *ps; - enum amdgpu_pm_state_type dpm_state; + enum amd_pm_state_type dpm_state; int ret; /* if dpm init failed */ @@ -584,22 +924,15 @@ force: amdgpu_dpm_print_power_state(adev, adev->pm.dpm.requested_ps); } - mutex_lock(&adev->ring_lock); - /* update whether vce is active */ ps->vce_active = adev->pm.dpm.vce_active; ret = amdgpu_dpm_pre_set_power_state(adev); if (ret) - goto done; + return; /* update display watermarks based on new power state */ amdgpu_display_bandwidth_update(adev); - /* update displays */ - amdgpu_dpm_display_configuration_changed(adev); - - adev->pm.dpm.current_active_crtcs = adev->pm.dpm.new_active_crtcs; - adev->pm.dpm.current_active_crtc_count = adev->pm.dpm.new_active_crtc_count; /* wait for the rings to drain */ for (i = 0; i < AMDGPU_MAX_RINGS; i++) { @@ -616,6 +949,12 @@ force: amdgpu_dpm_post_set_power_state(adev); + /* update displays */ + amdgpu_dpm_display_configuration_changed(adev); + + adev->pm.dpm.current_active_crtcs = adev->pm.dpm.new_active_crtcs; + adev->pm.dpm.current_active_crtc_count = adev->pm.dpm.new_active_crtc_count; + if (adev->pm.funcs->force_performance_level) { if (adev->pm.dpm.thermal_active) { enum amdgpu_dpm_forced_level level = adev->pm.dpm.forced_level; @@ -628,56 +967,58 @@ force: amdgpu_dpm_force_performance_level(adev, adev->pm.dpm.forced_level); } } - -done: - mutex_unlock(&adev->ring_lock); } void amdgpu_dpm_enable_uvd(struct amdgpu_device *adev, bool enable) { - if (adev->pm.funcs->powergate_uvd) { - mutex_lock(&adev->pm.mutex); - /* enable/disable UVD */ + if (adev->pp_enabled) amdgpu_dpm_powergate_uvd(adev, !enable); - mutex_unlock(&adev->pm.mutex); - } else { - if (enable) { + else { + if (adev->pm.funcs->powergate_uvd) { mutex_lock(&adev->pm.mutex); - adev->pm.dpm.uvd_active = true; - adev->pm.dpm.state = POWER_STATE_TYPE_INTERNAL_UVD; + /* enable/disable UVD */ + amdgpu_dpm_powergate_uvd(adev, !enable); mutex_unlock(&adev->pm.mutex); } else { - mutex_lock(&adev->pm.mutex); - adev->pm.dpm.uvd_active = false; - mutex_unlock(&adev->pm.mutex); + if (enable) { + mutex_lock(&adev->pm.mutex); + adev->pm.dpm.uvd_active = true; + adev->pm.dpm.state = POWER_STATE_TYPE_INTERNAL_UVD; + mutex_unlock(&adev->pm.mutex); + } else { + mutex_lock(&adev->pm.mutex); + adev->pm.dpm.uvd_active = false; + mutex_unlock(&adev->pm.mutex); + } + amdgpu_pm_compute_clocks(adev); } - amdgpu_pm_compute_clocks(adev); } } void amdgpu_dpm_enable_vce(struct amdgpu_device *adev, bool enable) { - if (adev->pm.funcs->powergate_vce) { - mutex_lock(&adev->pm.mutex); - /* enable/disable VCE */ + if (adev->pp_enabled) amdgpu_dpm_powergate_vce(adev, !enable); - - mutex_unlock(&adev->pm.mutex); - } else { - if (enable) { + else { + if (adev->pm.funcs->powergate_vce) { mutex_lock(&adev->pm.mutex); - adev->pm.dpm.vce_active = true; - /* XXX select vce level based on ring/task */ - adev->pm.dpm.vce_level = AMDGPU_VCE_LEVEL_AC_ALL; + amdgpu_dpm_powergate_vce(adev, !enable); mutex_unlock(&adev->pm.mutex); } else { - mutex_lock(&adev->pm.mutex); - adev->pm.dpm.vce_active = false; - mutex_unlock(&adev->pm.mutex); + if (enable) { + mutex_lock(&adev->pm.mutex); + adev->pm.dpm.vce_active = true; + /* XXX select vce level based on ring/task */ + adev->pm.dpm.vce_level = AMDGPU_VCE_LEVEL_AC_ALL; + mutex_unlock(&adev->pm.mutex); + } else { + mutex_lock(&adev->pm.mutex); + adev->pm.dpm.vce_active = false; + mutex_unlock(&adev->pm.mutex); + } + amdgpu_pm_compute_clocks(adev); } - - amdgpu_pm_compute_clocks(adev); } } @@ -685,10 +1026,13 @@ void amdgpu_pm_print_power_states(struct amdgpu_device *adev) { int i; - for (i = 0; i < adev->pm.dpm.num_ps; i++) { - printk("== power state %d ==\n", i); + if (adev->pp_enabled) + /* TO DO */ + return; + + for (i = 0; i < adev->pm.dpm.num_ps; i++) amdgpu_dpm_print_power_state(adev, &adev->pm.dpm.ps[i]); - } + } int amdgpu_pm_sysfs_init(struct amdgpu_device *adev) @@ -698,8 +1042,11 @@ int amdgpu_pm_sysfs_init(struct amdgpu_device *adev) if (adev->pm.sysfs_initialized) return 0; - if (adev->pm.funcs->get_temperature == NULL) - return 0; + if (!adev->pp_enabled) { + if (adev->pm.funcs->get_temperature == NULL) + return 0; + } + adev->pm.int_hwmon_dev = hwmon_device_register_with_groups(adev->dev, DRIVER_NAME, adev, hwmon_groups); @@ -720,6 +1067,44 @@ int amdgpu_pm_sysfs_init(struct amdgpu_device *adev) DRM_ERROR("failed to create device file for dpm state\n"); return ret; } + + if (adev->pp_enabled) { + ret = device_create_file(adev->dev, &dev_attr_pp_num_states); + if (ret) { + DRM_ERROR("failed to create device file pp_num_states\n"); + return ret; + } + ret = device_create_file(adev->dev, &dev_attr_pp_cur_state); + if (ret) { + DRM_ERROR("failed to create device file pp_cur_state\n"); + return ret; + } + ret = device_create_file(adev->dev, &dev_attr_pp_force_state); + if (ret) { + DRM_ERROR("failed to create device file pp_force_state\n"); + return ret; + } + ret = device_create_file(adev->dev, &dev_attr_pp_table); + if (ret) { + DRM_ERROR("failed to create device file pp_table\n"); + return ret; + } + ret = device_create_file(adev->dev, &dev_attr_pp_dpm_sclk); + if (ret) { + DRM_ERROR("failed to create device file pp_dpm_sclk\n"); + return ret; + } + ret = device_create_file(adev->dev, &dev_attr_pp_dpm_mclk); + if (ret) { + DRM_ERROR("failed to create device file pp_dpm_mclk\n"); + return ret; + } + ret = device_create_file(adev->dev, &dev_attr_pp_dpm_pcie); + if (ret) { + DRM_ERROR("failed to create device file pp_dpm_pcie\n"); + return ret; + } + } ret = amdgpu_debugfs_pm_init(adev); if (ret) { DRM_ERROR("Failed to register debugfs file for dpm!\n"); @@ -737,6 +1122,15 @@ void amdgpu_pm_sysfs_fini(struct amdgpu_device *adev) hwmon_device_unregister(adev->pm.int_hwmon_dev); device_remove_file(adev->dev, &dev_attr_power_dpm_state); device_remove_file(adev->dev, &dev_attr_power_dpm_force_performance_level); + if (adev->pp_enabled) { + device_remove_file(adev->dev, &dev_attr_pp_num_states); + device_remove_file(adev->dev, &dev_attr_pp_cur_state); + device_remove_file(adev->dev, &dev_attr_pp_force_state); + device_remove_file(adev->dev, &dev_attr_pp_table); + device_remove_file(adev->dev, &dev_attr_pp_dpm_sclk); + device_remove_file(adev->dev, &dev_attr_pp_dpm_mclk); + device_remove_file(adev->dev, &dev_attr_pp_dpm_pcie); + } } void amdgpu_pm_compute_clocks(struct amdgpu_device *adev) @@ -748,32 +1142,41 @@ void amdgpu_pm_compute_clocks(struct amdgpu_device *adev) if (!adev->pm.dpm_enabled) return; - mutex_lock(&adev->pm.mutex); + if (adev->pp_enabled) { + int i = 0; - /* update active crtc counts */ - adev->pm.dpm.new_active_crtcs = 0; - adev->pm.dpm.new_active_crtc_count = 0; - if (adev->mode_info.num_crtc && adev->mode_info.mode_config_initialized) { - list_for_each_entry(crtc, - &ddev->mode_config.crtc_list, head) { - amdgpu_crtc = to_amdgpu_crtc(crtc); - if (crtc->enabled) { - adev->pm.dpm.new_active_crtcs |= (1 << amdgpu_crtc->crtc_id); - adev->pm.dpm.new_active_crtc_count++; - } + amdgpu_display_bandwidth_update(adev); + for (i = 0; i < AMDGPU_MAX_RINGS; i++) { + struct amdgpu_ring *ring = adev->rings[i]; + if (ring && ring->ready) + amdgpu_fence_wait_empty(ring); } - } - /* update battery/ac status */ - if (power_supply_is_system_supplied() > 0) - adev->pm.dpm.ac_power = true; - else - adev->pm.dpm.ac_power = false; - - amdgpu_dpm_change_power_state_locked(adev); + amdgpu_dpm_dispatch_task(adev, AMD_PP_EVENT_DISPLAY_CONFIG_CHANGE, NULL, NULL); + } else { + mutex_lock(&adev->pm.mutex); + adev->pm.dpm.new_active_crtcs = 0; + adev->pm.dpm.new_active_crtc_count = 0; + if (adev->mode_info.num_crtc && adev->mode_info.mode_config_initialized) { + list_for_each_entry(crtc, + &ddev->mode_config.crtc_list, head) { + amdgpu_crtc = to_amdgpu_crtc(crtc); + if (crtc->enabled) { + adev->pm.dpm.new_active_crtcs |= (1 << amdgpu_crtc->crtc_id); + adev->pm.dpm.new_active_crtc_count++; + } + } + } + /* update battery/ac status */ + if (power_supply_is_system_supplied() > 0) + adev->pm.dpm.ac_power = true; + else + adev->pm.dpm.ac_power = false; - mutex_unlock(&adev->pm.mutex); + amdgpu_dpm_change_power_state_locked(adev); + mutex_unlock(&adev->pm.mutex); + } } /* @@ -786,8 +1189,18 @@ static int amdgpu_debugfs_pm_info(struct seq_file *m, void *data) struct drm_info_node *node = (struct drm_info_node *) m->private; struct drm_device *dev = node->minor->dev; struct amdgpu_device *adev = dev->dev_private; + struct drm_device *ddev = adev->ddev; - if (adev->pm.dpm_enabled) { + if (!adev->pm.dpm_enabled) { + seq_printf(m, "dpm not enabled\n"); + return 0; + } + if ((adev->flags & AMD_IS_PX) && + (ddev->switch_power_state != DRM_SWITCH_POWER_ON)) { + seq_printf(m, "PX asic powered off\n"); + } else if (adev->pp_enabled) { + amdgpu_dpm_debugfs_print_current_performance_level(adev, m); + } else { mutex_lock(&adev->pm.mutex); if (adev->pm.funcs->debugfs_print_current_performance_level) amdgpu_dpm_debugfs_print_current_performance_level(adev, m); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_powerplay.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_powerplay.c new file mode 100644 index 000000000000..3cb6d6c413c7 --- /dev/null +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_powerplay.c @@ -0,0 +1,330 @@ +/* + * Copyright 2015 Advanced Micro Devices, Inc. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice shall be included in + * all copies or substantial portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR + * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, + * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR + * OTHER DEALINGS IN THE SOFTWARE. + * + * Authors: AMD + * + */ +#include "atom.h" +#include "amdgpu.h" +#include "amd_shared.h" +#include <linux/module.h> +#include <linux/moduleparam.h> +#include "amdgpu_pm.h" +#include <drm/amdgpu_drm.h> +#include "amdgpu_powerplay.h" +#include "cik_dpm.h" +#include "vi_dpm.h" + +static int amdgpu_powerplay_init(struct amdgpu_device *adev) +{ + int ret = 0; + struct amd_powerplay *amd_pp; + + amd_pp = &(adev->powerplay); + + if (adev->pp_enabled) { +#ifdef CONFIG_DRM_AMD_POWERPLAY + struct amd_pp_init *pp_init; + + pp_init = kzalloc(sizeof(struct amd_pp_init), GFP_KERNEL); + + if (pp_init == NULL) + return -ENOMEM; + + pp_init->chip_family = adev->family; + pp_init->chip_id = adev->asic_type; + pp_init->device = amdgpu_cgs_create_device(adev); + + ret = amd_powerplay_init(pp_init, amd_pp); + kfree(pp_init); +#endif + } else { + amd_pp->pp_handle = (void *)adev; + + switch (adev->asic_type) { +#ifdef CONFIG_DRM_AMDGPU_CIK + case CHIP_BONAIRE: + case CHIP_HAWAII: + amd_pp->ip_funcs = &ci_dpm_ip_funcs; + break; + case CHIP_KABINI: + case CHIP_MULLINS: + case CHIP_KAVERI: + amd_pp->ip_funcs = &kv_dpm_ip_funcs; + break; +#endif + case CHIP_TOPAZ: + amd_pp->ip_funcs = &iceland_dpm_ip_funcs; + break; + case CHIP_TONGA: + amd_pp->ip_funcs = &tonga_dpm_ip_funcs; + break; + case CHIP_FIJI: + amd_pp->ip_funcs = &fiji_dpm_ip_funcs; + break; + case CHIP_CARRIZO: + case CHIP_STONEY: + amd_pp->ip_funcs = &cz_dpm_ip_funcs; + break; + default: + ret = -EINVAL; + break; + } + } + return ret; +} + +static int amdgpu_pp_early_init(void *handle) +{ + struct amdgpu_device *adev = (struct amdgpu_device *)handle; + int ret = 0; + +#ifdef CONFIG_DRM_AMD_POWERPLAY + switch (adev->asic_type) { + case CHIP_TONGA: + case CHIP_FIJI: + adev->pp_enabled = (amdgpu_powerplay == 0) ? false : true; + break; + case CHIP_CARRIZO: + case CHIP_STONEY: + adev->pp_enabled = (amdgpu_powerplay > 0) ? true : false; + break; + /* These chips don't have powerplay implemenations */ + case CHIP_BONAIRE: + case CHIP_HAWAII: + case CHIP_KABINI: + case CHIP_MULLINS: + case CHIP_KAVERI: + case CHIP_TOPAZ: + default: + adev->pp_enabled = false; + break; + } +#else + adev->pp_enabled = false; +#endif + + ret = amdgpu_powerplay_init(adev); + if (ret) + return ret; + + if (adev->powerplay.ip_funcs->early_init) + ret = adev->powerplay.ip_funcs->early_init( + adev->powerplay.pp_handle); + return ret; +} + + +static int amdgpu_pp_late_init(void *handle) +{ + int ret = 0; + struct amdgpu_device *adev = (struct amdgpu_device *)handle; + + if (adev->powerplay.ip_funcs->late_init) + ret = adev->powerplay.ip_funcs->late_init( + adev->powerplay.pp_handle); + +#ifdef CONFIG_DRM_AMD_POWERPLAY + if (adev->pp_enabled) { + amdgpu_pm_sysfs_init(adev); + amdgpu_dpm_dispatch_task(adev, AMD_PP_EVENT_COMPLETE_INIT, NULL, NULL); + } +#endif + return ret; +} + +static int amdgpu_pp_sw_init(void *handle) +{ + int ret = 0; + struct amdgpu_device *adev = (struct amdgpu_device *)handle; + + if (adev->powerplay.ip_funcs->sw_init) + ret = adev->powerplay.ip_funcs->sw_init( + adev->powerplay.pp_handle); + +#ifdef CONFIG_DRM_AMD_POWERPLAY + if (adev->pp_enabled) { + if (amdgpu_dpm == 0) + adev->pm.dpm_enabled = false; + else + adev->pm.dpm_enabled = true; + } +#endif + + return ret; +} + +static int amdgpu_pp_sw_fini(void *handle) +{ + int ret = 0; + struct amdgpu_device *adev = (struct amdgpu_device *)handle; + + if (adev->powerplay.ip_funcs->sw_fini) + ret = adev->powerplay.ip_funcs->sw_fini( + adev->powerplay.pp_handle); + if (ret) + return ret; + +#ifdef CONFIG_DRM_AMD_POWERPLAY + if (adev->pp_enabled) { + amdgpu_pm_sysfs_fini(adev); + amd_powerplay_fini(adev->powerplay.pp_handle); + } +#endif + + return ret; +} + +static int amdgpu_pp_hw_init(void *handle) +{ + int ret = 0; + struct amdgpu_device *adev = (struct amdgpu_device *)handle; + + if (adev->pp_enabled && adev->firmware.smu_load) + amdgpu_ucode_init_bo(adev); + + if (adev->powerplay.ip_funcs->hw_init) + ret = adev->powerplay.ip_funcs->hw_init( + adev->powerplay.pp_handle); + + return ret; +} + +static int amdgpu_pp_hw_fini(void *handle) +{ + int ret = 0; + struct amdgpu_device *adev = (struct amdgpu_device *)handle; + + if (adev->powerplay.ip_funcs->hw_fini) + ret = adev->powerplay.ip_funcs->hw_fini( + adev->powerplay.pp_handle); + + if (adev->pp_enabled && adev->firmware.smu_load) + amdgpu_ucode_fini_bo(adev); + + return ret; +} + +static int amdgpu_pp_suspend(void *handle) +{ + int ret = 0; + struct amdgpu_device *adev = (struct amdgpu_device *)handle; + + if (adev->powerplay.ip_funcs->suspend) + ret = adev->powerplay.ip_funcs->suspend( + adev->powerplay.pp_handle); + return ret; +} + +static int amdgpu_pp_resume(void *handle) +{ + int ret = 0; + struct amdgpu_device *adev = (struct amdgpu_device *)handle; + + if (adev->powerplay.ip_funcs->resume) + ret = adev->powerplay.ip_funcs->resume( + adev->powerplay.pp_handle); + return ret; +} + +static int amdgpu_pp_set_clockgating_state(void *handle, + enum amd_clockgating_state state) +{ + int ret = 0; + struct amdgpu_device *adev = (struct amdgpu_device *)handle; + + if (adev->powerplay.ip_funcs->set_clockgating_state) + ret = adev->powerplay.ip_funcs->set_clockgating_state( + adev->powerplay.pp_handle, state); + return ret; +} + +static int amdgpu_pp_set_powergating_state(void *handle, + enum amd_powergating_state state) +{ + int ret = 0; + struct amdgpu_device *adev = (struct amdgpu_device *)handle; + + if (adev->powerplay.ip_funcs->set_powergating_state) + ret = adev->powerplay.ip_funcs->set_powergating_state( + adev->powerplay.pp_handle, state); + return ret; +} + + +static bool amdgpu_pp_is_idle(void *handle) +{ + bool ret = true; + struct amdgpu_device *adev = (struct amdgpu_device *)handle; + + if (adev->powerplay.ip_funcs->is_idle) + ret = adev->powerplay.ip_funcs->is_idle( + adev->powerplay.pp_handle); + return ret; +} + +static int amdgpu_pp_wait_for_idle(void *handle) +{ + int ret = 0; + struct amdgpu_device *adev = (struct amdgpu_device *)handle; + + if (adev->powerplay.ip_funcs->wait_for_idle) + ret = adev->powerplay.ip_funcs->wait_for_idle( + adev->powerplay.pp_handle); + return ret; +} + +static int amdgpu_pp_soft_reset(void *handle) +{ + int ret = 0; + struct amdgpu_device *adev = (struct amdgpu_device *)handle; + + if (adev->powerplay.ip_funcs->soft_reset) + ret = adev->powerplay.ip_funcs->soft_reset( + adev->powerplay.pp_handle); + return ret; +} + +static void amdgpu_pp_print_status(void *handle) +{ + struct amdgpu_device *adev = (struct amdgpu_device *)handle; + + if (adev->powerplay.ip_funcs->print_status) + adev->powerplay.ip_funcs->print_status( + adev->powerplay.pp_handle); +} + +const struct amd_ip_funcs amdgpu_pp_ip_funcs = { + .early_init = amdgpu_pp_early_init, + .late_init = amdgpu_pp_late_init, + .sw_init = amdgpu_pp_sw_init, + .sw_fini = amdgpu_pp_sw_fini, + .hw_init = amdgpu_pp_hw_init, + .hw_fini = amdgpu_pp_hw_fini, + .suspend = amdgpu_pp_suspend, + .resume = amdgpu_pp_resume, + .is_idle = amdgpu_pp_is_idle, + .wait_for_idle = amdgpu_pp_wait_for_idle, + .soft_reset = amdgpu_pp_soft_reset, + .print_status = amdgpu_pp_print_status, + .set_clockgating_state = amdgpu_pp_set_clockgating_state, + .set_powergating_state = amdgpu_pp_set_powergating_state, +}; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_powerplay.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_powerplay.h new file mode 100644 index 000000000000..da5cf47cfd99 --- /dev/null +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_powerplay.h @@ -0,0 +1,33 @@ +/* + * Copyright 2015 Advanced Micro Devices, Inc. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice shall be included in + * all copies or substantial portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR + * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, + * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR + * OTHER DEALINGS IN THE SOFTWARE. + * + * Authors: AMD + * + */ + +#ifndef __AMDGPU_POPWERPLAY_H__ +#define __AMDGPU_POPWERPLAY_H__ + +#include "amd_shared.h" + +extern const struct amd_ip_funcs amdgpu_pp_ip_funcs; + +#endif /* __AMDSOC_DM_H__ */ diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_prime.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_prime.c index 59f735a933a9..be6388f73ba2 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_prime.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_prime.c @@ -73,10 +73,6 @@ struct drm_gem_object *amdgpu_gem_prime_import_sg_table(struct drm_device *dev, if (ret) return ERR_PTR(ret); - mutex_lock(&adev->gem.mutex); - list_add_tail(&bo->list, &adev->gem.objects); - mutex_unlock(&adev->gem.mutex); - return &bo->gem_base; } @@ -121,7 +117,7 @@ struct dma_buf *amdgpu_gem_prime_export(struct drm_device *dev, { struct amdgpu_bo *bo = gem_to_amdgpu_bo(gobj); - if (amdgpu_ttm_tt_has_userptr(bo->tbo.ttm)) + if (amdgpu_ttm_tt_get_usermm(bo->tbo.ttm)) return ERR_PTR(-EPERM); return drm_gem_prime_export(dev, gobj, flags); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c index 30dce235ddeb..972eed2ef787 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c @@ -49,30 +49,6 @@ static int amdgpu_debugfs_ring_init(struct amdgpu_device *adev, struct amdgpu_ring *ring); /** - * amdgpu_ring_free_size - update the free size - * - * @adev: amdgpu_device pointer - * @ring: amdgpu_ring structure holding ring information - * - * Update the free dw slots in the ring buffer (all asics). - */ -void amdgpu_ring_free_size(struct amdgpu_ring *ring) -{ - uint32_t rptr = amdgpu_ring_get_rptr(ring); - - /* This works because ring_size is a power of 2 */ - ring->ring_free_dw = rptr + (ring->ring_size / 4); - ring->ring_free_dw -= ring->wptr; - ring->ring_free_dw &= ring->ptr_mask; - if (!ring->ring_free_dw) { - /* this is an empty ring */ - ring->ring_free_dw = ring->ring_size / 4; - /* update lockup info to avoid false positive */ - amdgpu_ring_lockup_update(ring); - } -} - -/** * amdgpu_ring_alloc - allocate space on the ring buffer * * @adev: amdgpu_device pointer @@ -84,50 +60,18 @@ void amdgpu_ring_free_size(struct amdgpu_ring *ring) */ int amdgpu_ring_alloc(struct amdgpu_ring *ring, unsigned ndw) { - int r; - - /* make sure we aren't trying to allocate more space than there is on the ring */ - if (ndw > (ring->ring_size / 4)) - return -ENOMEM; /* Align requested size with padding so unlock_commit can * pad safely */ - amdgpu_ring_free_size(ring); ndw = (ndw + ring->align_mask) & ~ring->align_mask; - while (ndw > (ring->ring_free_dw - 1)) { - amdgpu_ring_free_size(ring); - if (ndw < ring->ring_free_dw) { - break; - } - r = amdgpu_fence_wait_next(ring); - if (r) - return r; - } - ring->count_dw = ndw; - ring->wptr_old = ring->wptr; - return 0; -} -/** - * amdgpu_ring_lock - lock the ring and allocate space on it - * - * @adev: amdgpu_device pointer - * @ring: amdgpu_ring structure holding ring information - * @ndw: number of dwords to allocate in the ring buffer - * - * Lock the ring and allocate @ndw dwords in the ring buffer - * (all asics). - * Returns 0 on success, error on failure. - */ -int amdgpu_ring_lock(struct amdgpu_ring *ring, unsigned ndw) -{ - int r; + /* Make sure we aren't trying to allocate more space + * than the maximum for one submission + */ + if (WARN_ON_ONCE(ndw > ring->max_dw)) + return -ENOMEM; - mutex_lock(ring->ring_lock); - r = amdgpu_ring_alloc(ring, ndw); - if (r) { - mutex_unlock(ring->ring_lock); - return r; - } + ring->count_dw = ndw; + ring->wptr_old = ring->wptr; return 0; } @@ -146,6 +90,19 @@ void amdgpu_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count) amdgpu_ring_write(ring, ring->nop); } +/** amdgpu_ring_generic_pad_ib - pad IB with NOP packets + * + * @ring: amdgpu_ring structure holding ring information + * @ib: IB to add NOP packets to + * + * This is the generic pad_ib function for rings except SDMA + */ +void amdgpu_ring_generic_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib) +{ + while (ib->length_dw & ring->align_mask) + ib->ptr[ib->length_dw++] = ring->nop; +} + /** * amdgpu_ring_commit - tell the GPU to execute the new * commands on the ring buffer @@ -170,20 +127,6 @@ void amdgpu_ring_commit(struct amdgpu_ring *ring) } /** - * amdgpu_ring_unlock_commit - tell the GPU to execute the new - * commands on the ring buffer and unlock it - * - * @ring: amdgpu_ring structure holding ring information - * - * Call amdgpu_ring_commit() then unlock the ring (all asics). - */ -void amdgpu_ring_unlock_commit(struct amdgpu_ring *ring) -{ - amdgpu_ring_commit(ring); - mutex_unlock(ring->ring_lock); -} - -/** * amdgpu_ring_undo - reset the wptr * * @ring: amdgpu_ring structure holding ring information @@ -196,59 +139,6 @@ void amdgpu_ring_undo(struct amdgpu_ring *ring) } /** - * amdgpu_ring_unlock_undo - reset the wptr and unlock the ring - * - * @ring: amdgpu_ring structure holding ring information - * - * Call amdgpu_ring_undo() then unlock the ring (all asics). - */ -void amdgpu_ring_unlock_undo(struct amdgpu_ring *ring) -{ - amdgpu_ring_undo(ring); - mutex_unlock(ring->ring_lock); -} - -/** - * amdgpu_ring_lockup_update - update lockup variables - * - * @ring: amdgpu_ring structure holding ring information - * - * Update the last rptr value and timestamp (all asics). - */ -void amdgpu_ring_lockup_update(struct amdgpu_ring *ring) -{ - atomic_set(&ring->last_rptr, amdgpu_ring_get_rptr(ring)); - atomic64_set(&ring->last_activity, jiffies_64); -} - -/** - * amdgpu_ring_test_lockup() - check if ring is lockedup by recording information - * @ring: amdgpu_ring structure holding ring information - * - */ -bool amdgpu_ring_test_lockup(struct amdgpu_ring *ring) -{ - uint32_t rptr = amdgpu_ring_get_rptr(ring); - uint64_t last = atomic64_read(&ring->last_activity); - uint64_t elapsed; - - if (rptr != atomic_read(&ring->last_rptr)) { - /* ring is still working, no lockup */ - amdgpu_ring_lockup_update(ring); - return false; - } - - elapsed = jiffies_to_msecs(jiffies_64 - last); - if (amdgpu_lockup_timeout && elapsed >= amdgpu_lockup_timeout) { - dev_err(ring->adev->dev, "ring %d stalled for more than %llumsec\n", - ring->idx, elapsed); - return true; - } - /* give a chance to the GPU ... */ - return false; -} - -/** * amdgpu_ring_backup - Back up the content of a ring * * @ring: the ring we want to back up @@ -260,43 +150,32 @@ unsigned amdgpu_ring_backup(struct amdgpu_ring *ring, { unsigned size, ptr, i; - /* just in case lock the ring */ - mutex_lock(ring->ring_lock); *data = NULL; - if (ring->ring_obj == NULL) { - mutex_unlock(ring->ring_lock); + if (ring->ring_obj == NULL) return 0; - } /* it doesn't make sense to save anything if all fences are signaled */ - if (!amdgpu_fence_count_emitted(ring)) { - mutex_unlock(ring->ring_lock); + if (!amdgpu_fence_count_emitted(ring)) return 0; - } ptr = le32_to_cpu(*ring->next_rptr_cpu_addr); size = ring->wptr + (ring->ring_size / 4); size -= ptr; size &= ring->ptr_mask; - if (size == 0) { - mutex_unlock(ring->ring_lock); + if (size == 0) return 0; - } /* and then save the content of the ring */ *data = kmalloc_array(size, sizeof(uint32_t), GFP_KERNEL); - if (!*data) { - mutex_unlock(ring->ring_lock); + if (!*data) return 0; - } for (i = 0; i < size; ++i) { (*data)[i] = ring->ring[ptr++]; ptr &= ring->ptr_mask; } - mutex_unlock(ring->ring_lock); return size; } @@ -318,7 +197,7 @@ int amdgpu_ring_restore(struct amdgpu_ring *ring, return 0; /* restore the saved ring content */ - r = amdgpu_ring_lock(ring, size); + r = amdgpu_ring_alloc(ring, size); if (r) return r; @@ -326,7 +205,7 @@ int amdgpu_ring_restore(struct amdgpu_ring *ring, amdgpu_ring_write(ring, data[i]); } - amdgpu_ring_unlock_commit(ring); + amdgpu_ring_commit(ring); kfree(data); return 0; } @@ -357,7 +236,8 @@ int amdgpu_ring_init(struct amdgpu_device *adev, struct amdgpu_ring *ring, ring->adev = adev; ring->idx = adev->num_rings++; adev->rings[ring->idx] = ring; - r = amdgpu_fence_driver_init_ring(ring); + r = amdgpu_fence_driver_init_ring(ring, + amdgpu_sched_hw_submission); if (r) return r; } @@ -394,7 +274,6 @@ int amdgpu_ring_init(struct amdgpu_device *adev, struct amdgpu_ring *ring, return r; } - ring->ring_lock = &adev->ring_lock; /* Align ring size */ rb_bufsz = order_base_2(ring_size / 8); ring_size = (1 << (rb_bufsz + 1)) * 4; @@ -431,12 +310,12 @@ int amdgpu_ring_init(struct amdgpu_device *adev, struct amdgpu_ring *ring, } } ring->ptr_mask = (ring->ring_size / 4) - 1; - ring->ring_free_dw = ring->ring_size / 4; + ring->max_dw = DIV_ROUND_UP(ring->ring_size / 4, + amdgpu_sched_hw_submission); if (amdgpu_debugfs_ring_init(adev, ring)) { DRM_ERROR("Failed to register debugfs file for rings !\n"); } - amdgpu_ring_lockup_update(ring); return 0; } @@ -453,15 +332,10 @@ void amdgpu_ring_fini(struct amdgpu_ring *ring) int r; struct amdgpu_bo *ring_obj; - if (ring->ring_lock == NULL) - return; - - mutex_lock(ring->ring_lock); ring_obj = ring->ring_obj; ring->ready = false; ring->ring = NULL; ring->ring_obj = NULL; - mutex_unlock(ring->ring_lock); amdgpu_wb_free(ring->adev, ring->fence_offs); amdgpu_wb_free(ring->adev, ring->rptr_offs); @@ -493,29 +367,18 @@ static int amdgpu_debugfs_ring_info(struct seq_file *m, void *data) struct amdgpu_ring *ring = (void *)(((uint8_t*)adev) + roffset); uint32_t rptr, wptr, rptr_next; - unsigned count, i, j; - - amdgpu_ring_free_size(ring); - count = (ring->ring_size / 4) - ring->ring_free_dw; + unsigned i; wptr = amdgpu_ring_get_wptr(ring); - seq_printf(m, "wptr: 0x%08x [%5d]\n", - wptr, wptr); + seq_printf(m, "wptr: 0x%08x [%5d]\n", wptr, wptr); rptr = amdgpu_ring_get_rptr(ring); - seq_printf(m, "rptr: 0x%08x [%5d]\n", - rptr, rptr); + rptr_next = le32_to_cpu(*ring->next_rptr_cpu_addr); - rptr_next = ~0; + seq_printf(m, "rptr: 0x%08x [%5d]\n", rptr, rptr); seq_printf(m, "driver's copy of the wptr: 0x%08x [%5d]\n", ring->wptr, ring->wptr); - seq_printf(m, "last semaphore signal addr : 0x%016llx\n", - ring->last_semaphore_signal_addr); - seq_printf(m, "last semaphore wait addr : 0x%016llx\n", - ring->last_semaphore_wait_addr); - seq_printf(m, "%u free dwords in ring\n", ring->ring_free_dw); - seq_printf(m, "%u dwords in ring\n", count); if (!ring->ready) return 0; @@ -524,11 +387,20 @@ static int amdgpu_debugfs_ring_info(struct seq_file *m, void *data) * packet that is the root issue */ i = (rptr + ring->ptr_mask + 1 - 32) & ring->ptr_mask; - for (j = 0; j <= (count + 32); j++) { + while (i != rptr) { + seq_printf(m, "r[%5d]=0x%08x", i, ring->ring[i]); + if (i == rptr) + seq_puts(m, " *"); + if (i == rptr_next) + seq_puts(m, " #"); + seq_puts(m, "\n"); + i = (i + 1) & ring->ptr_mask; + } + while (i != wptr) { seq_printf(m, "r[%5d]=0x%08x", i, ring->ring[i]); - if (rptr == i) + if (i == rptr) seq_puts(m, " *"); - if (rptr_next == i) + if (i == rptr_next) seq_puts(m, " #"); seq_puts(m, "\n"); i = (i + 1) & ring->ptr_mask; @@ -540,8 +412,8 @@ static int amdgpu_debugfs_ring_info(struct seq_file *m, void *data) static int amdgpu_gfx_index = offsetof(struct amdgpu_device, gfx.gfx_ring[0]); static int cayman_cp1_index = offsetof(struct amdgpu_device, gfx.compute_ring[0]); static int cayman_cp2_index = offsetof(struct amdgpu_device, gfx.compute_ring[1]); -static int amdgpu_dma1_index = offsetof(struct amdgpu_device, sdma[0].ring); -static int amdgpu_dma2_index = offsetof(struct amdgpu_device, sdma[1].ring); +static int amdgpu_dma1_index = offsetof(struct amdgpu_device, sdma.instance[0].ring); +static int amdgpu_dma2_index = offsetof(struct amdgpu_device, sdma.instance[1].ring); static int r600_uvd_index = offsetof(struct amdgpu_device, uvd.ring); static int si_vce1_index = offsetof(struct amdgpu_device, vce.ring[0]); static int si_vce2_index = offsetof(struct amdgpu_device, vce.ring[1]); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_sa.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_sa.c index e90712443fe9..8bf84efafb04 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_sa.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_sa.c @@ -60,9 +60,8 @@ int amdgpu_sa_bo_manager_init(struct amdgpu_device *adev, sa_manager->align = align; sa_manager->hole = &sa_manager->olist; INIT_LIST_HEAD(&sa_manager->olist); - for (i = 0; i < AMDGPU_MAX_RINGS; ++i) { + for (i = 0; i < AMDGPU_SA_NUM_FENCE_LISTS; ++i) INIT_LIST_HEAD(&sa_manager->flist[i]); - } r = amdgpu_bo_create(adev, size, align, true, domain, 0, NULL, NULL, &sa_manager->bo); @@ -139,25 +138,6 @@ int amdgpu_sa_bo_manager_suspend(struct amdgpu_device *adev, return r; } -static uint32_t amdgpu_sa_get_ring_from_fence(struct fence *f) -{ - struct amdgpu_fence *a_fence; - struct amd_sched_fence *s_fence; - - s_fence = to_amd_sched_fence(f); - if (s_fence) { - struct amdgpu_ring *ring; - - ring = container_of(s_fence->sched, struct amdgpu_ring, sched); - return ring->idx; - } - - a_fence = to_amdgpu_fence(f); - if (a_fence) - return a_fence->ring->idx; - return 0; -} - static void amdgpu_sa_bo_remove_locked(struct amdgpu_sa_bo *sa_bo) { struct amdgpu_sa_manager *sa_manager = sa_bo->manager; @@ -247,11 +227,9 @@ static bool amdgpu_sa_event(struct amdgpu_sa_manager *sa_manager, unsigned soffset, eoffset, wasted; int i; - for (i = 0; i < AMDGPU_MAX_RINGS; ++i) { - if (!list_empty(&sa_manager->flist[i])) { + for (i = 0; i < AMDGPU_SA_NUM_FENCE_LISTS; ++i) + if (!list_empty(&sa_manager->flist[i])) return true; - } - } soffset = amdgpu_sa_bo_hole_soffset(sa_manager); eoffset = amdgpu_sa_bo_hole_eoffset(sa_manager); @@ -284,12 +262,11 @@ static bool amdgpu_sa_bo_next_hole(struct amdgpu_sa_manager *sa_manager, /* go over all fence list and try to find the closest sa_bo * of the current last */ - for (i = 0; i < AMDGPU_MAX_RINGS; ++i) { + for (i = 0; i < AMDGPU_SA_NUM_FENCE_LISTS; ++i) { struct amdgpu_sa_bo *sa_bo; - if (list_empty(&sa_manager->flist[i])) { + if (list_empty(&sa_manager->flist[i])) continue; - } sa_bo = list_first_entry(&sa_manager->flist[i], struct amdgpu_sa_bo, flist); @@ -318,7 +295,9 @@ static bool amdgpu_sa_bo_next_hole(struct amdgpu_sa_manager *sa_manager, } if (best_bo) { - uint32_t idx = amdgpu_sa_get_ring_from_fence(best_bo->fence); + uint32_t idx = best_bo->fence->context; + + idx %= AMDGPU_SA_NUM_FENCE_LISTS; ++tries[idx]; sa_manager->hole = best_bo->olist.prev; @@ -330,18 +309,21 @@ static bool amdgpu_sa_bo_next_hole(struct amdgpu_sa_manager *sa_manager, return false; } -int amdgpu_sa_bo_new(struct amdgpu_device *adev, - struct amdgpu_sa_manager *sa_manager, +int amdgpu_sa_bo_new(struct amdgpu_sa_manager *sa_manager, struct amdgpu_sa_bo **sa_bo, unsigned size, unsigned align) { - struct fence *fences[AMDGPU_MAX_RINGS]; - unsigned tries[AMDGPU_MAX_RINGS]; + struct fence *fences[AMDGPU_SA_NUM_FENCE_LISTS]; + unsigned tries[AMDGPU_SA_NUM_FENCE_LISTS]; + unsigned count; int i, r; signed long t; - BUG_ON(align > sa_manager->align); - BUG_ON(size > sa_manager->size); + if (WARN_ON_ONCE(align > sa_manager->align)) + return -EINVAL; + + if (WARN_ON_ONCE(size > sa_manager->size)) + return -EINVAL; *sa_bo = kmalloc(sizeof(struct amdgpu_sa_bo), GFP_KERNEL); if ((*sa_bo) == NULL) { @@ -354,7 +336,7 @@ int amdgpu_sa_bo_new(struct amdgpu_device *adev, spin_lock(&sa_manager->wq.lock); do { - for (i = 0; i < AMDGPU_MAX_RINGS; ++i) { + for (i = 0; i < AMDGPU_SA_NUM_FENCE_LISTS; ++i) { fences[i] = NULL; tries[i] = 0; } @@ -371,13 +353,21 @@ int amdgpu_sa_bo_new(struct amdgpu_device *adev, /* see if we can skip over some allocations */ } while (amdgpu_sa_bo_next_hole(sa_manager, fences, tries)); - spin_unlock(&sa_manager->wq.lock); - t = amdgpu_fence_wait_any(adev, fences, AMDGPU_MAX_RINGS, - false, MAX_SCHEDULE_TIMEOUT); - r = (t > 0) ? 0 : t; - spin_lock(&sa_manager->wq.lock); - /* if we have nothing to wait for block */ - if (r == -ENOENT) { + for (i = 0, count = 0; i < AMDGPU_SA_NUM_FENCE_LISTS; ++i) + if (fences[i]) + fences[count++] = fence_get(fences[i]); + + if (count) { + spin_unlock(&sa_manager->wq.lock); + t = fence_wait_any_timeout(fences, count, false, + MAX_SCHEDULE_TIMEOUT); + for (i = 0; i < count; ++i) + fence_put(fences[i]); + + r = (t > 0) ? 0 : t; + spin_lock(&sa_manager->wq.lock); + } else { + /* if we have nothing to wait for block */ r = wait_event_interruptible_locked( sa_manager->wq, amdgpu_sa_event(sa_manager, size, align) @@ -405,8 +395,9 @@ void amdgpu_sa_bo_free(struct amdgpu_device *adev, struct amdgpu_sa_bo **sa_bo, spin_lock(&sa_manager->wq.lock); if (fence && !fence_is_signaled(fence)) { uint32_t idx; + (*sa_bo)->fence = fence_get(fence); - idx = amdgpu_sa_get_ring_from_fence(fence); + idx = fence->context % AMDGPU_SA_NUM_FENCE_LISTS; list_add_tail(&(*sa_bo)->flist, &sa_manager->flist[idx]); } else { amdgpu_sa_bo_remove_locked(*sa_bo); @@ -418,25 +409,6 @@ void amdgpu_sa_bo_free(struct amdgpu_device *adev, struct amdgpu_sa_bo **sa_bo, #if defined(CONFIG_DEBUG_FS) -static void amdgpu_sa_bo_dump_fence(struct fence *fence, struct seq_file *m) -{ - struct amdgpu_fence *a_fence = to_amdgpu_fence(fence); - struct amd_sched_fence *s_fence = to_amd_sched_fence(fence); - - if (a_fence) - seq_printf(m, " protected by 0x%016llx on ring %d", - a_fence->seq, a_fence->ring->idx); - - if (s_fence) { - struct amdgpu_ring *ring; - - - ring = container_of(s_fence->sched, struct amdgpu_ring, sched); - seq_printf(m, " protected by 0x%016x on ring %d", - s_fence->base.seqno, ring->idx); - } -} - void amdgpu_sa_bo_dump_debug_info(struct amdgpu_sa_manager *sa_manager, struct seq_file *m) { @@ -453,8 +425,11 @@ void amdgpu_sa_bo_dump_debug_info(struct amdgpu_sa_manager *sa_manager, } seq_printf(m, "[0x%010llx 0x%010llx] size %8lld", soffset, eoffset, eoffset - soffset); + if (i->fence) - amdgpu_sa_bo_dump_fence(i->fence, m); + seq_printf(m, " protected by 0x%08x on context %d", + i->fence->seqno, i->fence->context); + seq_printf(m, "\n"); } spin_unlock(&sa_manager->wq.lock); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c deleted file mode 100644 index 2e946b2cad88..000000000000 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c +++ /dev/null @@ -1,113 +0,0 @@ -/* - * Copyright 2015 Advanced Micro Devices, Inc. - * - * Permission is hereby granted, free of charge, to any person obtaining a - * copy of this software and associated documentation files (the "Software"), - * to deal in the Software without restriction, including without limitation - * the rights to use, copy, modify, merge, publish, distribute, sublicense, - * and/or sell copies of the Software, and to permit persons to whom the - * Software is furnished to do so, subject to the following conditions: - * - * The above copyright notice and this permission notice shall be included in - * all copies or substantial portions of the Software. - * - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR - * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, - * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL - * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR - * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, - * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR - * OTHER DEALINGS IN THE SOFTWARE. - * - * - */ -#include <linux/kthread.h> -#include <linux/wait.h> -#include <linux/sched.h> -#include <drm/drmP.h> -#include "amdgpu.h" - -static struct fence *amdgpu_sched_dependency(struct amd_sched_job *sched_job) -{ - struct amdgpu_job *job = to_amdgpu_job(sched_job); - return amdgpu_sync_get_fence(&job->ibs->sync); -} - -static struct fence *amdgpu_sched_run_job(struct amd_sched_job *sched_job) -{ - struct amdgpu_fence *fence = NULL; - struct amdgpu_job *job; - int r; - - if (!sched_job) { - DRM_ERROR("job is null\n"); - return NULL; - } - job = to_amdgpu_job(sched_job); - mutex_lock(&job->job_lock); - r = amdgpu_ib_schedule(job->adev, - job->num_ibs, - job->ibs, - job->base.owner); - if (r) { - DRM_ERROR("Error scheduling IBs (%d)\n", r); - goto err; - } - - fence = amdgpu_fence_ref(job->ibs[job->num_ibs - 1].fence); - -err: - if (job->free_job) - job->free_job(job); - - mutex_unlock(&job->job_lock); - fence_put(&job->base.s_fence->base); - kfree(job); - return fence ? &fence->base : NULL; -} - -struct amd_sched_backend_ops amdgpu_sched_ops = { - .dependency = amdgpu_sched_dependency, - .run_job = amdgpu_sched_run_job, -}; - -int amdgpu_sched_ib_submit_kernel_helper(struct amdgpu_device *adev, - struct amdgpu_ring *ring, - struct amdgpu_ib *ibs, - unsigned num_ibs, - int (*free_job)(struct amdgpu_job *), - void *owner, - struct fence **f) -{ - int r = 0; - if (amdgpu_enable_scheduler) { - struct amdgpu_job *job = - kzalloc(sizeof(struct amdgpu_job), GFP_KERNEL); - if (!job) - return -ENOMEM; - job->base.sched = &ring->sched; - job->base.s_entity = &adev->kernel_ctx.rings[ring->idx].entity; - job->adev = adev; - job->ibs = ibs; - job->num_ibs = num_ibs; - job->base.owner = owner; - mutex_init(&job->job_lock); - job->free_job = free_job; - mutex_lock(&job->job_lock); - r = amd_sched_entity_push_job(&job->base); - if (r) { - mutex_unlock(&job->job_lock); - kfree(job); - return r; - } - *f = fence_get(&job->base.s_fence->base); - mutex_unlock(&job->job_lock); - } else { - r = amdgpu_ib_schedule(adev, num_ibs, ibs, owner); - if (r) - return r; - *f = fence_get(&ibs[num_ibs - 1].fence->base); - } - - return 0; -} diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_semaphore.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_semaphore.c deleted file mode 100644 index ff3ca52ec6fe..000000000000 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_semaphore.c +++ /dev/null @@ -1,102 +0,0 @@ -/* - * Copyright 2011 Christian König. - * All Rights Reserved. - * - * Permission is hereby granted, free of charge, to any person obtaining a - * copy of this software and associated documentation files (the - * "Software"), to deal in the Software without restriction, including - * without limitation the rights to use, copy, modify, merge, publish, - * distribute, sub license, and/or sell copies of the Software, and to - * permit persons to whom the Software is furnished to do so, subject to - * the following conditions: - * - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR - * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, - * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL - * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, - * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR - * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE - * USE OR OTHER DEALINGS IN THE SOFTWARE. - * - * The above copyright notice and this permission notice (including the - * next paragraph) shall be included in all copies or substantial portions - * of the Software. - * - */ -/* - * Authors: - * Christian König <deathsimple@vodafone.de> - */ -#include <drm/drmP.h> -#include "amdgpu.h" -#include "amdgpu_trace.h" - -int amdgpu_semaphore_create(struct amdgpu_device *adev, - struct amdgpu_semaphore **semaphore) -{ - int r; - - *semaphore = kmalloc(sizeof(struct amdgpu_semaphore), GFP_KERNEL); - if (*semaphore == NULL) { - return -ENOMEM; - } - r = amdgpu_sa_bo_new(adev, &adev->ring_tmp_bo, - &(*semaphore)->sa_bo, 8, 8); - if (r) { - kfree(*semaphore); - *semaphore = NULL; - return r; - } - (*semaphore)->waiters = 0; - (*semaphore)->gpu_addr = amdgpu_sa_bo_gpu_addr((*semaphore)->sa_bo); - - *((uint64_t *)amdgpu_sa_bo_cpu_addr((*semaphore)->sa_bo)) = 0; - - return 0; -} - -bool amdgpu_semaphore_emit_signal(struct amdgpu_ring *ring, - struct amdgpu_semaphore *semaphore) -{ - trace_amdgpu_semaphore_signale(ring->idx, semaphore); - - if (amdgpu_ring_emit_semaphore(ring, semaphore, false)) { - --semaphore->waiters; - - /* for debugging lockup only, used by sysfs debug files */ - ring->last_semaphore_signal_addr = semaphore->gpu_addr; - return true; - } - return false; -} - -bool amdgpu_semaphore_emit_wait(struct amdgpu_ring *ring, - struct amdgpu_semaphore *semaphore) -{ - trace_amdgpu_semaphore_wait(ring->idx, semaphore); - - if (amdgpu_ring_emit_semaphore(ring, semaphore, true)) { - ++semaphore->waiters; - - /* for debugging lockup only, used by sysfs debug files */ - ring->last_semaphore_wait_addr = semaphore->gpu_addr; - return true; - } - return false; -} - -void amdgpu_semaphore_free(struct amdgpu_device *adev, - struct amdgpu_semaphore **semaphore, - struct fence *fence) -{ - if (semaphore == NULL || *semaphore == NULL) { - return; - } - if ((*semaphore)->waiters > 0) { - dev_err(adev->dev, "semaphore %p has more waiters than signalers," - " hardware lockup imminent!\n", *semaphore); - } - amdgpu_sa_bo_free(adev, &(*semaphore)->sa_bo, fence); - kfree(*semaphore); - *semaphore = NULL; -} diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_sync.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_sync.c index 4921de15b451..c48b4fce5e57 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_sync.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_sync.c @@ -37,6 +37,8 @@ struct amdgpu_sync_entry { struct fence *fence; }; +static struct kmem_cache *amdgpu_sync_slab; + /** * amdgpu_sync_create - zero init sync object * @@ -46,26 +48,22 @@ struct amdgpu_sync_entry { */ void amdgpu_sync_create(struct amdgpu_sync *sync) { - unsigned i; - - for (i = 0; i < AMDGPU_NUM_SYNCS; ++i) - sync->semaphores[i] = NULL; - - for (i = 0; i < AMDGPU_MAX_RINGS; ++i) - sync->sync_to[i] = NULL; - hash_init(sync->fences); sync->last_vm_update = NULL; } +/** + * amdgpu_sync_same_dev - test if fence belong to us + * + * @adev: amdgpu device to use for the test + * @f: fence to test + * + * Test if the fence was issued by us. + */ static bool amdgpu_sync_same_dev(struct amdgpu_device *adev, struct fence *f) { - struct amdgpu_fence *a_fence = to_amdgpu_fence(f); struct amd_sched_fence *s_fence = to_amd_sched_fence(f); - if (a_fence) - return a_fence->ring->adev == adev; - if (s_fence) { struct amdgpu_ring *ring; @@ -76,15 +74,38 @@ static bool amdgpu_sync_same_dev(struct amdgpu_device *adev, struct fence *f) return false; } -static bool amdgpu_sync_test_owner(struct fence *f, void *owner) +/** + * amdgpu_sync_get_owner - extract the owner of a fence + * + * @fence: fence get the owner from + * + * Extract who originally created the fence. + */ +static void *amdgpu_sync_get_owner(struct fence *f) { - struct amdgpu_fence *a_fence = to_amdgpu_fence(f); struct amd_sched_fence *s_fence = to_amd_sched_fence(f); + if (s_fence) - return s_fence->owner == owner; - if (a_fence) - return a_fence->owner == owner; - return false; + return s_fence->owner; + + return AMDGPU_FENCE_OWNER_UNDEFINED; +} + +/** + * amdgpu_sync_keep_later - Keep the later fence + * + * @keep: existing fence to test + * @fence: new fence + * + * Either keep the existing fence or the new one, depending which one is later. + */ +static void amdgpu_sync_keep_later(struct fence **keep, struct fence *fence) +{ + if (*keep && fence_is_later(*keep, fence)) + return; + + fence_put(*keep); + *keep = fence_get(fence); } /** @@ -98,76 +119,39 @@ int amdgpu_sync_fence(struct amdgpu_device *adev, struct amdgpu_sync *sync, struct fence *f) { struct amdgpu_sync_entry *e; - struct amdgpu_fence *fence; - struct amdgpu_fence *other; - struct fence *tmp, *later; if (!f) return 0; if (amdgpu_sync_same_dev(adev, f) && - amdgpu_sync_test_owner(f, AMDGPU_FENCE_OWNER_VM)) { - if (sync->last_vm_update) { - tmp = sync->last_vm_update; - BUG_ON(f->context != tmp->context); - later = (f->seqno - tmp->seqno <= INT_MAX) ? f : tmp; - sync->last_vm_update = fence_get(later); - fence_put(tmp); - } else - sync->last_vm_update = fence_get(f); - } - - fence = to_amdgpu_fence(f); - if (!fence || fence->ring->adev != adev) { - hash_for_each_possible(sync->fences, e, node, f->context) { - struct fence *new; - if (unlikely(e->fence->context != f->context)) - continue; - new = fence_get(fence_later(e->fence, f)); - if (new) { - fence_put(e->fence); - e->fence = new; - } - return 0; - } + amdgpu_sync_get_owner(f) == AMDGPU_FENCE_OWNER_VM) + amdgpu_sync_keep_later(&sync->last_vm_update, f); - e = kmalloc(sizeof(struct amdgpu_sync_entry), GFP_KERNEL); - if (!e) - return -ENOMEM; + hash_for_each_possible(sync->fences, e, node, f->context) { + if (unlikely(e->fence->context != f->context)) + continue; - hash_add(sync->fences, &e->node, f->context); - e->fence = fence_get(f); + amdgpu_sync_keep_later(&e->fence, f); return 0; } - other = sync->sync_to[fence->ring->idx]; - sync->sync_to[fence->ring->idx] = amdgpu_fence_ref( - amdgpu_fence_later(fence, other)); - amdgpu_fence_unref(&other); + e = kmem_cache_alloc(amdgpu_sync_slab, GFP_KERNEL); + if (!e) + return -ENOMEM; + hash_add(sync->fences, &e->node, f->context); + e->fence = fence_get(f); return 0; } -static void *amdgpu_sync_get_owner(struct fence *f) -{ - struct amdgpu_fence *a_fence = to_amdgpu_fence(f); - struct amd_sched_fence *s_fence = to_amd_sched_fence(f); - - if (s_fence) - return s_fence->owner; - else if (a_fence) - return a_fence->owner; - return AMDGPU_FENCE_OWNER_UNDEFINED; -} - /** - * amdgpu_sync_resv - use the semaphores to sync to a reservation object + * amdgpu_sync_resv - sync to a reservation object * * @sync: sync object to add fences from reservation object to * @resv: reservation object with embedded fence * @shared: true if we should only sync to the exclusive fence * - * Sync to the fence using the semaphore objects + * Sync to the fence */ int amdgpu_sync_resv(struct amdgpu_device *adev, struct amdgpu_sync *sync, @@ -199,8 +183,8 @@ int amdgpu_sync_resv(struct amdgpu_device *adev, * for other VM updates and moves. */ fence_owner = amdgpu_sync_get_owner(f); - if ((owner != AMDGPU_FENCE_OWNER_MOVE) && - (fence_owner != AMDGPU_FENCE_OWNER_MOVE) && + if ((owner != AMDGPU_FENCE_OWNER_UNDEFINED) && + (fence_owner != AMDGPU_FENCE_OWNER_UNDEFINED) && ((owner == AMDGPU_FENCE_OWNER_VM) != (fence_owner == AMDGPU_FENCE_OWNER_VM))) continue; @@ -232,7 +216,7 @@ struct fence *amdgpu_sync_get_fence(struct amdgpu_sync *sync) f = e->fence; hash_del(&e->node); - kfree(e); + kmem_cache_free(amdgpu_sync_slab, e); if (!fence_is_signaled(f)) return f; @@ -255,97 +239,7 @@ int amdgpu_sync_wait(struct amdgpu_sync *sync) hash_del(&e->node); fence_put(e->fence); - kfree(e); - } - - if (amdgpu_enable_semaphores) - return 0; - - for (i = 0; i < AMDGPU_MAX_RINGS; ++i) { - struct amdgpu_fence *fence = sync->sync_to[i]; - if (!fence) - continue; - - r = fence_wait(&fence->base, false); - if (r) - return r; - } - - return 0; -} - -/** - * amdgpu_sync_rings - sync ring to all registered fences - * - * @sync: sync object to use - * @ring: ring that needs sync - * - * Ensure that all registered fences are signaled before letting - * the ring continue. The caller must hold the ring lock. - */ -int amdgpu_sync_rings(struct amdgpu_sync *sync, - struct amdgpu_ring *ring) -{ - struct amdgpu_device *adev = ring->adev; - unsigned count = 0; - int i, r; - - for (i = 0; i < AMDGPU_MAX_RINGS; ++i) { - struct amdgpu_fence *fence = sync->sync_to[i]; - struct amdgpu_semaphore *semaphore; - struct amdgpu_ring *other = adev->rings[i]; - - /* check if we really need to sync */ - if (!amdgpu_fence_need_sync(fence, ring)) - continue; - - /* prevent GPU deadlocks */ - if (!other->ready) { - dev_err(adev->dev, "Syncing to a disabled ring!"); - return -EINVAL; - } - - if (amdgpu_enable_scheduler || !amdgpu_enable_semaphores || - (count >= AMDGPU_NUM_SYNCS)) { - /* not enough room, wait manually */ - r = fence_wait(&fence->base, false); - if (r) - return r; - continue; - } - r = amdgpu_semaphore_create(adev, &semaphore); - if (r) - return r; - - sync->semaphores[count++] = semaphore; - - /* allocate enough space for sync command */ - r = amdgpu_ring_alloc(other, 16); - if (r) - return r; - - /* emit the signal semaphore */ - if (!amdgpu_semaphore_emit_signal(other, semaphore)) { - /* signaling wasn't successful wait manually */ - amdgpu_ring_undo(other); - r = fence_wait(&fence->base, false); - if (r) - return r; - continue; - } - - /* we assume caller has already allocated space on waiters ring */ - if (!amdgpu_semaphore_emit_wait(ring, semaphore)) { - /* waiting wasn't successful wait manually */ - amdgpu_ring_undo(other); - r = fence_wait(&fence->base, false); - if (r) - return r; - continue; - } - - amdgpu_ring_commit(other); - amdgpu_fence_note_sync(fence, ring); + kmem_cache_free(amdgpu_sync_slab, e); } return 0; @@ -354,15 +248,11 @@ int amdgpu_sync_rings(struct amdgpu_sync *sync, /** * amdgpu_sync_free - free the sync object * - * @adev: amdgpu_device pointer * @sync: sync object to use - * @fence: fence to use for the free * - * Free the sync object by freeing all semaphores in it. + * Free the sync object. */ -void amdgpu_sync_free(struct amdgpu_device *adev, - struct amdgpu_sync *sync, - struct fence *fence) +void amdgpu_sync_free(struct amdgpu_sync *sync) { struct amdgpu_sync_entry *e; struct hlist_node *tmp; @@ -371,14 +261,34 @@ void amdgpu_sync_free(struct amdgpu_device *adev, hash_for_each_safe(sync->fences, i, tmp, e, node) { hash_del(&e->node); fence_put(e->fence); - kfree(e); + kmem_cache_free(amdgpu_sync_slab, e); } - for (i = 0; i < AMDGPU_NUM_SYNCS; ++i) - amdgpu_semaphore_free(adev, &sync->semaphores[i], fence); + fence_put(sync->last_vm_update); +} + +/** + * amdgpu_sync_init - init sync object subsystem + * + * Allocate the slab allocator. + */ +int amdgpu_sync_init(void) +{ + amdgpu_sync_slab = kmem_cache_create( + "amdgpu_sync", sizeof(struct amdgpu_sync_entry), 0, + SLAB_HWCACHE_ALIGN, NULL); + if (!amdgpu_sync_slab) + return -ENOMEM; - for (i = 0; i < AMDGPU_MAX_RINGS; ++i) - amdgpu_fence_unref(&sync->sync_to[i]); + return 0; +} - fence_put(sync->last_vm_update); +/** + * amdgpu_sync_fini - fini sync object subsystem + * + * Free the slab allocator. + */ +void amdgpu_sync_fini(void) +{ + kmem_cache_destroy(amdgpu_sync_slab); } diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_test.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_test.c index 4865615e9c06..05a53f4fc334 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_test.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_test.c @@ -238,144 +238,10 @@ void amdgpu_test_moves(struct amdgpu_device *adev) amdgpu_do_test_moves(adev); } -static int amdgpu_test_create_and_emit_fence(struct amdgpu_device *adev, - struct amdgpu_ring *ring, - struct fence **fence) -{ - uint32_t handle = ring->idx ^ 0xdeafbeef; - int r; - - if (ring == &adev->uvd.ring) { - r = amdgpu_uvd_get_create_msg(ring, handle, NULL); - if (r) { - DRM_ERROR("Failed to get dummy create msg\n"); - return r; - } - - r = amdgpu_uvd_get_destroy_msg(ring, handle, fence); - if (r) { - DRM_ERROR("Failed to get dummy destroy msg\n"); - return r; - } - - } else if (ring == &adev->vce.ring[0] || - ring == &adev->vce.ring[1]) { - r = amdgpu_vce_get_create_msg(ring, handle, NULL); - if (r) { - DRM_ERROR("Failed to get dummy create msg\n"); - return r; - } - - r = amdgpu_vce_get_destroy_msg(ring, handle, fence); - if (r) { - DRM_ERROR("Failed to get dummy destroy msg\n"); - return r; - } - } else { - struct amdgpu_fence *a_fence = NULL; - r = amdgpu_ring_lock(ring, 64); - if (r) { - DRM_ERROR("Failed to lock ring A %d\n", ring->idx); - return r; - } - amdgpu_fence_emit(ring, AMDGPU_FENCE_OWNER_UNDEFINED, &a_fence); - amdgpu_ring_unlock_commit(ring); - *fence = &a_fence->base; - } - return 0; -} - void amdgpu_test_ring_sync(struct amdgpu_device *adev, struct amdgpu_ring *ringA, struct amdgpu_ring *ringB) { - struct fence *fence1 = NULL, *fence2 = NULL; - struct amdgpu_semaphore *semaphore = NULL; - int r; - - r = amdgpu_semaphore_create(adev, &semaphore); - if (r) { - DRM_ERROR("Failed to create semaphore\n"); - goto out_cleanup; - } - - r = amdgpu_ring_lock(ringA, 64); - if (r) { - DRM_ERROR("Failed to lock ring A %d\n", ringA->idx); - goto out_cleanup; - } - amdgpu_semaphore_emit_wait(ringA, semaphore); - amdgpu_ring_unlock_commit(ringA); - - r = amdgpu_test_create_and_emit_fence(adev, ringA, &fence1); - if (r) - goto out_cleanup; - - r = amdgpu_ring_lock(ringA, 64); - if (r) { - DRM_ERROR("Failed to lock ring A %d\n", ringA->idx); - goto out_cleanup; - } - amdgpu_semaphore_emit_wait(ringA, semaphore); - amdgpu_ring_unlock_commit(ringA); - - r = amdgpu_test_create_and_emit_fence(adev, ringA, &fence2); - if (r) - goto out_cleanup; - - mdelay(1000); - - if (fence_is_signaled(fence1)) { - DRM_ERROR("Fence 1 signaled without waiting for semaphore.\n"); - goto out_cleanup; - } - - r = amdgpu_ring_lock(ringB, 64); - if (r) { - DRM_ERROR("Failed to lock ring B %p\n", ringB); - goto out_cleanup; - } - amdgpu_semaphore_emit_signal(ringB, semaphore); - amdgpu_ring_unlock_commit(ringB); - - r = fence_wait(fence1, false); - if (r) { - DRM_ERROR("Failed to wait for sync fence 1\n"); - goto out_cleanup; - } - - mdelay(1000); - - if (fence_is_signaled(fence2)) { - DRM_ERROR("Fence 2 signaled without waiting for semaphore.\n"); - goto out_cleanup; - } - - r = amdgpu_ring_lock(ringB, 64); - if (r) { - DRM_ERROR("Failed to lock ring B %p\n", ringB); - goto out_cleanup; - } - amdgpu_semaphore_emit_signal(ringB, semaphore); - amdgpu_ring_unlock_commit(ringB); - - r = fence_wait(fence2, false); - if (r) { - DRM_ERROR("Failed to wait for sync fence 1\n"); - goto out_cleanup; - } - -out_cleanup: - amdgpu_semaphore_free(adev, &semaphore, NULL); - - if (fence1) - fence_put(fence1); - - if (fence2) - fence_put(fence2); - - if (r) - printk(KERN_WARNING "Error while testing ring sync (%d).\n", r); } static void amdgpu_test_ring_sync2(struct amdgpu_device *adev, @@ -383,109 +249,6 @@ static void amdgpu_test_ring_sync2(struct amdgpu_device *adev, struct amdgpu_ring *ringB, struct amdgpu_ring *ringC) { - struct fence *fenceA = NULL, *fenceB = NULL; - struct amdgpu_semaphore *semaphore = NULL; - bool sigA, sigB; - int i, r; - - r = amdgpu_semaphore_create(adev, &semaphore); - if (r) { - DRM_ERROR("Failed to create semaphore\n"); - goto out_cleanup; - } - - r = amdgpu_ring_lock(ringA, 64); - if (r) { - DRM_ERROR("Failed to lock ring A %d\n", ringA->idx); - goto out_cleanup; - } - amdgpu_semaphore_emit_wait(ringA, semaphore); - amdgpu_ring_unlock_commit(ringA); - - r = amdgpu_test_create_and_emit_fence(adev, ringA, &fenceA); - if (r) - goto out_cleanup; - - r = amdgpu_ring_lock(ringB, 64); - if (r) { - DRM_ERROR("Failed to lock ring B %d\n", ringB->idx); - goto out_cleanup; - } - amdgpu_semaphore_emit_wait(ringB, semaphore); - amdgpu_ring_unlock_commit(ringB); - r = amdgpu_test_create_and_emit_fence(adev, ringB, &fenceB); - if (r) - goto out_cleanup; - - mdelay(1000); - - if (fence_is_signaled(fenceA)) { - DRM_ERROR("Fence A signaled without waiting for semaphore.\n"); - goto out_cleanup; - } - if (fence_is_signaled(fenceB)) { - DRM_ERROR("Fence B signaled without waiting for semaphore.\n"); - goto out_cleanup; - } - - r = amdgpu_ring_lock(ringC, 64); - if (r) { - DRM_ERROR("Failed to lock ring B %p\n", ringC); - goto out_cleanup; - } - amdgpu_semaphore_emit_signal(ringC, semaphore); - amdgpu_ring_unlock_commit(ringC); - - for (i = 0; i < 30; ++i) { - mdelay(100); - sigA = fence_is_signaled(fenceA); - sigB = fence_is_signaled(fenceB); - if (sigA || sigB) - break; - } - - if (!sigA && !sigB) { - DRM_ERROR("Neither fence A nor B has been signaled\n"); - goto out_cleanup; - } else if (sigA && sigB) { - DRM_ERROR("Both fence A and B has been signaled\n"); - goto out_cleanup; - } - - DRM_INFO("Fence %c was first signaled\n", sigA ? 'A' : 'B'); - - r = amdgpu_ring_lock(ringC, 64); - if (r) { - DRM_ERROR("Failed to lock ring B %p\n", ringC); - goto out_cleanup; - } - amdgpu_semaphore_emit_signal(ringC, semaphore); - amdgpu_ring_unlock_commit(ringC); - - mdelay(1000); - - r = fence_wait(fenceA, false); - if (r) { - DRM_ERROR("Failed to wait for sync fence A\n"); - goto out_cleanup; - } - r = fence_wait(fenceB, false); - if (r) { - DRM_ERROR("Failed to wait for sync fence B\n"); - goto out_cleanup; - } - -out_cleanup: - amdgpu_semaphore_free(adev, &semaphore, NULL); - - if (fenceA) - fence_put(fenceA); - - if (fenceB) - fence_put(fenceB); - - if (r) - printk(KERN_WARNING "Error while testing ring sync (%d).\n", r); } static bool amdgpu_test_sync_possible(struct amdgpu_ring *ringA, diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_trace.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_trace.h index 961d7265c286..26a5f4acf584 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_trace.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_trace.h @@ -38,29 +38,86 @@ TRACE_EVENT(amdgpu_cs, TP_fast_assign( __entry->bo_list = p->bo_list; - __entry->ring = p->ibs[i].ring->idx; - __entry->dw = p->ibs[i].length_dw; + __entry->ring = p->job->ring->idx; + __entry->dw = p->job->ibs[i].length_dw; __entry->fences = amdgpu_fence_count_emitted( - p->ibs[i].ring); + p->job->ring); ), TP_printk("bo_list=%p, ring=%u, dw=%u, fences=%u", __entry->bo_list, __entry->ring, __entry->dw, __entry->fences) ); +TRACE_EVENT(amdgpu_cs_ioctl, + TP_PROTO(struct amdgpu_job *job), + TP_ARGS(job), + TP_STRUCT__entry( + __field(struct amdgpu_device *, adev) + __field(struct amd_sched_job *, sched_job) + __field(struct amdgpu_ib *, ib) + __field(struct fence *, fence) + __field(char *, ring_name) + __field(u32, num_ibs) + ), + + TP_fast_assign( + __entry->adev = job->adev; + __entry->sched_job = &job->base; + __entry->ib = job->ibs; + __entry->fence = &job->base.s_fence->base; + __entry->ring_name = job->ring->name; + __entry->num_ibs = job->num_ibs; + ), + TP_printk("adev=%p, sched_job=%p, first ib=%p, sched fence=%p, ring name:%s, num_ibs:%u", + __entry->adev, __entry->sched_job, __entry->ib, + __entry->fence, __entry->ring_name, __entry->num_ibs) +); + +TRACE_EVENT(amdgpu_sched_run_job, + TP_PROTO(struct amdgpu_job *job), + TP_ARGS(job), + TP_STRUCT__entry( + __field(struct amdgpu_device *, adev) + __field(struct amd_sched_job *, sched_job) + __field(struct amdgpu_ib *, ib) + __field(struct fence *, fence) + __field(char *, ring_name) + __field(u32, num_ibs) + ), + + TP_fast_assign( + __entry->adev = job->adev; + __entry->sched_job = &job->base; + __entry->ib = job->ibs; + __entry->fence = &job->base.s_fence->base; + __entry->ring_name = job->ring->name; + __entry->num_ibs = job->num_ibs; + ), + TP_printk("adev=%p, sched_job=%p, first ib=%p, sched fence=%p, ring name:%s, num_ibs:%u", + __entry->adev, __entry->sched_job, __entry->ib, + __entry->fence, __entry->ring_name, __entry->num_ibs) +); + + TRACE_EVENT(amdgpu_vm_grab_id, - TP_PROTO(unsigned vmid, int ring), - TP_ARGS(vmid, ring), + TP_PROTO(struct amdgpu_vm *vm, int ring, unsigned vmid, + uint64_t pd_addr), + TP_ARGS(vm, ring, vmid, pd_addr), TP_STRUCT__entry( - __field(u32, vmid) + __field(struct amdgpu_vm *, vm) __field(u32, ring) + __field(u32, vmid) + __field(u64, pd_addr) ), TP_fast_assign( - __entry->vmid = vmid; + __entry->vm = vm; __entry->ring = ring; + __entry->vmid = vmid; + __entry->pd_addr = pd_addr; ), - TP_printk("vmid=%u, ring=%u", __entry->vmid, __entry->ring) + TP_printk("vm=%p, ring=%u, id=%u, pd_addr=%010Lx", __entry->vm, + __entry->ring, __entry->vmid, __entry->pd_addr) ); TRACE_EVENT(amdgpu_vm_bo_map, @@ -111,7 +168,7 @@ TRACE_EVENT(amdgpu_vm_bo_unmap, __entry->offset, __entry->flags) ); -TRACE_EVENT(amdgpu_vm_bo_update, +DECLARE_EVENT_CLASS(amdgpu_vm_mapping, TP_PROTO(struct amdgpu_bo_va_mapping *mapping), TP_ARGS(mapping), TP_STRUCT__entry( @@ -129,6 +186,16 @@ TRACE_EVENT(amdgpu_vm_bo_update, __entry->soffset, __entry->eoffset, __entry->flags) ); +DEFINE_EVENT(amdgpu_vm_mapping, amdgpu_vm_bo_update, + TP_PROTO(struct amdgpu_bo_va_mapping *mapping), + TP_ARGS(mapping) +); + +DEFINE_EVENT(amdgpu_vm_mapping, amdgpu_vm_bo_mapping, + TP_PROTO(struct amdgpu_bo_va_mapping *mapping), + TP_ARGS(mapping) +); + TRACE_EVENT(amdgpu_vm_set_page, TP_PROTO(uint64_t pe, uint64_t addr, unsigned count, uint32_t incr, uint32_t flags), @@ -167,8 +234,8 @@ TRACE_EVENT(amdgpu_vm_flush, __entry->ring = ring; __entry->id = id; ), - TP_printk("pd_addr=%010Lx, ring=%u, id=%u", - __entry->pd_addr, __entry->ring, __entry->id) + TP_printk("ring=%u, id=%u, pd_addr=%010Lx", + __entry->ring, __entry->id, __entry->pd_addr) ); TRACE_EVENT(amdgpu_bo_list_set, @@ -186,85 +253,6 @@ TRACE_EVENT(amdgpu_bo_list_set, TP_printk("list=%p, bo=%p", __entry->list, __entry->bo) ); -DECLARE_EVENT_CLASS(amdgpu_fence_request, - - TP_PROTO(struct drm_device *dev, int ring, u32 seqno), - - TP_ARGS(dev, ring, seqno), - - TP_STRUCT__entry( - __field(u32, dev) - __field(int, ring) - __field(u32, seqno) - ), - - TP_fast_assign( - __entry->dev = dev->primary->index; - __entry->ring = ring; - __entry->seqno = seqno; - ), - - TP_printk("dev=%u, ring=%d, seqno=%u", - __entry->dev, __entry->ring, __entry->seqno) -); - -DEFINE_EVENT(amdgpu_fence_request, amdgpu_fence_emit, - - TP_PROTO(struct drm_device *dev, int ring, u32 seqno), - - TP_ARGS(dev, ring, seqno) -); - -DEFINE_EVENT(amdgpu_fence_request, amdgpu_fence_wait_begin, - - TP_PROTO(struct drm_device *dev, int ring, u32 seqno), - - TP_ARGS(dev, ring, seqno) -); - -DEFINE_EVENT(amdgpu_fence_request, amdgpu_fence_wait_end, - - TP_PROTO(struct drm_device *dev, int ring, u32 seqno), - - TP_ARGS(dev, ring, seqno) -); - -DECLARE_EVENT_CLASS(amdgpu_semaphore_request, - - TP_PROTO(int ring, struct amdgpu_semaphore *sem), - - TP_ARGS(ring, sem), - - TP_STRUCT__entry( - __field(int, ring) - __field(signed, waiters) - __field(uint64_t, gpu_addr) - ), - - TP_fast_assign( - __entry->ring = ring; - __entry->waiters = sem->waiters; - __entry->gpu_addr = sem->gpu_addr; - ), - - TP_printk("ring=%u, waiters=%d, addr=%010Lx", __entry->ring, - __entry->waiters, __entry->gpu_addr) -); - -DEFINE_EVENT(amdgpu_semaphore_request, amdgpu_semaphore_signale, - - TP_PROTO(int ring, struct amdgpu_semaphore *sem), - - TP_ARGS(ring, sem) -); - -DEFINE_EVENT(amdgpu_semaphore_request, amdgpu_semaphore_wait, - - TP_PROTO(int ring, struct amdgpu_semaphore *sem), - - TP_ARGS(ring, sem) -); - #endif /* This part must be outside protection */ diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c index 364cbe975332..ab34190859a8 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c @@ -77,6 +77,8 @@ static void amdgpu_ttm_mem_global_release(struct drm_global_reference *ref) static int amdgpu_ttm_global_init(struct amdgpu_device *adev) { struct drm_global_reference *global_ref; + struct amdgpu_ring *ring; + struct amd_sched_rq *rq; int r; adev->mman.mem_global_referenced = false; @@ -106,13 +108,27 @@ static int amdgpu_ttm_global_init(struct amdgpu_device *adev) return r; } + ring = adev->mman.buffer_funcs_ring; + rq = &ring->sched.sched_rq[AMD_SCHED_PRIORITY_KERNEL]; + r = amd_sched_entity_init(&ring->sched, &adev->mman.entity, + rq, amdgpu_sched_jobs); + if (r != 0) { + DRM_ERROR("Failed setting up TTM BO move run queue.\n"); + drm_global_item_unref(&adev->mman.mem_global_ref); + drm_global_item_unref(&adev->mman.bo_global_ref.ref); + return r; + } + adev->mman.mem_global_referenced = true; + return 0; } static void amdgpu_ttm_global_fini(struct amdgpu_device *adev) { if (adev->mman.mem_global_referenced) { + amd_sched_entity_fini(adev->mman.entity.sched, + &adev->mman.entity); drm_global_item_unref(&adev->mman.bo_global_ref.ref); drm_global_item_unref(&adev->mman.mem_global_ref); adev->mman.mem_global_referenced = false; @@ -478,32 +494,32 @@ static void amdgpu_ttm_io_mem_free(struct ttm_bo_device *bdev, struct ttm_mem_re /* * TTM backend functions. */ +struct amdgpu_ttm_gup_task_list { + struct list_head list; + struct task_struct *task; +}; + struct amdgpu_ttm_tt { - struct ttm_dma_tt ttm; - struct amdgpu_device *adev; - u64 offset; - uint64_t userptr; - struct mm_struct *usermm; - uint32_t userflags; + struct ttm_dma_tt ttm; + struct amdgpu_device *adev; + u64 offset; + uint64_t userptr; + struct mm_struct *usermm; + uint32_t userflags; + spinlock_t guptasklock; + struct list_head guptasks; + atomic_t mmu_invalidations; }; -/* prepare the sg table with the user pages */ -static int amdgpu_ttm_tt_pin_userptr(struct ttm_tt *ttm) +int amdgpu_ttm_tt_get_user_pages(struct ttm_tt *ttm, struct page **pages) { - struct amdgpu_device *adev = amdgpu_get_adev(ttm->bdev); struct amdgpu_ttm_tt *gtt = (void *)ttm; - unsigned pinned = 0, nents; - int r; - int write = !(gtt->userflags & AMDGPU_GEM_USERPTR_READONLY); - enum dma_data_direction direction = write ? - DMA_BIDIRECTIONAL : DMA_TO_DEVICE; - - if (current->mm != gtt->usermm) - return -EPERM; + unsigned pinned = 0; + int r; if (gtt->userflags & AMDGPU_GEM_USERPTR_ANONONLY) { - /* check that we only pin down anonymous memory + /* check that we only use anonymous memory to prevent problems with writeback */ unsigned long end = gtt->userptr + ttm->num_pages * PAGE_SIZE; struct vm_area_struct *vma; @@ -516,10 +532,20 @@ static int amdgpu_ttm_tt_pin_userptr(struct ttm_tt *ttm) do { unsigned num_pages = ttm->num_pages - pinned; uint64_t userptr = gtt->userptr + pinned * PAGE_SIZE; - struct page **pages = ttm->pages + pinned; + struct page **p = pages + pinned; + struct amdgpu_ttm_gup_task_list guptask; + + guptask.task = current; + spin_lock(>t->guptasklock); + list_add(&guptask.list, >t->guptasks); + spin_unlock(>t->guptasklock); + + r = get_user_pages(userptr, num_pages, write, 0, p, NULL); + + spin_lock(>t->guptasklock); + list_del(&guptask.list); + spin_unlock(>t->guptasklock); - r = get_user_pages(current, current->mm, userptr, num_pages, - write, 0, pages, NULL); if (r < 0) goto release_pages; @@ -527,6 +553,25 @@ static int amdgpu_ttm_tt_pin_userptr(struct ttm_tt *ttm) } while (pinned < ttm->num_pages); + return 0; + +release_pages: + release_pages(pages, pinned, 0); + return r; +} + +/* prepare the sg table with the user pages */ +static int amdgpu_ttm_tt_pin_userptr(struct ttm_tt *ttm) +{ + struct amdgpu_device *adev = amdgpu_get_adev(ttm->bdev); + struct amdgpu_ttm_tt *gtt = (void *)ttm; + unsigned nents; + int r; + + int write = !(gtt->userflags & AMDGPU_GEM_USERPTR_READONLY); + enum dma_data_direction direction = write ? + DMA_BIDIRECTIONAL : DMA_TO_DEVICE; + r = sg_alloc_table_from_pages(ttm->sg, ttm->pages, ttm->num_pages, 0, ttm->num_pages << PAGE_SHIFT, GFP_KERNEL); @@ -545,9 +590,6 @@ static int amdgpu_ttm_tt_pin_userptr(struct ttm_tt *ttm) release_sg: kfree(ttm->sg); - -release_pages: - release_pages(ttm->pages, pinned, 0); return r; } @@ -587,9 +629,13 @@ static int amdgpu_ttm_backend_bind(struct ttm_tt *ttm, uint32_t flags = amdgpu_ttm_tt_pte_flags(gtt->adev, ttm, bo_mem); int r; - if (gtt->userptr) - amdgpu_ttm_tt_pin_userptr(ttm); - + if (gtt->userptr) { + r = amdgpu_ttm_tt_pin_userptr(ttm); + if (r) { + DRM_ERROR("failed to pin userptr\n"); + return r; + } + } gtt->offset = (unsigned long)(bo_mem->start << PAGE_SHIFT); if (!ttm->num_pages) { WARN(1, "nothing to bind %lu pages for mreg %p back %p!\n", @@ -708,7 +754,7 @@ static int amdgpu_ttm_tt_populate(struct ttm_tt *ttm) 0, PAGE_SIZE, PCI_DMA_BIDIRECTIONAL); if (pci_dma_mapping_error(adev->pdev, gtt->ttm.dma_address[i])) { - while (--i) { + while (i--) { pci_unmap_page(adev->pdev, gtt->ttm.dma_address[i], PAGE_SIZE, PCI_DMA_BIDIRECTIONAL); gtt->ttm.dma_address[i] = 0; @@ -766,17 +812,59 @@ int amdgpu_ttm_tt_set_userptr(struct ttm_tt *ttm, uint64_t addr, gtt->userptr = addr; gtt->usermm = current->mm; gtt->userflags = flags; + spin_lock_init(>t->guptasklock); + INIT_LIST_HEAD(>t->guptasks); + atomic_set(>t->mmu_invalidations, 0); + return 0; } -bool amdgpu_ttm_tt_has_userptr(struct ttm_tt *ttm) +struct mm_struct *amdgpu_ttm_tt_get_usermm(struct ttm_tt *ttm) { struct amdgpu_ttm_tt *gtt = (void *)ttm; if (gtt == NULL) + return NULL; + + return gtt->usermm; +} + +bool amdgpu_ttm_tt_affect_userptr(struct ttm_tt *ttm, unsigned long start, + unsigned long end) +{ + struct amdgpu_ttm_tt *gtt = (void *)ttm; + struct amdgpu_ttm_gup_task_list *entry; + unsigned long size; + + if (gtt == NULL || !gtt->userptr) + return false; + + size = (unsigned long)gtt->ttm.ttm.num_pages * PAGE_SIZE; + if (gtt->userptr > end || gtt->userptr + size <= start) return false; - return !!gtt->userptr; + spin_lock(>t->guptasklock); + list_for_each_entry(entry, >t->guptasks, list) { + if (entry->task == current) { + spin_unlock(>t->guptasklock); + return false; + } + } + spin_unlock(>t->guptasklock); + + atomic_inc(>t->mmu_invalidations); + + return true; +} + +bool amdgpu_ttm_tt_userptr_invalidated(struct ttm_tt *ttm, + int *last_invalidated) +{ + struct amdgpu_ttm_tt *gtt = (void *)ttm; + int prev_invalidated = *last_invalidated; + + *last_invalidated = atomic_read(>t->mmu_invalidations); + return prev_invalidated != *last_invalidated; } bool amdgpu_ttm_tt_is_readonly(struct ttm_tt *ttm) @@ -797,13 +885,14 @@ uint32_t amdgpu_ttm_tt_pte_flags(struct amdgpu_device *adev, struct ttm_tt *ttm, if (mem && mem->mem_type != TTM_PL_SYSTEM) flags |= AMDGPU_PTE_VALID; - if (mem && mem->mem_type == TTM_PL_TT) + if (mem && mem->mem_type == TTM_PL_TT) { flags |= AMDGPU_PTE_SYSTEM; - if (!ttm || ttm->caching_state == tt_cached) - flags |= AMDGPU_PTE_SNOOPED; + if (ttm->caching_state == tt_cached) + flags |= AMDGPU_PTE_SNOOPED; + } - if (adev->asic_type >= CHIP_TOPAZ) + if (adev->asic_type >= CHIP_TONGA) flags |= AMDGPU_PTE_EXECUTABLE; flags |= AMDGPU_PTE_READABLE; @@ -991,9 +1080,10 @@ int amdgpu_copy_buffer(struct amdgpu_ring *ring, struct fence **fence) { struct amdgpu_device *adev = ring->adev; + struct amdgpu_job *job; + uint32_t max_bytes; unsigned num_loops, num_dw; - struct amdgpu_ib *ib; unsigned i; int r; @@ -1005,20 +1095,12 @@ int amdgpu_copy_buffer(struct amdgpu_ring *ring, while (num_dw & 0x7) num_dw++; - ib = kzalloc(sizeof(struct amdgpu_ib), GFP_KERNEL); - if (!ib) - return -ENOMEM; - - r = amdgpu_ib_get(ring, NULL, num_dw * 4, ib); - if (r) { - kfree(ib); + r = amdgpu_job_alloc_with_ib(adev, num_dw * 4, &job); + if (r) return r; - } - - ib->length_dw = 0; if (resv) { - r = amdgpu_sync_resv(adev, &ib->sync, resv, + r = amdgpu_sync_resv(adev, &job->sync, resv, AMDGPU_FENCE_OWNER_UNDEFINED); if (r) { DRM_ERROR("sync failed (%d).\n", r); @@ -1029,31 +1111,25 @@ int amdgpu_copy_buffer(struct amdgpu_ring *ring, for (i = 0; i < num_loops; i++) { uint32_t cur_size_in_bytes = min(byte_count, max_bytes); - amdgpu_emit_copy_buffer(adev, ib, src_offset, dst_offset, - cur_size_in_bytes); + amdgpu_emit_copy_buffer(adev, &job->ibs[0], src_offset, + dst_offset, cur_size_in_bytes); src_offset += cur_size_in_bytes; dst_offset += cur_size_in_bytes; byte_count -= cur_size_in_bytes; } - amdgpu_vm_pad_ib(adev, ib); - WARN_ON(ib->length_dw > num_dw); - r = amdgpu_sched_ib_submit_kernel_helper(adev, ring, ib, 1, - &amdgpu_vm_free_job, - AMDGPU_FENCE_OWNER_MOVE, - fence); + amdgpu_ring_pad_ib(ring, &job->ibs[0]); + WARN_ON(job->ibs[0].length_dw > num_dw); + r = amdgpu_job_submit(job, ring, &adev->mman.entity, + AMDGPU_FENCE_OWNER_UNDEFINED, fence); if (r) goto error_free; - if (!amdgpu_enable_scheduler) { - amdgpu_ib_free(adev, ib); - kfree(ib); - } return 0; + error_free: - amdgpu_ib_free(adev, ib); - kfree(ib); + amdgpu_job_free(job); return r; } @@ -1072,6 +1148,11 @@ static int amdgpu_mm_dump_table(struct seq_file *m, void *data) spin_lock(&glob->lru_lock); ret = drm_mm_dump_table(m, mm); spin_unlock(&glob->lru_lock); + if (ttm_pl == TTM_PL_VRAM) + seq_printf(m, "man size:%llu pages, ram usage:%lluMB, vis usage:%lluMB\n", + adev->mman.bdev.man[ttm_pl].size, + (u64)atomic64_read(&adev->vram_usage) >> 20, + (u64)atomic64_read(&adev->vram_vis_usage) >> 20); return ret; } diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c index d0312364d950..c1a581044417 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c @@ -53,6 +53,7 @@ #define FIRMWARE_TONGA "amdgpu/tonga_uvd.bin" #define FIRMWARE_CARRIZO "amdgpu/carrizo_uvd.bin" #define FIRMWARE_FIJI "amdgpu/fiji_uvd.bin" +#define FIRMWARE_STONEY "amdgpu/stoney_uvd.bin" /** * amdgpu_uvd_cs_ctx - Command submission parser context @@ -83,12 +84,15 @@ MODULE_FIRMWARE(FIRMWARE_MULLINS); MODULE_FIRMWARE(FIRMWARE_TONGA); MODULE_FIRMWARE(FIRMWARE_CARRIZO); MODULE_FIRMWARE(FIRMWARE_FIJI); +MODULE_FIRMWARE(FIRMWARE_STONEY); static void amdgpu_uvd_note_usage(struct amdgpu_device *adev); static void amdgpu_uvd_idle_work_handler(struct work_struct *work); int amdgpu_uvd_sw_init(struct amdgpu_device *adev) { + struct amdgpu_ring *ring; + struct amd_sched_rq *rq; unsigned long bo_size; const char *fw_name; const struct common_firmware_header *hdr; @@ -124,6 +128,9 @@ int amdgpu_uvd_sw_init(struct amdgpu_device *adev) case CHIP_CARRIZO: fw_name = FIRMWARE_CARRIZO; break; + case CHIP_STONEY: + fw_name = FIRMWARE_STONEY; + break; default: return -EINVAL; } @@ -186,6 +193,15 @@ int amdgpu_uvd_sw_init(struct amdgpu_device *adev) amdgpu_bo_unreserve(adev->uvd.vcpu_bo); + ring = &adev->uvd.ring; + rq = &ring->sched.sched_rq[AMD_SCHED_PRIORITY_NORMAL]; + r = amd_sched_entity_init(&ring->sched, &adev->uvd.entity, + rq, amdgpu_sched_jobs); + if (r != 0) { + DRM_ERROR("Failed setting up UVD run queue.\n"); + return r; + } + for (i = 0; i < AMDGPU_MAX_UVD_HANDLES; ++i) { atomic_set(&adev->uvd.handles[i], 0); adev->uvd.filp[i] = NULL; @@ -205,6 +221,8 @@ int amdgpu_uvd_sw_fini(struct amdgpu_device *adev) if (adev->uvd.vcpu_bo == NULL) return 0; + amd_sched_entity_fini(&adev->uvd.ring.sched, &adev->uvd.entity); + r = amdgpu_bo_reserve(adev->uvd.vcpu_bo, false); if (!r) { amdgpu_bo_kunmap(adev->uvd.vcpu_bo); @@ -236,7 +254,7 @@ int amdgpu_uvd_suspend(struct amdgpu_device *adev) amdgpu_uvd_note_usage(adev); - r = amdgpu_uvd_get_destroy_msg(ring, handle, &fence); + r = amdgpu_uvd_get_destroy_msg(ring, handle, false, &fence); if (r) { DRM_ERROR("Error destroying UVD (%d)!\n", r); continue; @@ -290,7 +308,8 @@ void amdgpu_uvd_free_handles(struct amdgpu_device *adev, struct drm_file *filp) amdgpu_uvd_note_usage(adev); - r = amdgpu_uvd_get_destroy_msg(ring, handle, &fence); + r = amdgpu_uvd_get_destroy_msg(ring, handle, + false, &fence); if (r) { DRM_ERROR("Error destroying UVD (%d)!\n", r); continue; @@ -520,13 +539,6 @@ static int amdgpu_uvd_cs_msg(struct amdgpu_uvd_cs_ctx *ctx, return -EINVAL; } - r = reservation_object_wait_timeout_rcu(bo->tbo.resv, true, false, - MAX_SCHEDULE_TIMEOUT); - if (r < 0) { - DRM_ERROR("Failed waiting for UVD message (%ld)!\n", r); - return r; - } - r = amdgpu_bo_kmap(bo, &ptr); if (r) { DRM_ERROR("Failed mapping the UVD message (%ld)!\n", r); @@ -611,7 +623,6 @@ static int amdgpu_uvd_cs_pass2(struct amdgpu_uvd_cs_ctx *ctx) { struct amdgpu_bo_va_mapping *mapping; struct amdgpu_bo *bo; - struct amdgpu_ib *ib; uint32_t cmd, lo, hi; uint64_t start, end; uint64_t addr; @@ -633,9 +644,10 @@ static int amdgpu_uvd_cs_pass2(struct amdgpu_uvd_cs_ctx *ctx) addr -= ((uint64_t)mapping->it.start) * AMDGPU_GPU_PAGE_SIZE; start += addr; - ib = &ctx->parser->ibs[ctx->ib_idx]; - ib->ptr[ctx->data0] = start & 0xFFFFFFFF; - ib->ptr[ctx->data1] = start >> 32; + amdgpu_set_ib_value(ctx->parser, ctx->ib_idx, ctx->data0, + lower_32_bits(start)); + amdgpu_set_ib_value(ctx->parser, ctx->ib_idx, ctx->data1, + upper_32_bits(start)); cmd = amdgpu_get_ib_value(ctx->parser, ctx->ib_idx, ctx->idx) >> 1; if (cmd < 0x4) { @@ -697,7 +709,7 @@ static int amdgpu_uvd_cs_pass2(struct amdgpu_uvd_cs_ctx *ctx) static int amdgpu_uvd_cs_reg(struct amdgpu_uvd_cs_ctx *ctx, int (*cb)(struct amdgpu_uvd_cs_ctx *ctx)) { - struct amdgpu_ib *ib = &ctx->parser->ibs[ctx->ib_idx]; + struct amdgpu_ib *ib = &ctx->parser->job->ibs[ctx->ib_idx]; int i, r; ctx->idx++; @@ -743,7 +755,7 @@ static int amdgpu_uvd_cs_reg(struct amdgpu_uvd_cs_ctx *ctx, static int amdgpu_uvd_cs_packets(struct amdgpu_uvd_cs_ctx *ctx, int (*cb)(struct amdgpu_uvd_cs_ctx *ctx)) { - struct amdgpu_ib *ib = &ctx->parser->ibs[ctx->ib_idx]; + struct amdgpu_ib *ib = &ctx->parser->job->ibs[ctx->ib_idx]; int r; for (ctx->idx = 0 ; ctx->idx < ib->length_dw; ) { @@ -785,7 +797,7 @@ int amdgpu_uvd_ring_parse_cs(struct amdgpu_cs_parser *parser, uint32_t ib_idx) [0x00000003] = 2048, [0x00000004] = 0xFFFFFFFF, }; - struct amdgpu_ib *ib = &parser->ibs[ib_idx]; + struct amdgpu_ib *ib = &parser->job->ibs[ib_idx]; int r; if (ib->length_dw % 16) { @@ -818,22 +830,14 @@ int amdgpu_uvd_ring_parse_cs(struct amdgpu_cs_parser *parser, uint32_t ib_idx) return 0; } -static int amdgpu_uvd_free_job( - struct amdgpu_job *job) -{ - amdgpu_ib_free(job->adev, job->ibs); - kfree(job->ibs); - return 0; -} - -static int amdgpu_uvd_send_msg(struct amdgpu_ring *ring, - struct amdgpu_bo *bo, - struct fence **fence) +static int amdgpu_uvd_send_msg(struct amdgpu_ring *ring, struct amdgpu_bo *bo, + bool direct, struct fence **fence) { struct ttm_validate_buffer tv; struct ww_acquire_ctx ticket; struct list_head head; - struct amdgpu_ib *ib = NULL; + struct amdgpu_job *job; + struct amdgpu_ib *ib; struct fence *f = NULL; struct amdgpu_device *adev = ring->adev; uint64_t addr; @@ -857,15 +861,12 @@ static int amdgpu_uvd_send_msg(struct amdgpu_ring *ring, r = ttm_bo_validate(&bo->tbo, &bo->placement, true, false); if (r) goto err; - ib = kzalloc(sizeof(struct amdgpu_ib), GFP_KERNEL); - if (!ib) { - r = -ENOMEM; - goto err; - } - r = amdgpu_ib_get(ring, NULL, 64, ib); + + r = amdgpu_job_alloc_with_ib(adev, 64, &job); if (r) - goto err1; + goto err; + ib = &job->ibs[0]; addr = amdgpu_bo_gpu_offset(bo); ib->ptr[0] = PACKET0(mmUVD_GPCOM_VCPU_DATA0, 0); ib->ptr[1] = addr; @@ -877,12 +878,19 @@ static int amdgpu_uvd_send_msg(struct amdgpu_ring *ring, ib->ptr[i] = PACKET2(0); ib->length_dw = 16; - r = amdgpu_sched_ib_submit_kernel_helper(adev, ring, ib, 1, - &amdgpu_uvd_free_job, - AMDGPU_FENCE_OWNER_UNDEFINED, - &f); - if (r) - goto err2; + if (direct) { + r = amdgpu_ib_schedule(ring, 1, ib, NULL, &f); + job->fence = f; + if (r) + goto err_free; + + amdgpu_job_free(job); + } else { + r = amdgpu_job_submit(job, ring, &adev->uvd.entity, + AMDGPU_FENCE_OWNER_UNDEFINED, &f); + if (r) + goto err_free; + } ttm_eu_fence_buffer_objects(&ticket, &head, f); @@ -890,16 +898,12 @@ static int amdgpu_uvd_send_msg(struct amdgpu_ring *ring, *fence = fence_get(f); amdgpu_bo_unref(&bo); fence_put(f); - if (amdgpu_enable_scheduler) - return 0; - amdgpu_ib_free(ring->adev, ib); - kfree(ib); return 0; -err2: - amdgpu_ib_free(ring->adev, ib); -err1: - kfree(ib); + +err_free: + amdgpu_job_free(job); + err: ttm_eu_backoff_reservation(&ticket, &head); return r; @@ -954,11 +958,11 @@ int amdgpu_uvd_get_create_msg(struct amdgpu_ring *ring, uint32_t handle, amdgpu_bo_kunmap(bo); amdgpu_bo_unreserve(bo); - return amdgpu_uvd_send_msg(ring, bo, fence); + return amdgpu_uvd_send_msg(ring, bo, true, fence); } int amdgpu_uvd_get_destroy_msg(struct amdgpu_ring *ring, uint32_t handle, - struct fence **fence) + bool direct, struct fence **fence) { struct amdgpu_device *adev = ring->adev; struct amdgpu_bo *bo; @@ -996,7 +1000,7 @@ int amdgpu_uvd_get_destroy_msg(struct amdgpu_ring *ring, uint32_t handle, amdgpu_bo_kunmap(bo); amdgpu_bo_unreserve(bo); - return amdgpu_uvd_send_msg(ring, bo, fence); + return amdgpu_uvd_send_msg(ring, bo, direct, fence); } static void amdgpu_uvd_idle_work_handler(struct work_struct *work) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.h index 1724c2c86151..9a3b449081a7 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.h @@ -31,7 +31,7 @@ int amdgpu_uvd_resume(struct amdgpu_device *adev); int amdgpu_uvd_get_create_msg(struct amdgpu_ring *ring, uint32_t handle, struct fence **fence); int amdgpu_uvd_get_destroy_msg(struct amdgpu_ring *ring, uint32_t handle, - struct fence **fence); + bool direct, struct fence **fence); void amdgpu_uvd_free_handles(struct amdgpu_device *adev, struct drm_file *filp); int amdgpu_uvd_ring_parse_cs(struct amdgpu_cs_parser *parser, uint32_t ib_idx); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c index 74f2038ac747..4bec0c108cea 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c @@ -49,6 +49,7 @@ #define FIRMWARE_TONGA "amdgpu/tonga_vce.bin" #define FIRMWARE_CARRIZO "amdgpu/carrizo_vce.bin" #define FIRMWARE_FIJI "amdgpu/fiji_vce.bin" +#define FIRMWARE_STONEY "amdgpu/stoney_vce.bin" #ifdef CONFIG_DRM_AMDGPU_CIK MODULE_FIRMWARE(FIRMWARE_BONAIRE); @@ -60,6 +61,7 @@ MODULE_FIRMWARE(FIRMWARE_MULLINS); MODULE_FIRMWARE(FIRMWARE_TONGA); MODULE_FIRMWARE(FIRMWARE_CARRIZO); MODULE_FIRMWARE(FIRMWARE_FIJI); +MODULE_FIRMWARE(FIRMWARE_STONEY); static void amdgpu_vce_idle_work_handler(struct work_struct *work); @@ -72,6 +74,8 @@ static void amdgpu_vce_idle_work_handler(struct work_struct *work); */ int amdgpu_vce_sw_init(struct amdgpu_device *adev, unsigned long size) { + struct amdgpu_ring *ring; + struct amd_sched_rq *rq; const char *fw_name; const struct common_firmware_header *hdr; unsigned ucode_version, version_major, version_minor, binary_id; @@ -106,6 +110,9 @@ int amdgpu_vce_sw_init(struct amdgpu_device *adev, unsigned long size) case CHIP_FIJI: fw_name = FIRMWARE_FIJI; break; + case CHIP_STONEY: + fw_name = FIRMWARE_STONEY; + break; default: return -EINVAL; @@ -165,6 +172,16 @@ int amdgpu_vce_sw_init(struct amdgpu_device *adev, unsigned long size) return r; } + + ring = &adev->vce.ring[0]; + rq = &ring->sched.sched_rq[AMD_SCHED_PRIORITY_NORMAL]; + r = amd_sched_entity_init(&ring->sched, &adev->vce.entity, + rq, amdgpu_sched_jobs); + if (r != 0) { + DRM_ERROR("Failed setting up VCE run queue.\n"); + return r; + } + for (i = 0; i < AMDGPU_MAX_VCE_HANDLES; ++i) { atomic_set(&adev->vce.handles[i], 0); adev->vce.filp[i] = NULL; @@ -185,6 +202,8 @@ int amdgpu_vce_sw_fini(struct amdgpu_device *adev) if (adev->vce.vcpu_bo == NULL) return 0; + amd_sched_entity_fini(&adev->vce.ring[0].sched, &adev->vce.entity); + amdgpu_bo_unref(&adev->vce.vcpu_bo); amdgpu_ring_fini(&adev->vce.ring[0]); @@ -332,7 +351,7 @@ void amdgpu_vce_free_handles(struct amdgpu_device *adev, struct drm_file *filp) amdgpu_vce_note_usage(adev); - r = amdgpu_vce_get_destroy_msg(ring, handle, NULL); + r = amdgpu_vce_get_destroy_msg(ring, handle, false, NULL); if (r) DRM_ERROR("Error destroying VCE handle (%d)!\n", r); @@ -341,14 +360,6 @@ void amdgpu_vce_free_handles(struct amdgpu_device *adev, struct drm_file *filp) } } -static int amdgpu_vce_free_job( - struct amdgpu_job *job) -{ - amdgpu_ib_free(job->adev, job->ibs); - kfree(job->ibs); - return 0; -} - /** * amdgpu_vce_get_create_msg - generate a VCE create msg * @@ -363,21 +374,17 @@ int amdgpu_vce_get_create_msg(struct amdgpu_ring *ring, uint32_t handle, struct fence **fence) { const unsigned ib_size_dw = 1024; - struct amdgpu_ib *ib = NULL; + struct amdgpu_job *job; + struct amdgpu_ib *ib; struct fence *f = NULL; - struct amdgpu_device *adev = ring->adev; uint64_t dummy; int i, r; - ib = kzalloc(sizeof(struct amdgpu_ib), GFP_KERNEL); - if (!ib) - return -ENOMEM; - r = amdgpu_ib_get(ring, NULL, ib_size_dw * 4, ib); - if (r) { - DRM_ERROR("amdgpu: failed to get ib (%d).\n", r); - kfree(ib); + r = amdgpu_job_alloc_with_ib(ring->adev, ib_size_dw * 4, &job); + if (r) return r; - } + + ib = &job->ibs[0]; dummy = ib->gpu_addr + 1024; @@ -387,7 +394,10 @@ int amdgpu_vce_get_create_msg(struct amdgpu_ring *ring, uint32_t handle, ib->ptr[ib->length_dw++] = 0x00000001; /* session cmd */ ib->ptr[ib->length_dw++] = handle; - ib->ptr[ib->length_dw++] = 0x00000030; /* len */ + if ((ring->adev->vce.fw_version >> 24) >= 52) + ib->ptr[ib->length_dw++] = 0x00000040; /* len */ + else + ib->ptr[ib->length_dw++] = 0x00000030; /* len */ ib->ptr[ib->length_dw++] = 0x01000001; /* create cmd */ ib->ptr[ib->length_dw++] = 0x00000000; ib->ptr[ib->length_dw++] = 0x00000042; @@ -399,6 +409,12 @@ int amdgpu_vce_get_create_msg(struct amdgpu_ring *ring, uint32_t handle, ib->ptr[ib->length_dw++] = 0x00000100; ib->ptr[ib->length_dw++] = 0x0000000c; ib->ptr[ib->length_dw++] = 0x00000000; + if ((ring->adev->vce.fw_version >> 24) >= 52) { + ib->ptr[ib->length_dw++] = 0x00000000; + ib->ptr[ib->length_dw++] = 0x00000000; + ib->ptr[ib->length_dw++] = 0x00000000; + ib->ptr[ib->length_dw++] = 0x00000000; + } ib->ptr[ib->length_dw++] = 0x00000014; /* len */ ib->ptr[ib->length_dw++] = 0x05000005; /* feedback buffer */ @@ -409,20 +425,19 @@ int amdgpu_vce_get_create_msg(struct amdgpu_ring *ring, uint32_t handle, for (i = ib->length_dw; i < ib_size_dw; ++i) ib->ptr[i] = 0x0; - r = amdgpu_sched_ib_submit_kernel_helper(adev, ring, ib, 1, - &amdgpu_vce_free_job, - AMDGPU_FENCE_OWNER_UNDEFINED, - &f); + r = amdgpu_ib_schedule(ring, 1, ib, NULL, &f); + job->fence = f; if (r) goto err; + + amdgpu_job_free(job); if (fence) *fence = fence_get(f); fence_put(f); - if (amdgpu_enable_scheduler) - return 0; + return 0; + err: - amdgpu_ib_free(adev, ib); - kfree(ib); + amdgpu_job_free(job); return r; } @@ -437,26 +452,20 @@ err: * Close up a stream for HW test or if userspace failed to do so */ int amdgpu_vce_get_destroy_msg(struct amdgpu_ring *ring, uint32_t handle, - struct fence **fence) + bool direct, struct fence **fence) { const unsigned ib_size_dw = 1024; - struct amdgpu_ib *ib = NULL; + struct amdgpu_job *job; + struct amdgpu_ib *ib; struct fence *f = NULL; - struct amdgpu_device *adev = ring->adev; uint64_t dummy; int i, r; - ib = kzalloc(sizeof(struct amdgpu_ib), GFP_KERNEL); - if (!ib) - return -ENOMEM; - - r = amdgpu_ib_get(ring, NULL, ib_size_dw * 4, ib); - if (r) { - kfree(ib); - DRM_ERROR("amdgpu: failed to get ib (%d).\n", r); + r = amdgpu_job_alloc_with_ib(ring->adev, ib_size_dw * 4, &job); + if (r) return r; - } + ib = &job->ibs[0]; dummy = ib->gpu_addr + 1024; /* stitch together an VCE destroy msg */ @@ -476,20 +485,28 @@ int amdgpu_vce_get_destroy_msg(struct amdgpu_ring *ring, uint32_t handle, for (i = ib->length_dw; i < ib_size_dw; ++i) ib->ptr[i] = 0x0; - r = amdgpu_sched_ib_submit_kernel_helper(adev, ring, ib, 1, - &amdgpu_vce_free_job, - AMDGPU_FENCE_OWNER_UNDEFINED, - &f); - if (r) - goto err; + + if (direct) { + r = amdgpu_ib_schedule(ring, 1, ib, NULL, &f); + job->fence = f; + if (r) + goto err; + + amdgpu_job_free(job); + } else { + r = amdgpu_job_submit(job, ring, &ring->adev->vce.entity, + AMDGPU_FENCE_OWNER_UNDEFINED, &f); + if (r) + goto err; + } + if (fence) *fence = fence_get(f); fence_put(f); - if (amdgpu_enable_scheduler) - return 0; + return 0; + err: - amdgpu_ib_free(adev, ib); - kfree(ib); + amdgpu_job_free(job); return r; } @@ -507,7 +524,6 @@ static int amdgpu_vce_cs_reloc(struct amdgpu_cs_parser *p, uint32_t ib_idx, int lo, int hi, unsigned size, uint32_t index) { struct amdgpu_bo_va_mapping *mapping; - struct amdgpu_ib *ib = &p->ibs[ib_idx]; struct amdgpu_bo *bo; uint64_t addr; @@ -536,8 +552,8 @@ static int amdgpu_vce_cs_reloc(struct amdgpu_cs_parser *p, uint32_t ib_idx, addr += amdgpu_bo_gpu_offset(bo); addr -= ((uint64_t)size) * ((uint64_t)index); - ib->ptr[lo] = addr & 0xFFFFFFFF; - ib->ptr[hi] = addr >> 32; + amdgpu_set_ib_value(p, ib_idx, lo, lower_32_bits(addr)); + amdgpu_set_ib_value(p, ib_idx, hi, upper_32_bits(addr)); return 0; } @@ -592,7 +608,7 @@ static int amdgpu_vce_validate_handle(struct amdgpu_cs_parser *p, */ int amdgpu_vce_ring_parse_cs(struct amdgpu_cs_parser *p, uint32_t ib_idx) { - struct amdgpu_ib *ib = &p->ibs[ib_idx]; + struct amdgpu_ib *ib = &p->job->ibs[ib_idx]; unsigned fb_idx = 0, bs_idx = 0; int session_idx = -1; bool destroyed = false; @@ -729,30 +745,6 @@ out: } /** - * amdgpu_vce_ring_emit_semaphore - emit a semaphore command - * - * @ring: engine to use - * @semaphore: address of semaphore - * @emit_wait: true=emit wait, false=emit signal - * - */ -bool amdgpu_vce_ring_emit_semaphore(struct amdgpu_ring *ring, - struct amdgpu_semaphore *semaphore, - bool emit_wait) -{ - uint64_t addr = semaphore->gpu_addr; - - amdgpu_ring_write(ring, VCE_CMD_SEMAPHORE); - amdgpu_ring_write(ring, (addr >> 3) & 0x000FFFFF); - amdgpu_ring_write(ring, (addr >> 23) & 0x000FFFFF); - amdgpu_ring_write(ring, 0x01003000 | (emit_wait ? 1 : 0)); - if (!emit_wait) - amdgpu_ring_write(ring, VCE_CMD_END); - - return true; -} - -/** * amdgpu_vce_ring_emit_ib - execute indirect buffer * * @ring: engine to use @@ -800,14 +792,14 @@ int amdgpu_vce_ring_test_ring(struct amdgpu_ring *ring) unsigned i; int r; - r = amdgpu_ring_lock(ring, 16); + r = amdgpu_ring_alloc(ring, 16); if (r) { DRM_ERROR("amdgpu: vce failed to lock ring %d (%d).\n", ring->idx, r); return r; } amdgpu_ring_write(ring, VCE_CMD_END); - amdgpu_ring_unlock_commit(ring); + amdgpu_ring_commit(ring); for (i = 0; i < adev->usec_timeout; i++) { if (amdgpu_ring_get_rptr(ring) != rptr) @@ -848,7 +840,7 @@ int amdgpu_vce_ring_test_ib(struct amdgpu_ring *ring) goto error; } - r = amdgpu_vce_get_destroy_msg(ring, 1, &fence); + r = amdgpu_vce_get_destroy_msg(ring, 1, true, &fence); if (r) { DRM_ERROR("amdgpu: failed to get destroy ib (%d).\n", r); goto error; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.h index ba2da8ee5906..ef99d2370182 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.h @@ -31,12 +31,9 @@ int amdgpu_vce_resume(struct amdgpu_device *adev); int amdgpu_vce_get_create_msg(struct amdgpu_ring *ring, uint32_t handle, struct fence **fence); int amdgpu_vce_get_destroy_msg(struct amdgpu_ring *ring, uint32_t handle, - struct fence **fence); + bool direct, struct fence **fence); void amdgpu_vce_free_handles(struct amdgpu_device *adev, struct drm_file *filp); int amdgpu_vce_ring_parse_cs(struct amdgpu_cs_parser *p, uint32_t ib_idx); -bool amdgpu_vce_ring_emit_semaphore(struct amdgpu_ring *ring, - struct amdgpu_semaphore *semaphore, - bool emit_wait); void amdgpu_vce_ring_emit_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib); void amdgpu_vce_ring_emit_fence(struct amdgpu_ring *ring, u64 addr, u64 seq, unsigned flags); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c index 53d551f2d839..b6c011b83641 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c @@ -50,12 +50,15 @@ * SI supports 16. */ +/* Special value that no flush is necessary */ +#define AMDGPU_VM_NO_FLUSH (~0ll) + /** * amdgpu_vm_num_pde - return the number of page directory entries * * @adev: amdgpu_device pointer * - * Calculate the number of page directory entries (cayman+). + * Calculate the number of page directory entries. */ static unsigned amdgpu_vm_num_pdes(struct amdgpu_device *adev) { @@ -67,7 +70,7 @@ static unsigned amdgpu_vm_num_pdes(struct amdgpu_device *adev) * * @adev: amdgpu_device pointer * - * Calculate the size of the page directory in bytes (cayman+). + * Calculate the size of the page directory in bytes. */ static unsigned amdgpu_vm_directory_size(struct amdgpu_device *adev) { @@ -75,53 +78,76 @@ static unsigned amdgpu_vm_directory_size(struct amdgpu_device *adev) } /** - * amdgpu_vm_get_bos - add the vm BOs to a validation list + * amdgpu_vm_get_pd_bo - add the VM PD to a validation list * * @vm: vm providing the BOs - * @head: head of validation list + * @validated: head of validation list + * @entry: entry to add * * Add the page directory to the list of BOs to - * validate for command submission (cayman+). + * validate for command submission. */ -struct amdgpu_bo_list_entry *amdgpu_vm_get_bos(struct amdgpu_device *adev, - struct amdgpu_vm *vm, - struct list_head *head) +void amdgpu_vm_get_pd_bo(struct amdgpu_vm *vm, + struct list_head *validated, + struct amdgpu_bo_list_entry *entry) { - struct amdgpu_bo_list_entry *list; - unsigned i, idx; - - mutex_lock(&vm->mutex); - list = drm_malloc_ab(vm->max_pde_used + 2, - sizeof(struct amdgpu_bo_list_entry)); - if (!list) { - mutex_unlock(&vm->mutex); - return NULL; - } + entry->robj = vm->page_directory; + entry->priority = 0; + entry->tv.bo = &vm->page_directory->tbo; + entry->tv.shared = true; + entry->user_pages = NULL; + list_add(&entry->tv.head, validated); +} + +/** + * amdgpu_vm_get_bos - add the vm BOs to a duplicates list + * + * @vm: vm providing the BOs + * @duplicates: head of duplicates list + * + * Add the page directory to the BO duplicates list + * for command submission. + */ +void amdgpu_vm_get_pt_bos(struct amdgpu_vm *vm, struct list_head *duplicates) +{ + unsigned i; /* add the vm page table to the list */ - list[0].robj = vm->page_directory; - list[0].prefered_domains = AMDGPU_GEM_DOMAIN_VRAM; - list[0].allowed_domains = AMDGPU_GEM_DOMAIN_VRAM; - list[0].priority = 0; - list[0].tv.bo = &vm->page_directory->tbo; - list[0].tv.shared = true; - list_add(&list[0].tv.head, head); - - for (i = 0, idx = 1; i <= vm->max_pde_used; i++) { - if (!vm->page_tables[i].bo) + for (i = 0; i <= vm->max_pde_used; ++i) { + struct amdgpu_bo_list_entry *entry = &vm->page_tables[i].entry; + + if (!entry->robj) continue; - list[idx].robj = vm->page_tables[i].bo; - list[idx].prefered_domains = AMDGPU_GEM_DOMAIN_VRAM; - list[idx].allowed_domains = AMDGPU_GEM_DOMAIN_VRAM; - list[idx].priority = 0; - list[idx].tv.bo = &list[idx].robj->tbo; - list[idx].tv.shared = true; - list_add(&list[idx++].tv.head, head); + list_add(&entry->tv.head, duplicates); } - mutex_unlock(&vm->mutex); - return list; +} + +/** + * amdgpu_vm_move_pt_bos_in_lru - move the PT BOs to the LRU tail + * + * @adev: amdgpu device instance + * @vm: vm providing the BOs + * + * Move the PT BOs to the tail of the LRU. + */ +void amdgpu_vm_move_pt_bos_in_lru(struct amdgpu_device *adev, + struct amdgpu_vm *vm) +{ + struct ttm_bo_global *glob = adev->mman.bdev.glob; + unsigned i; + + spin_lock(&glob->lru_lock); + for (i = 0; i <= vm->max_pde_used; ++i) { + struct amdgpu_bo_list_entry *entry = &vm->page_tables[i].entry; + + if (!entry->robj) + continue; + + ttm_bo_move_to_lru_tail(&entry->robj->tbo); + } + spin_unlock(&glob->lru_lock); } /** @@ -130,128 +156,154 @@ struct amdgpu_bo_list_entry *amdgpu_vm_get_bos(struct amdgpu_device *adev, * @vm: vm to allocate id for * @ring: ring we want to submit job to * @sync: sync object where we add dependencies + * @fence: fence protecting ID from reuse * * Allocate an id for the vm, adding fences to the sync obj as necessary. - * - * Global mutex must be locked! */ int amdgpu_vm_grab_id(struct amdgpu_vm *vm, struct amdgpu_ring *ring, - struct amdgpu_sync *sync) + struct amdgpu_sync *sync, struct fence *fence, + unsigned *vm_id, uint64_t *vm_pd_addr) { - struct amdgpu_fence *best[AMDGPU_MAX_RINGS] = {}; - struct amdgpu_vm_id *vm_id = &vm->ids[ring->idx]; + uint64_t pd_addr = amdgpu_bo_gpu_offset(vm->page_directory); struct amdgpu_device *adev = ring->adev; + struct amdgpu_vm_id *id = &vm->ids[ring->idx]; + struct fence *updates = sync->last_vm_update; + int r; - unsigned choices[2] = {}; - unsigned i; + mutex_lock(&adev->vm_manager.lock); /* check if the id is still valid */ - if (vm_id->id && vm_id->last_id_use && - vm_id->last_id_use == adev->vm_manager.active[vm_id->id]) - return 0; + if (id->mgr_id) { + struct fence *flushed = id->flushed_updates; + bool is_later; + long owner; + + if (!flushed) + is_later = true; + else if (!updates) + is_later = false; + else + is_later = fence_is_later(updates, flushed); - /* we definately need to flush */ - vm_id->pd_gpu_addr = ~0ll; + owner = atomic_long_read(&id->mgr_id->owner); + if (!is_later && owner == (long)id && + pd_addr == id->pd_gpu_addr) { - /* skip over VMID 0, since it is the system VM */ - for (i = 1; i < adev->vm_manager.nvm; ++i) { - struct amdgpu_fence *fence = adev->vm_manager.active[i]; + r = amdgpu_sync_fence(ring->adev, sync, + id->mgr_id->active); + if (r) { + mutex_unlock(&adev->vm_manager.lock); + return r; + } - if (fence == NULL) { - /* found a free one */ - vm_id->id = i; - trace_amdgpu_vm_grab_id(i, ring->idx); - return 0; - } + fence_put(id->mgr_id->active); + id->mgr_id->active = fence_get(fence); + + list_move_tail(&id->mgr_id->list, + &adev->vm_manager.ids_lru); + + *vm_id = id->mgr_id - adev->vm_manager.ids; + *vm_pd_addr = AMDGPU_VM_NO_FLUSH; + trace_amdgpu_vm_grab_id(vm, ring->idx, *vm_id, + *vm_pd_addr); - if (amdgpu_fence_is_earlier(fence, best[fence->ring->idx])) { - best[fence->ring->idx] = fence; - choices[fence->ring == ring ? 0 : 1] = i; + mutex_unlock(&adev->vm_manager.lock); + return 0; } } - for (i = 0; i < 2; ++i) { - if (choices[i]) { - struct amdgpu_fence *fence; + id->mgr_id = list_first_entry(&adev->vm_manager.ids_lru, + struct amdgpu_vm_manager_id, + list); - fence = adev->vm_manager.active[choices[i]]; - vm_id->id = choices[i]; + r = amdgpu_sync_fence(ring->adev, sync, id->mgr_id->active); + if (!r) { + fence_put(id->mgr_id->active); + id->mgr_id->active = fence_get(fence); - trace_amdgpu_vm_grab_id(choices[i], ring->idx); - return amdgpu_sync_fence(ring->adev, sync, &fence->base); - } + fence_put(id->flushed_updates); + id->flushed_updates = fence_get(updates); + + id->pd_gpu_addr = pd_addr; + + list_move_tail(&id->mgr_id->list, &adev->vm_manager.ids_lru); + atomic_long_set(&id->mgr_id->owner, (long)id); + + *vm_id = id->mgr_id - adev->vm_manager.ids; + *vm_pd_addr = pd_addr; + trace_amdgpu_vm_grab_id(vm, ring->idx, *vm_id, *vm_pd_addr); } - /* should never happen */ - BUG(); - return -EINVAL; + mutex_unlock(&adev->vm_manager.lock); + return r; } /** * amdgpu_vm_flush - hardware flush the vm * * @ring: ring to use for flush - * @vm: vm we want to flush - * @updates: last vm update that we waited for - * - * Flush the vm (cayman+). + * @vm_id: vmid number to use + * @pd_addr: address of the page directory * - * Global and local mutex must be locked! + * Emit a VM flush when it is necessary. */ void amdgpu_vm_flush(struct amdgpu_ring *ring, - struct amdgpu_vm *vm, - struct fence *updates) + unsigned vm_id, uint64_t pd_addr, + uint32_t gds_base, uint32_t gds_size, + uint32_t gws_base, uint32_t gws_size, + uint32_t oa_base, uint32_t oa_size) { - uint64_t pd_addr = amdgpu_bo_gpu_offset(vm->page_directory); - struct amdgpu_vm_id *vm_id = &vm->ids[ring->idx]; - struct fence *flushed_updates = vm_id->flushed_updates; - bool is_earlier = false; - - if (flushed_updates && updates) { - BUG_ON(flushed_updates->context != updates->context); - is_earlier = (updates->seqno - flushed_updates->seqno <= - INT_MAX) ? true : false; + struct amdgpu_device *adev = ring->adev; + struct amdgpu_vm_manager_id *mgr_id = &adev->vm_manager.ids[vm_id]; + bool gds_switch_needed = ring->funcs->emit_gds_switch && ( + mgr_id->gds_base != gds_base || + mgr_id->gds_size != gds_size || + mgr_id->gws_base != gws_base || + mgr_id->gws_size != gws_size || + mgr_id->oa_base != oa_base || + mgr_id->oa_size != oa_size); + + if (ring->funcs->emit_pipeline_sync && ( + pd_addr != AMDGPU_VM_NO_FLUSH || gds_switch_needed)) + amdgpu_ring_emit_pipeline_sync(ring); + + if (pd_addr != AMDGPU_VM_NO_FLUSH) { + trace_amdgpu_vm_flush(pd_addr, ring->idx, vm_id); + amdgpu_ring_emit_vm_flush(ring, vm_id, pd_addr); } - if (pd_addr != vm_id->pd_gpu_addr || !flushed_updates || - is_earlier) { - - trace_amdgpu_vm_flush(pd_addr, ring->idx, vm_id->id); - if (is_earlier) { - vm_id->flushed_updates = fence_get(updates); - fence_put(flushed_updates); - } - if (!flushed_updates) - vm_id->flushed_updates = fence_get(updates); - vm_id->pd_gpu_addr = pd_addr; - amdgpu_ring_emit_vm_flush(ring, vm_id->id, vm_id->pd_gpu_addr); + if (gds_switch_needed) { + mgr_id->gds_base = gds_base; + mgr_id->gds_size = gds_size; + mgr_id->gws_base = gws_base; + mgr_id->gws_size = gws_size; + mgr_id->oa_base = oa_base; + mgr_id->oa_size = oa_size; + amdgpu_ring_emit_gds_switch(ring, vm_id, + gds_base, gds_size, + gws_base, gws_size, + oa_base, oa_size); } } /** - * amdgpu_vm_fence - remember fence for vm + * amdgpu_vm_reset_id - reset VMID to zero * - * @adev: amdgpu_device pointer - * @vm: vm we want to fence - * @fence: fence to remember - * - * Fence the vm (cayman+). - * Set the fence used to protect page table and id. + * @adev: amdgpu device structure + * @vm_id: vmid number to use * - * Global and local mutex must be locked! + * Reset saved GDW, GWS and OA to force switch on next flush. */ -void amdgpu_vm_fence(struct amdgpu_device *adev, - struct amdgpu_vm *vm, - struct amdgpu_fence *fence) +void amdgpu_vm_reset_id(struct amdgpu_device *adev, unsigned vm_id) { - unsigned ridx = fence->ring->idx; - unsigned vm_id = vm->ids[ridx].id; - - amdgpu_fence_unref(&adev->vm_manager.active[vm_id]); - adev->vm_manager.active[vm_id] = amdgpu_fence_ref(fence); - - amdgpu_fence_unref(&vm->ids[ridx].last_id_use); - vm->ids[ridx].last_id_use = amdgpu_fence_ref(fence); + struct amdgpu_vm_manager_id *mgr_id = &adev->vm_manager.ids[vm_id]; + + mgr_id->gds_base = 0; + mgr_id->gds_size = 0; + mgr_id->gws_base = 0; + mgr_id->gws_size = 0; + mgr_id->oa_base = 0; + mgr_id->oa_size = 0; } /** @@ -260,7 +312,7 @@ void amdgpu_vm_fence(struct amdgpu_device *adev, * @vm: requested vm * @bo: requested buffer object * - * Find @bo inside the requested vm (cayman+). + * Find @bo inside the requested vm. * Search inside the @bos vm list for the requested vm * Returns the found bo_va or NULL if none is found * @@ -283,32 +335,40 @@ struct amdgpu_bo_va *amdgpu_vm_bo_find(struct amdgpu_vm *vm, * amdgpu_vm_update_pages - helper to call the right asic function * * @adev: amdgpu_device pointer + * @gtt: GART instance to use for mapping + * @gtt_flags: GTT hw access flags * @ib: indirect buffer to fill with commands * @pe: addr of the page entry * @addr: dst addr to write into pe * @count: number of page entries to update * @incr: increase next addr by incr bytes * @flags: hw access flags - * @gtt_flags: GTT hw access flags * * Traces the parameters and calls the right asic functions * to setup the page table using the DMA. */ static void amdgpu_vm_update_pages(struct amdgpu_device *adev, + struct amdgpu_gart *gtt, + uint32_t gtt_flags, struct amdgpu_ib *ib, uint64_t pe, uint64_t addr, unsigned count, uint32_t incr, - uint32_t flags, uint32_t gtt_flags) + uint32_t flags) { trace_amdgpu_vm_set_page(pe, addr, count, incr, flags); - if ((flags & AMDGPU_PTE_SYSTEM) && (flags == gtt_flags)) { - uint64_t src = adev->gart.table_addr + (addr >> 12) * 8; + if ((gtt == &adev->gart) && (flags == gtt_flags)) { + uint64_t src = gtt->table_addr + (addr >> 12) * 8; amdgpu_vm_copy_pte(adev, ib, pe, src, count); - } else if ((flags & AMDGPU_PTE_SYSTEM) || (count < 3)) { - amdgpu_vm_write_pte(adev, ib, pe, addr, - count, incr, flags); + } else if (gtt) { + dma_addr_t *pages_addr = gtt->pages_addr; + amdgpu_vm_write_pte(adev, ib, pages_addr, pe, addr, + count, incr, flags); + + } else if (count < 3) { + amdgpu_vm_write_pte(adev, ib, NULL, pe, addr, + count, incr, flags); } else { amdgpu_vm_set_pte_pde(adev, ib, pe, addr, @@ -316,34 +376,26 @@ static void amdgpu_vm_update_pages(struct amdgpu_device *adev, } } -int amdgpu_vm_free_job(struct amdgpu_job *job) -{ - int i; - for (i = 0; i < job->num_ibs; i++) - amdgpu_ib_free(job->adev, &job->ibs[i]); - kfree(job->ibs); - return 0; -} - /** * amdgpu_vm_clear_bo - initially clear the page dir/table * * @adev: amdgpu_device pointer * @bo: bo to clear + * + * need to reserve bo first before calling it. */ static int amdgpu_vm_clear_bo(struct amdgpu_device *adev, + struct amdgpu_vm *vm, struct amdgpu_bo *bo) { - struct amdgpu_ring *ring = adev->vm_manager.vm_pte_funcs_ring; + struct amdgpu_ring *ring; struct fence *fence = NULL; - struct amdgpu_ib *ib; + struct amdgpu_job *job; unsigned entries; uint64_t addr; int r; - r = amdgpu_bo_reserve(bo, false); - if (r) - return r; + ring = container_of(vm->entity.sched, struct amdgpu_ring, sched); r = reservation_object_reserve_shared(bo->tbo.resv); if (r) @@ -351,63 +403,62 @@ static int amdgpu_vm_clear_bo(struct amdgpu_device *adev, r = ttm_bo_validate(&bo->tbo, &bo->placement, true, false); if (r) - goto error_unreserve; + goto error; addr = amdgpu_bo_gpu_offset(bo); entries = amdgpu_bo_size(bo) / 8; - ib = kzalloc(sizeof(struct amdgpu_ib), GFP_KERNEL); - if (!ib) - goto error_unreserve; + r = amdgpu_job_alloc_with_ib(adev, 64, &job); + if (r) + goto error; + + amdgpu_vm_update_pages(adev, NULL, 0, &job->ibs[0], addr, 0, entries, + 0, 0); + amdgpu_ring_pad_ib(ring, &job->ibs[0]); - r = amdgpu_ib_get(ring, NULL, entries * 2 + 64, ib); + WARN_ON(job->ibs[0].length_dw > 64); + r = amdgpu_job_submit(job, ring, &vm->entity, + AMDGPU_FENCE_OWNER_VM, &fence); if (r) goto error_free; - ib->length_dw = 0; - - amdgpu_vm_update_pages(adev, ib, addr, 0, entries, 0, 0, 0); - amdgpu_vm_pad_ib(adev, ib); - WARN_ON(ib->length_dw > 64); - r = amdgpu_sched_ib_submit_kernel_helper(adev, ring, ib, 1, - &amdgpu_vm_free_job, - AMDGPU_FENCE_OWNER_VM, - &fence); - if (!r) - amdgpu_bo_fence(bo, fence, true); + amdgpu_bo_fence(bo, fence, true); fence_put(fence); - if (amdgpu_enable_scheduler) { - amdgpu_bo_unreserve(bo); - return 0; - } + return 0; + error_free: - amdgpu_ib_free(adev, ib); - kfree(ib); + amdgpu_job_free(job); -error_unreserve: - amdgpu_bo_unreserve(bo); +error: return r; } /** - * amdgpu_vm_map_gart - get the physical address of a gart page + * amdgpu_vm_map_gart - Resolve gart mapping of addr * - * @adev: amdgpu_device pointer + * @pages_addr: optional DMA address to use for lookup * @addr: the unmapped addr * * Look up the physical address of the page that the pte resolves - * to (cayman+). - * Returns the physical address of the page. + * to and return the pointer for the page table entry. */ -uint64_t amdgpu_vm_map_gart(struct amdgpu_device *adev, uint64_t addr) +uint64_t amdgpu_vm_map_gart(const dma_addr_t *pages_addr, uint64_t addr) { uint64_t result; - /* page table offset */ - result = adev->gart.pages_addr[addr >> PAGE_SHIFT]; + if (pages_addr) { + /* page table offset */ + result = pages_addr[addr >> PAGE_SHIFT]; + + /* in case cpu page size != gpu page size*/ + result |= addr & (~PAGE_MASK); + + } else { + /* No mapping required */ + result = addr; + } - /* in case cpu page size != gpu page size*/ - result |= addr & (~PAGE_MASK); + result &= 0xFFFFFFFFFFFFF000ULL; return result; } @@ -421,49 +472,41 @@ uint64_t amdgpu_vm_map_gart(struct amdgpu_device *adev, uint64_t addr) * @end: end of GPU address range * * Allocates new page tables if necessary - * and updates the page directory (cayman+). + * and updates the page directory. * Returns 0 for success, error for failure. - * - * Global and local mutex must be locked! */ int amdgpu_vm_update_page_directory(struct amdgpu_device *adev, struct amdgpu_vm *vm) { - struct amdgpu_ring *ring = adev->vm_manager.vm_pte_funcs_ring; + struct amdgpu_ring *ring; struct amdgpu_bo *pd = vm->page_directory; uint64_t pd_addr = amdgpu_bo_gpu_offset(pd); uint32_t incr = AMDGPU_VM_PTE_COUNT * 8; uint64_t last_pde = ~0, last_pt = ~0; unsigned count = 0, pt_idx, ndw; + struct amdgpu_job *job; struct amdgpu_ib *ib; struct fence *fence = NULL; int r; + ring = container_of(vm->entity.sched, struct amdgpu_ring, sched); + /* padding, etc. */ ndw = 64; /* assume the worst case */ ndw += vm->max_pde_used * 6; - /* update too big for an IB */ - if (ndw > 0xfffff) - return -ENOMEM; - - ib = kzalloc(sizeof(struct amdgpu_ib), GFP_KERNEL); - if (!ib) - return -ENOMEM; - - r = amdgpu_ib_get(ring, NULL, ndw * 4, ib); - if (r) { - kfree(ib); + r = amdgpu_job_alloc_with_ib(adev, ndw * 4, &job); + if (r) return r; - } - ib->length_dw = 0; + + ib = &job->ibs[0]; /* walk over the address space and update the page directory */ for (pt_idx = 0; pt_idx <= vm->max_pde_used; ++pt_idx) { - struct amdgpu_bo *bo = vm->page_tables[pt_idx].bo; + struct amdgpu_bo *bo = vm->page_tables[pt_idx].entry.robj; uint64_t pde, pt; if (bo == NULL) @@ -479,9 +522,10 @@ int amdgpu_vm_update_page_directory(struct amdgpu_device *adev, ((last_pt + incr * count) != pt)) { if (count) { - amdgpu_vm_update_pages(adev, ib, last_pde, - last_pt, count, incr, - AMDGPU_PTE_VALID, 0); + amdgpu_vm_update_pages(adev, NULL, 0, ib, + last_pde, last_pt, + count, incr, + AMDGPU_PTE_VALID); } count = 1; @@ -493,17 +537,16 @@ int amdgpu_vm_update_page_directory(struct amdgpu_device *adev, } if (count) - amdgpu_vm_update_pages(adev, ib, last_pde, last_pt, count, - incr, AMDGPU_PTE_VALID, 0); + amdgpu_vm_update_pages(adev, NULL, 0, ib, last_pde, last_pt, + count, incr, AMDGPU_PTE_VALID); if (ib->length_dw != 0) { - amdgpu_vm_pad_ib(adev, ib); - amdgpu_sync_resv(adev, &ib->sync, pd->tbo.resv, AMDGPU_FENCE_OWNER_VM); + amdgpu_ring_pad_ib(ring, ib); + amdgpu_sync_resv(adev, &job->sync, pd->tbo.resv, + AMDGPU_FENCE_OWNER_VM); WARN_ON(ib->length_dw > ndw); - r = amdgpu_sched_ib_submit_kernel_helper(adev, ring, ib, 1, - &amdgpu_vm_free_job, - AMDGPU_FENCE_OWNER_VM, - &fence); + r = amdgpu_job_submit(job, ring, &vm->entity, + AMDGPU_FENCE_OWNER_VM, &fence); if (r) goto error_free; @@ -511,18 +554,15 @@ int amdgpu_vm_update_page_directory(struct amdgpu_device *adev, fence_put(vm->page_directory_fence); vm->page_directory_fence = fence_get(fence); fence_put(fence); - } - if (!amdgpu_enable_scheduler || ib->length_dw == 0) { - amdgpu_ib_free(adev, ib); - kfree(ib); + } else { + amdgpu_job_free(job); } return 0; error_free: - amdgpu_ib_free(adev, ib); - kfree(ib); + amdgpu_job_free(job); return r; } @@ -530,20 +570,20 @@ error_free: * amdgpu_vm_frag_ptes - add fragment information to PTEs * * @adev: amdgpu_device pointer + * @gtt: GART instance to use for mapping + * @gtt_flags: GTT hw mapping flags * @ib: IB for the update * @pe_start: first PTE to handle * @pe_end: last PTE to handle * @addr: addr those PTEs should point to * @flags: hw mapping flags - * @gtt_flags: GTT hw mapping flags - * - * Global and local mutex must be locked! */ static void amdgpu_vm_frag_ptes(struct amdgpu_device *adev, + struct amdgpu_gart *gtt, + uint32_t gtt_flags, struct amdgpu_ib *ib, uint64_t pe_start, uint64_t pe_end, - uint64_t addr, uint32_t flags, - uint32_t gtt_flags) + uint64_t addr, uint32_t flags) { /** * The MC L1 TLB supports variable sized pages, based on a fragment @@ -573,36 +613,39 @@ static void amdgpu_vm_frag_ptes(struct amdgpu_device *adev, unsigned count; + /* Abort early if there isn't anything to do */ + if (pe_start == pe_end) + return; + /* system pages are non continuously */ - if ((flags & AMDGPU_PTE_SYSTEM) || !(flags & AMDGPU_PTE_VALID) || - (frag_start >= frag_end)) { + if (gtt || !(flags & AMDGPU_PTE_VALID) || (frag_start >= frag_end)) { count = (pe_end - pe_start) / 8; - amdgpu_vm_update_pages(adev, ib, pe_start, addr, count, - AMDGPU_GPU_PAGE_SIZE, flags, gtt_flags); + amdgpu_vm_update_pages(adev, gtt, gtt_flags, ib, pe_start, + addr, count, AMDGPU_GPU_PAGE_SIZE, + flags); return; } /* handle the 4K area at the beginning */ if (pe_start != frag_start) { count = (frag_start - pe_start) / 8; - amdgpu_vm_update_pages(adev, ib, pe_start, addr, count, - AMDGPU_GPU_PAGE_SIZE, flags, gtt_flags); + amdgpu_vm_update_pages(adev, NULL, 0, ib, pe_start, addr, + count, AMDGPU_GPU_PAGE_SIZE, flags); addr += AMDGPU_GPU_PAGE_SIZE * count; } /* handle the area in the middle */ count = (frag_end - frag_start) / 8; - amdgpu_vm_update_pages(adev, ib, frag_start, addr, count, - AMDGPU_GPU_PAGE_SIZE, flags | frag_flags, - gtt_flags); + amdgpu_vm_update_pages(adev, NULL, 0, ib, frag_start, addr, count, + AMDGPU_GPU_PAGE_SIZE, flags | frag_flags); /* handle the 4K area at the end */ if (frag_end != pe_end) { addr += AMDGPU_GPU_PAGE_SIZE * count; count = (pe_end - frag_end) / 8; - amdgpu_vm_update_pages(adev, ib, frag_end, addr, count, - AMDGPU_GPU_PAGE_SIZE, flags, gtt_flags); + amdgpu_vm_update_pages(adev, NULL, 0, ib, frag_end, addr, + count, AMDGPU_GPU_PAGE_SIZE, flags); } } @@ -610,122 +653,105 @@ static void amdgpu_vm_frag_ptes(struct amdgpu_device *adev, * amdgpu_vm_update_ptes - make sure that page tables are valid * * @adev: amdgpu_device pointer + * @gtt: GART instance to use for mapping + * @gtt_flags: GTT hw mapping flags * @vm: requested vm * @start: start of GPU address range * @end: end of GPU address range * @dst: destination address to map to * @flags: mapping flags * - * Update the page tables in the range @start - @end (cayman+). - * - * Global and local mutex must be locked! + * Update the page tables in the range @start - @end. */ -static int amdgpu_vm_update_ptes(struct amdgpu_device *adev, - struct amdgpu_vm *vm, - struct amdgpu_ib *ib, - uint64_t start, uint64_t end, - uint64_t dst, uint32_t flags, - uint32_t gtt_flags) +static void amdgpu_vm_update_ptes(struct amdgpu_device *adev, + struct amdgpu_gart *gtt, + uint32_t gtt_flags, + struct amdgpu_vm *vm, + struct amdgpu_ib *ib, + uint64_t start, uint64_t end, + uint64_t dst, uint32_t flags) { - uint64_t mask = AMDGPU_VM_PTE_COUNT - 1; - uint64_t last_pte = ~0, last_dst = ~0; - void *owner = AMDGPU_FENCE_OWNER_VM; - unsigned count = 0; - uint64_t addr; + const uint64_t mask = AMDGPU_VM_PTE_COUNT - 1; - /* sync to everything on unmapping */ - if (!(flags & AMDGPU_PTE_VALID)) - owner = AMDGPU_FENCE_OWNER_UNDEFINED; + uint64_t last_pe_start = ~0, last_pe_end = ~0, last_dst = ~0; + uint64_t addr; /* walk over the address space and update the page tables */ for (addr = start; addr < end; ) { uint64_t pt_idx = addr >> amdgpu_vm_block_size; - struct amdgpu_bo *pt = vm->page_tables[pt_idx].bo; + struct amdgpu_bo *pt = vm->page_tables[pt_idx].entry.robj; unsigned nptes; - uint64_t pte; - int r; - - amdgpu_sync_resv(adev, &ib->sync, pt->tbo.resv, owner); - r = reservation_object_reserve_shared(pt->tbo.resv); - if (r) - return r; + uint64_t pe_start; if ((addr & ~mask) == (end & ~mask)) nptes = end - addr; else nptes = AMDGPU_VM_PTE_COUNT - (addr & mask); - pte = amdgpu_bo_gpu_offset(pt); - pte += (addr & mask) * 8; + pe_start = amdgpu_bo_gpu_offset(pt); + pe_start += (addr & mask) * 8; - if ((last_pte + 8 * count) != pte) { + if (last_pe_end != pe_start) { - if (count) { - amdgpu_vm_frag_ptes(adev, ib, last_pte, - last_pte + 8 * count, - last_dst, flags, - gtt_flags); - } + amdgpu_vm_frag_ptes(adev, gtt, gtt_flags, ib, + last_pe_start, last_pe_end, + last_dst, flags); - count = nptes; - last_pte = pte; + last_pe_start = pe_start; + last_pe_end = pe_start + 8 * nptes; last_dst = dst; } else { - count += nptes; + last_pe_end += 8 * nptes; } addr += nptes; dst += nptes * AMDGPU_GPU_PAGE_SIZE; } - if (count) { - amdgpu_vm_frag_ptes(adev, ib, last_pte, - last_pte + 8 * count, - last_dst, flags, gtt_flags); - } - - return 0; + amdgpu_vm_frag_ptes(adev, gtt, gtt_flags, ib, + last_pe_start, last_pe_end, + last_dst, flags); } /** * amdgpu_vm_bo_update_mapping - update a mapping in the vm page table * * @adev: amdgpu_device pointer + * @gtt: GART instance to use for mapping + * @gtt_flags: flags as they are used for GTT * @vm: requested vm - * @mapping: mapped range and flags to use for the update + * @start: start of mapped range + * @last: last mapped entry + * @flags: flags for the entries * @addr: addr to set the area to - * @gtt_flags: flags as they are used for GTT * @fence: optional resulting fence * - * Fill in the page table entries for @mapping. + * Fill in the page table entries between @start and @last. * Returns 0 for success, -EINVAL for failure. - * - * Object have to be reserved and mutex must be locked! */ static int amdgpu_vm_bo_update_mapping(struct amdgpu_device *adev, + struct amdgpu_gart *gtt, + uint32_t gtt_flags, struct amdgpu_vm *vm, - struct amdgpu_bo_va_mapping *mapping, - uint64_t addr, uint32_t gtt_flags, + uint64_t start, uint64_t last, + uint32_t flags, uint64_t addr, struct fence **fence) { - struct amdgpu_ring *ring = adev->vm_manager.vm_pte_funcs_ring; + struct amdgpu_ring *ring; + void *owner = AMDGPU_FENCE_OWNER_VM; unsigned nptes, ncmds, ndw; - uint32_t flags = gtt_flags; + struct amdgpu_job *job; struct amdgpu_ib *ib; struct fence *f = NULL; int r; - /* normally,bo_va->flags only contians READABLE and WIRTEABLE bit go here - * but in case of something, we filter the flags in first place - */ - if (!(mapping->flags & AMDGPU_PTE_READABLE)) - flags &= ~AMDGPU_PTE_READABLE; - if (!(mapping->flags & AMDGPU_PTE_WRITEABLE)) - flags &= ~AMDGPU_PTE_WRITEABLE; + ring = container_of(vm->entity.sched, struct amdgpu_ring, sched); - trace_amdgpu_vm_bo_update(mapping); + /* sync to everything on unmapping */ + if (!(flags & AMDGPU_PTE_VALID)) + owner = AMDGPU_FENCE_OWNER_UNDEFINED; - nptes = mapping->it.last - mapping->it.start + 1; + nptes = last - start + 1; /* * reserve space for one command every (1 << BLOCK_SIZE) @@ -736,11 +762,11 @@ static int amdgpu_vm_bo_update_mapping(struct amdgpu_device *adev, /* padding, etc. */ ndw = 64; - if ((flags & AMDGPU_PTE_SYSTEM) && (flags == gtt_flags)) { + if ((gtt == &adev->gart) && (flags == gtt_flags)) { /* only copy commands needed */ ndw += ncmds * 7; - } else if (flags & AMDGPU_PTE_SYSTEM) { + } else if (gtt) { /* header for write data commands */ ndw += ncmds * 4; @@ -755,38 +781,28 @@ static int amdgpu_vm_bo_update_mapping(struct amdgpu_device *adev, ndw += 2 * 10; } - /* update too big for an IB */ - if (ndw > 0xfffff) - return -ENOMEM; - - ib = kzalloc(sizeof(struct amdgpu_ib), GFP_KERNEL); - if (!ib) - return -ENOMEM; - - r = amdgpu_ib_get(ring, NULL, ndw * 4, ib); - if (r) { - kfree(ib); + r = amdgpu_job_alloc_with_ib(adev, ndw * 4, &job); + if (r) return r; - } - ib->length_dw = 0; + ib = &job->ibs[0]; - r = amdgpu_vm_update_ptes(adev, vm, ib, mapping->it.start, - mapping->it.last + 1, addr + mapping->offset, - flags, gtt_flags); + r = amdgpu_sync_resv(adev, &job->sync, vm->page_directory->tbo.resv, + owner); + if (r) + goto error_free; - if (r) { - amdgpu_ib_free(adev, ib); - kfree(ib); - return r; - } + r = reservation_object_reserve_shared(vm->page_directory->tbo.resv); + if (r) + goto error_free; + + amdgpu_vm_update_ptes(adev, gtt, gtt_flags, vm, ib, start, last + 1, + addr, flags); - amdgpu_vm_pad_ib(adev, ib); + amdgpu_ring_pad_ib(ring, ib); WARN_ON(ib->length_dw > ndw); - r = amdgpu_sched_ib_submit_kernel_helper(adev, ring, ib, 1, - &amdgpu_vm_free_job, - AMDGPU_FENCE_OWNER_VM, - &f); + r = amdgpu_job_submit(job, ring, &vm->entity, + AMDGPU_FENCE_OWNER_VM, &f); if (r) goto error_free; @@ -796,19 +812,76 @@ static int amdgpu_vm_bo_update_mapping(struct amdgpu_device *adev, *fence = fence_get(f); } fence_put(f); - if (!amdgpu_enable_scheduler) { - amdgpu_ib_free(adev, ib); - kfree(ib); - } return 0; error_free: - amdgpu_ib_free(adev, ib); - kfree(ib); + amdgpu_job_free(job); return r; } /** + * amdgpu_vm_bo_split_mapping - split a mapping into smaller chunks + * + * @adev: amdgpu_device pointer + * @gtt: GART instance to use for mapping + * @vm: requested vm + * @mapping: mapped range and flags to use for the update + * @addr: addr to set the area to + * @gtt_flags: flags as they are used for GTT + * @fence: optional resulting fence + * + * Split the mapping into smaller chunks so that each update fits + * into a SDMA IB. + * Returns 0 for success, -EINVAL for failure. + */ +static int amdgpu_vm_bo_split_mapping(struct amdgpu_device *adev, + struct amdgpu_gart *gtt, + uint32_t gtt_flags, + struct amdgpu_vm *vm, + struct amdgpu_bo_va_mapping *mapping, + uint64_t addr, struct fence **fence) +{ + const uint64_t max_size = 64ULL * 1024ULL * 1024ULL / AMDGPU_GPU_PAGE_SIZE; + + uint64_t start = mapping->it.start; + uint32_t flags = gtt_flags; + int r; + + /* normally,bo_va->flags only contians READABLE and WIRTEABLE bit go here + * but in case of something, we filter the flags in first place + */ + if (!(mapping->flags & AMDGPU_PTE_READABLE)) + flags &= ~AMDGPU_PTE_READABLE; + if (!(mapping->flags & AMDGPU_PTE_WRITEABLE)) + flags &= ~AMDGPU_PTE_WRITEABLE; + + trace_amdgpu_vm_bo_update(mapping); + + addr += mapping->offset; + + if (!gtt || ((gtt == &adev->gart) && (flags == gtt_flags))) + return amdgpu_vm_bo_update_mapping(adev, gtt, gtt_flags, vm, + start, mapping->it.last, + flags, addr, fence); + + while (start != mapping->it.last + 1) { + uint64_t last; + + last = min((uint64_t)mapping->it.last, start + max_size - 1); + r = amdgpu_vm_bo_update_mapping(adev, gtt, gtt_flags, vm, + start, last, flags, addr, + fence); + if (r) + return r; + + start = last + 1; + addr += max_size * AMDGPU_GPU_PAGE_SIZE; + } + + return 0; +} + +/** * amdgpu_vm_bo_update - update all BO mappings in the vm page table * * @adev: amdgpu_device pointer @@ -826,14 +899,25 @@ int amdgpu_vm_bo_update(struct amdgpu_device *adev, { struct amdgpu_vm *vm = bo_va->vm; struct amdgpu_bo_va_mapping *mapping; + struct amdgpu_gart *gtt = NULL; uint32_t flags; uint64_t addr; int r; if (mem) { addr = (u64)mem->start << PAGE_SHIFT; - if (mem->mem_type != TTM_PL_TT) + switch (mem->mem_type) { + case TTM_PL_TT: + gtt = &bo_va->bo->adev->gart; + break; + + case TTM_PL_VRAM: addr += adev->vm_manager.vram_base_offset; + break; + + default: + break; + } } else { addr = 0; } @@ -846,12 +930,20 @@ int amdgpu_vm_bo_update(struct amdgpu_device *adev, spin_unlock(&vm->status_lock); list_for_each_entry(mapping, &bo_va->invalids, list) { - r = amdgpu_vm_bo_update_mapping(adev, vm, mapping, addr, - flags, &bo_va->last_pt_update); + r = amdgpu_vm_bo_split_mapping(adev, gtt, flags, vm, mapping, addr, + &bo_va->last_pt_update); if (r) return r; } + if (trace_amdgpu_vm_bo_mapping_enabled()) { + list_for_each_entry(mapping, &bo_va->valids, list) + trace_amdgpu_vm_bo_mapping(mapping); + + list_for_each_entry(mapping, &bo_va->invalids, list) + trace_amdgpu_vm_bo_mapping(mapping); + } + spin_lock(&vm->status_lock); list_splice_init(&bo_va->invalids, &bo_va->valids); list_del_init(&bo_va->vm_status); @@ -884,7 +976,8 @@ int amdgpu_vm_clear_freed(struct amdgpu_device *adev, struct amdgpu_bo_va_mapping, list); list_del(&mapping->list); - r = amdgpu_vm_bo_update_mapping(adev, vm, mapping, 0, 0, NULL); + r = amdgpu_vm_bo_split_mapping(adev, NULL, 0, vm, mapping, + 0, NULL); kfree(mapping); if (r) return r; @@ -938,7 +1031,7 @@ int amdgpu_vm_clear_invalids(struct amdgpu_device *adev, * @vm: requested vm * @bo: amdgpu buffer object * - * Add @bo into the requested vm (cayman+). + * Add @bo into the requested vm. * Add @bo to the list of bos associated with the vm * Returns newly added bo_va or NULL for failure * @@ -962,9 +1055,7 @@ struct amdgpu_bo_va *amdgpu_vm_bo_add(struct amdgpu_device *adev, INIT_LIST_HEAD(&bo_va->invalids); INIT_LIST_HEAD(&bo_va->vm_status); - mutex_lock(&vm->mutex); list_add_tail(&bo_va->bo_list, &bo->va); - mutex_unlock(&vm->mutex); return bo_va; } @@ -981,7 +1072,7 @@ struct amdgpu_bo_va *amdgpu_vm_bo_add(struct amdgpu_device *adev, * Add a mapping of the BO at the specefied addr into the VM. * Returns 0 for success, error for failure. * - * Object has to be reserved and gets unreserved by this function! + * Object has to be reserved and unreserved outside! */ int amdgpu_vm_bo_map(struct amdgpu_device *adev, struct amdgpu_bo_va *bo_va, @@ -997,32 +1088,25 @@ int amdgpu_vm_bo_map(struct amdgpu_device *adev, /* validate the parameters */ if (saddr & AMDGPU_GPU_PAGE_MASK || offset & AMDGPU_GPU_PAGE_MASK || - size == 0 || size & AMDGPU_GPU_PAGE_MASK) { - amdgpu_bo_unreserve(bo_va->bo); + size == 0 || size & AMDGPU_GPU_PAGE_MASK) return -EINVAL; - } /* make sure object fit at this offset */ - eaddr = saddr + size; - if ((saddr >= eaddr) || (offset + size > amdgpu_bo_size(bo_va->bo))) { - amdgpu_bo_unreserve(bo_va->bo); + eaddr = saddr + size - 1; + if ((saddr >= eaddr) || (offset + size > amdgpu_bo_size(bo_va->bo))) return -EINVAL; - } last_pfn = eaddr / AMDGPU_GPU_PAGE_SIZE; - if (last_pfn > adev->vm_manager.max_pfn) { - dev_err(adev->dev, "va above limit (0x%08X > 0x%08X)\n", + if (last_pfn >= adev->vm_manager.max_pfn) { + dev_err(adev->dev, "va above limit (0x%08X >= 0x%08X)\n", last_pfn, adev->vm_manager.max_pfn); - amdgpu_bo_unreserve(bo_va->bo); return -EINVAL; } - mutex_lock(&vm->mutex); - saddr /= AMDGPU_GPU_PAGE_SIZE; eaddr /= AMDGPU_GPU_PAGE_SIZE; - it = interval_tree_iter_first(&vm->va, saddr, eaddr - 1); + it = interval_tree_iter_first(&vm->va, saddr, eaddr); if (it) { struct amdgpu_bo_va_mapping *tmp; tmp = container_of(it, struct amdgpu_bo_va_mapping, it); @@ -1030,27 +1114,24 @@ int amdgpu_vm_bo_map(struct amdgpu_device *adev, dev_err(adev->dev, "bo %p va 0x%010Lx-0x%010Lx conflict with " "0x%010lx-0x%010lx\n", bo_va->bo, saddr, eaddr, tmp->it.start, tmp->it.last + 1); - amdgpu_bo_unreserve(bo_va->bo); r = -EINVAL; - goto error_unlock; + goto error; } mapping = kmalloc(sizeof(*mapping), GFP_KERNEL); if (!mapping) { - amdgpu_bo_unreserve(bo_va->bo); r = -ENOMEM; - goto error_unlock; + goto error; } INIT_LIST_HEAD(&mapping->list); mapping->it.start = saddr; - mapping->it.last = eaddr - 1; + mapping->it.last = eaddr; mapping->offset = offset; mapping->flags = flags; list_add(&mapping->list, &bo_va->invalids); interval_tree_insert(&mapping->it, &vm->va); - trace_amdgpu_vm_bo_map(bo_va, mapping); /* Make sure the page tables are allocated */ saddr >>= amdgpu_vm_block_size; @@ -1061,61 +1142,52 @@ int amdgpu_vm_bo_map(struct amdgpu_device *adev, if (eaddr > vm->max_pde_used) vm->max_pde_used = eaddr; - amdgpu_bo_unreserve(bo_va->bo); - /* walk over the address space and allocate the page tables */ for (pt_idx = saddr; pt_idx <= eaddr; ++pt_idx) { struct reservation_object *resv = vm->page_directory->tbo.resv; + struct amdgpu_bo_list_entry *entry; struct amdgpu_bo *pt; - if (vm->page_tables[pt_idx].bo) + entry = &vm->page_tables[pt_idx].entry; + if (entry->robj) continue; - /* drop mutex to allocate and clear page table */ - mutex_unlock(&vm->mutex); - - ww_mutex_lock(&resv->lock, NULL); r = amdgpu_bo_create(adev, AMDGPU_VM_PTE_COUNT * 8, AMDGPU_GPU_PAGE_SIZE, true, AMDGPU_GEM_DOMAIN_VRAM, AMDGPU_GEM_CREATE_NO_CPU_ACCESS, NULL, resv, &pt); - ww_mutex_unlock(&resv->lock); if (r) goto error_free; - r = amdgpu_vm_clear_bo(adev, pt); + /* Keep a reference to the page table to avoid freeing + * them up in the wrong order. + */ + pt->parent = amdgpu_bo_ref(vm->page_directory); + + r = amdgpu_vm_clear_bo(adev, vm, pt); if (r) { amdgpu_bo_unref(&pt); goto error_free; } - /* aquire mutex again */ - mutex_lock(&vm->mutex); - if (vm->page_tables[pt_idx].bo) { - /* someone else allocated the pt in the meantime */ - mutex_unlock(&vm->mutex); - amdgpu_bo_unref(&pt); - mutex_lock(&vm->mutex); - continue; - } - + entry->robj = pt; + entry->priority = 0; + entry->tv.bo = &entry->robj->tbo; + entry->tv.shared = true; + entry->user_pages = NULL; vm->page_tables[pt_idx].addr = 0; - vm->page_tables[pt_idx].bo = pt; } - mutex_unlock(&vm->mutex); return 0; error_free: - mutex_lock(&vm->mutex); list_del(&mapping->list); interval_tree_remove(&mapping->it, &vm->va); trace_amdgpu_vm_bo_unmap(bo_va, mapping); kfree(mapping); -error_unlock: - mutex_unlock(&vm->mutex); +error: return r; } @@ -1129,7 +1201,7 @@ error_unlock: * Remove a mapping of the BO at the specefied addr from the VM. * Returns 0 for success, error for failure. * - * Object has to be reserved and gets unreserved by this function! + * Object has to be reserved and unreserved outside! */ int amdgpu_vm_bo_unmap(struct amdgpu_device *adev, struct amdgpu_bo_va *bo_va, @@ -1154,13 +1226,10 @@ int amdgpu_vm_bo_unmap(struct amdgpu_device *adev, break; } - if (&mapping->list == &bo_va->invalids) { - amdgpu_bo_unreserve(bo_va->bo); + if (&mapping->list == &bo_va->invalids) return -ENOENT; - } } - mutex_lock(&vm->mutex); list_del(&mapping->list); interval_tree_remove(&mapping->it, &vm->va); trace_amdgpu_vm_bo_unmap(bo_va, mapping); @@ -1169,8 +1238,6 @@ int amdgpu_vm_bo_unmap(struct amdgpu_device *adev, list_add(&mapping->list, &vm->freed); else kfree(mapping); - mutex_unlock(&vm->mutex); - amdgpu_bo_unreserve(bo_va->bo); return 0; } @@ -1181,7 +1248,7 @@ int amdgpu_vm_bo_unmap(struct amdgpu_device *adev, * @adev: amdgpu_device pointer * @bo_va: requested bo_va * - * Remove @bo_va->bo from the requested vm (cayman+). + * Remove @bo_va->bo from the requested vm. * * Object have to be reserved! */ @@ -1193,8 +1260,6 @@ void amdgpu_vm_bo_rmv(struct amdgpu_device *adev, list_del(&bo_va->bo_list); - mutex_lock(&vm->mutex); - spin_lock(&vm->status_lock); list_del(&bo_va->vm_status); spin_unlock(&vm->status_lock); @@ -1213,8 +1278,6 @@ void amdgpu_vm_bo_rmv(struct amdgpu_device *adev, fence_put(bo_va->last_pt_update); kfree(bo_va); - - mutex_unlock(&vm->mutex); } /** @@ -1224,7 +1287,7 @@ void amdgpu_vm_bo_rmv(struct amdgpu_device *adev, * @vm: requested vm * @bo: amdgpu buffer object * - * Mark @bo as invalid (cayman+). + * Mark @bo as invalid. */ void amdgpu_vm_bo_invalidate(struct amdgpu_device *adev, struct amdgpu_bo *bo) @@ -1245,21 +1308,22 @@ void amdgpu_vm_bo_invalidate(struct amdgpu_device *adev, * @adev: amdgpu_device pointer * @vm: requested vm * - * Init @vm fields (cayman+). + * Init @vm fields. */ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm) { const unsigned align = min(AMDGPU_VM_PTB_ALIGN_SIZE, AMDGPU_VM_PTE_COUNT * 8); - unsigned pd_size, pd_entries, pts_size; + unsigned pd_size, pd_entries; + unsigned ring_instance; + struct amdgpu_ring *ring; + struct amd_sched_rq *rq; int i, r; for (i = 0; i < AMDGPU_MAX_RINGS; ++i) { - vm->ids[i].id = 0; + vm->ids[i].mgr_id = NULL; vm->ids[i].flushed_updates = NULL; - vm->ids[i].last_id_use = NULL; } - mutex_init(&vm->mutex); vm->va = RB_ROOT; spin_lock_init(&vm->status_lock); INIT_LIST_HEAD(&vm->invalidated); @@ -1270,13 +1334,23 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm) pd_entries = amdgpu_vm_num_pdes(adev); /* allocate page table array */ - pts_size = pd_entries * sizeof(struct amdgpu_vm_pt); - vm->page_tables = kzalloc(pts_size, GFP_KERNEL); + vm->page_tables = drm_calloc_large(pd_entries, sizeof(struct amdgpu_vm_pt)); if (vm->page_tables == NULL) { DRM_ERROR("Cannot allocate memory for page table array\n"); return -ENOMEM; } + /* create scheduler entity for page table updates */ + + ring_instance = atomic_inc_return(&adev->vm_manager.vm_pte_next_ring); + ring_instance %= adev->vm_manager.vm_pte_num_rings; + ring = adev->vm_manager.vm_pte_rings[ring_instance]; + rq = &ring->sched.sched_rq[AMD_SCHED_PRIORITY_KERNEL]; + r = amd_sched_entity_init(&ring->sched, &vm->entity, + rq, amdgpu_sched_jobs); + if (r) + return r; + vm->page_directory_fence = NULL; r = amdgpu_bo_create(adev, pd_size, align, true, @@ -1284,16 +1358,27 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm) AMDGPU_GEM_CREATE_NO_CPU_ACCESS, NULL, NULL, &vm->page_directory); if (r) - return r; + goto error_free_sched_entity; - r = amdgpu_vm_clear_bo(adev, vm->page_directory); - if (r) { - amdgpu_bo_unref(&vm->page_directory); - vm->page_directory = NULL; - return r; - } + r = amdgpu_bo_reserve(vm->page_directory, false); + if (r) + goto error_free_page_directory; + + r = amdgpu_vm_clear_bo(adev, vm, vm->page_directory); + amdgpu_bo_unreserve(vm->page_directory); + if (r) + goto error_free_page_directory; return 0; + +error_free_page_directory: + amdgpu_bo_unref(&vm->page_directory); + vm->page_directory = NULL; + +error_free_sched_entity: + amd_sched_entity_fini(&ring->sched, &vm->entity); + + return r; } /** @@ -1302,7 +1387,7 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm) * @adev: amdgpu_device pointer * @vm: requested vm * - * Tear down @vm (cayman+). + * Tear down @vm. * Unbind the VM and remove all bos from the vm bo list */ void amdgpu_vm_fini(struct amdgpu_device *adev, struct amdgpu_vm *vm) @@ -1310,6 +1395,8 @@ void amdgpu_vm_fini(struct amdgpu_device *adev, struct amdgpu_vm *vm) struct amdgpu_bo_va_mapping *mapping, *tmp; int i; + amd_sched_entity_fini(vm->entity.sched, &vm->entity); + if (!RB_EMPTY_ROOT(&vm->va)) { dev_err(adev->dev, "still active bo inside vm\n"); } @@ -1324,16 +1411,56 @@ void amdgpu_vm_fini(struct amdgpu_device *adev, struct amdgpu_vm *vm) } for (i = 0; i < amdgpu_vm_num_pdes(adev); i++) - amdgpu_bo_unref(&vm->page_tables[i].bo); - kfree(vm->page_tables); + amdgpu_bo_unref(&vm->page_tables[i].entry.robj); + drm_free_large(vm->page_tables); amdgpu_bo_unref(&vm->page_directory); fence_put(vm->page_directory_fence); for (i = 0; i < AMDGPU_MAX_RINGS; ++i) { - fence_put(vm->ids[i].flushed_updates); - amdgpu_fence_unref(&vm->ids[i].last_id_use); + struct amdgpu_vm_id *id = &vm->ids[i]; + + if (id->mgr_id) + atomic_long_cmpxchg(&id->mgr_id->owner, + (long)id, 0); + fence_put(id->flushed_updates); + } +} + +/** + * amdgpu_vm_manager_init - init the VM manager + * + * @adev: amdgpu_device pointer + * + * Initialize the VM manager structures + */ +void amdgpu_vm_manager_init(struct amdgpu_device *adev) +{ + unsigned i; + + INIT_LIST_HEAD(&adev->vm_manager.ids_lru); + + /* skip over VMID 0, since it is the system VM */ + for (i = 1; i < adev->vm_manager.num_ids; ++i) { + amdgpu_vm_reset_id(adev, i); + list_add_tail(&adev->vm_manager.ids[i].list, + &adev->vm_manager.ids_lru); } - mutex_destroy(&vm->mutex); + atomic_set(&adev->vm_manager.vm_pte_next_ring, 0); +} + +/** + * amdgpu_vm_manager_fini - cleanup VM manager + * + * @adev: amdgpu_device pointer + * + * Cleanup the VM manager and free resources. + */ +void amdgpu_vm_manager_fini(struct amdgpu_device *adev) +{ + unsigned i; + + for (i = 0; i < AMDGPU_NUM_VM; ++i) + fence_put(adev->vm_manager.ids[i].active); } diff --git a/drivers/gpu/drm/amd/amdgpu/atom.c b/drivers/gpu/drm/amd/amdgpu/atom.c index a0346a90d805..1b50e6c13fb3 100644 --- a/drivers/gpu/drm/amd/amdgpu/atom.c +++ b/drivers/gpu/drm/amd/amdgpu/atom.c @@ -685,6 +685,27 @@ static void atom_op_div(atom_exec_context *ctx, int *ptr, int arg) } } +static void atom_op_div32(atom_exec_context *ctx, int *ptr, int arg) +{ + uint64_t val64; + uint8_t attr = U8((*ptr)++); + uint32_t dst, src; + SDEBUG(" src1: "); + dst = atom_get_dst(ctx, arg, attr, ptr, NULL, 1); + SDEBUG(" src2: "); + src = atom_get_src(ctx, attr, ptr); + if (src != 0) { + val64 = dst; + val64 |= ((uint64_t)ctx->ctx->divmul[1]) << 32; + do_div(val64, src); + ctx->ctx->divmul[0] = lower_32_bits(val64); + ctx->ctx->divmul[1] = upper_32_bits(val64); + } else { + ctx->ctx->divmul[0] = 0; + ctx->ctx->divmul[1] = 0; + } +} + static void atom_op_eot(atom_exec_context *ctx, int *ptr, int arg) { /* functionally, a nop */ @@ -788,6 +809,20 @@ static void atom_op_mul(atom_exec_context *ctx, int *ptr, int arg) ctx->ctx->divmul[0] = dst * src; } +static void atom_op_mul32(atom_exec_context *ctx, int *ptr, int arg) +{ + uint64_t val64; + uint8_t attr = U8((*ptr)++); + uint32_t dst, src; + SDEBUG(" src1: "); + dst = atom_get_dst(ctx, arg, attr, ptr, NULL, 1); + SDEBUG(" src2: "); + src = atom_get_src(ctx, attr, ptr); + val64 = (uint64_t)dst * (uint64_t)src; + ctx->ctx->divmul[0] = lower_32_bits(val64); + ctx->ctx->divmul[1] = upper_32_bits(val64); +} + static void atom_op_nop(atom_exec_context *ctx, int *ptr, int arg) { /* nothing */ @@ -1022,7 +1057,15 @@ static void atom_op_xor(atom_exec_context *ctx, int *ptr, int arg) static void atom_op_debug(atom_exec_context *ctx, int *ptr, int arg) { - printk(KERN_INFO "unimplemented!\n"); + uint8_t val = U8((*ptr)++); + SDEBUG("DEBUG output: 0x%02X\n", val); +} + +static void atom_op_processds(atom_exec_context *ctx, int *ptr, int arg) +{ + uint16_t val = U16(*ptr); + (*ptr) += val + 2; + SDEBUG("PROCESSDS output: 0x%02X\n", val); } static struct { @@ -1151,7 +1194,13 @@ static struct { atom_op_shr, ATOM_ARG_FB}, { atom_op_shr, ATOM_ARG_PLL}, { atom_op_shr, ATOM_ARG_MC}, { -atom_op_debug, 0},}; + atom_op_debug, 0}, { + atom_op_processds, 0}, { + atom_op_mul32, ATOM_ARG_PS}, { + atom_op_mul32, ATOM_ARG_WS}, { + atom_op_div32, ATOM_ARG_PS}, { + atom_op_div32, ATOM_ARG_WS}, +}; static int amdgpu_atom_execute_table_locked(struct atom_context *ctx, int index, uint32_t * params) { diff --git a/drivers/gpu/drm/amd/amdgpu/atom.h b/drivers/gpu/drm/amd/amdgpu/atom.h index 09d0f8230708..fece8f45dc7a 100644 --- a/drivers/gpu/drm/amd/amdgpu/atom.h +++ b/drivers/gpu/drm/amd/amdgpu/atom.h @@ -60,7 +60,7 @@ #define ATOM_CT_PS_MASK 0x7F #define ATOM_CT_CODE_PTR 6 -#define ATOM_OP_CNT 123 +#define ATOM_OP_CNT 127 #define ATOM_OP_EOT 91 #define ATOM_CASE_MAGIC 0x63 diff --git a/drivers/gpu/drm/amd/amdgpu/atombios_dp.c b/drivers/gpu/drm/amd/amdgpu/atombios_dp.c index 92b6acadfc52..bf731e9f643e 100644 --- a/drivers/gpu/drm/amd/amdgpu/atombios_dp.c +++ b/drivers/gpu/drm/amd/amdgpu/atombios_dp.c @@ -243,7 +243,7 @@ static void amdgpu_atombios_dp_get_adjust_train(const u8 link_status[DP_LINK_STA /* convert bits per color to bits per pixel */ /* get bpc from the EDID */ -static int amdgpu_atombios_dp_convert_bpc_to_bpp(int bpc) +static unsigned amdgpu_atombios_dp_convert_bpc_to_bpp(int bpc) { if (bpc == 0) return 24; @@ -251,64 +251,44 @@ static int amdgpu_atombios_dp_convert_bpc_to_bpp(int bpc) return bpc * 3; } -/* get the max pix clock supported by the link rate and lane num */ -static int amdgpu_atombios_dp_get_max_dp_pix_clock(int link_rate, - int lane_num, - int bpp) -{ - return (link_rate * lane_num * 8) / bpp; -} - /***** amdgpu specific DP functions *****/ -/* First get the min lane# when low rate is used according to pixel clock - * (prefer low rate), second check max lane# supported by DP panel, - * if the max lane# < low rate lane# then use max lane# instead. - */ -static int amdgpu_atombios_dp_get_dp_lane_number(struct drm_connector *connector, +static int amdgpu_atombios_dp_get_dp_link_config(struct drm_connector *connector, const u8 dpcd[DP_DPCD_SIZE], - int pix_clock) + unsigned pix_clock, + unsigned *dp_lanes, unsigned *dp_rate) { - int bpp = amdgpu_atombios_dp_convert_bpc_to_bpp(amdgpu_connector_get_monitor_bpc(connector)); - int max_link_rate = drm_dp_max_link_rate(dpcd); - int max_lane_num = drm_dp_max_lane_count(dpcd); - int lane_num; - int max_dp_pix_clock; - - for (lane_num = 1; lane_num < max_lane_num; lane_num <<= 1) { - max_dp_pix_clock = amdgpu_atombios_dp_get_max_dp_pix_clock(max_link_rate, lane_num, bpp); - if (pix_clock <= max_dp_pix_clock) - break; - } - - return lane_num; -} - -static int amdgpu_atombios_dp_get_dp_link_clock(struct drm_connector *connector, - const u8 dpcd[DP_DPCD_SIZE], - int pix_clock) -{ - int bpp = amdgpu_atombios_dp_convert_bpc_to_bpp(amdgpu_connector_get_monitor_bpc(connector)); - int lane_num, max_pix_clock; + unsigned bpp = + amdgpu_atombios_dp_convert_bpc_to_bpp(amdgpu_connector_get_monitor_bpc(connector)); + static const unsigned link_rates[3] = { 162000, 270000, 540000 }; + unsigned max_link_rate = drm_dp_max_link_rate(dpcd); + unsigned max_lane_num = drm_dp_max_lane_count(dpcd); + unsigned lane_num, i, max_pix_clock; if (amdgpu_connector_encoder_get_dp_bridge_encoder_id(connector) == - ENCODER_OBJECT_ID_NUTMEG) - return 270000; - - lane_num = amdgpu_atombios_dp_get_dp_lane_number(connector, dpcd, pix_clock); - max_pix_clock = amdgpu_atombios_dp_get_max_dp_pix_clock(162000, lane_num, bpp); - if (pix_clock <= max_pix_clock) - return 162000; - max_pix_clock = amdgpu_atombios_dp_get_max_dp_pix_clock(270000, lane_num, bpp); - if (pix_clock <= max_pix_clock) - return 270000; - if (amdgpu_connector_is_dp12_capable(connector)) { - max_pix_clock = amdgpu_atombios_dp_get_max_dp_pix_clock(540000, lane_num, bpp); - if (pix_clock <= max_pix_clock) - return 540000; + ENCODER_OBJECT_ID_NUTMEG) { + for (lane_num = 1; lane_num <= max_lane_num; lane_num <<= 1) { + max_pix_clock = (lane_num * 270000 * 8) / bpp; + if (max_pix_clock >= pix_clock) { + *dp_lanes = lane_num; + *dp_rate = 270000; + return 0; + } + } + } else { + for (lane_num = 1; lane_num <= max_lane_num; lane_num <<= 1) { + for (i = 0; i < ARRAY_SIZE(link_rates) && link_rates[i] <= max_link_rate; i++) { + max_pix_clock = (lane_num * link_rates[i] * 8) / bpp; + if (max_pix_clock >= pix_clock) { + *dp_lanes = lane_num; + *dp_rate = link_rates[i]; + return 0; + } + } + } } - return drm_dp_max_link_rate(dpcd); + return -EINVAL; } static u8 amdgpu_atombios_dp_encoder_service(struct amdgpu_device *adev, @@ -422,6 +402,7 @@ void amdgpu_atombios_dp_set_link_config(struct drm_connector *connector, { struct amdgpu_connector *amdgpu_connector = to_amdgpu_connector(connector); struct amdgpu_connector_atom_dig *dig_connector; + int ret; if (!amdgpu_connector->con_priv) return; @@ -429,10 +410,14 @@ void amdgpu_atombios_dp_set_link_config(struct drm_connector *connector, if ((dig_connector->dp_sink_type == CONNECTOR_OBJECT_ID_DISPLAYPORT) || (dig_connector->dp_sink_type == CONNECTOR_OBJECT_ID_eDP)) { - dig_connector->dp_clock = - amdgpu_atombios_dp_get_dp_link_clock(connector, dig_connector->dpcd, mode->clock); - dig_connector->dp_lane_count = - amdgpu_atombios_dp_get_dp_lane_number(connector, dig_connector->dpcd, mode->clock); + ret = amdgpu_atombios_dp_get_dp_link_config(connector, dig_connector->dpcd, + mode->clock, + &dig_connector->dp_lane_count, + &dig_connector->dp_clock); + if (ret) { + dig_connector->dp_clock = 0; + dig_connector->dp_lane_count = 0; + } } } @@ -441,14 +426,17 @@ int amdgpu_atombios_dp_mode_valid_helper(struct drm_connector *connector, { struct amdgpu_connector *amdgpu_connector = to_amdgpu_connector(connector); struct amdgpu_connector_atom_dig *dig_connector; - int dp_clock; + unsigned dp_lanes, dp_clock; + int ret; if (!amdgpu_connector->con_priv) return MODE_CLOCK_HIGH; dig_connector = amdgpu_connector->con_priv; - dp_clock = - amdgpu_atombios_dp_get_dp_link_clock(connector, dig_connector->dpcd, mode->clock); + ret = amdgpu_atombios_dp_get_dp_link_config(connector, dig_connector->dpcd, + mode->clock, &dp_lanes, &dp_clock); + if (ret) + return MODE_CLOCK_HIGH; if ((dp_clock == 540000) && (!amdgpu_connector_is_dp12_capable(connector))) diff --git a/drivers/gpu/drm/amd/amdgpu/ci_dpm.c b/drivers/gpu/drm/amd/amdgpu/ci_dpm.c index a1a35a5df8e7..1f9109d3348b 100644 --- a/drivers/gpu/drm/amd/amdgpu/ci_dpm.c +++ b/drivers/gpu/drm/amd/amdgpu/ci_dpm.c @@ -31,6 +31,7 @@ #include "ci_dpm.h" #include "gfx_v7_0.h" #include "atom.h" +#include "amd_pcie.h" #include <linux/seq_file.h> #include "smu/smu_7_0_1_d.h" @@ -1395,7 +1396,6 @@ static void ci_thermal_stop_thermal_controller(struct amdgpu_device *adev) ci_fan_ctrl_set_default_mode(adev); } -#if 0 static int ci_read_smc_soft_register(struct amdgpu_device *adev, u16 reg_offset, u32 *value) { @@ -1405,7 +1405,6 @@ static int ci_read_smc_soft_register(struct amdgpu_device *adev, pi->soft_regs_start + reg_offset, value, pi->sram_end); } -#endif static int ci_write_smc_soft_register(struct amdgpu_device *adev, u16 reg_offset, u32 value) @@ -3018,7 +3017,6 @@ static int ci_populate_single_memory_level(struct amdgpu_device *adev, &memory_level->MinVddcPhases); memory_level->EnabledForThrottle = 1; - memory_level->EnabledForActivity = 1; memory_level->UpH = 0; memory_level->DownH = 100; memory_level->VoltageDownH = 0; @@ -3377,7 +3375,6 @@ static int ci_populate_single_graphic_level(struct amdgpu_device *adev, graphic_level->SpllSpreadSpectrum2 = cpu_to_be32(graphic_level->SpllSpreadSpectrum2); graphic_level->CcPwrDynRm = cpu_to_be32(graphic_level->CcPwrDynRm); graphic_level->CcPwrDynRm1 = cpu_to_be32(graphic_level->CcPwrDynRm1); - graphic_level->EnabledForActivity = 1; return 0; } @@ -3408,6 +3405,7 @@ static int ci_populate_all_graphic_levels(struct amdgpu_device *adev) pi->smc_state_table.GraphicsLevel[i].DisplayWatermark = PPSMC_DISPLAY_WATERMARK_HIGH; } + pi->smc_state_table.GraphicsLevel[0].EnabledForActivity = 1; pi->smc_state_table.GraphicsDpmLevelCount = (u8)dpm_table->sclk_table.count; pi->dpm_level_enable_mask.sclk_dpm_enable_mask = @@ -3451,6 +3449,8 @@ static int ci_populate_all_memory_levels(struct amdgpu_device *adev) return ret; } + pi->smc_state_table.MemoryLevel[0].EnabledForActivity = 1; + if ((dpm_table->mclk_table.count >= 2) && ((adev->pdev->device == 0x67B0) || (adev->pdev->device == 0x67B1))) { pi->smc_state_table.MemoryLevel[1].MinVddc = @@ -4382,26 +4382,6 @@ static int ci_dpm_force_performance_level(struct amdgpu_device *adev, } } } - if ((!pi->pcie_dpm_key_disabled) && - pi->dpm_level_enable_mask.pcie_dpm_enable_mask) { - levels = 0; - tmp = pi->dpm_level_enable_mask.pcie_dpm_enable_mask; - while (tmp >>= 1) - levels++; - if (levels) { - ret = ci_dpm_force_state_pcie(adev, level); - if (ret) - return ret; - for (i = 0; i < adev->usec_timeout; i++) { - tmp = (RREG32_SMC(ixTARGET_AND_CURRENT_PROFILE_INDEX_1) & - TARGET_AND_CURRENT_PROFILE_INDEX_1__CURR_PCIE_INDEX_MASK) >> - TARGET_AND_CURRENT_PROFILE_INDEX_1__CURR_PCIE_INDEX__SHIFT; - if (tmp == levels) - break; - udelay(1); - } - } - } } else if (level == AMDGPU_DPM_FORCED_LEVEL_LOW) { if ((!pi->sclk_dpm_key_disabled) && pi->dpm_level_enable_mask.sclk_dpm_enable_mask) { @@ -5396,30 +5376,6 @@ static int ci_dpm_enable(struct amdgpu_device *adev) ci_update_current_ps(adev, boot_ps); - if (adev->irq.installed && - amdgpu_is_internal_thermal_sensor(adev->pm.int_thermal_type)) { -#if 0 - PPSMC_Result result; -#endif - ret = ci_thermal_set_temperature_range(adev, CISLANDS_TEMP_RANGE_MIN, - CISLANDS_TEMP_RANGE_MAX); - if (ret) { - DRM_ERROR("ci_thermal_set_temperature_range failed\n"); - return ret; - } - amdgpu_irq_get(adev, &adev->pm.dpm.thermal.irq, - AMDGPU_THERMAL_IRQ_LOW_TO_HIGH); - amdgpu_irq_get(adev, &adev->pm.dpm.thermal.irq, - AMDGPU_THERMAL_IRQ_HIGH_TO_LOW); - -#if 0 - result = amdgpu_ci_send_msg_to_smc(adev, PPSMC_MSG_EnableThermalInterrupt); - - if (result != PPSMC_Result_OK) - DRM_DEBUG_KMS("Could not enable thermal interrupts.\n"); -#endif - } - return 0; } @@ -5837,18 +5793,16 @@ static int ci_dpm_init(struct amdgpu_device *adev) u8 frev, crev; struct ci_power_info *pi; int ret; - u32 mask; pi = kzalloc(sizeof(struct ci_power_info), GFP_KERNEL); if (pi == NULL) return -ENOMEM; adev->pm.dpm.priv = pi; - ret = drm_pcie_get_speed_cap_mask(adev->ddev, &mask); - if (ret) - pi->sys_pcie_mask = 0; - else - pi->sys_pcie_mask = mask; + pi->sys_pcie_mask = + (adev->pm.pcie_gen_mask & CAIL_PCIE_LINK_SPEED_SUPPORT_MASK) >> + CAIL_PCIE_LINK_SPEED_SUPPORT_SHIFT; + pi->force_pcie_gen = AMDGPU_PCIE_GEN_INVALID; pi->pcie_gen_performance.max = AMDGPU_PCIE_GEN1; @@ -6084,11 +6038,23 @@ ci_dpm_debugfs_print_current_performance_level(struct amdgpu_device *adev, struct amdgpu_ps *rps = &pi->current_rps; u32 sclk = ci_get_average_sclk_freq(adev); u32 mclk = ci_get_average_mclk_freq(adev); + u32 activity_percent = 50; + int ret; + + ret = ci_read_smc_soft_register(adev, offsetof(SMU7_SoftRegisters, AverageGraphicsA), + &activity_percent); + + if (ret == 0) { + activity_percent += 0x80; + activity_percent >>= 8; + activity_percent = activity_percent > 100 ? 100 : activity_percent; + } seq_printf(m, "uvd %sabled\n", pi->uvd_enabled ? "en" : "dis"); seq_printf(m, "vce %sabled\n", rps->vce_active ? "en" : "dis"); seq_printf(m, "power level avg sclk: %u mclk: %u\n", sclk, mclk); + seq_printf(m, "GPU load: %u %%\n", activity_percent); } static void ci_dpm_print_power_state(struct amdgpu_device *adev, @@ -6569,12 +6535,12 @@ static int ci_dpm_set_interrupt_state(struct amdgpu_device *adev, switch (state) { case AMDGPU_IRQ_STATE_DISABLE: cg_thermal_int = RREG32_SMC(ixCG_THERMAL_INT); - cg_thermal_int &= ~CG_THERMAL_INT_CTRL__THERM_INTH_MASK_MASK; + cg_thermal_int |= CG_THERMAL_INT_CTRL__THERM_INTH_MASK_MASK; WREG32_SMC(ixCG_THERMAL_INT, cg_thermal_int); break; case AMDGPU_IRQ_STATE_ENABLE: cg_thermal_int = RREG32_SMC(ixCG_THERMAL_INT); - cg_thermal_int |= CG_THERMAL_INT_CTRL__THERM_INTH_MASK_MASK; + cg_thermal_int &= ~CG_THERMAL_INT_CTRL__THERM_INTH_MASK_MASK; WREG32_SMC(ixCG_THERMAL_INT, cg_thermal_int); break; default: @@ -6586,12 +6552,12 @@ static int ci_dpm_set_interrupt_state(struct amdgpu_device *adev, switch (state) { case AMDGPU_IRQ_STATE_DISABLE: cg_thermal_int = RREG32_SMC(ixCG_THERMAL_INT); - cg_thermal_int &= ~CG_THERMAL_INT_CTRL__THERM_INTL_MASK_MASK; + cg_thermal_int |= CG_THERMAL_INT_CTRL__THERM_INTL_MASK_MASK; WREG32_SMC(ixCG_THERMAL_INT, cg_thermal_int); break; case AMDGPU_IRQ_STATE_ENABLE: cg_thermal_int = RREG32_SMC(ixCG_THERMAL_INT); - cg_thermal_int |= CG_THERMAL_INT_CTRL__THERM_INTL_MASK_MASK; + cg_thermal_int &= ~CG_THERMAL_INT_CTRL__THERM_INTL_MASK_MASK; WREG32_SMC(ixCG_THERMAL_INT, cg_thermal_int); break; default: diff --git a/drivers/gpu/drm/amd/amdgpu/cik.c b/drivers/gpu/drm/amd/amdgpu/cik.c index 484710cfdf82..bddc9ba11495 100644 --- a/drivers/gpu/drm/amd/amdgpu/cik.c +++ b/drivers/gpu/drm/amd/amdgpu/cik.c @@ -32,6 +32,7 @@ #include "amdgpu_vce.h" #include "cikd.h" #include "atom.h" +#include "amd_pcie.h" #include "cik.h" #include "gmc_v7_0.h" @@ -65,6 +66,7 @@ #include "oss/oss_2_0_sh_mask.h" #include "amdgpu_amdkfd.h" +#include "amdgpu_powerplay.h" /* * Indirect registers accessor @@ -929,6 +931,37 @@ static bool cik_read_disabled_bios(struct amdgpu_device *adev) return r; } +static bool cik_read_bios_from_rom(struct amdgpu_device *adev, + u8 *bios, u32 length_bytes) +{ + u32 *dw_ptr; + unsigned long flags; + u32 i, length_dw; + + if (bios == NULL) + return false; + if (length_bytes == 0) + return false; + /* APU vbios image is part of sbios image */ + if (adev->flags & AMD_IS_APU) + return false; + + dw_ptr = (u32 *)bios; + length_dw = ALIGN(length_bytes, 4) / 4; + /* take the smc lock since we are using the smc index */ + spin_lock_irqsave(&adev->smc_idx_lock, flags); + /* set rom index to 0 */ + WREG32(mmSMC_IND_INDEX_0, ixROM_INDEX); + WREG32(mmSMC_IND_DATA_0, 0); + /* set index to data for continous read */ + WREG32(mmSMC_IND_INDEX_0, ixROM_DATA); + for (i = 0; i < length_dw; i++) + dw_ptr[i] = RREG32(mmSMC_IND_DATA_0); + spin_unlock_irqrestore(&adev->smc_idx_lock, flags); + + return true; +} + static struct amdgpu_allowed_register_entry cik_allowed_read_registers[] = { {mmGRBM_STATUS, false}, {mmGB_ADDR_CONFIG, false}, @@ -1026,257 +1059,6 @@ static int cik_read_register(struct amdgpu_device *adev, u32 se_num, return -EINVAL; } -static void cik_print_gpu_status_regs(struct amdgpu_device *adev) -{ - dev_info(adev->dev, " GRBM_STATUS=0x%08X\n", - RREG32(mmGRBM_STATUS)); - dev_info(adev->dev, " GRBM_STATUS2=0x%08X\n", - RREG32(mmGRBM_STATUS2)); - dev_info(adev->dev, " GRBM_STATUS_SE0=0x%08X\n", - RREG32(mmGRBM_STATUS_SE0)); - dev_info(adev->dev, " GRBM_STATUS_SE1=0x%08X\n", - RREG32(mmGRBM_STATUS_SE1)); - dev_info(adev->dev, " GRBM_STATUS_SE2=0x%08X\n", - RREG32(mmGRBM_STATUS_SE2)); - dev_info(adev->dev, " GRBM_STATUS_SE3=0x%08X\n", - RREG32(mmGRBM_STATUS_SE3)); - dev_info(adev->dev, " SRBM_STATUS=0x%08X\n", - RREG32(mmSRBM_STATUS)); - dev_info(adev->dev, " SRBM_STATUS2=0x%08X\n", - RREG32(mmSRBM_STATUS2)); - dev_info(adev->dev, " SDMA0_STATUS_REG = 0x%08X\n", - RREG32(mmSDMA0_STATUS_REG + SDMA0_REGISTER_OFFSET)); - dev_info(adev->dev, " SDMA1_STATUS_REG = 0x%08X\n", - RREG32(mmSDMA0_STATUS_REG + SDMA1_REGISTER_OFFSET)); - dev_info(adev->dev, " CP_STAT = 0x%08x\n", RREG32(mmCP_STAT)); - dev_info(adev->dev, " CP_STALLED_STAT1 = 0x%08x\n", - RREG32(mmCP_STALLED_STAT1)); - dev_info(adev->dev, " CP_STALLED_STAT2 = 0x%08x\n", - RREG32(mmCP_STALLED_STAT2)); - dev_info(adev->dev, " CP_STALLED_STAT3 = 0x%08x\n", - RREG32(mmCP_STALLED_STAT3)); - dev_info(adev->dev, " CP_CPF_BUSY_STAT = 0x%08x\n", - RREG32(mmCP_CPF_BUSY_STAT)); - dev_info(adev->dev, " CP_CPF_STALLED_STAT1 = 0x%08x\n", - RREG32(mmCP_CPF_STALLED_STAT1)); - dev_info(adev->dev, " CP_CPF_STATUS = 0x%08x\n", RREG32(mmCP_CPF_STATUS)); - dev_info(adev->dev, " CP_CPC_BUSY_STAT = 0x%08x\n", RREG32(mmCP_CPC_BUSY_STAT)); - dev_info(adev->dev, " CP_CPC_STALLED_STAT1 = 0x%08x\n", - RREG32(mmCP_CPC_STALLED_STAT1)); - dev_info(adev->dev, " CP_CPC_STATUS = 0x%08x\n", RREG32(mmCP_CPC_STATUS)); -} - -/** - * cik_gpu_check_soft_reset - check which blocks are busy - * - * @adev: amdgpu_device pointer - * - * Check which blocks are busy and return the relevant reset - * mask to be used by cik_gpu_soft_reset(). - * Returns a mask of the blocks to be reset. - */ -u32 amdgpu_cik_gpu_check_soft_reset(struct amdgpu_device *adev) -{ - u32 reset_mask = 0; - u32 tmp; - - /* GRBM_STATUS */ - tmp = RREG32(mmGRBM_STATUS); - if (tmp & (GRBM_STATUS__PA_BUSY_MASK | GRBM_STATUS__SC_BUSY_MASK | - GRBM_STATUS__BCI_BUSY_MASK | GRBM_STATUS__SX_BUSY_MASK | - GRBM_STATUS__TA_BUSY_MASK | GRBM_STATUS__VGT_BUSY_MASK | - GRBM_STATUS__DB_BUSY_MASK | GRBM_STATUS__CB_BUSY_MASK | - GRBM_STATUS__GDS_BUSY_MASK | GRBM_STATUS__SPI_BUSY_MASK | - GRBM_STATUS__IA_BUSY_MASK | GRBM_STATUS__IA_BUSY_NO_DMA_MASK)) - reset_mask |= AMDGPU_RESET_GFX; - - if (tmp & (GRBM_STATUS__CP_BUSY_MASK | GRBM_STATUS__CP_COHERENCY_BUSY_MASK)) - reset_mask |= AMDGPU_RESET_CP; - - /* GRBM_STATUS2 */ - tmp = RREG32(mmGRBM_STATUS2); - if (tmp & GRBM_STATUS2__RLC_BUSY_MASK) - reset_mask |= AMDGPU_RESET_RLC; - - /* SDMA0_STATUS_REG */ - tmp = RREG32(mmSDMA0_STATUS_REG + SDMA0_REGISTER_OFFSET); - if (!(tmp & SDMA0_STATUS_REG__IDLE_MASK)) - reset_mask |= AMDGPU_RESET_DMA; - - /* SDMA1_STATUS_REG */ - tmp = RREG32(mmSDMA0_STATUS_REG + SDMA1_REGISTER_OFFSET); - if (!(tmp & SDMA0_STATUS_REG__IDLE_MASK)) - reset_mask |= AMDGPU_RESET_DMA1; - - /* SRBM_STATUS2 */ - tmp = RREG32(mmSRBM_STATUS2); - if (tmp & SRBM_STATUS2__SDMA_BUSY_MASK) - reset_mask |= AMDGPU_RESET_DMA; - - if (tmp & SRBM_STATUS2__SDMA1_BUSY_MASK) - reset_mask |= AMDGPU_RESET_DMA1; - - /* SRBM_STATUS */ - tmp = RREG32(mmSRBM_STATUS); - - if (tmp & SRBM_STATUS__IH_BUSY_MASK) - reset_mask |= AMDGPU_RESET_IH; - - if (tmp & SRBM_STATUS__SEM_BUSY_MASK) - reset_mask |= AMDGPU_RESET_SEM; - - if (tmp & SRBM_STATUS__GRBM_RQ_PENDING_MASK) - reset_mask |= AMDGPU_RESET_GRBM; - - if (tmp & SRBM_STATUS__VMC_BUSY_MASK) - reset_mask |= AMDGPU_RESET_VMC; - - if (tmp & (SRBM_STATUS__MCB_BUSY_MASK | SRBM_STATUS__MCB_NON_DISPLAY_BUSY_MASK | - SRBM_STATUS__MCC_BUSY_MASK | SRBM_STATUS__MCD_BUSY_MASK)) - reset_mask |= AMDGPU_RESET_MC; - - if (amdgpu_display_is_display_hung(adev)) - reset_mask |= AMDGPU_RESET_DISPLAY; - - /* Skip MC reset as it's mostly likely not hung, just busy */ - if (reset_mask & AMDGPU_RESET_MC) { - DRM_DEBUG("MC busy: 0x%08X, clearing.\n", reset_mask); - reset_mask &= ~AMDGPU_RESET_MC; - } - - return reset_mask; -} - -/** - * cik_gpu_soft_reset - soft reset GPU - * - * @adev: amdgpu_device pointer - * @reset_mask: mask of which blocks to reset - * - * Soft reset the blocks specified in @reset_mask. - */ -static void cik_gpu_soft_reset(struct amdgpu_device *adev, u32 reset_mask) -{ - struct amdgpu_mode_mc_save save; - u32 grbm_soft_reset = 0, srbm_soft_reset = 0; - u32 tmp; - - if (reset_mask == 0) - return; - - dev_info(adev->dev, "GPU softreset: 0x%08X\n", reset_mask); - - cik_print_gpu_status_regs(adev); - dev_info(adev->dev, " VM_CONTEXT1_PROTECTION_FAULT_ADDR 0x%08X\n", - RREG32(mmVM_CONTEXT1_PROTECTION_FAULT_ADDR)); - dev_info(adev->dev, " VM_CONTEXT1_PROTECTION_FAULT_STATUS 0x%08X\n", - RREG32(mmVM_CONTEXT1_PROTECTION_FAULT_STATUS)); - - /* disable CG/PG */ - - /* stop the rlc */ - gfx_v7_0_rlc_stop(adev); - - /* Disable GFX parsing/prefetching */ - WREG32(mmCP_ME_CNTL, CP_ME_CNTL__ME_HALT_MASK | CP_ME_CNTL__PFP_HALT_MASK | CP_ME_CNTL__CE_HALT_MASK); - - /* Disable MEC parsing/prefetching */ - WREG32(mmCP_MEC_CNTL, CP_MEC_CNTL__MEC_ME1_HALT_MASK | CP_MEC_CNTL__MEC_ME2_HALT_MASK); - - if (reset_mask & AMDGPU_RESET_DMA) { - /* sdma0 */ - tmp = RREG32(mmSDMA0_F32_CNTL + SDMA0_REGISTER_OFFSET); - tmp |= SDMA0_F32_CNTL__HALT_MASK; - WREG32(mmSDMA0_F32_CNTL + SDMA0_REGISTER_OFFSET, tmp); - } - if (reset_mask & AMDGPU_RESET_DMA1) { - /* sdma1 */ - tmp = RREG32(mmSDMA0_F32_CNTL + SDMA1_REGISTER_OFFSET); - tmp |= SDMA0_F32_CNTL__HALT_MASK; - WREG32(mmSDMA0_F32_CNTL + SDMA1_REGISTER_OFFSET, tmp); - } - - gmc_v7_0_mc_stop(adev, &save); - if (amdgpu_asic_wait_for_mc_idle(adev)) { - dev_warn(adev->dev, "Wait for MC idle timedout !\n"); - } - - if (reset_mask & (AMDGPU_RESET_GFX | AMDGPU_RESET_COMPUTE | AMDGPU_RESET_CP)) - grbm_soft_reset = GRBM_SOFT_RESET__SOFT_RESET_CP_MASK | - GRBM_SOFT_RESET__SOFT_RESET_GFX_MASK; - - if (reset_mask & AMDGPU_RESET_CP) { - grbm_soft_reset |= GRBM_SOFT_RESET__SOFT_RESET_CP_MASK; - - srbm_soft_reset |= SRBM_SOFT_RESET__SOFT_RESET_GRBM_MASK; - } - - if (reset_mask & AMDGPU_RESET_DMA) - srbm_soft_reset |= SRBM_SOFT_RESET__SOFT_RESET_SDMA_MASK; - - if (reset_mask & AMDGPU_RESET_DMA1) - srbm_soft_reset |= SRBM_SOFT_RESET__SOFT_RESET_SDMA1_MASK; - - if (reset_mask & AMDGPU_RESET_DISPLAY) - srbm_soft_reset |= SRBM_SOFT_RESET__SOFT_RESET_DC_MASK; - - if (reset_mask & AMDGPU_RESET_RLC) - grbm_soft_reset |= GRBM_SOFT_RESET__SOFT_RESET_RLC_MASK; - - if (reset_mask & AMDGPU_RESET_SEM) - srbm_soft_reset |= SRBM_SOFT_RESET__SOFT_RESET_SEM_MASK; - - if (reset_mask & AMDGPU_RESET_IH) - srbm_soft_reset |= SRBM_SOFT_RESET__SOFT_RESET_IH_MASK; - - if (reset_mask & AMDGPU_RESET_GRBM) - srbm_soft_reset |= SRBM_SOFT_RESET__SOFT_RESET_GRBM_MASK; - - if (reset_mask & AMDGPU_RESET_VMC) - srbm_soft_reset |= SRBM_SOFT_RESET__SOFT_RESET_VMC_MASK; - - if (!(adev->flags & AMD_IS_APU)) { - if (reset_mask & AMDGPU_RESET_MC) - srbm_soft_reset |= SRBM_SOFT_RESET__SOFT_RESET_MC_MASK; - } - - if (grbm_soft_reset) { - tmp = RREG32(mmGRBM_SOFT_RESET); - tmp |= grbm_soft_reset; - dev_info(adev->dev, "GRBM_SOFT_RESET=0x%08X\n", tmp); - WREG32(mmGRBM_SOFT_RESET, tmp); - tmp = RREG32(mmGRBM_SOFT_RESET); - - udelay(50); - - tmp &= ~grbm_soft_reset; - WREG32(mmGRBM_SOFT_RESET, tmp); - tmp = RREG32(mmGRBM_SOFT_RESET); - } - - if (srbm_soft_reset) { - tmp = RREG32(mmSRBM_SOFT_RESET); - tmp |= srbm_soft_reset; - dev_info(adev->dev, "SRBM_SOFT_RESET=0x%08X\n", tmp); - WREG32(mmSRBM_SOFT_RESET, tmp); - tmp = RREG32(mmSRBM_SOFT_RESET); - - udelay(50); - - tmp &= ~srbm_soft_reset; - WREG32(mmSRBM_SOFT_RESET, tmp); - tmp = RREG32(mmSRBM_SOFT_RESET); - } - - /* Wait a little for things to settle down */ - udelay(50); - - gmc_v7_0_mc_resume(adev, &save); - udelay(50); - - cik_print_gpu_status_regs(adev); -} - struct kv_reset_save_regs { u32 gmcon_reng_execute; u32 gmcon_misc; @@ -1372,45 +1154,11 @@ static void kv_restore_regs_for_reset(struct amdgpu_device *adev, static void cik_gpu_pci_config_reset(struct amdgpu_device *adev) { - struct amdgpu_mode_mc_save save; struct kv_reset_save_regs kv_save = { 0 }; - u32 tmp, i; + u32 i; dev_info(adev->dev, "GPU pci config reset\n"); - /* disable dpm? */ - - /* disable cg/pg */ - - /* Disable GFX parsing/prefetching */ - WREG32(mmCP_ME_CNTL, CP_ME_CNTL__ME_HALT_MASK | - CP_ME_CNTL__PFP_HALT_MASK | CP_ME_CNTL__CE_HALT_MASK); - - /* Disable MEC parsing/prefetching */ - WREG32(mmCP_MEC_CNTL, - CP_MEC_CNTL__MEC_ME1_HALT_MASK | CP_MEC_CNTL__MEC_ME2_HALT_MASK); - - /* sdma0 */ - tmp = RREG32(mmSDMA0_F32_CNTL + SDMA0_REGISTER_OFFSET); - tmp |= SDMA0_F32_CNTL__HALT_MASK; - WREG32(mmSDMA0_F32_CNTL + SDMA0_REGISTER_OFFSET, tmp); - /* sdma1 */ - tmp = RREG32(mmSDMA0_F32_CNTL + SDMA1_REGISTER_OFFSET); - tmp |= SDMA0_F32_CNTL__HALT_MASK; - WREG32(mmSDMA0_F32_CNTL + SDMA1_REGISTER_OFFSET, tmp); - /* XXX other engines? */ - - /* halt the rlc, disable cp internal ints */ - gfx_v7_0_rlc_stop(adev); - - udelay(50); - - /* disable mem access */ - gmc_v7_0_mc_stop(adev, &save); - if (amdgpu_asic_wait_for_mc_idle(adev)) { - dev_warn(adev->dev, "Wait for MC idle timed out !\n"); - } - if (adev->flags & AMD_IS_APU) kv_save_regs_for_reset(adev, &kv_save); @@ -1456,26 +1204,11 @@ static void cik_set_bios_scratch_engine_hung(struct amdgpu_device *adev, bool hu */ static int cik_asic_reset(struct amdgpu_device *adev) { - u32 reset_mask; + cik_set_bios_scratch_engine_hung(adev, true); - reset_mask = amdgpu_cik_gpu_check_soft_reset(adev); + cik_gpu_pci_config_reset(adev); - if (reset_mask) - cik_set_bios_scratch_engine_hung(adev, true); - - /* try soft reset */ - cik_gpu_soft_reset(adev, reset_mask); - - reset_mask = amdgpu_cik_gpu_check_soft_reset(adev); - - /* try pci config reset */ - if (reset_mask && amdgpu_hard_reset) - cik_gpu_pci_config_reset(adev); - - reset_mask = amdgpu_cik_gpu_check_soft_reset(adev); - - if (!reset_mask) - cik_set_bios_scratch_engine_hung(adev, false); + cik_set_bios_scratch_engine_hung(adev, false); return 0; } @@ -1563,8 +1296,8 @@ static void cik_pcie_gen3_enable(struct amdgpu_device *adev) { struct pci_dev *root = adev->pdev->bus->self; int bridge_pos, gpu_pos; - u32 speed_cntl, mask, current_data_rate; - int ret, i; + u32 speed_cntl, current_data_rate; + int i; u16 tmp16; if (pci_is_root_bus(adev->pdev->bus)) @@ -1576,23 +1309,20 @@ static void cik_pcie_gen3_enable(struct amdgpu_device *adev) if (adev->flags & AMD_IS_APU) return; - ret = drm_pcie_get_speed_cap_mask(adev->ddev, &mask); - if (ret != 0) - return; - - if (!(mask & (DRM_PCIE_SPEED_50 | DRM_PCIE_SPEED_80))) + if (!(adev->pm.pcie_gen_mask & (CAIL_PCIE_LINK_SPEED_SUPPORT_GEN2 | + CAIL_PCIE_LINK_SPEED_SUPPORT_GEN3))) return; speed_cntl = RREG32_PCIE(ixPCIE_LC_SPEED_CNTL); current_data_rate = (speed_cntl & PCIE_LC_SPEED_CNTL__LC_CURRENT_DATA_RATE_MASK) >> PCIE_LC_SPEED_CNTL__LC_CURRENT_DATA_RATE__SHIFT; - if (mask & DRM_PCIE_SPEED_80) { + if (adev->pm.pcie_gen_mask & CAIL_PCIE_LINK_SPEED_SUPPORT_GEN3) { if (current_data_rate == 2) { DRM_INFO("PCIE gen 3 link speeds already enabled\n"); return; } DRM_INFO("enabling PCIE gen 3 link speeds, disable with amdgpu.pcie_gen2=0\n"); - } else if (mask & DRM_PCIE_SPEED_50) { + } else if (adev->pm.pcie_gen_mask & CAIL_PCIE_LINK_SPEED_SUPPORT_GEN2) { if (current_data_rate == 1) { DRM_INFO("PCIE gen 2 link speeds already enabled\n"); return; @@ -1608,7 +1338,7 @@ static void cik_pcie_gen3_enable(struct amdgpu_device *adev) if (!gpu_pos) return; - if (mask & DRM_PCIE_SPEED_80) { + if (adev->pm.pcie_gen_mask & CAIL_PCIE_LINK_SPEED_SUPPORT_GEN3) { /* re-try equalization if gen3 is not already enabled */ if (current_data_rate != 2) { u16 bridge_cfg, gpu_cfg; @@ -1703,9 +1433,9 @@ static void cik_pcie_gen3_enable(struct amdgpu_device *adev) pci_read_config_word(adev->pdev, gpu_pos + PCI_EXP_LNKCTL2, &tmp16); tmp16 &= ~0xf; - if (mask & DRM_PCIE_SPEED_80) + if (adev->pm.pcie_gen_mask & CAIL_PCIE_LINK_SPEED_SUPPORT_GEN3) tmp16 |= 3; /* gen3 */ - else if (mask & DRM_PCIE_SPEED_50) + else if (adev->pm.pcie_gen_mask & CAIL_PCIE_LINK_SPEED_SUPPORT_GEN2) tmp16 |= 2; /* gen2 */ else tmp16 |= 1; /* gen1 */ @@ -1732,6 +1462,9 @@ static void cik_program_aspm(struct amdgpu_device *adev) if (amdgpu_aspm == 0) return; + if (pci_is_root_bus(adev->pdev->bus)) + return; + /* XXX double check APUs */ if (adev->flags & AMD_IS_APU) return; @@ -1922,7 +1655,7 @@ static const struct amdgpu_ip_block_version bonaire_ip_blocks[] = .major = 7, .minor = 0, .rev = 0, - .funcs = &ci_dpm_ip_funcs, + .funcs = &amdgpu_pp_ip_funcs, }, { .type = AMD_IP_BLOCK_TYPE_DCE, @@ -1990,7 +1723,7 @@ static const struct amdgpu_ip_block_version hawaii_ip_blocks[] = .major = 7, .minor = 0, .rev = 0, - .funcs = &ci_dpm_ip_funcs, + .funcs = &amdgpu_pp_ip_funcs, }, { .type = AMD_IP_BLOCK_TYPE_DCE, @@ -2058,7 +1791,7 @@ static const struct amdgpu_ip_block_version kabini_ip_blocks[] = .major = 7, .minor = 0, .rev = 0, - .funcs = &kv_dpm_ip_funcs, + .funcs = &amdgpu_pp_ip_funcs, }, { .type = AMD_IP_BLOCK_TYPE_DCE, @@ -2126,7 +1859,7 @@ static const struct amdgpu_ip_block_version mullins_ip_blocks[] = .major = 7, .minor = 0, .rev = 0, - .funcs = &kv_dpm_ip_funcs, + .funcs = &amdgpu_pp_ip_funcs, }, { .type = AMD_IP_BLOCK_TYPE_DCE, @@ -2194,7 +1927,7 @@ static const struct amdgpu_ip_block_version kaveri_ip_blocks[] = .major = 7, .minor = 0, .rev = 0, - .funcs = &kv_dpm_ip_funcs, + .funcs = &amdgpu_pp_ip_funcs, }, { .type = AMD_IP_BLOCK_TYPE_DCE, @@ -2267,6 +2000,7 @@ int cik_set_ip_blocks(struct amdgpu_device *adev) static const struct amdgpu_asic_funcs cik_asic_funcs = { .read_disabled_bios = &cik_read_disabled_bios, + .read_bios_from_rom = &cik_read_bios_from_rom, .read_register = &cik_read_register, .reset = &cik_asic_reset, .set_vga_state = &cik_vga_set_state, @@ -2294,79 +2028,77 @@ static int cik_common_early_init(void *handle) adev->asic_funcs = &cik_asic_funcs; - adev->has_uvd = true; - adev->rev_id = cik_get_rev_id(adev); adev->external_rev_id = 0xFF; switch (adev->asic_type) { case CHIP_BONAIRE: adev->cg_flags = - AMDGPU_CG_SUPPORT_GFX_MGCG | - AMDGPU_CG_SUPPORT_GFX_MGLS | - /*AMDGPU_CG_SUPPORT_GFX_CGCG |*/ - AMDGPU_CG_SUPPORT_GFX_CGLS | - AMDGPU_CG_SUPPORT_GFX_CGTS | - AMDGPU_CG_SUPPORT_GFX_CGTS_LS | - AMDGPU_CG_SUPPORT_GFX_CP_LS | - AMDGPU_CG_SUPPORT_MC_LS | - AMDGPU_CG_SUPPORT_MC_MGCG | - AMDGPU_CG_SUPPORT_SDMA_MGCG | - AMDGPU_CG_SUPPORT_SDMA_LS | - AMDGPU_CG_SUPPORT_BIF_LS | - AMDGPU_CG_SUPPORT_VCE_MGCG | - AMDGPU_CG_SUPPORT_UVD_MGCG | - AMDGPU_CG_SUPPORT_HDP_LS | - AMDGPU_CG_SUPPORT_HDP_MGCG; + AMD_CG_SUPPORT_GFX_MGCG | + AMD_CG_SUPPORT_GFX_MGLS | + /*AMD_CG_SUPPORT_GFX_CGCG |*/ + AMD_CG_SUPPORT_GFX_CGLS | + AMD_CG_SUPPORT_GFX_CGTS | + AMD_CG_SUPPORT_GFX_CGTS_LS | + AMD_CG_SUPPORT_GFX_CP_LS | + AMD_CG_SUPPORT_MC_LS | + AMD_CG_SUPPORT_MC_MGCG | + AMD_CG_SUPPORT_SDMA_MGCG | + AMD_CG_SUPPORT_SDMA_LS | + AMD_CG_SUPPORT_BIF_LS | + AMD_CG_SUPPORT_VCE_MGCG | + AMD_CG_SUPPORT_UVD_MGCG | + AMD_CG_SUPPORT_HDP_LS | + AMD_CG_SUPPORT_HDP_MGCG; adev->pg_flags = 0; adev->external_rev_id = adev->rev_id + 0x14; break; case CHIP_HAWAII: adev->cg_flags = - AMDGPU_CG_SUPPORT_GFX_MGCG | - AMDGPU_CG_SUPPORT_GFX_MGLS | - /*AMDGPU_CG_SUPPORT_GFX_CGCG |*/ - AMDGPU_CG_SUPPORT_GFX_CGLS | - AMDGPU_CG_SUPPORT_GFX_CGTS | - AMDGPU_CG_SUPPORT_GFX_CP_LS | - AMDGPU_CG_SUPPORT_MC_LS | - AMDGPU_CG_SUPPORT_MC_MGCG | - AMDGPU_CG_SUPPORT_SDMA_MGCG | - AMDGPU_CG_SUPPORT_SDMA_LS | - AMDGPU_CG_SUPPORT_BIF_LS | - AMDGPU_CG_SUPPORT_VCE_MGCG | - AMDGPU_CG_SUPPORT_UVD_MGCG | - AMDGPU_CG_SUPPORT_HDP_LS | - AMDGPU_CG_SUPPORT_HDP_MGCG; + AMD_CG_SUPPORT_GFX_MGCG | + AMD_CG_SUPPORT_GFX_MGLS | + /*AMD_CG_SUPPORT_GFX_CGCG |*/ + AMD_CG_SUPPORT_GFX_CGLS | + AMD_CG_SUPPORT_GFX_CGTS | + AMD_CG_SUPPORT_GFX_CP_LS | + AMD_CG_SUPPORT_MC_LS | + AMD_CG_SUPPORT_MC_MGCG | + AMD_CG_SUPPORT_SDMA_MGCG | + AMD_CG_SUPPORT_SDMA_LS | + AMD_CG_SUPPORT_BIF_LS | + AMD_CG_SUPPORT_VCE_MGCG | + AMD_CG_SUPPORT_UVD_MGCG | + AMD_CG_SUPPORT_HDP_LS | + AMD_CG_SUPPORT_HDP_MGCG; adev->pg_flags = 0; adev->external_rev_id = 0x28; break; case CHIP_KAVERI: adev->cg_flags = - AMDGPU_CG_SUPPORT_GFX_MGCG | - AMDGPU_CG_SUPPORT_GFX_MGLS | - /*AMDGPU_CG_SUPPORT_GFX_CGCG |*/ - AMDGPU_CG_SUPPORT_GFX_CGLS | - AMDGPU_CG_SUPPORT_GFX_CGTS | - AMDGPU_CG_SUPPORT_GFX_CGTS_LS | - AMDGPU_CG_SUPPORT_GFX_CP_LS | - AMDGPU_CG_SUPPORT_SDMA_MGCG | - AMDGPU_CG_SUPPORT_SDMA_LS | - AMDGPU_CG_SUPPORT_BIF_LS | - AMDGPU_CG_SUPPORT_VCE_MGCG | - AMDGPU_CG_SUPPORT_UVD_MGCG | - AMDGPU_CG_SUPPORT_HDP_LS | - AMDGPU_CG_SUPPORT_HDP_MGCG; + AMD_CG_SUPPORT_GFX_MGCG | + AMD_CG_SUPPORT_GFX_MGLS | + /*AMD_CG_SUPPORT_GFX_CGCG |*/ + AMD_CG_SUPPORT_GFX_CGLS | + AMD_CG_SUPPORT_GFX_CGTS | + AMD_CG_SUPPORT_GFX_CGTS_LS | + AMD_CG_SUPPORT_GFX_CP_LS | + AMD_CG_SUPPORT_SDMA_MGCG | + AMD_CG_SUPPORT_SDMA_LS | + AMD_CG_SUPPORT_BIF_LS | + AMD_CG_SUPPORT_VCE_MGCG | + AMD_CG_SUPPORT_UVD_MGCG | + AMD_CG_SUPPORT_HDP_LS | + AMD_CG_SUPPORT_HDP_MGCG; adev->pg_flags = - /*AMDGPU_PG_SUPPORT_GFX_PG | - AMDGPU_PG_SUPPORT_GFX_SMG | - AMDGPU_PG_SUPPORT_GFX_DMG |*/ - AMDGPU_PG_SUPPORT_UVD | - /*AMDGPU_PG_SUPPORT_VCE | - AMDGPU_PG_SUPPORT_CP | - AMDGPU_PG_SUPPORT_GDS | - AMDGPU_PG_SUPPORT_RLC_SMU_HS | - AMDGPU_PG_SUPPORT_ACP | - AMDGPU_PG_SUPPORT_SAMU |*/ + /*AMD_PG_SUPPORT_GFX_PG | + AMD_PG_SUPPORT_GFX_SMG | + AMD_PG_SUPPORT_GFX_DMG |*/ + AMD_PG_SUPPORT_UVD | + /*AMD_PG_SUPPORT_VCE | + AMD_PG_SUPPORT_CP | + AMD_PG_SUPPORT_GDS | + AMD_PG_SUPPORT_RLC_SMU_HS | + AMD_PG_SUPPORT_ACP | + AMD_PG_SUPPORT_SAMU |*/ 0; if (adev->pdev->device == 0x1312 || adev->pdev->device == 0x1316 || @@ -2378,29 +2110,29 @@ static int cik_common_early_init(void *handle) case CHIP_KABINI: case CHIP_MULLINS: adev->cg_flags = - AMDGPU_CG_SUPPORT_GFX_MGCG | - AMDGPU_CG_SUPPORT_GFX_MGLS | - /*AMDGPU_CG_SUPPORT_GFX_CGCG |*/ - AMDGPU_CG_SUPPORT_GFX_CGLS | - AMDGPU_CG_SUPPORT_GFX_CGTS | - AMDGPU_CG_SUPPORT_GFX_CGTS_LS | - AMDGPU_CG_SUPPORT_GFX_CP_LS | - AMDGPU_CG_SUPPORT_SDMA_MGCG | - AMDGPU_CG_SUPPORT_SDMA_LS | - AMDGPU_CG_SUPPORT_BIF_LS | - AMDGPU_CG_SUPPORT_VCE_MGCG | - AMDGPU_CG_SUPPORT_UVD_MGCG | - AMDGPU_CG_SUPPORT_HDP_LS | - AMDGPU_CG_SUPPORT_HDP_MGCG; + AMD_CG_SUPPORT_GFX_MGCG | + AMD_CG_SUPPORT_GFX_MGLS | + /*AMD_CG_SUPPORT_GFX_CGCG |*/ + AMD_CG_SUPPORT_GFX_CGLS | + AMD_CG_SUPPORT_GFX_CGTS | + AMD_CG_SUPPORT_GFX_CGTS_LS | + AMD_CG_SUPPORT_GFX_CP_LS | + AMD_CG_SUPPORT_SDMA_MGCG | + AMD_CG_SUPPORT_SDMA_LS | + AMD_CG_SUPPORT_BIF_LS | + AMD_CG_SUPPORT_VCE_MGCG | + AMD_CG_SUPPORT_UVD_MGCG | + AMD_CG_SUPPORT_HDP_LS | + AMD_CG_SUPPORT_HDP_MGCG; adev->pg_flags = - /*AMDGPU_PG_SUPPORT_GFX_PG | - AMDGPU_PG_SUPPORT_GFX_SMG | */ - AMDGPU_PG_SUPPORT_UVD | - /*AMDGPU_PG_SUPPORT_VCE | - AMDGPU_PG_SUPPORT_CP | - AMDGPU_PG_SUPPORT_GDS | - AMDGPU_PG_SUPPORT_RLC_SMU_HS | - AMDGPU_PG_SUPPORT_SAMU |*/ + /*AMD_PG_SUPPORT_GFX_PG | + AMD_PG_SUPPORT_GFX_SMG | */ + AMD_PG_SUPPORT_UVD | + /*AMD_PG_SUPPORT_VCE | + AMD_PG_SUPPORT_CP | + AMD_PG_SUPPORT_GDS | + AMD_PG_SUPPORT_RLC_SMU_HS | + AMD_PG_SUPPORT_SAMU |*/ 0; if (adev->asic_type == CHIP_KABINI) { if (adev->rev_id == 0) @@ -2417,6 +2149,8 @@ static int cik_common_early_init(void *handle) return -EINVAL; } + amdgpu_get_pcie_info(adev); + return 0; } diff --git a/drivers/gpu/drm/amd/amdgpu/cik_ih.c b/drivers/gpu/drm/amd/amdgpu/cik_ih.c index 8993c50cb89f..30c9b3beeef9 100644 --- a/drivers/gpu/drm/amd/amdgpu/cik_ih.c +++ b/drivers/gpu/drm/amd/amdgpu/cik_ih.c @@ -274,6 +274,11 @@ static void cik_ih_set_rptr(struct amdgpu_device *adev) static int cik_ih_early_init(void *handle) { struct amdgpu_device *adev = (struct amdgpu_device *)handle; + int ret; + + ret = amdgpu_irq_add_domain(adev); + if (ret) + return ret; cik_ih_set_interrupt_funcs(adev); @@ -300,6 +305,7 @@ static int cik_ih_sw_fini(void *handle) amdgpu_irq_fini(adev); amdgpu_ih_ring_fini(adev); + amdgpu_irq_remove_domain(adev); return 0; } diff --git a/drivers/gpu/drm/amd/amdgpu/cik_sdma.c b/drivers/gpu/drm/amd/amdgpu/cik_sdma.c index 9ea9de457da3..d3ac3298fba8 100644 --- a/drivers/gpu/drm/amd/amdgpu/cik_sdma.c +++ b/drivers/gpu/drm/amd/amdgpu/cik_sdma.c @@ -96,7 +96,7 @@ static int cik_sdma_init_microcode(struct amdgpu_device *adev) { const char *chip_name; char fw_name[30]; - int err, i; + int err = 0, i; DRM_DEBUG("\n"); @@ -119,24 +119,24 @@ static int cik_sdma_init_microcode(struct amdgpu_device *adev) default: BUG(); } - for (i = 0; i < SDMA_MAX_INSTANCE; i++) { + for (i = 0; i < adev->sdma.num_instances; i++) { if (i == 0) snprintf(fw_name, sizeof(fw_name), "radeon/%s_sdma.bin", chip_name); else snprintf(fw_name, sizeof(fw_name), "radeon/%s_sdma1.bin", chip_name); - err = request_firmware(&adev->sdma[i].fw, fw_name, adev->dev); + err = request_firmware(&adev->sdma.instance[i].fw, fw_name, adev->dev); if (err) goto out; - err = amdgpu_ucode_validate(adev->sdma[i].fw); + err = amdgpu_ucode_validate(adev->sdma.instance[i].fw); } out: if (err) { printk(KERN_ERR "cik_sdma: Failed to load firmware \"%s\"\n", fw_name); - for (i = 0; i < SDMA_MAX_INSTANCE; i++) { - release_firmware(adev->sdma[i].fw); - adev->sdma[i].fw = NULL; + for (i = 0; i < adev->sdma.num_instances; i++) { + release_firmware(adev->sdma.instance[i].fw); + adev->sdma.instance[i].fw = NULL; } } return err; @@ -168,7 +168,7 @@ static uint32_t cik_sdma_ring_get_rptr(struct amdgpu_ring *ring) static uint32_t cik_sdma_ring_get_wptr(struct amdgpu_ring *ring) { struct amdgpu_device *adev = ring->adev; - u32 me = (ring == &adev->sdma[0].ring) ? 0 : 1; + u32 me = (ring == &adev->sdma.instance[0].ring) ? 0 : 1; return (RREG32(mmSDMA0_GFX_RB_WPTR + sdma_offsets[me]) & 0x3fffc) >> 2; } @@ -183,14 +183,14 @@ static uint32_t cik_sdma_ring_get_wptr(struct amdgpu_ring *ring) static void cik_sdma_ring_set_wptr(struct amdgpu_ring *ring) { struct amdgpu_device *adev = ring->adev; - u32 me = (ring == &adev->sdma[0].ring) ? 0 : 1; + u32 me = (ring == &adev->sdma.instance[0].ring) ? 0 : 1; WREG32(mmSDMA0_GFX_RB_WPTR + sdma_offsets[me], (ring->wptr << 2) & 0x3fffc); } static void cik_sdma_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count) { - struct amdgpu_sdma *sdma = amdgpu_get_sdma_instance(ring); + struct amdgpu_sdma_instance *sdma = amdgpu_get_sdma_instance(ring); int i; for (i = 0; i < count; i++) @@ -212,7 +212,7 @@ static void cik_sdma_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count) static void cik_sdma_ring_emit_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib) { - u32 extra_bits = (ib->vm ? ib->vm->ids[ring->idx].id : 0) & 0xf; + u32 extra_bits = ib->vm_id & 0xf; u32 next_rptr = ring->wptr + 5; while ((next_rptr & 7) != 4) @@ -248,7 +248,7 @@ static void cik_sdma_ring_emit_hdp_flush(struct amdgpu_ring *ring) SDMA_POLL_REG_MEM_EXTRA_FUNC(3)); /* == */ u32 ref_and_mask; - if (ring == &ring->adev->sdma[0].ring) + if (ring == &ring->adev->sdma.instance[0].ring) ref_and_mask = GPU_HDP_FLUSH_DONE__SDMA0_MASK; else ref_and_mask = GPU_HDP_FLUSH_DONE__SDMA1_MASK; @@ -261,6 +261,13 @@ static void cik_sdma_ring_emit_hdp_flush(struct amdgpu_ring *ring) amdgpu_ring_write(ring, (0xfff << 16) | 10); /* retry count, poll interval */ } +static void cik_sdma_ring_emit_hdp_invalidate(struct amdgpu_ring *ring) +{ + amdgpu_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_SRBM_WRITE, 0, 0xf000)); + amdgpu_ring_write(ring, mmHDP_DEBUG0); + amdgpu_ring_write(ring, 1); +} + /** * cik_sdma_ring_emit_fence - emit a fence on the DMA ring * @@ -295,30 +302,6 @@ static void cik_sdma_ring_emit_fence(struct amdgpu_ring *ring, u64 addr, u64 seq } /** - * cik_sdma_ring_emit_semaphore - emit a semaphore on the dma ring - * - * @ring: amdgpu_ring structure holding ring information - * @semaphore: amdgpu semaphore object - * @emit_wait: wait or signal semaphore - * - * Add a DMA semaphore packet to the ring wait on or signal - * other rings (CIK). - */ -static bool cik_sdma_ring_emit_semaphore(struct amdgpu_ring *ring, - struct amdgpu_semaphore *semaphore, - bool emit_wait) -{ - u64 addr = semaphore->gpu_addr; - u32 extra_bits = emit_wait ? 0 : SDMA_SEMAPHORE_EXTRA_S; - - amdgpu_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_SEMAPHORE, 0, extra_bits)); - amdgpu_ring_write(ring, addr & 0xfffffff8); - amdgpu_ring_write(ring, upper_32_bits(addr) & 0xffffffff); - - return true; -} - -/** * cik_sdma_gfx_stop - stop the gfx async dma engines * * @adev: amdgpu_device pointer @@ -327,8 +310,8 @@ static bool cik_sdma_ring_emit_semaphore(struct amdgpu_ring *ring, */ static void cik_sdma_gfx_stop(struct amdgpu_device *adev) { - struct amdgpu_ring *sdma0 = &adev->sdma[0].ring; - struct amdgpu_ring *sdma1 = &adev->sdma[1].ring; + struct amdgpu_ring *sdma0 = &adev->sdma.instance[0].ring; + struct amdgpu_ring *sdma1 = &adev->sdma.instance[1].ring; u32 rb_cntl; int i; @@ -336,7 +319,7 @@ static void cik_sdma_gfx_stop(struct amdgpu_device *adev) (adev->mman.buffer_funcs_ring == sdma1)) amdgpu_ttm_set_active_vram_size(adev, adev->mc.visible_vram_size); - for (i = 0; i < SDMA_MAX_INSTANCE; i++) { + for (i = 0; i < adev->sdma.num_instances; i++) { rb_cntl = RREG32(mmSDMA0_GFX_RB_CNTL + sdma_offsets[i]); rb_cntl &= ~SDMA0_GFX_RB_CNTL__RB_ENABLE_MASK; WREG32(mmSDMA0_GFX_RB_CNTL + sdma_offsets[i], rb_cntl); @@ -376,7 +359,7 @@ static void cik_sdma_enable(struct amdgpu_device *adev, bool enable) cik_sdma_rlc_stop(adev); } - for (i = 0; i < SDMA_MAX_INSTANCE; i++) { + for (i = 0; i < adev->sdma.num_instances; i++) { me_cntl = RREG32(mmSDMA0_F32_CNTL + sdma_offsets[i]); if (enable) me_cntl &= ~SDMA0_F32_CNTL__HALT_MASK; @@ -402,8 +385,8 @@ static int cik_sdma_gfx_resume(struct amdgpu_device *adev) u32 wb_offset; int i, j, r; - for (i = 0; i < SDMA_MAX_INSTANCE; i++) { - ring = &adev->sdma[i].ring; + for (i = 0; i < adev->sdma.num_instances; i++) { + ring = &adev->sdma.instance[i].ring; wb_offset = (ring->rptr_offs * 4); mutex_lock(&adev->srbm_mutex); @@ -417,6 +400,9 @@ static int cik_sdma_gfx_resume(struct amdgpu_device *adev) cik_srbm_select(adev, 0, 0, 0, 0); mutex_unlock(&adev->srbm_mutex); + WREG32(mmSDMA0_TILING_CONFIG + sdma_offsets[i], + adev->gfx.config.gb_addr_config & 0x70); + WREG32(mmSDMA0_SEM_INCOMPLETE_TIMER_CNTL + sdma_offsets[i], 0); WREG32(mmSDMA0_SEM_WAIT_FAIL_TIMER_CNTL + sdma_offsets[i], 0); @@ -502,26 +488,25 @@ static int cik_sdma_load_microcode(struct amdgpu_device *adev) u32 fw_size; int i, j; - if (!adev->sdma[0].fw || !adev->sdma[1].fw) - return -EINVAL; - /* halt the MEs */ cik_sdma_enable(adev, false); - for (i = 0; i < SDMA_MAX_INSTANCE; i++) { - hdr = (const struct sdma_firmware_header_v1_0 *)adev->sdma[i].fw->data; + for (i = 0; i < adev->sdma.num_instances; i++) { + if (!adev->sdma.instance[i].fw) + return -EINVAL; + hdr = (const struct sdma_firmware_header_v1_0 *)adev->sdma.instance[i].fw->data; amdgpu_ucode_print_sdma_hdr(&hdr->header); fw_size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4; - adev->sdma[i].fw_version = le32_to_cpu(hdr->header.ucode_version); - adev->sdma[i].feature_version = le32_to_cpu(hdr->ucode_feature_version); - if (adev->sdma[i].feature_version >= 20) - adev->sdma[i].burst_nop = true; + adev->sdma.instance[i].fw_version = le32_to_cpu(hdr->header.ucode_version); + adev->sdma.instance[i].feature_version = le32_to_cpu(hdr->ucode_feature_version); + if (adev->sdma.instance[i].feature_version >= 20) + adev->sdma.instance[i].burst_nop = true; fw_data = (const __le32 *) - (adev->sdma[i].fw->data + le32_to_cpu(hdr->header.ucode_array_offset_bytes)); + (adev->sdma.instance[i].fw->data + le32_to_cpu(hdr->header.ucode_array_offset_bytes)); WREG32(mmSDMA0_UCODE_ADDR + sdma_offsets[i], 0); for (j = 0; j < fw_size; j++) WREG32(mmSDMA0_UCODE_DATA + sdma_offsets[i], le32_to_cpup(fw_data++)); - WREG32(mmSDMA0_UCODE_ADDR + sdma_offsets[i], adev->sdma[i].fw_version); + WREG32(mmSDMA0_UCODE_ADDR + sdma_offsets[i], adev->sdma.instance[i].fw_version); } return 0; @@ -585,7 +570,7 @@ static int cik_sdma_ring_test_ring(struct amdgpu_ring *ring) tmp = 0xCAFEDEAD; adev->wb.wb[index] = cpu_to_le32(tmp); - r = amdgpu_ring_lock(ring, 5); + r = amdgpu_ring_alloc(ring, 5); if (r) { DRM_ERROR("amdgpu: dma failed to lock ring %d (%d).\n", ring->idx, r); amdgpu_wb_free(adev, index); @@ -596,7 +581,7 @@ static int cik_sdma_ring_test_ring(struct amdgpu_ring *ring) amdgpu_ring_write(ring, upper_32_bits(gpu_addr)); amdgpu_ring_write(ring, 1); /* number of DWs to follow */ amdgpu_ring_write(ring, 0xDEADBEEF); - amdgpu_ring_unlock_commit(ring); + amdgpu_ring_commit(ring); for (i = 0; i < adev->usec_timeout; i++) { tmp = le32_to_cpu(adev->wb.wb[index]); @@ -646,7 +631,7 @@ static int cik_sdma_ring_test_ib(struct amdgpu_ring *ring) tmp = 0xCAFEDEAD; adev->wb.wb[index] = cpu_to_le32(tmp); memset(&ib, 0, sizeof(ib)); - r = amdgpu_ib_get(ring, NULL, 256, &ib); + r = amdgpu_ib_get(adev, NULL, 256, &ib); if (r) { DRM_ERROR("amdgpu: failed to get ib (%d).\n", r); goto err0; @@ -658,9 +643,7 @@ static int cik_sdma_ring_test_ib(struct amdgpu_ring *ring) ib.ptr[3] = 1; ib.ptr[4] = 0xDEADBEEF; ib.length_dw = 5; - r = amdgpu_sched_ib_submit_kernel_helper(adev, ring, &ib, 1, NULL, - AMDGPU_FENCE_OWNER_UNDEFINED, - &f); + r = amdgpu_ib_schedule(ring, 1, &ib, NULL, &f); if (r) goto err1; @@ -686,7 +669,8 @@ static int cik_sdma_ring_test_ib(struct amdgpu_ring *ring) err1: fence_put(f); - amdgpu_ib_free(adev, &ib); + amdgpu_ib_free(adev, &ib, NULL); + fence_put(f); err0: amdgpu_wb_free(adev, index); return r; @@ -739,7 +723,7 @@ static void cik_sdma_vm_copy_pte(struct amdgpu_ib *ib, * Update PTEs by writing them manually using sDMA (CIK). */ static void cik_sdma_vm_write_pte(struct amdgpu_ib *ib, - uint64_t pe, + const dma_addr_t *pages_addr, uint64_t pe, uint64_t addr, unsigned count, uint32_t incr, uint32_t flags) { @@ -758,14 +742,7 @@ static void cik_sdma_vm_write_pte(struct amdgpu_ib *ib, ib->ptr[ib->length_dw++] = upper_32_bits(pe); ib->ptr[ib->length_dw++] = ndw; for (; ndw > 0; ndw -= 2, --count, pe += 8) { - if (flags & AMDGPU_PTE_SYSTEM) { - value = amdgpu_vm_map_gart(ib->ring->adev, addr); - value &= 0xFFFFFFFFFFFFF000ULL; - } else if (flags & AMDGPU_PTE_VALID) { - value = addr; - } else { - value = 0; - } + value = amdgpu_vm_map_gart(pages_addr, addr); addr += incr; value |= flags; ib->ptr[ib->length_dw++] = value; @@ -828,9 +805,9 @@ static void cik_sdma_vm_set_pte_pde(struct amdgpu_ib *ib, * @ib: indirect buffer to fill with padding * */ -static void cik_sdma_vm_pad_ib(struct amdgpu_ib *ib) +static void cik_sdma_ring_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib) { - struct amdgpu_sdma *sdma = amdgpu_get_sdma_instance(ib->ring); + struct amdgpu_sdma_instance *sdma = amdgpu_get_sdma_instance(ring); u32 pad_count; int i; @@ -846,6 +823,30 @@ static void cik_sdma_vm_pad_ib(struct amdgpu_ib *ib) } /** + * cik_sdma_ring_emit_pipeline_sync - sync the pipeline + * + * @ring: amdgpu_ring pointer + * + * Make sure all previous operations are completed (CIK). + */ +static void cik_sdma_ring_emit_pipeline_sync(struct amdgpu_ring *ring) +{ + uint32_t seq = ring->fence_drv.sync_seq; + uint64_t addr = ring->fence_drv.gpu_addr; + + /* wait for idle */ + amdgpu_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_POLL_REG_MEM, 0, + SDMA_POLL_REG_MEM_EXTRA_OP(0) | + SDMA_POLL_REG_MEM_EXTRA_FUNC(3) | /* equal */ + SDMA_POLL_REG_MEM_EXTRA_M)); + amdgpu_ring_write(ring, addr & 0xfffffffc); + amdgpu_ring_write(ring, upper_32_bits(addr) & 0xffffffff); + amdgpu_ring_write(ring, seq); /* reference */ + amdgpu_ring_write(ring, 0xfffffff); /* mask */ + amdgpu_ring_write(ring, (0xfff << 16) | 4); /* retry count, poll interval */ +} + +/** * cik_sdma_ring_emit_vm_flush - cik vm flush using sDMA * * @ring: amdgpu_ring pointer @@ -886,7 +887,7 @@ static void cik_enable_sdma_mgcg(struct amdgpu_device *adev, { u32 orig, data; - if (enable && (adev->cg_flags & AMDGPU_CG_SUPPORT_SDMA_MGCG)) { + if (enable && (adev->cg_flags & AMD_CG_SUPPORT_SDMA_MGCG)) { WREG32(mmSDMA0_CLK_CTRL + SDMA0_REGISTER_OFFSET, 0x00000100); WREG32(mmSDMA0_CLK_CTRL + SDMA1_REGISTER_OFFSET, 0x00000100); } else { @@ -907,7 +908,7 @@ static void cik_enable_sdma_mgls(struct amdgpu_device *adev, { u32 orig, data; - if (enable && (adev->cg_flags & AMDGPU_CG_SUPPORT_SDMA_LS)) { + if (enable && (adev->cg_flags & AMD_CG_SUPPORT_SDMA_LS)) { orig = data = RREG32(mmSDMA0_POWER_CNTL + SDMA0_REGISTER_OFFSET); data |= 0x100; if (orig != data) @@ -934,6 +935,8 @@ static int cik_sdma_early_init(void *handle) { struct amdgpu_device *adev = (struct amdgpu_device *)handle; + adev->sdma.num_instances = SDMA_MAX_INSTANCE; + cik_sdma_set_ring_funcs(adev); cik_sdma_set_irq_funcs(adev); cik_sdma_set_buffer_funcs(adev); @@ -946,7 +949,7 @@ static int cik_sdma_sw_init(void *handle) { struct amdgpu_ring *ring; struct amdgpu_device *adev = (struct amdgpu_device *)handle; - int r; + int r, i; r = cik_sdma_init_microcode(adev); if (r) { @@ -955,43 +958,33 @@ static int cik_sdma_sw_init(void *handle) } /* SDMA trap event */ - r = amdgpu_irq_add_id(adev, 224, &adev->sdma_trap_irq); + r = amdgpu_irq_add_id(adev, 224, &adev->sdma.trap_irq); if (r) return r; /* SDMA Privileged inst */ - r = amdgpu_irq_add_id(adev, 241, &adev->sdma_illegal_inst_irq); + r = amdgpu_irq_add_id(adev, 241, &adev->sdma.illegal_inst_irq); if (r) return r; /* SDMA Privileged inst */ - r = amdgpu_irq_add_id(adev, 247, &adev->sdma_illegal_inst_irq); - if (r) - return r; - - ring = &adev->sdma[0].ring; - ring->ring_obj = NULL; - - ring = &adev->sdma[1].ring; - ring->ring_obj = NULL; - - ring = &adev->sdma[0].ring; - sprintf(ring->name, "sdma0"); - r = amdgpu_ring_init(adev, ring, 256 * 1024, - SDMA_PACKET(SDMA_OPCODE_NOP, 0, 0), 0xf, - &adev->sdma_trap_irq, AMDGPU_SDMA_IRQ_TRAP0, - AMDGPU_RING_TYPE_SDMA); + r = amdgpu_irq_add_id(adev, 247, &adev->sdma.illegal_inst_irq); if (r) return r; - ring = &adev->sdma[1].ring; - sprintf(ring->name, "sdma1"); - r = amdgpu_ring_init(adev, ring, 256 * 1024, - SDMA_PACKET(SDMA_OPCODE_NOP, 0, 0), 0xf, - &adev->sdma_trap_irq, AMDGPU_SDMA_IRQ_TRAP1, - AMDGPU_RING_TYPE_SDMA); - if (r) - return r; + for (i = 0; i < adev->sdma.num_instances; i++) { + ring = &adev->sdma.instance[i].ring; + ring->ring_obj = NULL; + sprintf(ring->name, "sdma%d", i); + r = amdgpu_ring_init(adev, ring, 256 * 1024, + SDMA_PACKET(SDMA_OPCODE_NOP, 0, 0), 0xf, + &adev->sdma.trap_irq, + (i == 0) ? + AMDGPU_SDMA_IRQ_TRAP0 : AMDGPU_SDMA_IRQ_TRAP1, + AMDGPU_RING_TYPE_SDMA); + if (r) + return r; + } return r; } @@ -999,9 +992,10 @@ static int cik_sdma_sw_init(void *handle) static int cik_sdma_sw_fini(void *handle) { struct amdgpu_device *adev = (struct amdgpu_device *)handle; + int i; - amdgpu_ring_fini(&adev->sdma[0].ring); - amdgpu_ring_fini(&adev->sdma[1].ring); + for (i = 0; i < adev->sdma.num_instances; i++) + amdgpu_ring_fini(&adev->sdma.instance[i].ring); return 0; } @@ -1078,7 +1072,7 @@ static void cik_sdma_print_status(void *handle) dev_info(adev->dev, "CIK SDMA registers\n"); dev_info(adev->dev, " SRBM_STATUS2=0x%08X\n", RREG32(mmSRBM_STATUS2)); - for (i = 0; i < SDMA_MAX_INSTANCE; i++) { + for (i = 0; i < adev->sdma.num_instances; i++) { dev_info(adev->dev, " SDMA%d_STATUS_REG=0x%08X\n", i, RREG32(mmSDMA0_STATUS_REG + sdma_offsets[i])); dev_info(adev->dev, " SDMA%d_ME_CNTL=0x%08X\n", @@ -1105,6 +1099,8 @@ static void cik_sdma_print_status(void *handle) i, RREG32(mmSDMA0_GFX_RB_BASE + sdma_offsets[i])); dev_info(adev->dev, " SDMA%d_GFX_RB_BASE_HI=0x%08X\n", i, RREG32(mmSDMA0_GFX_RB_BASE_HI + sdma_offsets[i])); + dev_info(adev->dev, " SDMA%d_TILING_CONFIG=0x%08X\n", + i, RREG32(mmSDMA0_TILING_CONFIG + sdma_offsets[i])); mutex_lock(&adev->srbm_mutex); for (j = 0; j < 16; j++) { cik_srbm_select(adev, 0, 0, 0, j); @@ -1223,7 +1219,7 @@ static int cik_sdma_process_trap_irq(struct amdgpu_device *adev, case 0: switch (queue_id) { case 0: - amdgpu_fence_process(&adev->sdma[0].ring); + amdgpu_fence_process(&adev->sdma.instance[0].ring); break; case 1: /* XXX compute */ @@ -1236,7 +1232,7 @@ static int cik_sdma_process_trap_irq(struct amdgpu_device *adev, case 1: switch (queue_id) { case 0: - amdgpu_fence_process(&adev->sdma[1].ring); + amdgpu_fence_process(&adev->sdma.instance[1].ring); break; case 1: /* XXX compute */ @@ -1298,24 +1294,6 @@ const struct amd_ip_funcs cik_sdma_ip_funcs = { .set_powergating_state = cik_sdma_set_powergating_state, }; -/** - * cik_sdma_ring_is_lockup - Check if the DMA engine is locked up - * - * @ring: amdgpu_ring structure holding ring information - * - * Check if the async DMA engine is locked up (CIK). - * Returns true if the engine appears to be locked up, false if not. - */ -static bool cik_sdma_ring_is_lockup(struct amdgpu_ring *ring) -{ - - if (cik_sdma_is_idle(ring->adev)) { - amdgpu_ring_lockup_update(ring); - return false; - } - return amdgpu_ring_test_lockup(ring); -} - static const struct amdgpu_ring_funcs cik_sdma_ring_funcs = { .get_rptr = cik_sdma_ring_get_rptr, .get_wptr = cik_sdma_ring_get_wptr, @@ -1323,19 +1301,22 @@ static const struct amdgpu_ring_funcs cik_sdma_ring_funcs = { .parse_cs = NULL, .emit_ib = cik_sdma_ring_emit_ib, .emit_fence = cik_sdma_ring_emit_fence, - .emit_semaphore = cik_sdma_ring_emit_semaphore, + .emit_pipeline_sync = cik_sdma_ring_emit_pipeline_sync, .emit_vm_flush = cik_sdma_ring_emit_vm_flush, .emit_hdp_flush = cik_sdma_ring_emit_hdp_flush, + .emit_hdp_invalidate = cik_sdma_ring_emit_hdp_invalidate, .test_ring = cik_sdma_ring_test_ring, .test_ib = cik_sdma_ring_test_ib, - .is_lockup = cik_sdma_ring_is_lockup, .insert_nop = cik_sdma_ring_insert_nop, + .pad_ib = cik_sdma_ring_pad_ib, }; static void cik_sdma_set_ring_funcs(struct amdgpu_device *adev) { - adev->sdma[0].ring.funcs = &cik_sdma_ring_funcs; - adev->sdma[1].ring.funcs = &cik_sdma_ring_funcs; + int i; + + for (i = 0; i < adev->sdma.num_instances; i++) + adev->sdma.instance[i].ring.funcs = &cik_sdma_ring_funcs; } static const struct amdgpu_irq_src_funcs cik_sdma_trap_irq_funcs = { @@ -1349,9 +1330,9 @@ static const struct amdgpu_irq_src_funcs cik_sdma_illegal_inst_irq_funcs = { static void cik_sdma_set_irq_funcs(struct amdgpu_device *adev) { - adev->sdma_trap_irq.num_types = AMDGPU_SDMA_IRQ_LAST; - adev->sdma_trap_irq.funcs = &cik_sdma_trap_irq_funcs; - adev->sdma_illegal_inst_irq.funcs = &cik_sdma_illegal_inst_irq_funcs; + adev->sdma.trap_irq.num_types = AMDGPU_SDMA_IRQ_LAST; + adev->sdma.trap_irq.funcs = &cik_sdma_trap_irq_funcs; + adev->sdma.illegal_inst_irq.funcs = &cik_sdma_illegal_inst_irq_funcs; } /** @@ -1416,7 +1397,7 @@ static void cik_sdma_set_buffer_funcs(struct amdgpu_device *adev) { if (adev->mman.buffer_funcs == NULL) { adev->mman.buffer_funcs = &cik_sdma_buffer_funcs; - adev->mman.buffer_funcs_ring = &adev->sdma[0].ring; + adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].ring; } } @@ -1424,14 +1405,18 @@ static const struct amdgpu_vm_pte_funcs cik_sdma_vm_pte_funcs = { .copy_pte = cik_sdma_vm_copy_pte, .write_pte = cik_sdma_vm_write_pte, .set_pte_pde = cik_sdma_vm_set_pte_pde, - .pad_ib = cik_sdma_vm_pad_ib, }; static void cik_sdma_set_vm_pte_funcs(struct amdgpu_device *adev) { + unsigned i; + if (adev->vm_manager.vm_pte_funcs == NULL) { adev->vm_manager.vm_pte_funcs = &cik_sdma_vm_pte_funcs; - adev->vm_manager.vm_pte_funcs_ring = &adev->sdma[0].ring; - adev->vm_manager.vm_pte_funcs_ring->is_pte_ring = true; + for (i = 0; i < adev->sdma.num_instances; i++) + adev->vm_manager.vm_pte_rings[i] = + &adev->sdma.instance[i].ring; + + adev->vm_manager.vm_pte_num_rings = adev->sdma.num_instances; } } diff --git a/drivers/gpu/drm/amd/amdgpu/cikd.h b/drivers/gpu/drm/amd/amdgpu/cikd.h index 7f6d457f250a..60d4493206dd 100644 --- a/drivers/gpu/drm/amd/amdgpu/cikd.h +++ b/drivers/gpu/drm/amd/amdgpu/cikd.h @@ -46,9 +46,6 @@ #define BONAIRE_GB_ADDR_CONFIG_GOLDEN 0x12010001 #define HAWAII_GB_ADDR_CONFIG_GOLDEN 0x12011003 -#define CIK_RB_BITMAP_WIDTH_PER_SH 2 -#define HAWAII_RB_BITMAP_WIDTH_PER_SH 4 - #define AMDGPU_NUM_OF_VMIDS 8 #define PIPEID(x) ((x) << 0) diff --git a/drivers/gpu/drm/amd/amdgpu/cz_dpm.c b/drivers/gpu/drm/amd/amdgpu/cz_dpm.c index 2e3373ed4c94..e7ef2261ff4a 100644 --- a/drivers/gpu/drm/amd/amdgpu/cz_dpm.c +++ b/drivers/gpu/drm/amd/amdgpu/cz_dpm.c @@ -445,13 +445,13 @@ static int cz_dpm_init(struct amdgpu_device *adev) pi->gfx_pg_threshold = 500; pi->caps_fps = true; /* uvd */ - pi->caps_uvd_pg = (adev->pg_flags & AMDGPU_PG_SUPPORT_UVD) ? true : false; + pi->caps_uvd_pg = (adev->pg_flags & AMD_PG_SUPPORT_UVD) ? true : false; pi->caps_uvd_dpm = true; /* vce */ - pi->caps_vce_pg = (adev->pg_flags & AMDGPU_PG_SUPPORT_VCE) ? true : false; + pi->caps_vce_pg = (adev->pg_flags & AMD_PG_SUPPORT_VCE) ? true : false; pi->caps_vce_dpm = true; /* acp */ - pi->caps_acp_pg = (adev->pg_flags & AMDGPU_PG_SUPPORT_ACP) ? true : false; + pi->caps_acp_pg = (adev->pg_flags & AMD_PG_SUPPORT_ACP) ? true : false; pi->caps_acp_dpm = true; pi->caps_stable_power_state = false; @@ -1078,6 +1078,37 @@ static uint32_t cz_get_eclk_level(struct amdgpu_device *adev, return i; } +static uint32_t cz_get_uvd_level(struct amdgpu_device *adev, + uint32_t clock, uint16_t msg) +{ + int i = 0; + struct amdgpu_uvd_clock_voltage_dependency_table *table = + &adev->pm.dpm.dyn_state.uvd_clock_voltage_dependency_table; + + switch (msg) { + case PPSMC_MSG_SetUvdSoftMin: + case PPSMC_MSG_SetUvdHardMin: + for (i = 0; i < table->count; i++) + if (clock <= table->entries[i].vclk) + break; + if (i == table->count) + i = table->count - 1; + break; + case PPSMC_MSG_SetUvdSoftMax: + case PPSMC_MSG_SetUvdHardMax: + for (i = table->count - 1; i >= 0; i--) + if (clock >= table->entries[i].vclk) + break; + if (i < 0) + i = 0; + break; + default: + break; + } + + return i; +} + static int cz_program_bootup_state(struct amdgpu_device *adev) { struct cz_power_info *pi = cz_get_pi(adev); @@ -1264,6 +1295,7 @@ static void cz_apply_state_adjust_rules(struct amdgpu_device *adev, static int cz_dpm_enable(struct amdgpu_device *adev) { + const char *chip_name; int ret = 0; /* renable will hang up SMU, so check first */ @@ -1272,21 +1304,33 @@ static int cz_dpm_enable(struct amdgpu_device *adev) cz_program_voting_clients(adev); + switch (adev->asic_type) { + case CHIP_CARRIZO: + chip_name = "carrizo"; + break; + case CHIP_STONEY: + chip_name = "stoney"; + break; + default: + BUG(); + } + + ret = cz_start_dpm(adev); if (ret) { - DRM_ERROR("Carrizo DPM enable failed\n"); + DRM_ERROR("%s DPM enable failed\n", chip_name); return -EINVAL; } ret = cz_program_bootup_state(adev); if (ret) { - DRM_ERROR("Carrizo bootup state program failed\n"); + DRM_ERROR("%s bootup state program failed\n", chip_name); return -EINVAL; } ret = cz_enable_didt(adev, true); if (ret) { - DRM_ERROR("Carrizo enable di/dt failed\n"); + DRM_ERROR("%s enable di/dt failed\n", chip_name); return -EINVAL; } @@ -1353,7 +1397,7 @@ static int cz_dpm_disable(struct amdgpu_device *adev) ret = cz_enable_didt(adev, false); if (ret) { - DRM_ERROR("Carrizo disable di/dt failed\n"); + DRM_ERROR("disable di/dt failed\n"); return -EINVAL; } @@ -1726,6 +1770,200 @@ static int cz_dpm_unforce_dpm_levels(struct amdgpu_device *adev) return 0; } +static int cz_dpm_uvd_force_highest(struct amdgpu_device *adev) +{ + struct cz_power_info *pi = cz_get_pi(adev); + int ret = 0; + + if (pi->uvd_dpm.soft_min_clk != pi->uvd_dpm.soft_max_clk) { + pi->uvd_dpm.soft_min_clk = + pi->uvd_dpm.soft_max_clk; + ret = cz_send_msg_to_smc_with_parameter(adev, + PPSMC_MSG_SetUvdSoftMin, + cz_get_uvd_level(adev, + pi->uvd_dpm.soft_min_clk, + PPSMC_MSG_SetUvdSoftMin)); + if (ret) + return ret; + } + + return ret; +} + +static int cz_dpm_uvd_force_lowest(struct amdgpu_device *adev) +{ + struct cz_power_info *pi = cz_get_pi(adev); + int ret = 0; + + if (pi->uvd_dpm.soft_max_clk != pi->uvd_dpm.soft_min_clk) { + pi->uvd_dpm.soft_max_clk = pi->uvd_dpm.soft_min_clk; + ret = cz_send_msg_to_smc_with_parameter(adev, + PPSMC_MSG_SetUvdSoftMax, + cz_get_uvd_level(adev, + pi->uvd_dpm.soft_max_clk, + PPSMC_MSG_SetUvdSoftMax)); + if (ret) + return ret; + } + + return ret; +} + +static uint32_t cz_dpm_get_max_uvd_level(struct amdgpu_device *adev) +{ + struct cz_power_info *pi = cz_get_pi(adev); + + if (!pi->max_uvd_level) { + cz_send_msg_to_smc(adev, PPSMC_MSG_GetMaxUvdLevel); + pi->max_uvd_level = cz_get_argument(adev) + 1; + } + + if (pi->max_uvd_level > CZ_MAX_HARDWARE_POWERLEVELS) { + DRM_ERROR("Invalid max uvd level!\n"); + return -EINVAL; + } + + return pi->max_uvd_level; +} + +static int cz_dpm_unforce_uvd_dpm_levels(struct amdgpu_device *adev) +{ + struct cz_power_info *pi = cz_get_pi(adev); + struct amdgpu_uvd_clock_voltage_dependency_table *dep_table = + &adev->pm.dpm.dyn_state.uvd_clock_voltage_dependency_table; + uint32_t level = 0; + int ret = 0; + + pi->uvd_dpm.soft_min_clk = dep_table->entries[0].vclk; + level = cz_dpm_get_max_uvd_level(adev) - 1; + if (level < dep_table->count) + pi->uvd_dpm.soft_max_clk = dep_table->entries[level].vclk; + else + pi->uvd_dpm.soft_max_clk = + dep_table->entries[dep_table->count - 1].vclk; + + /* get min/max sclk soft value + * notify SMU to execute */ + ret = cz_send_msg_to_smc_with_parameter(adev, + PPSMC_MSG_SetUvdSoftMin, + cz_get_uvd_level(adev, + pi->uvd_dpm.soft_min_clk, + PPSMC_MSG_SetUvdSoftMin)); + if (ret) + return ret; + + ret = cz_send_msg_to_smc_with_parameter(adev, + PPSMC_MSG_SetUvdSoftMax, + cz_get_uvd_level(adev, + pi->uvd_dpm.soft_max_clk, + PPSMC_MSG_SetUvdSoftMax)); + if (ret) + return ret; + + DRM_DEBUG("DPM uvd unforce state min=%d, max=%d.\n", + pi->uvd_dpm.soft_min_clk, + pi->uvd_dpm.soft_max_clk); + + return 0; +} + +static int cz_dpm_vce_force_highest(struct amdgpu_device *adev) +{ + struct cz_power_info *pi = cz_get_pi(adev); + int ret = 0; + + if (pi->vce_dpm.soft_min_clk != pi->vce_dpm.soft_max_clk) { + pi->vce_dpm.soft_min_clk = + pi->vce_dpm.soft_max_clk; + ret = cz_send_msg_to_smc_with_parameter(adev, + PPSMC_MSG_SetEclkSoftMin, + cz_get_eclk_level(adev, + pi->vce_dpm.soft_min_clk, + PPSMC_MSG_SetEclkSoftMin)); + if (ret) + return ret; + } + + return ret; +} + +static int cz_dpm_vce_force_lowest(struct amdgpu_device *adev) +{ + struct cz_power_info *pi = cz_get_pi(adev); + int ret = 0; + + if (pi->vce_dpm.soft_max_clk != pi->vce_dpm.soft_min_clk) { + pi->vce_dpm.soft_max_clk = pi->vce_dpm.soft_min_clk; + ret = cz_send_msg_to_smc_with_parameter(adev, + PPSMC_MSG_SetEclkSoftMax, + cz_get_uvd_level(adev, + pi->vce_dpm.soft_max_clk, + PPSMC_MSG_SetEclkSoftMax)); + if (ret) + return ret; + } + + return ret; +} + +static uint32_t cz_dpm_get_max_vce_level(struct amdgpu_device *adev) +{ + struct cz_power_info *pi = cz_get_pi(adev); + + if (!pi->max_vce_level) { + cz_send_msg_to_smc(adev, PPSMC_MSG_GetMaxEclkLevel); + pi->max_vce_level = cz_get_argument(adev) + 1; + } + + if (pi->max_vce_level > CZ_MAX_HARDWARE_POWERLEVELS) { + DRM_ERROR("Invalid max vce level!\n"); + return -EINVAL; + } + + return pi->max_vce_level; +} + +static int cz_dpm_unforce_vce_dpm_levels(struct amdgpu_device *adev) +{ + struct cz_power_info *pi = cz_get_pi(adev); + struct amdgpu_vce_clock_voltage_dependency_table *dep_table = + &adev->pm.dpm.dyn_state.vce_clock_voltage_dependency_table; + uint32_t level = 0; + int ret = 0; + + pi->vce_dpm.soft_min_clk = dep_table->entries[0].ecclk; + level = cz_dpm_get_max_vce_level(adev) - 1; + if (level < dep_table->count) + pi->vce_dpm.soft_max_clk = dep_table->entries[level].ecclk; + else + pi->vce_dpm.soft_max_clk = + dep_table->entries[dep_table->count - 1].ecclk; + + /* get min/max sclk soft value + * notify SMU to execute */ + ret = cz_send_msg_to_smc_with_parameter(adev, + PPSMC_MSG_SetEclkSoftMin, + cz_get_eclk_level(adev, + pi->vce_dpm.soft_min_clk, + PPSMC_MSG_SetEclkSoftMin)); + if (ret) + return ret; + + ret = cz_send_msg_to_smc_with_parameter(adev, + PPSMC_MSG_SetEclkSoftMax, + cz_get_eclk_level(adev, + pi->vce_dpm.soft_max_clk, + PPSMC_MSG_SetEclkSoftMax)); + if (ret) + return ret; + + DRM_DEBUG("DPM vce unforce state min=%d, max=%d.\n", + pi->vce_dpm.soft_min_clk, + pi->vce_dpm.soft_max_clk); + + return 0; +} + static int cz_dpm_force_dpm_level(struct amdgpu_device *adev, enum amdgpu_dpm_forced_level level) { @@ -1733,25 +1971,70 @@ static int cz_dpm_force_dpm_level(struct amdgpu_device *adev, switch (level) { case AMDGPU_DPM_FORCED_LEVEL_HIGH: + /* sclk */ ret = cz_dpm_unforce_dpm_levels(adev); if (ret) return ret; ret = cz_dpm_force_highest(adev); if (ret) return ret; + + /* uvd */ + ret = cz_dpm_unforce_uvd_dpm_levels(adev); + if (ret) + return ret; + ret = cz_dpm_uvd_force_highest(adev); + if (ret) + return ret; + + /* vce */ + ret = cz_dpm_unforce_vce_dpm_levels(adev); + if (ret) + return ret; + ret = cz_dpm_vce_force_highest(adev); + if (ret) + return ret; break; case AMDGPU_DPM_FORCED_LEVEL_LOW: + /* sclk */ ret = cz_dpm_unforce_dpm_levels(adev); if (ret) return ret; ret = cz_dpm_force_lowest(adev); if (ret) return ret; + + /* uvd */ + ret = cz_dpm_unforce_uvd_dpm_levels(adev); + if (ret) + return ret; + ret = cz_dpm_uvd_force_lowest(adev); + if (ret) + return ret; + + /* vce */ + ret = cz_dpm_unforce_vce_dpm_levels(adev); + if (ret) + return ret; + ret = cz_dpm_vce_force_lowest(adev); + if (ret) + return ret; break; case AMDGPU_DPM_FORCED_LEVEL_AUTO: + /* sclk */ ret = cz_dpm_unforce_dpm_levels(adev); if (ret) return ret; + + /* uvd */ + ret = cz_dpm_unforce_uvd_dpm_levels(adev); + if (ret) + return ret; + + /* vce */ + ret = cz_dpm_unforce_vce_dpm_levels(adev); + if (ret) + return ret; break; default: break; @@ -1892,7 +2175,8 @@ static int cz_update_vce_dpm(struct amdgpu_device *adev) pi->vce_dpm.hard_min_clk = table->entries[table->count-1].ecclk; } else { /* non-stable p-state cases. without vce.Arbiter.EcclkHardMin */ - pi->vce_dpm.hard_min_clk = table->entries[0].ecclk; + /* leave it as set by user */ + /*pi->vce_dpm.hard_min_clk = table->entries[0].ecclk;*/ } cz_send_msg_to_smc_with_parameter(adev, @@ -1918,8 +2202,7 @@ static void cz_dpm_powergate_vce(struct amdgpu_device *adev, bool gate) AMD_PG_STATE_GATE); cz_enable_vce_dpm(adev, false); - /* TODO: to figure out why vce can't be poweroff. */ - /* cz_send_msg_to_smc(adev, PPSMC_MSG_VCEPowerOFF); */ + cz_send_msg_to_smc(adev, PPSMC_MSG_VCEPowerOFF); pi->vce_power_gated = true; } else { cz_send_msg_to_smc(adev, PPSMC_MSG_VCEPowerON); @@ -1942,10 +2225,8 @@ static void cz_dpm_powergate_vce(struct amdgpu_device *adev, bool gate) } } else { /*pi->caps_vce_pg*/ cz_update_vce_dpm(adev); - cz_enable_vce_dpm(adev, true); + cz_enable_vce_dpm(adev, !gate); } - - return; } const struct amd_ip_funcs cz_dpm_ip_funcs = { diff --git a/drivers/gpu/drm/amd/amdgpu/cz_dpm.h b/drivers/gpu/drm/amd/amdgpu/cz_dpm.h index 99e1afc89629..5df8c1faab51 100644 --- a/drivers/gpu/drm/amd/amdgpu/cz_dpm.h +++ b/drivers/gpu/drm/amd/amdgpu/cz_dpm.h @@ -183,6 +183,8 @@ struct cz_power_info { uint32_t voltage_drop_threshold; uint32_t gfx_pg_threshold; uint32_t max_sclk_level; + uint32_t max_uvd_level; + uint32_t max_vce_level; /* flags */ bool didt_enabled; bool video_start; diff --git a/drivers/gpu/drm/amd/amdgpu/cz_ih.c b/drivers/gpu/drm/amd/amdgpu/cz_ih.c index bc751bfbcae2..c79638f8e732 100644 --- a/drivers/gpu/drm/amd/amdgpu/cz_ih.c +++ b/drivers/gpu/drm/amd/amdgpu/cz_ih.c @@ -253,8 +253,14 @@ static void cz_ih_set_rptr(struct amdgpu_device *adev) static int cz_ih_early_init(void *handle) { struct amdgpu_device *adev = (struct amdgpu_device *)handle; + int ret; + + ret = amdgpu_irq_add_domain(adev); + if (ret) + return ret; cz_ih_set_interrupt_funcs(adev); + return 0; } @@ -278,6 +284,7 @@ static int cz_ih_sw_fini(void *handle) amdgpu_irq_fini(adev); amdgpu_ih_ring_fini(adev); + amdgpu_irq_remove_domain(adev); return 0; } diff --git a/drivers/gpu/drm/amd/amdgpu/cz_ppsmc.h b/drivers/gpu/drm/amd/amdgpu/cz_ppsmc.h deleted file mode 100644 index 273616ab43db..000000000000 --- a/drivers/gpu/drm/amd/amdgpu/cz_ppsmc.h +++ /dev/null @@ -1,185 +0,0 @@ -/* - * Copyright 2014 Advanced Micro Devices, Inc. - * - * Permission is hereby granted, free of charge, to any person obtaining a - * copy of this software and associated documentation files (the "Software"), - * to deal in the Software without restriction, including without limitation - * the rights to use, copy, modify, merge, publish, distribute, sublicense, - * and/or sell copies of the Software, and to permit persons to whom the - * Software is furnished to do so, subject to the following conditions: - * - * The above copyright notice and this permission notice shall be included in - * all copies or substantial portions of the Software. - * - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR - * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, - * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL - * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR - * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, - * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR - * OTHER DEALINGS IN THE SOFTWARE. - * - */ - -#ifndef CZ_PP_SMC_H -#define CZ_PP_SMC_H - -#pragma pack(push, 1) - -/* Fan control algorithm:*/ -#define FDO_MODE_HARDWARE 0 -#define FDO_MODE_PIECE_WISE_LINEAR 1 - -enum FAN_CONTROL { - FAN_CONTROL_FUZZY, - FAN_CONTROL_TABLE -}; - -enum DPM_ARRAY { - DPM_ARRAY_HARD_MAX, - DPM_ARRAY_HARD_MIN, - DPM_ARRAY_SOFT_MAX, - DPM_ARRAY_SOFT_MIN -}; - -/* - * Return codes for driver to SMC communication. - * Leave these #define-s, enums might not be exactly 8-bits on the microcontroller. - */ -#define PPSMC_Result_OK ((uint16_t)0x01) -#define PPSMC_Result_NoMore ((uint16_t)0x02) -#define PPSMC_Result_NotNow ((uint16_t)0x03) -#define PPSMC_Result_Failed ((uint16_t)0xFF) -#define PPSMC_Result_UnknownCmd ((uint16_t)0xFE) -#define PPSMC_Result_UnknownVT ((uint16_t)0xFD) - -#define PPSMC_isERROR(x) ((uint16_t)0x80 & (x)) - -/* - * Supported driver messages - */ -#define PPSMC_MSG_Test ((uint16_t) 0x1) -#define PPSMC_MSG_GetFeatureStatus ((uint16_t) 0x2) -#define PPSMC_MSG_EnableAllSmuFeatures ((uint16_t) 0x3) -#define PPSMC_MSG_DisableAllSmuFeatures ((uint16_t) 0x4) -#define PPSMC_MSG_OptimizeBattery ((uint16_t) 0x5) -#define PPSMC_MSG_MaximizePerf ((uint16_t) 0x6) -#define PPSMC_MSG_UVDPowerOFF ((uint16_t) 0x7) -#define PPSMC_MSG_UVDPowerON ((uint16_t) 0x8) -#define PPSMC_MSG_VCEPowerOFF ((uint16_t) 0x9) -#define PPSMC_MSG_VCEPowerON ((uint16_t) 0xA) -#define PPSMC_MSG_ACPPowerOFF ((uint16_t) 0xB) -#define PPSMC_MSG_ACPPowerON ((uint16_t) 0xC) -#define PPSMC_MSG_SDMAPowerOFF ((uint16_t) 0xD) -#define PPSMC_MSG_SDMAPowerON ((uint16_t) 0xE) -#define PPSMC_MSG_XDMAPowerOFF ((uint16_t) 0xF) -#define PPSMC_MSG_XDMAPowerON ((uint16_t) 0x10) -#define PPSMC_MSG_SetMinDeepSleepSclk ((uint16_t) 0x11) -#define PPSMC_MSG_SetSclkSoftMin ((uint16_t) 0x12) -#define PPSMC_MSG_SetSclkSoftMax ((uint16_t) 0x13) -#define PPSMC_MSG_SetSclkHardMin ((uint16_t) 0x14) -#define PPSMC_MSG_SetSclkHardMax ((uint16_t) 0x15) -#define PPSMC_MSG_SetLclkSoftMin ((uint16_t) 0x16) -#define PPSMC_MSG_SetLclkSoftMax ((uint16_t) 0x17) -#define PPSMC_MSG_SetLclkHardMin ((uint16_t) 0x18) -#define PPSMC_MSG_SetLclkHardMax ((uint16_t) 0x19) -#define PPSMC_MSG_SetUvdSoftMin ((uint16_t) 0x1A) -#define PPSMC_MSG_SetUvdSoftMax ((uint16_t) 0x1B) -#define PPSMC_MSG_SetUvdHardMin ((uint16_t) 0x1C) -#define PPSMC_MSG_SetUvdHardMax ((uint16_t) 0x1D) -#define PPSMC_MSG_SetEclkSoftMin ((uint16_t) 0x1E) -#define PPSMC_MSG_SetEclkSoftMax ((uint16_t) 0x1F) -#define PPSMC_MSG_SetEclkHardMin ((uint16_t) 0x20) -#define PPSMC_MSG_SetEclkHardMax ((uint16_t) 0x21) -#define PPSMC_MSG_SetAclkSoftMin ((uint16_t) 0x22) -#define PPSMC_MSG_SetAclkSoftMax ((uint16_t) 0x23) -#define PPSMC_MSG_SetAclkHardMin ((uint16_t) 0x24) -#define PPSMC_MSG_SetAclkHardMax ((uint16_t) 0x25) -#define PPSMC_MSG_SetNclkSoftMin ((uint16_t) 0x26) -#define PPSMC_MSG_SetNclkSoftMax ((uint16_t) 0x27) -#define PPSMC_MSG_SetNclkHardMin ((uint16_t) 0x28) -#define PPSMC_MSG_SetNclkHardMax ((uint16_t) 0x29) -#define PPSMC_MSG_SetPstateSoftMin ((uint16_t) 0x2A) -#define PPSMC_MSG_SetPstateSoftMax ((uint16_t) 0x2B) -#define PPSMC_MSG_SetPstateHardMin ((uint16_t) 0x2C) -#define PPSMC_MSG_SetPstateHardMax ((uint16_t) 0x2D) -#define PPSMC_MSG_DisableLowMemoryPstate ((uint16_t) 0x2E) -#define PPSMC_MSG_EnableLowMemoryPstate ((uint16_t) 0x2F) -#define PPSMC_MSG_UcodeAddressLow ((uint16_t) 0x30) -#define PPSMC_MSG_UcodeAddressHigh ((uint16_t) 0x31) -#define PPSMC_MSG_UcodeLoadStatus ((uint16_t) 0x32) -#define PPSMC_MSG_DriverDramAddrHi ((uint16_t) 0x33) -#define PPSMC_MSG_DriverDramAddrLo ((uint16_t) 0x34) -#define PPSMC_MSG_CondExecDramAddrHi ((uint16_t) 0x35) -#define PPSMC_MSG_CondExecDramAddrLo ((uint16_t) 0x36) -#define PPSMC_MSG_LoadUcodes ((uint16_t) 0x37) -#define PPSMC_MSG_DriverResetMode ((uint16_t) 0x38) -#define PPSMC_MSG_PowerStateNotify ((uint16_t) 0x39) -#define PPSMC_MSG_SetDisplayPhyConfig ((uint16_t) 0x3A) -#define PPSMC_MSG_GetMaxSclkLevel ((uint16_t) 0x3B) -#define PPSMC_MSG_GetMaxLclkLevel ((uint16_t) 0x3C) -#define PPSMC_MSG_GetMaxUvdLevel ((uint16_t) 0x3D) -#define PPSMC_MSG_GetMaxEclkLevel ((uint16_t) 0x3E) -#define PPSMC_MSG_GetMaxAclkLevel ((uint16_t) 0x3F) -#define PPSMC_MSG_GetMaxNclkLevel ((uint16_t) 0x40) -#define PPSMC_MSG_GetMaxPstate ((uint16_t) 0x41) -#define PPSMC_MSG_DramAddrHiVirtual ((uint16_t) 0x42) -#define PPSMC_MSG_DramAddrLoVirtual ((uint16_t) 0x43) -#define PPSMC_MSG_DramAddrHiPhysical ((uint16_t) 0x44) -#define PPSMC_MSG_DramAddrLoPhysical ((uint16_t) 0x45) -#define PPSMC_MSG_DramBufferSize ((uint16_t) 0x46) -#define PPSMC_MSG_SetMmPwrLogDramAddrHi ((uint16_t) 0x47) -#define PPSMC_MSG_SetMmPwrLogDramAddrLo ((uint16_t) 0x48) -#define PPSMC_MSG_SetClkTableAddrHi ((uint16_t) 0x49) -#define PPSMC_MSG_SetClkTableAddrLo ((uint16_t) 0x4A) -#define PPSMC_MSG_GetConservativePowerLimit ((uint16_t) 0x4B) - -#define PPSMC_MSG_InitJobs ((uint16_t) 0x252) -#define PPSMC_MSG_ExecuteJob ((uint16_t) 0x254) - -#define PPSMC_MSG_NBDPM_Enable ((uint16_t) 0x140) -#define PPSMC_MSG_NBDPM_Disable ((uint16_t) 0x141) - -#define PPSMC_MSG_DPM_FPS_Mode ((uint16_t) 0x15d) -#define PPSMC_MSG_DPM_Activity_Mode ((uint16_t) 0x15e) - -#define PPSMC_MSG_PmStatusLogStart ((uint16_t) 0x170) -#define PPSMC_MSG_PmStatusLogSample ((uint16_t) 0x171) - -#define PPSMC_MSG_AllowLowSclkInterrupt ((uint16_t) 0x184) -#define PPSMC_MSG_MmPowerMonitorStart ((uint16_t) 0x18F) -#define PPSMC_MSG_MmPowerMonitorStop ((uint16_t) 0x190) -#define PPSMC_MSG_MmPowerMonitorRestart ((uint16_t) 0x191) - -#define PPSMC_MSG_SetClockGateMask ((uint16_t) 0x260) -#define PPSMC_MSG_SetFpsThresholdLo ((uint16_t) 0x264) -#define PPSMC_MSG_SetFpsThresholdHi ((uint16_t) 0x265) -#define PPSMC_MSG_SetLowSclkIntrThreshold ((uint16_t) 0x266) - -#define PPSMC_MSG_ClkTableXferToDram ((uint16_t) 0x267) -#define PPSMC_MSG_ClkTableXferToSmu ((uint16_t) 0x268) -#define PPSMC_MSG_GetAverageGraphicsActivity ((uint16_t) 0x269) -#define PPSMC_MSG_GetAverageGioActivity ((uint16_t) 0x26A) -#define PPSMC_MSG_SetLoggerBufferSize ((uint16_t) 0x26B) -#define PPSMC_MSG_SetLoggerAddressHigh ((uint16_t) 0x26C) -#define PPSMC_MSG_SetLoggerAddressLow ((uint16_t) 0x26D) -#define PPSMC_MSG_SetWatermarkFrequency ((uint16_t) 0x26E) - -/* REMOVE LATER*/ -#define PPSMC_MSG_DPM_ForceState ((uint16_t) 0x104) - -/* Feature Enable Masks*/ -#define NB_DPM_MASK 0x00000800 -#define VDDGFX_MASK 0x00800000 -#define VCE_DPM_MASK 0x00400000 -#define ACP_DPM_MASK 0x00040000 -#define UVD_DPM_MASK 0x00010000 -#define GFX_CU_PG_MASK 0x00004000 -#define SCLK_DPM_MASK 0x00080000 - -#if !defined(SMC_MICROCODE) -#pragma pack(pop) - -#endif - -#endif diff --git a/drivers/gpu/drm/amd/amdgpu/cz_smc.c b/drivers/gpu/drm/amd/amdgpu/cz_smc.c index e33180d3314a..ac7fee7b7eca 100644 --- a/drivers/gpu/drm/amd/amdgpu/cz_smc.c +++ b/drivers/gpu/drm/amd/amdgpu/cz_smc.c @@ -312,13 +312,16 @@ int cz_smu_start(struct amdgpu_device *adev) UCODE_ID_CP_MEC_JT1_MASK | UCODE_ID_CP_MEC_JT2_MASK; + if (adev->asic_type == CHIP_STONEY) + fw_to_check &= ~(UCODE_ID_SDMA1_MASK | UCODE_ID_CP_MEC_JT2_MASK); + cz_smu_request_load_fw(adev); ret = cz_smu_check_fw_load_finish(adev, fw_to_check); if (ret) return ret; /* manually load MEC firmware for CZ */ - if (adev->asic_type == CHIP_CARRIZO) { + if (adev->asic_type == CHIP_CARRIZO || adev->asic_type == CHIP_STONEY) { ret = cz_load_mec_firmware(adev); if (ret) { dev_err(adev->dev, "(%d) Mec Firmware load failed\n", ret); @@ -336,6 +339,9 @@ int cz_smu_start(struct amdgpu_device *adev) AMDGPU_CPMEC2_UCODE_LOADED | AMDGPU_CPRLC_UCODE_LOADED; + if (adev->asic_type == CHIP_STONEY) + adev->smu.fw_flags &= ~(AMDGPU_SDMA1_UCODE_LOADED | AMDGPU_CPMEC2_UCODE_LOADED); + return ret; } @@ -601,8 +607,13 @@ static int cz_smu_construct_toc_for_vddgfx_exit(struct amdgpu_device *adev) CZ_SCRATCH_ENTRY_UCODE_ID_CP_ME, false); cz_smu_populate_single_ucode_load_task(adev, CZ_SCRATCH_ENTRY_UCODE_ID_CP_MEC_JT1, false); - cz_smu_populate_single_ucode_load_task(adev, + if (adev->asic_type == CHIP_STONEY) { + cz_smu_populate_single_ucode_load_task(adev, + CZ_SCRATCH_ENTRY_UCODE_ID_CP_MEC_JT1, false); + } else { + cz_smu_populate_single_ucode_load_task(adev, CZ_SCRATCH_ENTRY_UCODE_ID_CP_MEC_JT2, false); + } cz_smu_populate_single_ucode_load_task(adev, CZ_SCRATCH_ENTRY_UCODE_ID_RLC_G, false); } @@ -642,8 +653,13 @@ static int cz_smu_construct_toc_for_bootup(struct amdgpu_device *adev) if (adev->firmware.smu_load) { cz_smu_populate_single_ucode_load_task(adev, CZ_SCRATCH_ENTRY_UCODE_ID_SDMA0, false); - cz_smu_populate_single_ucode_load_task(adev, + if (adev->asic_type == CHIP_STONEY) { + cz_smu_populate_single_ucode_load_task(adev, + CZ_SCRATCH_ENTRY_UCODE_ID_SDMA0, false); + } else { + cz_smu_populate_single_ucode_load_task(adev, CZ_SCRATCH_ENTRY_UCODE_ID_SDMA1, false); + } cz_smu_populate_single_ucode_load_task(adev, CZ_SCRATCH_ENTRY_UCODE_ID_CP_CE, false); cz_smu_populate_single_ucode_load_task(adev, @@ -652,8 +668,13 @@ static int cz_smu_construct_toc_for_bootup(struct amdgpu_device *adev) CZ_SCRATCH_ENTRY_UCODE_ID_CP_ME, false); cz_smu_populate_single_ucode_load_task(adev, CZ_SCRATCH_ENTRY_UCODE_ID_CP_MEC_JT1, false); - cz_smu_populate_single_ucode_load_task(adev, + if (adev->asic_type == CHIP_STONEY) { + cz_smu_populate_single_ucode_load_task(adev, + CZ_SCRATCH_ENTRY_UCODE_ID_CP_MEC_JT1, false); + } else { + cz_smu_populate_single_ucode_load_task(adev, CZ_SCRATCH_ENTRY_UCODE_ID_CP_MEC_JT2, false); + } cz_smu_populate_single_ucode_load_task(adev, CZ_SCRATCH_ENTRY_UCODE_ID_RLC_G, true); } @@ -888,10 +909,18 @@ int cz_smu_init(struct amdgpu_device *adev) CZ_SCRATCH_ENTRY_UCODE_ID_SDMA0, &priv->driver_buffer[priv->driver_buffer_length++])) goto smu_init_failed; - if (cz_smu_populate_single_firmware_entry(adev, - CZ_SCRATCH_ENTRY_UCODE_ID_SDMA1, - &priv->driver_buffer[priv->driver_buffer_length++])) - goto smu_init_failed; + + if (adev->asic_type == CHIP_STONEY) { + if (cz_smu_populate_single_firmware_entry(adev, + CZ_SCRATCH_ENTRY_UCODE_ID_SDMA0, + &priv->driver_buffer[priv->driver_buffer_length++])) + goto smu_init_failed; + } else { + if (cz_smu_populate_single_firmware_entry(adev, + CZ_SCRATCH_ENTRY_UCODE_ID_SDMA1, + &priv->driver_buffer[priv->driver_buffer_length++])) + goto smu_init_failed; + } if (cz_smu_populate_single_firmware_entry(adev, CZ_SCRATCH_ENTRY_UCODE_ID_CP_CE, &priv->driver_buffer[priv->driver_buffer_length++])) @@ -908,10 +937,17 @@ int cz_smu_init(struct amdgpu_device *adev) CZ_SCRATCH_ENTRY_UCODE_ID_CP_MEC_JT1, &priv->driver_buffer[priv->driver_buffer_length++])) goto smu_init_failed; - if (cz_smu_populate_single_firmware_entry(adev, - CZ_SCRATCH_ENTRY_UCODE_ID_CP_MEC_JT2, - &priv->driver_buffer[priv->driver_buffer_length++])) - goto smu_init_failed; + if (adev->asic_type == CHIP_STONEY) { + if (cz_smu_populate_single_firmware_entry(adev, + CZ_SCRATCH_ENTRY_UCODE_ID_CP_MEC_JT1, + &priv->driver_buffer[priv->driver_buffer_length++])) + goto smu_init_failed; + } else { + if (cz_smu_populate_single_firmware_entry(adev, + CZ_SCRATCH_ENTRY_UCODE_ID_CP_MEC_JT2, + &priv->driver_buffer[priv->driver_buffer_length++])) + goto smu_init_failed; + } if (cz_smu_populate_single_firmware_entry(adev, CZ_SCRATCH_ENTRY_UCODE_ID_RLC_G, &priv->driver_buffer[priv->driver_buffer_length++])) diff --git a/drivers/gpu/drm/amd/amdgpu/dce_v10_0.c b/drivers/gpu/drm/amd/amdgpu/dce_v10_0.c index d4c82b625727..6de2ce535e37 100644 --- a/drivers/gpu/drm/amd/amdgpu/dce_v10_0.c +++ b/drivers/gpu/drm/amd/amdgpu/dce_v10_0.c @@ -280,46 +280,22 @@ static void dce_v10_0_pageflip_interrupt_fini(struct amdgpu_device *adev) * @crtc_id: crtc to cleanup pageflip on * @crtc_base: new address of the crtc (GPU MC address) * - * Does the actual pageflip (evergreen+). - * During vblank we take the crtc lock and wait for the update_pending - * bit to go high, when it does, we release the lock, and allow the - * double buffered update to take place. - * Returns the current update pending status. + * Triggers the actual pageflip by updating the primary + * surface base address. */ static void dce_v10_0_page_flip(struct amdgpu_device *adev, int crtc_id, u64 crtc_base) { struct amdgpu_crtc *amdgpu_crtc = adev->mode_info.crtcs[crtc_id]; - u32 tmp = RREG32(mmGRPH_UPDATE + amdgpu_crtc->crtc_offset); - int i; - - /* Lock the graphics update lock */ - tmp = REG_SET_FIELD(tmp, GRPH_UPDATE, GRPH_UPDATE_LOCK, 1); - WREG32(mmGRPH_UPDATE + amdgpu_crtc->crtc_offset, tmp); - - /* update the scanout addresses */ - WREG32(mmGRPH_SECONDARY_SURFACE_ADDRESS_HIGH + amdgpu_crtc->crtc_offset, - upper_32_bits(crtc_base)); - WREG32(mmGRPH_SECONDARY_SURFACE_ADDRESS + amdgpu_crtc->crtc_offset, - lower_32_bits(crtc_base)); + /* update the primary scanout address */ WREG32(mmGRPH_PRIMARY_SURFACE_ADDRESS_HIGH + amdgpu_crtc->crtc_offset, upper_32_bits(crtc_base)); + /* writing to the low address triggers the update */ WREG32(mmGRPH_PRIMARY_SURFACE_ADDRESS + amdgpu_crtc->crtc_offset, lower_32_bits(crtc_base)); - - /* Wait for update_pending to go high. */ - for (i = 0; i < adev->usec_timeout; i++) { - if (RREG32(mmGRPH_UPDATE + amdgpu_crtc->crtc_offset) & - GRPH_UPDATE__GRPH_SURFACE_UPDATE_PENDING_MASK) - break; - udelay(1); - } - DRM_DEBUG("Update pending now high. Unlocking vupdate_lock.\n"); - - /* Unlock the lock, so double-buffering can take place inside vblank */ - tmp = REG_SET_FIELD(tmp, GRPH_UPDATE, GRPH_UPDATE_LOCK, 0); - WREG32(mmGRPH_UPDATE + amdgpu_crtc->crtc_offset, tmp); + /* post the write */ + RREG32(mmGRPH_PRIMARY_SURFACE_ADDRESS + amdgpu_crtc->crtc_offset); } static int dce_v10_0_crtc_get_scanoutpos(struct amdgpu_device *adev, int crtc, @@ -1274,7 +1250,7 @@ static void dce_v10_0_program_watermarks(struct amdgpu_device *adev, u32 pixel_period; u32 line_time = 0; u32 latency_watermark_a = 0, latency_watermark_b = 0; - u32 tmp, wm_mask; + u32 tmp, wm_mask, lb_vblank_lead_lines = 0; if (amdgpu_crtc->base.enabled && num_heads && mode) { pixel_period = 1000000 / (u32)mode->clock; @@ -1357,6 +1333,7 @@ static void dce_v10_0_program_watermarks(struct amdgpu_device *adev, (adev->mode_info.disp_priority == 2)) { DRM_DEBUG_KMS("force priority to high\n"); } + lb_vblank_lead_lines = DIV_ROUND_UP(lb_size, mode->crtc_hdisplay); } /* select wm A */ @@ -1381,6 +1358,8 @@ static void dce_v10_0_program_watermarks(struct amdgpu_device *adev, amdgpu_crtc->line_time = line_time; amdgpu_crtc->wm_high = latency_watermark_a; amdgpu_crtc->wm_low = latency_watermark_b; + /* Save number of lines the linebuffer leads before the scanout */ + amdgpu_crtc->lb_vblank_lead_lines = lb_vblank_lead_lines; } /** @@ -1689,6 +1668,9 @@ static void dce_v10_0_audio_fini(struct amdgpu_device *adev) { int i; + if (!amdgpu_audio) + return; + if (!adev->mode_info.audio.enabled) return; @@ -1994,7 +1976,7 @@ static void dce_v10_0_afmt_enable(struct drm_encoder *encoder, bool enable) enable ? "En" : "Dis", dig->afmt->offset, amdgpu_encoder->encoder_id); } -static void dce_v10_0_afmt_init(struct amdgpu_device *adev) +static int dce_v10_0_afmt_init(struct amdgpu_device *adev) { int i; @@ -2007,8 +1989,16 @@ static void dce_v10_0_afmt_init(struct amdgpu_device *adev) if (adev->mode_info.afmt[i]) { adev->mode_info.afmt[i]->offset = dig_offsets[i]; adev->mode_info.afmt[i]->id = i; + } else { + int j; + for (j = 0; j < i; j++) { + kfree(adev->mode_info.afmt[j]); + adev->mode_info.afmt[j] = NULL; + } + return -ENOMEM; } } + return 0; } static void dce_v10_0_afmt_fini(struct amdgpu_device *adev) @@ -2085,8 +2075,7 @@ static int dce_v10_0_crtc_do_set_base(struct drm_crtc *crtc, if (atomic) { amdgpu_fb = to_amdgpu_framebuffer(fb); target_fb = fb; - } - else { + } else { amdgpu_fb = to_amdgpu_framebuffer(crtc->primary->fb); target_fb = crtc->primary->fb; } @@ -2100,9 +2089,9 @@ static int dce_v10_0_crtc_do_set_base(struct drm_crtc *crtc, if (unlikely(r != 0)) return r; - if (atomic) + if (atomic) { fb_location = amdgpu_bo_gpu_offset(rbo); - else { + } else { r = amdgpu_bo_pin(rbo, AMDGPU_GEM_DOMAIN_VRAM, &fb_location); if (unlikely(r != 0)) { amdgpu_bo_unreserve(rbo); @@ -2517,26 +2506,19 @@ static void dce_v10_0_show_cursor(struct drm_crtc *crtc) struct amdgpu_device *adev = crtc->dev->dev_private; u32 tmp; + WREG32(mmCUR_SURFACE_ADDRESS_HIGH + amdgpu_crtc->crtc_offset, + upper_32_bits(amdgpu_crtc->cursor_addr)); + WREG32(mmCUR_SURFACE_ADDRESS + amdgpu_crtc->crtc_offset, + lower_32_bits(amdgpu_crtc->cursor_addr)); + tmp = RREG32_IDX(mmCUR_CONTROL + amdgpu_crtc->crtc_offset); tmp = REG_SET_FIELD(tmp, CUR_CONTROL, CURSOR_EN, 1); tmp = REG_SET_FIELD(tmp, CUR_CONTROL, CURSOR_MODE, 2); WREG32_IDX(mmCUR_CONTROL + amdgpu_crtc->crtc_offset, tmp); } -static void dce_v10_0_set_cursor(struct drm_crtc *crtc, struct drm_gem_object *obj, - uint64_t gpu_addr) -{ - struct amdgpu_crtc *amdgpu_crtc = to_amdgpu_crtc(crtc); - struct amdgpu_device *adev = crtc->dev->dev_private; - - WREG32(mmCUR_SURFACE_ADDRESS_HIGH + amdgpu_crtc->crtc_offset, - upper_32_bits(gpu_addr)); - WREG32(mmCUR_SURFACE_ADDRESS + amdgpu_crtc->crtc_offset, - lower_32_bits(gpu_addr)); -} - -static int dce_v10_0_crtc_cursor_move(struct drm_crtc *crtc, - int x, int y) +static int dce_v10_0_cursor_move_locked(struct drm_crtc *crtc, + int x, int y) { struct amdgpu_crtc *amdgpu_crtc = to_amdgpu_crtc(crtc); struct amdgpu_device *adev = crtc->dev->dev_private; @@ -2556,26 +2538,40 @@ static int dce_v10_0_crtc_cursor_move(struct drm_crtc *crtc, y = 0; } - dce_v10_0_lock_cursor(crtc, true); WREG32(mmCUR_POSITION + amdgpu_crtc->crtc_offset, (x << 16) | y); WREG32(mmCUR_HOT_SPOT + amdgpu_crtc->crtc_offset, (xorigin << 16) | yorigin); WREG32(mmCUR_SIZE + amdgpu_crtc->crtc_offset, ((amdgpu_crtc->cursor_width - 1) << 16) | (amdgpu_crtc->cursor_height - 1)); - dce_v10_0_lock_cursor(crtc, false); + + amdgpu_crtc->cursor_x = x; + amdgpu_crtc->cursor_y = y; return 0; } -static int dce_v10_0_crtc_cursor_set(struct drm_crtc *crtc, - struct drm_file *file_priv, - uint32_t handle, - uint32_t width, - uint32_t height) +static int dce_v10_0_crtc_cursor_move(struct drm_crtc *crtc, + int x, int y) +{ + int ret; + + dce_v10_0_lock_cursor(crtc, true); + ret = dce_v10_0_cursor_move_locked(crtc, x, y); + dce_v10_0_lock_cursor(crtc, false); + + return ret; +} + +static int dce_v10_0_crtc_cursor_set2(struct drm_crtc *crtc, + struct drm_file *file_priv, + uint32_t handle, + uint32_t width, + uint32_t height, + int32_t hot_x, + int32_t hot_y) { struct amdgpu_crtc *amdgpu_crtc = to_amdgpu_crtc(crtc); struct drm_gem_object *obj; - struct amdgpu_bo *robj; - uint64_t gpu_addr; + struct amdgpu_bo *aobj; int ret; if (!handle) { @@ -2597,41 +2593,71 @@ static int dce_v10_0_crtc_cursor_set(struct drm_crtc *crtc, return -ENOENT; } - robj = gem_to_amdgpu_bo(obj); - ret = amdgpu_bo_reserve(robj, false); - if (unlikely(ret != 0)) - goto fail; - ret = amdgpu_bo_pin_restricted(robj, AMDGPU_GEM_DOMAIN_VRAM, - 0, 0, &gpu_addr); - amdgpu_bo_unreserve(robj); - if (ret) - goto fail; + aobj = gem_to_amdgpu_bo(obj); + ret = amdgpu_bo_reserve(aobj, false); + if (ret != 0) { + drm_gem_object_unreference_unlocked(obj); + return ret; + } + + ret = amdgpu_bo_pin(aobj, AMDGPU_GEM_DOMAIN_VRAM, &amdgpu_crtc->cursor_addr); + amdgpu_bo_unreserve(aobj); + if (ret) { + DRM_ERROR("Failed to pin new cursor BO (%d)\n", ret); + drm_gem_object_unreference_unlocked(obj); + return ret; + } amdgpu_crtc->cursor_width = width; amdgpu_crtc->cursor_height = height; dce_v10_0_lock_cursor(crtc, true); - dce_v10_0_set_cursor(crtc, obj, gpu_addr); + + if (hot_x != amdgpu_crtc->cursor_hot_x || + hot_y != amdgpu_crtc->cursor_hot_y) { + int x, y; + + x = amdgpu_crtc->cursor_x + amdgpu_crtc->cursor_hot_x - hot_x; + y = amdgpu_crtc->cursor_y + amdgpu_crtc->cursor_hot_y - hot_y; + + dce_v10_0_cursor_move_locked(crtc, x, y); + + amdgpu_crtc->cursor_hot_x = hot_x; + amdgpu_crtc->cursor_hot_y = hot_y; + } + dce_v10_0_show_cursor(crtc); dce_v10_0_lock_cursor(crtc, false); unpin: if (amdgpu_crtc->cursor_bo) { - robj = gem_to_amdgpu_bo(amdgpu_crtc->cursor_bo); - ret = amdgpu_bo_reserve(robj, false); + struct amdgpu_bo *aobj = gem_to_amdgpu_bo(amdgpu_crtc->cursor_bo); + ret = amdgpu_bo_reserve(aobj, false); if (likely(ret == 0)) { - amdgpu_bo_unpin(robj); - amdgpu_bo_unreserve(robj); + amdgpu_bo_unpin(aobj); + amdgpu_bo_unreserve(aobj); } drm_gem_object_unreference_unlocked(amdgpu_crtc->cursor_bo); } amdgpu_crtc->cursor_bo = obj; return 0; -fail: - drm_gem_object_unreference_unlocked(obj); +} - return ret; +static void dce_v10_0_cursor_reset(struct drm_crtc *crtc) +{ + struct amdgpu_crtc *amdgpu_crtc = to_amdgpu_crtc(crtc); + + if (amdgpu_crtc->cursor_bo) { + dce_v10_0_lock_cursor(crtc, true); + + dce_v10_0_cursor_move_locked(crtc, amdgpu_crtc->cursor_x, + amdgpu_crtc->cursor_y); + + dce_v10_0_show_cursor(crtc); + + dce_v10_0_lock_cursor(crtc, false); + } } static void dce_v10_0_crtc_gamma_set(struct drm_crtc *crtc, u16 *red, u16 *green, @@ -2654,12 +2680,11 @@ static void dce_v10_0_crtc_destroy(struct drm_crtc *crtc) struct amdgpu_crtc *amdgpu_crtc = to_amdgpu_crtc(crtc); drm_crtc_cleanup(crtc); - destroy_workqueue(amdgpu_crtc->pflip_queue); kfree(amdgpu_crtc); } static const struct drm_crtc_funcs dce_v10_0_crtc_funcs = { - .cursor_set = dce_v10_0_crtc_cursor_set, + .cursor_set2 = dce_v10_0_crtc_cursor_set2, .cursor_move = dce_v10_0_crtc_cursor_move, .gamma_set = dce_v10_0_crtc_gamma_set, .set_config = amdgpu_crtc_set_config, @@ -2685,13 +2710,13 @@ static void dce_v10_0_crtc_dpms(struct drm_crtc *crtc, int mode) type = amdgpu_crtc_idx_to_irq_type(adev, amdgpu_crtc->crtc_id); amdgpu_irq_update(adev, &adev->crtc_irq, type); amdgpu_irq_update(adev, &adev->pageflip_irq, type); - drm_vblank_post_modeset(dev, amdgpu_crtc->crtc_id); + drm_vblank_on(dev, amdgpu_crtc->crtc_id); dce_v10_0_crtc_load_lut(crtc); break; case DRM_MODE_DPMS_STANDBY: case DRM_MODE_DPMS_SUSPEND: case DRM_MODE_DPMS_OFF: - drm_vblank_pre_modeset(dev, amdgpu_crtc->crtc_id); + drm_vblank_off(dev, amdgpu_crtc->crtc_id); if (amdgpu_crtc->enabled) { dce_v10_0_vga_enable(crtc, true); amdgpu_atombios_crtc_blank(crtc, ATOM_ENABLE); @@ -2793,6 +2818,7 @@ static int dce_v10_0_crtc_mode_set(struct drm_crtc *crtc, dce_v10_0_crtc_do_set_base(crtc, old_fb, x, y, 0); amdgpu_atombios_crtc_overscan_setup(crtc, mode, adjusted_mode); amdgpu_atombios_crtc_scaler_setup(crtc); + dce_v10_0_cursor_reset(crtc); /* update the hw version fpr dpm */ amdgpu_crtc->hw_mode = *adjusted_mode; @@ -2873,7 +2899,6 @@ static int dce_v10_0_crtc_init(struct amdgpu_device *adev, int index) drm_mode_crtc_set_gamma_size(&amdgpu_crtc->base, 256); amdgpu_crtc->crtc_id = index; - amdgpu_crtc->pflip_queue = create_singlethread_workqueue("amdgpu-pageflip-queue"); adev->mode_info.crtcs[index] = amdgpu_crtc; amdgpu_crtc->max_cursor_width = 128; @@ -2965,8 +2990,6 @@ static int dce_v10_0_sw_init(void *handle) if (r) return r; - adev->mode_info.mode_config_initialized = true; - adev->ddev->mode_config.funcs = &amdgpu_mode_funcs; adev->ddev->mode_config.max_width = 16384; @@ -2997,7 +3020,9 @@ static int dce_v10_0_sw_init(void *handle) return -EINVAL; /* setup afmt */ - dce_v10_0_afmt_init(adev); + r = dce_v10_0_afmt_init(adev); + if (r) + return r; r = dce_v10_0_audio_init(adev); if (r) @@ -3005,7 +3030,8 @@ static int dce_v10_0_sw_init(void *handle) drm_kms_helper_poll_init(adev->ddev); - return r; + adev->mode_info.mode_config_initialized = true; + return 0; } static int dce_v10_0_sw_fini(void *handle) @@ -3071,24 +3097,18 @@ static int dce_v10_0_suspend(void *handle) amdgpu_atombios_scratch_regs_save(adev); - dce_v10_0_hpd_fini(adev); - - dce_v10_0_pageflip_interrupt_fini(adev); - - return 0; + return dce_v10_0_hw_fini(handle); } static int dce_v10_0_resume(void *handle) { struct amdgpu_device *adev = (struct amdgpu_device *)handle; + int ret; - dce_v10_0_init_golden_registers(adev); + ret = dce_v10_0_hw_init(handle); amdgpu_atombios_scratch_regs_restore(adev); - /* init dig PHYs, disp eng pll */ - amdgpu_atombios_encoder_init_dig(adev); - amdgpu_atombios_crtc_set_disp_eng_pll(adev, adev->clock.default_dispclk); /* turn on the BL */ if (adev->mode_info.bl_encoder) { u8 bl_level = amdgpu_display_backlight_get_level(adev, @@ -3097,12 +3117,7 @@ static int dce_v10_0_resume(void *handle) bl_level); } - /* initialize hpd */ - dce_v10_0_hpd_init(adev); - - dce_v10_0_pageflip_interrupt_init(adev); - - return 0; + return ret; } static bool dce_v10_0_is_idle(void *handle) @@ -3294,37 +3309,20 @@ static int dce_v10_0_set_pageflip_irq_state(struct amdgpu_device *adev, unsigned type, enum amdgpu_interrupt_state state) { - u32 reg, reg_block; - /* now deal with page flip IRQ */ - switch (type) { - case AMDGPU_PAGEFLIP_IRQ_D1: - reg_block = CRTC0_REGISTER_OFFSET; - break; - case AMDGPU_PAGEFLIP_IRQ_D2: - reg_block = CRTC1_REGISTER_OFFSET; - break; - case AMDGPU_PAGEFLIP_IRQ_D3: - reg_block = CRTC2_REGISTER_OFFSET; - break; - case AMDGPU_PAGEFLIP_IRQ_D4: - reg_block = CRTC3_REGISTER_OFFSET; - break; - case AMDGPU_PAGEFLIP_IRQ_D5: - reg_block = CRTC4_REGISTER_OFFSET; - break; - case AMDGPU_PAGEFLIP_IRQ_D6: - reg_block = CRTC5_REGISTER_OFFSET; - break; - default: - DRM_ERROR("invalid pageflip crtc %d\n", type); - return -EINVAL; + u32 reg; + + if (type >= adev->mode_info.num_crtc) { + DRM_ERROR("invalid pageflip crtc %d\n", type); + return -EINVAL; } - reg = RREG32(mmGRPH_INTERRUPT_CONTROL + reg_block); + reg = RREG32(mmGRPH_INTERRUPT_CONTROL + crtc_offsets[type]); if (state == AMDGPU_IRQ_STATE_DISABLE) - WREG32(mmGRPH_INTERRUPT_CONTROL + reg_block, reg & ~GRPH_INTERRUPT_CONTROL__GRPH_PFLIP_INT_MASK_MASK); + WREG32(mmGRPH_INTERRUPT_CONTROL + crtc_offsets[type], + reg & ~GRPH_INTERRUPT_CONTROL__GRPH_PFLIP_INT_MASK_MASK); else - WREG32(mmGRPH_INTERRUPT_CONTROL + reg_block, reg | GRPH_INTERRUPT_CONTROL__GRPH_PFLIP_INT_MASK_MASK); + WREG32(mmGRPH_INTERRUPT_CONTROL + crtc_offsets[type], + reg | GRPH_INTERRUPT_CONTROL__GRPH_PFLIP_INT_MASK_MASK); return 0; } @@ -3333,7 +3331,6 @@ static int dce_v10_0_pageflip_irq(struct amdgpu_device *adev, struct amdgpu_irq_src *source, struct amdgpu_iv_entry *entry) { - int reg_block; unsigned long flags; unsigned crtc_id; struct amdgpu_crtc *amdgpu_crtc; @@ -3342,33 +3339,15 @@ static int dce_v10_0_pageflip_irq(struct amdgpu_device *adev, crtc_id = (entry->src_id - 8) >> 1; amdgpu_crtc = adev->mode_info.crtcs[crtc_id]; - /* ack the interrupt */ - switch(crtc_id){ - case AMDGPU_PAGEFLIP_IRQ_D1: - reg_block = CRTC0_REGISTER_OFFSET; - break; - case AMDGPU_PAGEFLIP_IRQ_D2: - reg_block = CRTC1_REGISTER_OFFSET; - break; - case AMDGPU_PAGEFLIP_IRQ_D3: - reg_block = CRTC2_REGISTER_OFFSET; - break; - case AMDGPU_PAGEFLIP_IRQ_D4: - reg_block = CRTC3_REGISTER_OFFSET; - break; - case AMDGPU_PAGEFLIP_IRQ_D5: - reg_block = CRTC4_REGISTER_OFFSET; - break; - case AMDGPU_PAGEFLIP_IRQ_D6: - reg_block = CRTC5_REGISTER_OFFSET; - break; - default: - DRM_ERROR("invalid pageflip crtc %d\n", crtc_id); - return -EINVAL; + if (crtc_id >= adev->mode_info.num_crtc) { + DRM_ERROR("invalid pageflip crtc %d\n", crtc_id); + return -EINVAL; } - if (RREG32(mmGRPH_INTERRUPT_STATUS + reg_block) & GRPH_INTERRUPT_STATUS__GRPH_PFLIP_INT_OCCURRED_MASK) - WREG32(mmGRPH_INTERRUPT_STATUS + reg_block, GRPH_INTERRUPT_STATUS__GRPH_PFLIP_INT_CLEAR_MASK); + if (RREG32(mmGRPH_INTERRUPT_STATUS + crtc_offsets[crtc_id]) & + GRPH_INTERRUPT_STATUS__GRPH_PFLIP_INT_OCCURRED_MASK) + WREG32(mmGRPH_INTERRUPT_STATUS + crtc_offsets[crtc_id], + GRPH_INTERRUPT_STATUS__GRPH_PFLIP_INT_CLEAR_MASK); /* IRQ could occur when in initial stage */ if (amdgpu_crtc == NULL) @@ -3396,7 +3375,7 @@ static int dce_v10_0_pageflip_irq(struct amdgpu_device *adev, spin_unlock_irqrestore(&adev->ddev->event_lock, flags); drm_vblank_put(adev->ddev, amdgpu_crtc->crtc_id); - queue_work(amdgpu_crtc->pflip_queue, &works->unpin_work); + schedule_work(&works->unpin_work); return 0; } @@ -3654,16 +3633,8 @@ dce_v10_0_ext_dpms(struct drm_encoder *encoder, int mode) } -static bool dce_v10_0_ext_mode_fixup(struct drm_encoder *encoder, - const struct drm_display_mode *mode, - struct drm_display_mode *adjusted_mode) -{ - return true; -} - static const struct drm_encoder_helper_funcs dce_v10_0_ext_helper_funcs = { .dpms = dce_v10_0_ext_dpms, - .mode_fixup = dce_v10_0_ext_mode_fixup, .prepare = dce_v10_0_ext_prepare, .mode_set = dce_v10_0_ext_mode_set, .commit = dce_v10_0_ext_commit, @@ -3759,7 +3730,7 @@ static void dce_v10_0_encoder_add(struct amdgpu_device *adev, case ENCODER_OBJECT_ID_INTERNAL_KLDSCP_DAC1: case ENCODER_OBJECT_ID_INTERNAL_KLDSCP_DAC2: drm_encoder_init(dev, encoder, &dce_v10_0_encoder_funcs, - DRM_MODE_ENCODER_DAC); + DRM_MODE_ENCODER_DAC, NULL); drm_encoder_helper_add(encoder, &dce_v10_0_dac_helper_funcs); break; case ENCODER_OBJECT_ID_INTERNAL_KLDSCP_DVO1: @@ -3770,15 +3741,15 @@ static void dce_v10_0_encoder_add(struct amdgpu_device *adev, if (amdgpu_encoder->devices & (ATOM_DEVICE_LCD_SUPPORT)) { amdgpu_encoder->rmx_type = RMX_FULL; drm_encoder_init(dev, encoder, &dce_v10_0_encoder_funcs, - DRM_MODE_ENCODER_LVDS); + DRM_MODE_ENCODER_LVDS, NULL); amdgpu_encoder->enc_priv = amdgpu_atombios_encoder_get_lcd_info(amdgpu_encoder); } else if (amdgpu_encoder->devices & (ATOM_DEVICE_CRT_SUPPORT)) { drm_encoder_init(dev, encoder, &dce_v10_0_encoder_funcs, - DRM_MODE_ENCODER_DAC); + DRM_MODE_ENCODER_DAC, NULL); amdgpu_encoder->enc_priv = amdgpu_atombios_encoder_get_dig_info(amdgpu_encoder); } else { drm_encoder_init(dev, encoder, &dce_v10_0_encoder_funcs, - DRM_MODE_ENCODER_TMDS); + DRM_MODE_ENCODER_TMDS, NULL); amdgpu_encoder->enc_priv = amdgpu_atombios_encoder_get_dig_info(amdgpu_encoder); } drm_encoder_helper_add(encoder, &dce_v10_0_dig_helper_funcs); @@ -3796,13 +3767,13 @@ static void dce_v10_0_encoder_add(struct amdgpu_device *adev, amdgpu_encoder->is_ext_encoder = true; if (amdgpu_encoder->devices & (ATOM_DEVICE_LCD_SUPPORT)) drm_encoder_init(dev, encoder, &dce_v10_0_encoder_funcs, - DRM_MODE_ENCODER_LVDS); + DRM_MODE_ENCODER_LVDS, NULL); else if (amdgpu_encoder->devices & (ATOM_DEVICE_CRT_SUPPORT)) drm_encoder_init(dev, encoder, &dce_v10_0_encoder_funcs, - DRM_MODE_ENCODER_DAC); + DRM_MODE_ENCODER_DAC, NULL); else drm_encoder_init(dev, encoder, &dce_v10_0_encoder_funcs, - DRM_MODE_ENCODER_TMDS); + DRM_MODE_ENCODER_TMDS, NULL); drm_encoder_helper_add(encoder, &dce_v10_0_ext_helper_funcs); break; } diff --git a/drivers/gpu/drm/amd/amdgpu/dce_v11_0.c b/drivers/gpu/drm/amd/amdgpu/dce_v11_0.c index 7e1cf5e4eebf..e9ccc6b787f3 100644 --- a/drivers/gpu/drm/amd/amdgpu/dce_v11_0.c +++ b/drivers/gpu/drm/amd/amdgpu/dce_v11_0.c @@ -126,6 +126,13 @@ static const u32 cz_mgcg_cgcg_init[] = mmXDMA_MEM_POWER_CNTL, 0x00000101, 0x00000000, }; +static const u32 stoney_golden_settings_a11[] = +{ + mmCRTC_DOUBLE_BUFFER_CONTROL, 0x00010101, 0x00010000, + mmFBC_MISC, 0x1f311fff, 0x14302000, +}; + + static void dce_v11_0_init_golden_registers(struct amdgpu_device *adev) { switch (adev->asic_type) { @@ -137,6 +144,11 @@ static void dce_v11_0_init_golden_registers(struct amdgpu_device *adev) cz_golden_settings_a11, (const u32)ARRAY_SIZE(cz_golden_settings_a11)); break; + case CHIP_STONEY: + amdgpu_program_register_sequence(adev, + stoney_golden_settings_a11, + (const u32)ARRAY_SIZE(stoney_golden_settings_a11)); + break; default: break; } @@ -199,9 +211,9 @@ static bool dce_v11_0_is_counter_moving(struct amdgpu_device *adev, int crtc) */ static void dce_v11_0_vblank_wait(struct amdgpu_device *adev, int crtc) { - unsigned i = 0; + unsigned i = 100; - if (crtc >= adev->mode_info.num_crtc) + if (crtc < 0 || crtc >= adev->mode_info.num_crtc) return; if (!(RREG32(mmCRTC_CONTROL + crtc_offsets[crtc]) & CRTC_CONTROL__CRTC_MASTER_EN_MASK)) @@ -211,14 +223,16 @@ static void dce_v11_0_vblank_wait(struct amdgpu_device *adev, int crtc) * wait for another frame. */ while (dce_v11_0_is_in_vblank(adev, crtc)) { - if (i++ % 100 == 0) { + if (i++ == 100) { + i = 0; if (!dce_v11_0_is_counter_moving(adev, crtc)) break; } } while (!dce_v11_0_is_in_vblank(adev, crtc)) { - if (i++ % 100 == 0) { + if (i++ == 100) { + i = 0; if (!dce_v11_0_is_counter_moving(adev, crtc)) break; } @@ -227,7 +241,7 @@ static void dce_v11_0_vblank_wait(struct amdgpu_device *adev, int crtc) static u32 dce_v11_0_vblank_get_counter(struct amdgpu_device *adev, int crtc) { - if (crtc >= adev->mode_info.num_crtc) + if (crtc < 0 || crtc >= adev->mode_info.num_crtc) return 0; else return RREG32(mmCRTC_STATUS_FRAME_COUNT + crtc_offsets[crtc]); @@ -258,46 +272,22 @@ static void dce_v11_0_pageflip_interrupt_fini(struct amdgpu_device *adev) * @crtc_id: crtc to cleanup pageflip on * @crtc_base: new address of the crtc (GPU MC address) * - * Does the actual pageflip (evergreen+). - * During vblank we take the crtc lock and wait for the update_pending - * bit to go high, when it does, we release the lock, and allow the - * double buffered update to take place. - * Returns the current update pending status. + * Triggers the actual pageflip by updating the primary + * surface base address. */ static void dce_v11_0_page_flip(struct amdgpu_device *adev, int crtc_id, u64 crtc_base) { struct amdgpu_crtc *amdgpu_crtc = adev->mode_info.crtcs[crtc_id]; - u32 tmp = RREG32(mmGRPH_UPDATE + amdgpu_crtc->crtc_offset); - int i; - - /* Lock the graphics update lock */ - tmp = REG_SET_FIELD(tmp, GRPH_UPDATE, GRPH_UPDATE_LOCK, 1); - WREG32(mmGRPH_UPDATE + amdgpu_crtc->crtc_offset, tmp); /* update the scanout addresses */ - WREG32(mmGRPH_SECONDARY_SURFACE_ADDRESS_HIGH + amdgpu_crtc->crtc_offset, - upper_32_bits(crtc_base)); - WREG32(mmGRPH_SECONDARY_SURFACE_ADDRESS + amdgpu_crtc->crtc_offset, - lower_32_bits(crtc_base)); - WREG32(mmGRPH_PRIMARY_SURFACE_ADDRESS_HIGH + amdgpu_crtc->crtc_offset, upper_32_bits(crtc_base)); + /* writing to the low address triggers the update */ WREG32(mmGRPH_PRIMARY_SURFACE_ADDRESS + amdgpu_crtc->crtc_offset, lower_32_bits(crtc_base)); - - /* Wait for update_pending to go high. */ - for (i = 0; i < adev->usec_timeout; i++) { - if (RREG32(mmGRPH_UPDATE + amdgpu_crtc->crtc_offset) & - GRPH_UPDATE__GRPH_SURFACE_UPDATE_PENDING_MASK) - break; - udelay(1); - } - DRM_DEBUG("Update pending now high. Unlocking vupdate_lock.\n"); - - /* Unlock the lock, so double-buffering can take place inside vblank */ - tmp = REG_SET_FIELD(tmp, GRPH_UPDATE, GRPH_UPDATE_LOCK, 0); - WREG32(mmGRPH_UPDATE + amdgpu_crtc->crtc_offset, tmp); + /* post the write */ + RREG32(mmGRPH_PRIMARY_SURFACE_ADDRESS + amdgpu_crtc->crtc_offset); } static int dce_v11_0_crtc_get_scanoutpos(struct amdgpu_device *adev, int crtc, @@ -1250,7 +1240,7 @@ static void dce_v11_0_program_watermarks(struct amdgpu_device *adev, u32 pixel_period; u32 line_time = 0; u32 latency_watermark_a = 0, latency_watermark_b = 0; - u32 tmp, wm_mask; + u32 tmp, wm_mask, lb_vblank_lead_lines = 0; if (amdgpu_crtc->base.enabled && num_heads && mode) { pixel_period = 1000000 / (u32)mode->clock; @@ -1333,6 +1323,7 @@ static void dce_v11_0_program_watermarks(struct amdgpu_device *adev, (adev->mode_info.disp_priority == 2)) { DRM_DEBUG_KMS("force priority to high\n"); } + lb_vblank_lead_lines = DIV_ROUND_UP(lb_size, mode->crtc_hdisplay); } /* select wm A */ @@ -1357,6 +1348,8 @@ static void dce_v11_0_program_watermarks(struct amdgpu_device *adev, amdgpu_crtc->line_time = line_time; amdgpu_crtc->wm_high = latency_watermark_a; amdgpu_crtc->wm_low = latency_watermark_b; + /* Save number of lines the linebuffer leads before the scanout */ + amdgpu_crtc->lb_vblank_lead_lines = lb_vblank_lead_lines; } /** @@ -1665,6 +1658,9 @@ static void dce_v11_0_audio_fini(struct amdgpu_device *adev) { int i; + if (!amdgpu_audio) + return; + if (!adev->mode_info.audio.enabled) return; @@ -1970,7 +1966,7 @@ static void dce_v11_0_afmt_enable(struct drm_encoder *encoder, bool enable) enable ? "En" : "Dis", dig->afmt->offset, amdgpu_encoder->encoder_id); } -static void dce_v11_0_afmt_init(struct amdgpu_device *adev) +static int dce_v11_0_afmt_init(struct amdgpu_device *adev) { int i; @@ -1983,8 +1979,16 @@ static void dce_v11_0_afmt_init(struct amdgpu_device *adev) if (adev->mode_info.afmt[i]) { adev->mode_info.afmt[i]->offset = dig_offsets[i]; adev->mode_info.afmt[i]->id = i; + } else { + int j; + for (j = 0; j < i; j++) { + kfree(adev->mode_info.afmt[j]); + adev->mode_info.afmt[j] = NULL; + } + return -ENOMEM; } } + return 0; } static void dce_v11_0_afmt_fini(struct amdgpu_device *adev) @@ -2061,8 +2065,7 @@ static int dce_v11_0_crtc_do_set_base(struct drm_crtc *crtc, if (atomic) { amdgpu_fb = to_amdgpu_framebuffer(fb); target_fb = fb; - } - else { + } else { amdgpu_fb = to_amdgpu_framebuffer(crtc->primary->fb); target_fb = crtc->primary->fb; } @@ -2076,9 +2079,9 @@ static int dce_v11_0_crtc_do_set_base(struct drm_crtc *crtc, if (unlikely(r != 0)) return r; - if (atomic) + if (atomic) { fb_location = amdgpu_bo_gpu_offset(rbo); - else { + } else { r = amdgpu_bo_pin(rbo, AMDGPU_GEM_DOMAIN_VRAM, &fb_location); if (unlikely(r != 0)) { amdgpu_bo_unreserve(rbo); @@ -2443,7 +2446,7 @@ static u32 dce_v11_0_pick_pll(struct drm_crtc *crtc) /* XXX need to determine what plls are available on each DCE11 part */ pll_in_use = amdgpu_pll_get_use_mask(crtc); - if (adev->asic_type == CHIP_CARRIZO) { + if (adev->asic_type == CHIP_CARRIZO || adev->asic_type == CHIP_STONEY) { if (!(pll_in_use & (1 << ATOM_PPLL1))) return ATOM_PPLL1; if (!(pll_in_use & (1 << ATOM_PPLL0))) @@ -2494,26 +2497,19 @@ static void dce_v11_0_show_cursor(struct drm_crtc *crtc) struct amdgpu_device *adev = crtc->dev->dev_private; u32 tmp; + WREG32(mmCUR_SURFACE_ADDRESS_HIGH + amdgpu_crtc->crtc_offset, + upper_32_bits(amdgpu_crtc->cursor_addr)); + WREG32(mmCUR_SURFACE_ADDRESS + amdgpu_crtc->crtc_offset, + lower_32_bits(amdgpu_crtc->cursor_addr)); + tmp = RREG32_IDX(mmCUR_CONTROL + amdgpu_crtc->crtc_offset); tmp = REG_SET_FIELD(tmp, CUR_CONTROL, CURSOR_EN, 1); tmp = REG_SET_FIELD(tmp, CUR_CONTROL, CURSOR_MODE, 2); WREG32_IDX(mmCUR_CONTROL + amdgpu_crtc->crtc_offset, tmp); } -static void dce_v11_0_set_cursor(struct drm_crtc *crtc, struct drm_gem_object *obj, - uint64_t gpu_addr) -{ - struct amdgpu_crtc *amdgpu_crtc = to_amdgpu_crtc(crtc); - struct amdgpu_device *adev = crtc->dev->dev_private; - - WREG32(mmCUR_SURFACE_ADDRESS_HIGH + amdgpu_crtc->crtc_offset, - upper_32_bits(gpu_addr)); - WREG32(mmCUR_SURFACE_ADDRESS + amdgpu_crtc->crtc_offset, - lower_32_bits(gpu_addr)); -} - -static int dce_v11_0_crtc_cursor_move(struct drm_crtc *crtc, - int x, int y) +static int dce_v11_0_cursor_move_locked(struct drm_crtc *crtc, + int x, int y) { struct amdgpu_crtc *amdgpu_crtc = to_amdgpu_crtc(crtc); struct amdgpu_device *adev = crtc->dev->dev_private; @@ -2533,26 +2529,40 @@ static int dce_v11_0_crtc_cursor_move(struct drm_crtc *crtc, y = 0; } - dce_v11_0_lock_cursor(crtc, true); WREG32(mmCUR_POSITION + amdgpu_crtc->crtc_offset, (x << 16) | y); WREG32(mmCUR_HOT_SPOT + amdgpu_crtc->crtc_offset, (xorigin << 16) | yorigin); WREG32(mmCUR_SIZE + amdgpu_crtc->crtc_offset, ((amdgpu_crtc->cursor_width - 1) << 16) | (amdgpu_crtc->cursor_height - 1)); - dce_v11_0_lock_cursor(crtc, false); + + amdgpu_crtc->cursor_x = x; + amdgpu_crtc->cursor_y = y; return 0; } -static int dce_v11_0_crtc_cursor_set(struct drm_crtc *crtc, - struct drm_file *file_priv, - uint32_t handle, - uint32_t width, - uint32_t height) +static int dce_v11_0_crtc_cursor_move(struct drm_crtc *crtc, + int x, int y) +{ + int ret; + + dce_v11_0_lock_cursor(crtc, true); + ret = dce_v11_0_cursor_move_locked(crtc, x, y); + dce_v11_0_lock_cursor(crtc, false); + + return ret; +} + +static int dce_v11_0_crtc_cursor_set2(struct drm_crtc *crtc, + struct drm_file *file_priv, + uint32_t handle, + uint32_t width, + uint32_t height, + int32_t hot_x, + int32_t hot_y) { struct amdgpu_crtc *amdgpu_crtc = to_amdgpu_crtc(crtc); struct drm_gem_object *obj; - struct amdgpu_bo *robj; - uint64_t gpu_addr; + struct amdgpu_bo *aobj; int ret; if (!handle) { @@ -2574,41 +2584,71 @@ static int dce_v11_0_crtc_cursor_set(struct drm_crtc *crtc, return -ENOENT; } - robj = gem_to_amdgpu_bo(obj); - ret = amdgpu_bo_reserve(robj, false); - if (unlikely(ret != 0)) - goto fail; - ret = amdgpu_bo_pin_restricted(robj, AMDGPU_GEM_DOMAIN_VRAM, - 0, 0, &gpu_addr); - amdgpu_bo_unreserve(robj); - if (ret) - goto fail; + aobj = gem_to_amdgpu_bo(obj); + ret = amdgpu_bo_reserve(aobj, false); + if (ret != 0) { + drm_gem_object_unreference_unlocked(obj); + return ret; + } + + ret = amdgpu_bo_pin(aobj, AMDGPU_GEM_DOMAIN_VRAM, &amdgpu_crtc->cursor_addr); + amdgpu_bo_unreserve(aobj); + if (ret) { + DRM_ERROR("Failed to pin new cursor BO (%d)\n", ret); + drm_gem_object_unreference_unlocked(obj); + return ret; + } amdgpu_crtc->cursor_width = width; amdgpu_crtc->cursor_height = height; dce_v11_0_lock_cursor(crtc, true); - dce_v11_0_set_cursor(crtc, obj, gpu_addr); + + if (hot_x != amdgpu_crtc->cursor_hot_x || + hot_y != amdgpu_crtc->cursor_hot_y) { + int x, y; + + x = amdgpu_crtc->cursor_x + amdgpu_crtc->cursor_hot_x - hot_x; + y = amdgpu_crtc->cursor_y + amdgpu_crtc->cursor_hot_y - hot_y; + + dce_v11_0_cursor_move_locked(crtc, x, y); + + amdgpu_crtc->cursor_hot_x = hot_x; + amdgpu_crtc->cursor_hot_y = hot_y; + } + dce_v11_0_show_cursor(crtc); dce_v11_0_lock_cursor(crtc, false); unpin: if (amdgpu_crtc->cursor_bo) { - robj = gem_to_amdgpu_bo(amdgpu_crtc->cursor_bo); - ret = amdgpu_bo_reserve(robj, false); + struct amdgpu_bo *aobj = gem_to_amdgpu_bo(amdgpu_crtc->cursor_bo); + ret = amdgpu_bo_reserve(aobj, false); if (likely(ret == 0)) { - amdgpu_bo_unpin(robj); - amdgpu_bo_unreserve(robj); + amdgpu_bo_unpin(aobj); + amdgpu_bo_unreserve(aobj); } drm_gem_object_unreference_unlocked(amdgpu_crtc->cursor_bo); } amdgpu_crtc->cursor_bo = obj; return 0; -fail: - drm_gem_object_unreference_unlocked(obj); +} - return ret; +static void dce_v11_0_cursor_reset(struct drm_crtc *crtc) +{ + struct amdgpu_crtc *amdgpu_crtc = to_amdgpu_crtc(crtc); + + if (amdgpu_crtc->cursor_bo) { + dce_v11_0_lock_cursor(crtc, true); + + dce_v11_0_cursor_move_locked(crtc, amdgpu_crtc->cursor_x, + amdgpu_crtc->cursor_y); + + dce_v11_0_show_cursor(crtc); + + dce_v11_0_lock_cursor(crtc, false); + } } static void dce_v11_0_crtc_gamma_set(struct drm_crtc *crtc, u16 *red, u16 *green, @@ -2631,12 +2671,11 @@ static void dce_v11_0_crtc_destroy(struct drm_crtc *crtc) struct amdgpu_crtc *amdgpu_crtc = to_amdgpu_crtc(crtc); drm_crtc_cleanup(crtc); - destroy_workqueue(amdgpu_crtc->pflip_queue); kfree(amdgpu_crtc); } static const struct drm_crtc_funcs dce_v11_0_crtc_funcs = { - .cursor_set = dce_v11_0_crtc_cursor_set, + .cursor_set2 = dce_v11_0_crtc_cursor_set2, .cursor_move = dce_v11_0_crtc_cursor_move, .gamma_set = dce_v11_0_crtc_gamma_set, .set_config = amdgpu_crtc_set_config, @@ -2662,13 +2701,13 @@ static void dce_v11_0_crtc_dpms(struct drm_crtc *crtc, int mode) type = amdgpu_crtc_idx_to_irq_type(adev, amdgpu_crtc->crtc_id); amdgpu_irq_update(adev, &adev->crtc_irq, type); amdgpu_irq_update(adev, &adev->pageflip_irq, type); - drm_vblank_post_modeset(dev, amdgpu_crtc->crtc_id); + drm_vblank_on(dev, amdgpu_crtc->crtc_id); dce_v11_0_crtc_load_lut(crtc); break; case DRM_MODE_DPMS_STANDBY: case DRM_MODE_DPMS_SUSPEND: case DRM_MODE_DPMS_OFF: - drm_vblank_pre_modeset(dev, amdgpu_crtc->crtc_id); + drm_vblank_off(dev, amdgpu_crtc->crtc_id); if (amdgpu_crtc->enabled) { dce_v11_0_vga_enable(crtc, true); amdgpu_atombios_crtc_blank(crtc, ATOM_ENABLE); @@ -2770,6 +2809,7 @@ static int dce_v11_0_crtc_mode_set(struct drm_crtc *crtc, dce_v11_0_crtc_do_set_base(crtc, old_fb, x, y, 0); amdgpu_atombios_crtc_overscan_setup(crtc, mode, adjusted_mode); amdgpu_atombios_crtc_scaler_setup(crtc); + dce_v11_0_cursor_reset(crtc); /* update the hw version fpr dpm */ amdgpu_crtc->hw_mode = *adjusted_mode; @@ -2850,7 +2890,6 @@ static int dce_v11_0_crtc_init(struct amdgpu_device *adev, int index) drm_mode_crtc_set_gamma_size(&amdgpu_crtc->base, 256); amdgpu_crtc->crtc_id = index; - amdgpu_crtc->pflip_queue = create_singlethread_workqueue("amdgpu-pageflip-queue"); adev->mode_info.crtcs[index] = amdgpu_crtc; amdgpu_crtc->max_cursor_width = 128; @@ -2911,6 +2950,11 @@ static int dce_v11_0_early_init(void *handle) adev->mode_info.num_hpd = 6; adev->mode_info.num_dig = 9; break; + case CHIP_STONEY: + adev->mode_info.num_crtc = 2; + adev->mode_info.num_hpd = 6; + adev->mode_info.num_dig = 9; + break; default: /* FIXME: not supported yet */ return -EINVAL; @@ -2927,7 +2971,7 @@ static int dce_v11_0_sw_init(void *handle) for (i = 0; i < adev->mode_info.num_crtc; i++) { r = amdgpu_irq_add_id(adev, i + 1, &adev->crtc_irq); if (r) - return r; + return r; } for (i = 8; i < 20; i += 2) { @@ -2939,9 +2983,7 @@ static int dce_v11_0_sw_init(void *handle) /* HPD hotplug */ r = amdgpu_irq_add_id(adev, 42, &adev->hpd_irq); if (r) - return r; - - adev->mode_info.mode_config_initialized = true; + return r; adev->ddev->mode_config.funcs = &amdgpu_mode_funcs; @@ -2960,6 +3002,7 @@ static int dce_v11_0_sw_init(void *handle) adev->ddev->mode_config.max_width = 16384; adev->ddev->mode_config.max_height = 16384; + /* allocate crtcs */ for (i = 0; i < adev->mode_info.num_crtc; i++) { r = dce_v11_0_crtc_init(adev, i); @@ -2973,7 +3016,9 @@ static int dce_v11_0_sw_init(void *handle) return -EINVAL; /* setup afmt */ - dce_v11_0_afmt_init(adev); + r = dce_v11_0_afmt_init(adev); + if (r) + return r; r = dce_v11_0_audio_init(adev); if (r) @@ -2981,7 +3026,8 @@ static int dce_v11_0_sw_init(void *handle) drm_kms_helper_poll_init(adev->ddev); - return r; + adev->mode_info.mode_config_initialized = true; + return 0; } static int dce_v11_0_sw_fini(void *handle) @@ -3009,6 +3055,7 @@ static int dce_v11_0_hw_init(void *handle) dce_v11_0_init_golden_registers(adev); /* init dig PHYs, disp eng pll */ + amdgpu_atombios_crtc_powergate_init(adev); amdgpu_atombios_encoder_init_dig(adev); amdgpu_atombios_crtc_set_disp_eng_pll(adev, adev->clock.default_dispclk); @@ -3046,25 +3093,18 @@ static int dce_v11_0_suspend(void *handle) amdgpu_atombios_scratch_regs_save(adev); - dce_v11_0_hpd_fini(adev); - - dce_v11_0_pageflip_interrupt_fini(adev); - - return 0; + return dce_v11_0_hw_fini(handle); } static int dce_v11_0_resume(void *handle) { struct amdgpu_device *adev = (struct amdgpu_device *)handle; + int ret; - dce_v11_0_init_golden_registers(adev); + ret = dce_v11_0_hw_init(handle); amdgpu_atombios_scratch_regs_restore(adev); - /* init dig PHYs, disp eng pll */ - amdgpu_atombios_crtc_powergate_init(adev); - amdgpu_atombios_encoder_init_dig(adev); - amdgpu_atombios_crtc_set_disp_eng_pll(adev, adev->clock.default_dispclk); /* turn on the BL */ if (adev->mode_info.bl_encoder) { u8 bl_level = amdgpu_display_backlight_get_level(adev, @@ -3073,12 +3113,7 @@ static int dce_v11_0_resume(void *handle) bl_level); } - /* initialize hpd */ - dce_v11_0_hpd_init(adev); - - dce_v11_0_pageflip_interrupt_init(adev); - - return 0; + return ret; } static bool dce_v11_0_is_idle(void *handle) @@ -3270,37 +3305,20 @@ static int dce_v11_0_set_pageflip_irq_state(struct amdgpu_device *adev, unsigned type, enum amdgpu_interrupt_state state) { - u32 reg, reg_block; - /* now deal with page flip IRQ */ - switch (type) { - case AMDGPU_PAGEFLIP_IRQ_D1: - reg_block = CRTC0_REGISTER_OFFSET; - break; - case AMDGPU_PAGEFLIP_IRQ_D2: - reg_block = CRTC1_REGISTER_OFFSET; - break; - case AMDGPU_PAGEFLIP_IRQ_D3: - reg_block = CRTC2_REGISTER_OFFSET; - break; - case AMDGPU_PAGEFLIP_IRQ_D4: - reg_block = CRTC3_REGISTER_OFFSET; - break; - case AMDGPU_PAGEFLIP_IRQ_D5: - reg_block = CRTC4_REGISTER_OFFSET; - break; - case AMDGPU_PAGEFLIP_IRQ_D6: - reg_block = CRTC5_REGISTER_OFFSET; - break; - default: - DRM_ERROR("invalid pageflip crtc %d\n", type); - return -EINVAL; + u32 reg; + + if (type >= adev->mode_info.num_crtc) { + DRM_ERROR("invalid pageflip crtc %d\n", type); + return -EINVAL; } - reg = RREG32(mmGRPH_INTERRUPT_CONTROL + reg_block); + reg = RREG32(mmGRPH_INTERRUPT_CONTROL + crtc_offsets[type]); if (state == AMDGPU_IRQ_STATE_DISABLE) - WREG32(mmGRPH_INTERRUPT_CONTROL + reg_block, reg & ~GRPH_INTERRUPT_CONTROL__GRPH_PFLIP_INT_MASK_MASK); + WREG32(mmGRPH_INTERRUPT_CONTROL + crtc_offsets[type], + reg & ~GRPH_INTERRUPT_CONTROL__GRPH_PFLIP_INT_MASK_MASK); else - WREG32(mmGRPH_INTERRUPT_CONTROL + reg_block, reg | GRPH_INTERRUPT_CONTROL__GRPH_PFLIP_INT_MASK_MASK); + WREG32(mmGRPH_INTERRUPT_CONTROL + crtc_offsets[type], + reg | GRPH_INTERRUPT_CONTROL__GRPH_PFLIP_INT_MASK_MASK); return 0; } @@ -3309,7 +3327,6 @@ static int dce_v11_0_pageflip_irq(struct amdgpu_device *adev, struct amdgpu_irq_src *source, struct amdgpu_iv_entry *entry) { - int reg_block; unsigned long flags; unsigned crtc_id; struct amdgpu_crtc *amdgpu_crtc; @@ -3318,33 +3335,15 @@ static int dce_v11_0_pageflip_irq(struct amdgpu_device *adev, crtc_id = (entry->src_id - 8) >> 1; amdgpu_crtc = adev->mode_info.crtcs[crtc_id]; - /* ack the interrupt */ - switch(crtc_id){ - case AMDGPU_PAGEFLIP_IRQ_D1: - reg_block = CRTC0_REGISTER_OFFSET; - break; - case AMDGPU_PAGEFLIP_IRQ_D2: - reg_block = CRTC1_REGISTER_OFFSET; - break; - case AMDGPU_PAGEFLIP_IRQ_D3: - reg_block = CRTC2_REGISTER_OFFSET; - break; - case AMDGPU_PAGEFLIP_IRQ_D4: - reg_block = CRTC3_REGISTER_OFFSET; - break; - case AMDGPU_PAGEFLIP_IRQ_D5: - reg_block = CRTC4_REGISTER_OFFSET; - break; - case AMDGPU_PAGEFLIP_IRQ_D6: - reg_block = CRTC5_REGISTER_OFFSET; - break; - default: - DRM_ERROR("invalid pageflip crtc %d\n", crtc_id); - return -EINVAL; + if (crtc_id >= adev->mode_info.num_crtc) { + DRM_ERROR("invalid pageflip crtc %d\n", crtc_id); + return -EINVAL; } - if (RREG32(mmGRPH_INTERRUPT_STATUS + reg_block) & GRPH_INTERRUPT_STATUS__GRPH_PFLIP_INT_OCCURRED_MASK) - WREG32(mmGRPH_INTERRUPT_STATUS + reg_block, GRPH_INTERRUPT_STATUS__GRPH_PFLIP_INT_CLEAR_MASK); + if (RREG32(mmGRPH_INTERRUPT_STATUS + crtc_offsets[crtc_id]) & + GRPH_INTERRUPT_STATUS__GRPH_PFLIP_INT_OCCURRED_MASK) + WREG32(mmGRPH_INTERRUPT_STATUS + crtc_offsets[crtc_id], + GRPH_INTERRUPT_STATUS__GRPH_PFLIP_INT_CLEAR_MASK); /* IRQ could occur when in initial stage */ if(amdgpu_crtc == NULL) @@ -3372,7 +3371,7 @@ static int dce_v11_0_pageflip_irq(struct amdgpu_device *adev, spin_unlock_irqrestore(&adev->ddev->event_lock, flags); drm_vblank_put(adev->ddev, amdgpu_crtc->crtc_id); - queue_work(amdgpu_crtc->pflip_queue, &works->unpin_work); + schedule_work(&works->unpin_work); return 0; } @@ -3397,7 +3396,7 @@ static void dce_v11_0_crtc_vblank_int_ack(struct amdgpu_device *adev, { u32 tmp; - if (crtc >= adev->mode_info.num_crtc) { + if (crtc < 0 || crtc >= adev->mode_info.num_crtc) { DRM_DEBUG("invalid crtc %d\n", crtc); return; } @@ -3412,7 +3411,7 @@ static void dce_v11_0_crtc_vline_int_ack(struct amdgpu_device *adev, { u32 tmp; - if (crtc >= adev->mode_info.num_crtc) { + if (crtc < 0 || crtc >= adev->mode_info.num_crtc) { DRM_DEBUG("invalid crtc %d\n", crtc); return; } @@ -3630,16 +3629,8 @@ dce_v11_0_ext_dpms(struct drm_encoder *encoder, int mode) } -static bool dce_v11_0_ext_mode_fixup(struct drm_encoder *encoder, - const struct drm_display_mode *mode, - struct drm_display_mode *adjusted_mode) -{ - return true; -} - static const struct drm_encoder_helper_funcs dce_v11_0_ext_helper_funcs = { .dpms = dce_v11_0_ext_dpms, - .mode_fixup = dce_v11_0_ext_mode_fixup, .prepare = dce_v11_0_ext_prepare, .mode_set = dce_v11_0_ext_mode_set, .commit = dce_v11_0_ext_commit, @@ -3735,7 +3726,7 @@ static void dce_v11_0_encoder_add(struct amdgpu_device *adev, case ENCODER_OBJECT_ID_INTERNAL_KLDSCP_DAC1: case ENCODER_OBJECT_ID_INTERNAL_KLDSCP_DAC2: drm_encoder_init(dev, encoder, &dce_v11_0_encoder_funcs, - DRM_MODE_ENCODER_DAC); + DRM_MODE_ENCODER_DAC, NULL); drm_encoder_helper_add(encoder, &dce_v11_0_dac_helper_funcs); break; case ENCODER_OBJECT_ID_INTERNAL_KLDSCP_DVO1: @@ -3746,15 +3737,15 @@ static void dce_v11_0_encoder_add(struct amdgpu_device *adev, if (amdgpu_encoder->devices & (ATOM_DEVICE_LCD_SUPPORT)) { amdgpu_encoder->rmx_type = RMX_FULL; drm_encoder_init(dev, encoder, &dce_v11_0_encoder_funcs, - DRM_MODE_ENCODER_LVDS); + DRM_MODE_ENCODER_LVDS, NULL); amdgpu_encoder->enc_priv = amdgpu_atombios_encoder_get_lcd_info(amdgpu_encoder); } else if (amdgpu_encoder->devices & (ATOM_DEVICE_CRT_SUPPORT)) { drm_encoder_init(dev, encoder, &dce_v11_0_encoder_funcs, - DRM_MODE_ENCODER_DAC); + DRM_MODE_ENCODER_DAC, NULL); amdgpu_encoder->enc_priv = amdgpu_atombios_encoder_get_dig_info(amdgpu_encoder); } else { drm_encoder_init(dev, encoder, &dce_v11_0_encoder_funcs, - DRM_MODE_ENCODER_TMDS); + DRM_MODE_ENCODER_TMDS, NULL); amdgpu_encoder->enc_priv = amdgpu_atombios_encoder_get_dig_info(amdgpu_encoder); } drm_encoder_helper_add(encoder, &dce_v11_0_dig_helper_funcs); @@ -3772,13 +3763,13 @@ static void dce_v11_0_encoder_add(struct amdgpu_device *adev, amdgpu_encoder->is_ext_encoder = true; if (amdgpu_encoder->devices & (ATOM_DEVICE_LCD_SUPPORT)) drm_encoder_init(dev, encoder, &dce_v11_0_encoder_funcs, - DRM_MODE_ENCODER_LVDS); + DRM_MODE_ENCODER_LVDS, NULL); else if (amdgpu_encoder->devices & (ATOM_DEVICE_CRT_SUPPORT)) drm_encoder_init(dev, encoder, &dce_v11_0_encoder_funcs, - DRM_MODE_ENCODER_DAC); + DRM_MODE_ENCODER_DAC, NULL); else drm_encoder_init(dev, encoder, &dce_v11_0_encoder_funcs, - DRM_MODE_ENCODER_TMDS); + DRM_MODE_ENCODER_TMDS, NULL); drm_encoder_helper_add(encoder, &dce_v11_0_ext_helper_funcs); break; } diff --git a/drivers/gpu/drm/amd/amdgpu/dce_v8_0.c b/drivers/gpu/drm/amd/amdgpu/dce_v8_0.c index 34b9c2a9d8d4..e56b55d8c280 100644 --- a/drivers/gpu/drm/amd/amdgpu/dce_v8_0.c +++ b/drivers/gpu/drm/amd/amdgpu/dce_v8_0.c @@ -229,46 +229,22 @@ static void dce_v8_0_pageflip_interrupt_fini(struct amdgpu_device *adev) * @crtc_id: crtc to cleanup pageflip on * @crtc_base: new address of the crtc (GPU MC address) * - * Does the actual pageflip (evergreen+). - * During vblank we take the crtc lock and wait for the update_pending - * bit to go high, when it does, we release the lock, and allow the - * double buffered update to take place. - * Returns the current update pending status. + * Triggers the actual pageflip by updating the primary + * surface base address. */ static void dce_v8_0_page_flip(struct amdgpu_device *adev, int crtc_id, u64 crtc_base) { struct amdgpu_crtc *amdgpu_crtc = adev->mode_info.crtcs[crtc_id]; - u32 tmp = RREG32(mmGRPH_UPDATE + amdgpu_crtc->crtc_offset); - int i; - - /* Lock the graphics update lock */ - tmp |= GRPH_UPDATE__GRPH_UPDATE_LOCK_MASK; - WREG32(mmGRPH_UPDATE + amdgpu_crtc->crtc_offset, tmp); - - /* update the scanout addresses */ - WREG32(mmGRPH_SECONDARY_SURFACE_ADDRESS_HIGH + amdgpu_crtc->crtc_offset, - upper_32_bits(crtc_base)); - WREG32(mmGRPH_SECONDARY_SURFACE_ADDRESS + amdgpu_crtc->crtc_offset, - (u32)crtc_base); + /* update the primary scanout addresses */ WREG32(mmGRPH_PRIMARY_SURFACE_ADDRESS_HIGH + amdgpu_crtc->crtc_offset, upper_32_bits(crtc_base)); + /* writing to the low address triggers the update */ WREG32(mmGRPH_PRIMARY_SURFACE_ADDRESS + amdgpu_crtc->crtc_offset, - (u32)crtc_base); - - /* Wait for update_pending to go high. */ - for (i = 0; i < adev->usec_timeout; i++) { - if (RREG32(mmGRPH_UPDATE + amdgpu_crtc->crtc_offset) & - GRPH_UPDATE__GRPH_SURFACE_UPDATE_PENDING_MASK) - break; - udelay(1); - } - DRM_DEBUG("Update pending now high. Unlocking vupdate_lock.\n"); - - /* Unlock the lock, so double-buffering can take place inside vblank */ - tmp &= ~GRPH_UPDATE__GRPH_UPDATE_LOCK_MASK; - WREG32(mmGRPH_UPDATE + amdgpu_crtc->crtc_offset, tmp); + lower_32_bits(crtc_base)); + /* post the write */ + RREG32(mmGRPH_PRIMARY_SURFACE_ADDRESS + amdgpu_crtc->crtc_offset); } static int dce_v8_0_crtc_get_scanoutpos(struct amdgpu_device *adev, int crtc, @@ -1217,7 +1193,7 @@ static void dce_v8_0_program_watermarks(struct amdgpu_device *adev, u32 pixel_period; u32 line_time = 0; u32 latency_watermark_a = 0, latency_watermark_b = 0; - u32 tmp, wm_mask; + u32 tmp, wm_mask, lb_vblank_lead_lines = 0; if (amdgpu_crtc->base.enabled && num_heads && mode) { pixel_period = 1000000 / (u32)mode->clock; @@ -1300,6 +1276,7 @@ static void dce_v8_0_program_watermarks(struct amdgpu_device *adev, (adev->mode_info.disp_priority == 2)) { DRM_DEBUG_KMS("force priority to high\n"); } + lb_vblank_lead_lines = DIV_ROUND_UP(lb_size, mode->crtc_hdisplay); } /* select wm A */ @@ -1326,6 +1303,8 @@ static void dce_v8_0_program_watermarks(struct amdgpu_device *adev, amdgpu_crtc->line_time = line_time; amdgpu_crtc->wm_high = latency_watermark_a; amdgpu_crtc->wm_low = latency_watermark_b; + /* Save number of lines the linebuffer leads before the scanout */ + amdgpu_crtc->lb_vblank_lead_lines = lb_vblank_lead_lines; } /** @@ -1660,6 +1639,9 @@ static void dce_v8_0_audio_fini(struct amdgpu_device *adev) { int i; + if (!amdgpu_audio) + return; + if (!adev->mode_info.audio.enabled) return; @@ -1931,7 +1913,7 @@ static void dce_v8_0_afmt_enable(struct drm_encoder *encoder, bool enable) enable ? "En" : "Dis", dig->afmt->offset, amdgpu_encoder->encoder_id); } -static void dce_v8_0_afmt_init(struct amdgpu_device *adev) +static int dce_v8_0_afmt_init(struct amdgpu_device *adev) { int i; @@ -1944,8 +1926,16 @@ static void dce_v8_0_afmt_init(struct amdgpu_device *adev) if (adev->mode_info.afmt[i]) { adev->mode_info.afmt[i]->offset = dig_offsets[i]; adev->mode_info.afmt[i]->id = i; + } else { + int j; + for (j = 0; j < i; j++) { + kfree(adev->mode_info.afmt[j]); + adev->mode_info.afmt[j] = NULL; + } + return -ENOMEM; } } + return 0; } static void dce_v8_0_afmt_fini(struct amdgpu_device *adev) @@ -2022,8 +2012,7 @@ static int dce_v8_0_crtc_do_set_base(struct drm_crtc *crtc, if (atomic) { amdgpu_fb = to_amdgpu_framebuffer(fb); target_fb = fb; - } - else { + } else { amdgpu_fb = to_amdgpu_framebuffer(crtc->primary->fb); target_fb = crtc->primary->fb; } @@ -2037,9 +2026,9 @@ static int dce_v8_0_crtc_do_set_base(struct drm_crtc *crtc, if (unlikely(r != 0)) return r; - if (atomic) + if (atomic) { fb_location = amdgpu_bo_gpu_offset(rbo); - else { + } else { r = amdgpu_bo_pin(rbo, AMDGPU_GEM_DOMAIN_VRAM, &fb_location); if (unlikely(r != 0)) { amdgpu_bo_unreserve(rbo); @@ -2429,26 +2418,19 @@ static void dce_v8_0_show_cursor(struct drm_crtc *crtc) struct amdgpu_crtc *amdgpu_crtc = to_amdgpu_crtc(crtc); struct amdgpu_device *adev = crtc->dev->dev_private; + WREG32(mmCUR_SURFACE_ADDRESS_HIGH + amdgpu_crtc->crtc_offset, + upper_32_bits(amdgpu_crtc->cursor_addr)); + WREG32(mmCUR_SURFACE_ADDRESS + amdgpu_crtc->crtc_offset, + lower_32_bits(amdgpu_crtc->cursor_addr)); + WREG32_IDX(mmCUR_CONTROL + amdgpu_crtc->crtc_offset, CUR_CONTROL__CURSOR_EN_MASK | (CURSOR_24_8_PRE_MULT << CUR_CONTROL__CURSOR_MODE__SHIFT) | (CURSOR_URGENT_1_2 << CUR_CONTROL__CURSOR_URGENT_CONTROL__SHIFT)); } -static void dce_v8_0_set_cursor(struct drm_crtc *crtc, struct drm_gem_object *obj, - uint64_t gpu_addr) -{ - struct amdgpu_crtc *amdgpu_crtc = to_amdgpu_crtc(crtc); - struct amdgpu_device *adev = crtc->dev->dev_private; - - WREG32(mmCUR_SURFACE_ADDRESS_HIGH + amdgpu_crtc->crtc_offset, - upper_32_bits(gpu_addr)); - WREG32(mmCUR_SURFACE_ADDRESS + amdgpu_crtc->crtc_offset, - gpu_addr & 0xffffffff); -} - -static int dce_v8_0_crtc_cursor_move(struct drm_crtc *crtc, - int x, int y) +static int dce_v8_0_cursor_move_locked(struct drm_crtc *crtc, + int x, int y) { struct amdgpu_crtc *amdgpu_crtc = to_amdgpu_crtc(crtc); struct amdgpu_device *adev = crtc->dev->dev_private; @@ -2468,26 +2450,40 @@ static int dce_v8_0_crtc_cursor_move(struct drm_crtc *crtc, y = 0; } - dce_v8_0_lock_cursor(crtc, true); WREG32(mmCUR_POSITION + amdgpu_crtc->crtc_offset, (x << 16) | y); WREG32(mmCUR_HOT_SPOT + amdgpu_crtc->crtc_offset, (xorigin << 16) | yorigin); WREG32(mmCUR_SIZE + amdgpu_crtc->crtc_offset, ((amdgpu_crtc->cursor_width - 1) << 16) | (amdgpu_crtc->cursor_height - 1)); - dce_v8_0_lock_cursor(crtc, false); + + amdgpu_crtc->cursor_x = x; + amdgpu_crtc->cursor_y = y; return 0; } -static int dce_v8_0_crtc_cursor_set(struct drm_crtc *crtc, - struct drm_file *file_priv, - uint32_t handle, - uint32_t width, - uint32_t height) +static int dce_v8_0_crtc_cursor_move(struct drm_crtc *crtc, + int x, int y) +{ + int ret; + + dce_v8_0_lock_cursor(crtc, true); + ret = dce_v8_0_cursor_move_locked(crtc, x, y); + dce_v8_0_lock_cursor(crtc, false); + + return ret; +} + +static int dce_v8_0_crtc_cursor_set2(struct drm_crtc *crtc, + struct drm_file *file_priv, + uint32_t handle, + uint32_t width, + uint32_t height, + int32_t hot_x, + int32_t hot_y) { struct amdgpu_crtc *amdgpu_crtc = to_amdgpu_crtc(crtc); struct drm_gem_object *obj; - struct amdgpu_bo *robj; - uint64_t gpu_addr; + struct amdgpu_bo *aobj; int ret; if (!handle) { @@ -2509,41 +2505,71 @@ static int dce_v8_0_crtc_cursor_set(struct drm_crtc *crtc, return -ENOENT; } - robj = gem_to_amdgpu_bo(obj); - ret = amdgpu_bo_reserve(robj, false); - if (unlikely(ret != 0)) - goto fail; - ret = amdgpu_bo_pin_restricted(robj, AMDGPU_GEM_DOMAIN_VRAM, - 0, 0, &gpu_addr); - amdgpu_bo_unreserve(robj); - if (ret) - goto fail; + aobj = gem_to_amdgpu_bo(obj); + ret = amdgpu_bo_reserve(aobj, false); + if (ret != 0) { + drm_gem_object_unreference_unlocked(obj); + return ret; + } + + ret = amdgpu_bo_pin(aobj, AMDGPU_GEM_DOMAIN_VRAM, &amdgpu_crtc->cursor_addr); + amdgpu_bo_unreserve(aobj); + if (ret) { + DRM_ERROR("Failed to pin new cursor BO (%d)\n", ret); + drm_gem_object_unreference_unlocked(obj); + return ret; + } amdgpu_crtc->cursor_width = width; amdgpu_crtc->cursor_height = height; dce_v8_0_lock_cursor(crtc, true); - dce_v8_0_set_cursor(crtc, obj, gpu_addr); + + if (hot_x != amdgpu_crtc->cursor_hot_x || + hot_y != amdgpu_crtc->cursor_hot_y) { + int x, y; + + x = amdgpu_crtc->cursor_x + amdgpu_crtc->cursor_hot_x - hot_x; + y = amdgpu_crtc->cursor_y + amdgpu_crtc->cursor_hot_y - hot_y; + + dce_v8_0_cursor_move_locked(crtc, x, y); + + amdgpu_crtc->cursor_hot_x = hot_x; + amdgpu_crtc->cursor_hot_y = hot_y; + } + dce_v8_0_show_cursor(crtc); dce_v8_0_lock_cursor(crtc, false); unpin: if (amdgpu_crtc->cursor_bo) { - robj = gem_to_amdgpu_bo(amdgpu_crtc->cursor_bo); - ret = amdgpu_bo_reserve(robj, false); + struct amdgpu_bo *aobj = gem_to_amdgpu_bo(amdgpu_crtc->cursor_bo); + ret = amdgpu_bo_reserve(aobj, false); if (likely(ret == 0)) { - amdgpu_bo_unpin(robj); - amdgpu_bo_unreserve(robj); + amdgpu_bo_unpin(aobj); + amdgpu_bo_unreserve(aobj); } drm_gem_object_unreference_unlocked(amdgpu_crtc->cursor_bo); } amdgpu_crtc->cursor_bo = obj; return 0; -fail: - drm_gem_object_unreference_unlocked(obj); +} - return ret; +static void dce_v8_0_cursor_reset(struct drm_crtc *crtc) +{ + struct amdgpu_crtc *amdgpu_crtc = to_amdgpu_crtc(crtc); + + if (amdgpu_crtc->cursor_bo) { + dce_v8_0_lock_cursor(crtc, true); + + dce_v8_0_cursor_move_locked(crtc, amdgpu_crtc->cursor_x, + amdgpu_crtc->cursor_y); + + dce_v8_0_show_cursor(crtc); + + dce_v8_0_lock_cursor(crtc, false); + } } static void dce_v8_0_crtc_gamma_set(struct drm_crtc *crtc, u16 *red, u16 *green, @@ -2566,12 +2592,11 @@ static void dce_v8_0_crtc_destroy(struct drm_crtc *crtc) struct amdgpu_crtc *amdgpu_crtc = to_amdgpu_crtc(crtc); drm_crtc_cleanup(crtc); - destroy_workqueue(amdgpu_crtc->pflip_queue); kfree(amdgpu_crtc); } static const struct drm_crtc_funcs dce_v8_0_crtc_funcs = { - .cursor_set = dce_v8_0_crtc_cursor_set, + .cursor_set2 = dce_v8_0_crtc_cursor_set2, .cursor_move = dce_v8_0_crtc_cursor_move, .gamma_set = dce_v8_0_crtc_gamma_set, .set_config = amdgpu_crtc_set_config, @@ -2597,13 +2622,13 @@ static void dce_v8_0_crtc_dpms(struct drm_crtc *crtc, int mode) type = amdgpu_crtc_idx_to_irq_type(adev, amdgpu_crtc->crtc_id); amdgpu_irq_update(adev, &adev->crtc_irq, type); amdgpu_irq_update(adev, &adev->pageflip_irq, type); - drm_vblank_post_modeset(dev, amdgpu_crtc->crtc_id); + drm_vblank_on(dev, amdgpu_crtc->crtc_id); dce_v8_0_crtc_load_lut(crtc); break; case DRM_MODE_DPMS_STANDBY: case DRM_MODE_DPMS_SUSPEND: case DRM_MODE_DPMS_OFF: - drm_vblank_pre_modeset(dev, amdgpu_crtc->crtc_id); + drm_vblank_off(dev, amdgpu_crtc->crtc_id); if (amdgpu_crtc->enabled) { dce_v8_0_vga_enable(crtc, true); amdgpu_atombios_crtc_blank(crtc, ATOM_ENABLE); @@ -2712,6 +2737,7 @@ static int dce_v8_0_crtc_mode_set(struct drm_crtc *crtc, dce_v8_0_crtc_do_set_base(crtc, old_fb, x, y, 0); amdgpu_atombios_crtc_overscan_setup(crtc, mode, adjusted_mode); amdgpu_atombios_crtc_scaler_setup(crtc); + dce_v8_0_cursor_reset(crtc); /* update the hw version fpr dpm */ amdgpu_crtc->hw_mode = *adjusted_mode; @@ -2792,7 +2818,6 @@ static int dce_v8_0_crtc_init(struct amdgpu_device *adev, int index) drm_mode_crtc_set_gamma_size(&amdgpu_crtc->base, 256); amdgpu_crtc->crtc_id = index; - amdgpu_crtc->pflip_queue = create_singlethread_workqueue("amdgpu-pageflip-queue"); adev->mode_info.crtcs[index] = amdgpu_crtc; amdgpu_crtc->max_cursor_width = CIK_CURSOR_WIDTH; @@ -2875,8 +2900,6 @@ static int dce_v8_0_sw_init(void *handle) if (r) return r; - adev->mode_info.mode_config_initialized = true; - adev->ddev->mode_config.funcs = &amdgpu_mode_funcs; adev->ddev->mode_config.max_width = 16384; @@ -2907,7 +2930,9 @@ static int dce_v8_0_sw_init(void *handle) return -EINVAL; /* setup afmt */ - dce_v8_0_afmt_init(adev); + r = dce_v8_0_afmt_init(adev); + if (r) + return r; r = dce_v8_0_audio_init(adev); if (r) @@ -2915,7 +2940,8 @@ static int dce_v8_0_sw_init(void *handle) drm_kms_helper_poll_init(adev->ddev); - return r; + adev->mode_info.mode_config_initialized = true; + return 0; } static int dce_v8_0_sw_fini(void *handle) @@ -2979,22 +3005,18 @@ static int dce_v8_0_suspend(void *handle) amdgpu_atombios_scratch_regs_save(adev); - dce_v8_0_hpd_fini(adev); - - dce_v8_0_pageflip_interrupt_fini(adev); - - return 0; + return dce_v8_0_hw_fini(handle); } static int dce_v8_0_resume(void *handle) { struct amdgpu_device *adev = (struct amdgpu_device *)handle; + int ret; + + ret = dce_v8_0_hw_init(handle); amdgpu_atombios_scratch_regs_restore(adev); - /* init dig PHYs, disp eng pll */ - amdgpu_atombios_encoder_init_dig(adev); - amdgpu_atombios_crtc_set_disp_eng_pll(adev, adev->clock.default_dispclk); /* turn on the BL */ if (adev->mode_info.bl_encoder) { u8 bl_level = amdgpu_display_backlight_get_level(adev, @@ -3003,12 +3025,7 @@ static int dce_v8_0_resume(void *handle) bl_level); } - /* initialize hpd */ - dce_v8_0_hpd_init(adev); - - dce_v8_0_pageflip_interrupt_init(adev); - - return 0; + return ret; } static bool dce_v8_0_is_idle(void *handle) @@ -3301,37 +3318,20 @@ static int dce_v8_0_set_pageflip_interrupt_state(struct amdgpu_device *adev, unsigned type, enum amdgpu_interrupt_state state) { - u32 reg, reg_block; - /* now deal with page flip IRQ */ - switch (type) { - case AMDGPU_PAGEFLIP_IRQ_D1: - reg_block = CRTC0_REGISTER_OFFSET; - break; - case AMDGPU_PAGEFLIP_IRQ_D2: - reg_block = CRTC1_REGISTER_OFFSET; - break; - case AMDGPU_PAGEFLIP_IRQ_D3: - reg_block = CRTC2_REGISTER_OFFSET; - break; - case AMDGPU_PAGEFLIP_IRQ_D4: - reg_block = CRTC3_REGISTER_OFFSET; - break; - case AMDGPU_PAGEFLIP_IRQ_D5: - reg_block = CRTC4_REGISTER_OFFSET; - break; - case AMDGPU_PAGEFLIP_IRQ_D6: - reg_block = CRTC5_REGISTER_OFFSET; - break; - default: - DRM_ERROR("invalid pageflip crtc %d\n", type); - return -EINVAL; + u32 reg; + + if (type >= adev->mode_info.num_crtc) { + DRM_ERROR("invalid pageflip crtc %d\n", type); + return -EINVAL; } - reg = RREG32(mmGRPH_INTERRUPT_CONTROL + reg_block); + reg = RREG32(mmGRPH_INTERRUPT_CONTROL + crtc_offsets[type]); if (state == AMDGPU_IRQ_STATE_DISABLE) - WREG32(mmGRPH_INTERRUPT_CONTROL + reg_block, reg & ~GRPH_INTERRUPT_CONTROL__GRPH_PFLIP_INT_MASK_MASK); + WREG32(mmGRPH_INTERRUPT_CONTROL + crtc_offsets[type], + reg & ~GRPH_INTERRUPT_CONTROL__GRPH_PFLIP_INT_MASK_MASK); else - WREG32(mmGRPH_INTERRUPT_CONTROL + reg_block, reg | GRPH_INTERRUPT_CONTROL__GRPH_PFLIP_INT_MASK_MASK); + WREG32(mmGRPH_INTERRUPT_CONTROL + crtc_offsets[type], + reg | GRPH_INTERRUPT_CONTROL__GRPH_PFLIP_INT_MASK_MASK); return 0; } @@ -3340,7 +3340,6 @@ static int dce_v8_0_pageflip_irq(struct amdgpu_device *adev, struct amdgpu_irq_src *source, struct amdgpu_iv_entry *entry) { - int reg_block; unsigned long flags; unsigned crtc_id; struct amdgpu_crtc *amdgpu_crtc; @@ -3349,33 +3348,15 @@ static int dce_v8_0_pageflip_irq(struct amdgpu_device *adev, crtc_id = (entry->src_id - 8) >> 1; amdgpu_crtc = adev->mode_info.crtcs[crtc_id]; - /* ack the interrupt */ - switch(crtc_id){ - case AMDGPU_PAGEFLIP_IRQ_D1: - reg_block = CRTC0_REGISTER_OFFSET; - break; - case AMDGPU_PAGEFLIP_IRQ_D2: - reg_block = CRTC1_REGISTER_OFFSET; - break; - case AMDGPU_PAGEFLIP_IRQ_D3: - reg_block = CRTC2_REGISTER_OFFSET; - break; - case AMDGPU_PAGEFLIP_IRQ_D4: - reg_block = CRTC3_REGISTER_OFFSET; - break; - case AMDGPU_PAGEFLIP_IRQ_D5: - reg_block = CRTC4_REGISTER_OFFSET; - break; - case AMDGPU_PAGEFLIP_IRQ_D6: - reg_block = CRTC5_REGISTER_OFFSET; - break; - default: - DRM_ERROR("invalid pageflip crtc %d\n", crtc_id); - return -EINVAL; + if (crtc_id >= adev->mode_info.num_crtc) { + DRM_ERROR("invalid pageflip crtc %d\n", crtc_id); + return -EINVAL; } - if (RREG32(mmGRPH_INTERRUPT_STATUS + reg_block) & GRPH_INTERRUPT_STATUS__GRPH_PFLIP_INT_OCCURRED_MASK) - WREG32(mmGRPH_INTERRUPT_STATUS + reg_block, GRPH_INTERRUPT_STATUS__GRPH_PFLIP_INT_CLEAR_MASK); + if (RREG32(mmGRPH_INTERRUPT_STATUS + crtc_offsets[crtc_id]) & + GRPH_INTERRUPT_STATUS__GRPH_PFLIP_INT_OCCURRED_MASK) + WREG32(mmGRPH_INTERRUPT_STATUS + crtc_offsets[crtc_id], + GRPH_INTERRUPT_STATUS__GRPH_PFLIP_INT_CLEAR_MASK); /* IRQ could occur when in initial stage */ if (amdgpu_crtc == NULL) @@ -3403,7 +3384,7 @@ static int dce_v8_0_pageflip_irq(struct amdgpu_device *adev, spin_unlock_irqrestore(&adev->ddev->event_lock, flags); drm_vblank_put(adev->ddev, amdgpu_crtc->crtc_id); - queue_work(amdgpu_crtc->pflip_queue, &works->unpin_work); + schedule_work(&works->unpin_work); return 0; } @@ -3582,16 +3563,8 @@ dce_v8_0_ext_dpms(struct drm_encoder *encoder, int mode) } -static bool dce_v8_0_ext_mode_fixup(struct drm_encoder *encoder, - const struct drm_display_mode *mode, - struct drm_display_mode *adjusted_mode) -{ - return true; -} - static const struct drm_encoder_helper_funcs dce_v8_0_ext_helper_funcs = { .dpms = dce_v8_0_ext_dpms, - .mode_fixup = dce_v8_0_ext_mode_fixup, .prepare = dce_v8_0_ext_prepare, .mode_set = dce_v8_0_ext_mode_set, .commit = dce_v8_0_ext_commit, @@ -3687,7 +3660,7 @@ static void dce_v8_0_encoder_add(struct amdgpu_device *adev, case ENCODER_OBJECT_ID_INTERNAL_KLDSCP_DAC1: case ENCODER_OBJECT_ID_INTERNAL_KLDSCP_DAC2: drm_encoder_init(dev, encoder, &dce_v8_0_encoder_funcs, - DRM_MODE_ENCODER_DAC); + DRM_MODE_ENCODER_DAC, NULL); drm_encoder_helper_add(encoder, &dce_v8_0_dac_helper_funcs); break; case ENCODER_OBJECT_ID_INTERNAL_KLDSCP_DVO1: @@ -3698,15 +3671,15 @@ static void dce_v8_0_encoder_add(struct amdgpu_device *adev, if (amdgpu_encoder->devices & (ATOM_DEVICE_LCD_SUPPORT)) { amdgpu_encoder->rmx_type = RMX_FULL; drm_encoder_init(dev, encoder, &dce_v8_0_encoder_funcs, - DRM_MODE_ENCODER_LVDS); + DRM_MODE_ENCODER_LVDS, NULL); amdgpu_encoder->enc_priv = amdgpu_atombios_encoder_get_lcd_info(amdgpu_encoder); } else if (amdgpu_encoder->devices & (ATOM_DEVICE_CRT_SUPPORT)) { drm_encoder_init(dev, encoder, &dce_v8_0_encoder_funcs, - DRM_MODE_ENCODER_DAC); + DRM_MODE_ENCODER_DAC, NULL); amdgpu_encoder->enc_priv = amdgpu_atombios_encoder_get_dig_info(amdgpu_encoder); } else { drm_encoder_init(dev, encoder, &dce_v8_0_encoder_funcs, - DRM_MODE_ENCODER_TMDS); + DRM_MODE_ENCODER_TMDS, NULL); amdgpu_encoder->enc_priv = amdgpu_atombios_encoder_get_dig_info(amdgpu_encoder); } drm_encoder_helper_add(encoder, &dce_v8_0_dig_helper_funcs); @@ -3724,13 +3697,13 @@ static void dce_v8_0_encoder_add(struct amdgpu_device *adev, amdgpu_encoder->is_ext_encoder = true; if (amdgpu_encoder->devices & (ATOM_DEVICE_LCD_SUPPORT)) drm_encoder_init(dev, encoder, &dce_v8_0_encoder_funcs, - DRM_MODE_ENCODER_LVDS); + DRM_MODE_ENCODER_LVDS, NULL); else if (amdgpu_encoder->devices & (ATOM_DEVICE_CRT_SUPPORT)) drm_encoder_init(dev, encoder, &dce_v8_0_encoder_funcs, - DRM_MODE_ENCODER_DAC); + DRM_MODE_ENCODER_DAC, NULL); else drm_encoder_init(dev, encoder, &dce_v8_0_encoder_funcs, - DRM_MODE_ENCODER_TMDS); + DRM_MODE_ENCODER_TMDS, NULL); drm_encoder_helper_add(encoder, &dce_v8_0_ext_helper_funcs); break; } diff --git a/drivers/gpu/drm/amd/amdgpu/fiji_dpm.c b/drivers/gpu/drm/amd/amdgpu/fiji_dpm.c index 8f9845d9a986..4b0e45a27129 100644 --- a/drivers/gpu/drm/amd/amdgpu/fiji_dpm.c +++ b/drivers/gpu/drm/amd/amdgpu/fiji_dpm.c @@ -24,7 +24,7 @@ #include <linux/firmware.h> #include "drmP.h" #include "amdgpu.h" -#include "fiji_smumgr.h" +#include "fiji_smum.h" MODULE_FIRMWARE("amdgpu/fiji_smc.bin"); diff --git a/drivers/gpu/drm/amd/amdgpu/fiji_ppsmc.h b/drivers/gpu/drm/amd/amdgpu/fiji_ppsmc.h deleted file mode 100644 index 3c4824082990..000000000000 --- a/drivers/gpu/drm/amd/amdgpu/fiji_ppsmc.h +++ /dev/null @@ -1,182 +0,0 @@ -/* - * Copyright 2014 Advanced Micro Devices, Inc. - * - * Permission is hereby granted, free of charge, to any person obtaining a - * copy of this software and associated documentation files (the "Software"), - * to deal in the Software without restriction, including without limitation - * the rights to use, copy, modify, merge, publish, distribute, sublicense, - * and/or sell copies of the Software, and to permit persons to whom the - * Software is furnished to do so, subject to the following conditions: - * - * The above copyright notice and this permission notice shall be included in - * all copies or substantial portions of the Software. - * - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR - * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, - * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL - * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR - * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, - * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR - * OTHER DEALINGS IN THE SOFTWARE. - * - */ - -#ifndef FIJI_PP_SMC_H -#define FIJI_PP_SMC_H - -#pragma pack(push, 1) - -#define PPSMC_SWSTATE_FLAG_DC 0x01 -#define PPSMC_SWSTATE_FLAG_UVD 0x02 -#define PPSMC_SWSTATE_FLAG_VCE 0x04 - -#define PPSMC_THERMAL_PROTECT_TYPE_INTERNAL 0x00 -#define PPSMC_THERMAL_PROTECT_TYPE_EXTERNAL 0x01 -#define PPSMC_THERMAL_PROTECT_TYPE_NONE 0xff - -#define PPSMC_SYSTEMFLAG_GPIO_DC 0x01 -#define PPSMC_SYSTEMFLAG_STEPVDDC 0x02 -#define PPSMC_SYSTEMFLAG_GDDR5 0x04 - -#define PPSMC_SYSTEMFLAG_DISABLE_BABYSTEP 0x08 - -#define PPSMC_SYSTEMFLAG_REGULATOR_HOT 0x10 -#define PPSMC_SYSTEMFLAG_REGULATOR_HOT_ANALOG 0x20 - -#define PPSMC_EXTRAFLAGS_AC2DC_ACTION_MASK 0x07 -#define PPSMC_EXTRAFLAGS_AC2DC_DONT_WAIT_FOR_VBLANK 0x08 - -#define PPSMC_EXTRAFLAGS_AC2DC_ACTION_GOTODPMLOWSTATE 0x00 -#define PPSMC_EXTRAFLAGS_AC2DC_ACTION_GOTOINITIALSTATE 0x01 - -#define PPSMC_DPM2FLAGS_TDPCLMP 0x01 -#define PPSMC_DPM2FLAGS_PWRSHFT 0x02 -#define PPSMC_DPM2FLAGS_OCP 0x04 - -#define PPSMC_DISPLAY_WATERMARK_LOW 0 -#define PPSMC_DISPLAY_WATERMARK_HIGH 1 - -#define PPSMC_STATEFLAG_AUTO_PULSE_SKIP 0x01 -#define PPSMC_STATEFLAG_POWERBOOST 0x02 -#define PPSMC_STATEFLAG_PSKIP_ON_TDP_FAULT 0x04 -#define PPSMC_STATEFLAG_POWERSHIFT 0x08 -#define PPSMC_STATEFLAG_SLOW_READ_MARGIN 0x10 -#define PPSMC_STATEFLAG_DEEPSLEEP_THROTTLE 0x20 -#define PPSMC_STATEFLAG_DEEPSLEEP_BYPASS 0x40 - -#define FDO_MODE_HARDWARE 0 -#define FDO_MODE_PIECE_WISE_LINEAR 1 - -enum FAN_CONTROL { - FAN_CONTROL_FUZZY, - FAN_CONTROL_TABLE -}; - -//Gemini Modes -#define PPSMC_GeminiModeNone 0 //Single GPU board -#define PPSMC_GeminiModeMaster 1 //Master GPU on a Gemini board -#define PPSMC_GeminiModeSlave 2 //Slave GPU on a Gemini board - -#define PPSMC_Result_OK ((uint16_t)0x01) -#define PPSMC_Result_NoMore ((uint16_t)0x02) -#define PPSMC_Result_NotNow ((uint16_t)0x03) -#define PPSMC_Result_Failed ((uint16_t)0xFF) -#define PPSMC_Result_UnknownCmd ((uint16_t)0xFE) -#define PPSMC_Result_UnknownVT ((uint16_t)0xFD) - -typedef uint16_t PPSMC_Result; - -#define PPSMC_isERROR(x) ((uint16_t)0x80 & (x)) - -#define PPSMC_MSG_Halt ((uint16_t)0x10) -#define PPSMC_MSG_Resume ((uint16_t)0x11) -#define PPSMC_MSG_EnableDPMLevel ((uint16_t)0x12) -#define PPSMC_MSG_ZeroLevelsDisabled ((uint16_t)0x13) -#define PPSMC_MSG_OneLevelsDisabled ((uint16_t)0x14) -#define PPSMC_MSG_TwoLevelsDisabled ((uint16_t)0x15) -#define PPSMC_MSG_EnableThermalInterrupt ((uint16_t)0x16) -#define PPSMC_MSG_RunningOnAC ((uint16_t)0x17) -#define PPSMC_MSG_LevelUp ((uint16_t)0x18) -#define PPSMC_MSG_LevelDown ((uint16_t)0x19) -#define PPSMC_MSG_ResetDPMCounters ((uint16_t)0x1a) -#define PPSMC_MSG_SwitchToSwState ((uint16_t)0x20) -#define PPSMC_MSG_SwitchToSwStateLast ((uint16_t)0x3f) -#define PPSMC_MSG_SwitchToInitialState ((uint16_t)0x40) -#define PPSMC_MSG_NoForcedLevel ((uint16_t)0x41) -#define PPSMC_MSG_ForceHigh ((uint16_t)0x42) -#define PPSMC_MSG_ForceMediumOrHigh ((uint16_t)0x43) -#define PPSMC_MSG_SwitchToMinimumPower ((uint16_t)0x51) -#define PPSMC_MSG_ResumeFromMinimumPower ((uint16_t)0x52) -#define PPSMC_MSG_EnableCac ((uint16_t)0x53) -#define PPSMC_MSG_DisableCac ((uint16_t)0x54) -#define PPSMC_DPMStateHistoryStart ((uint16_t)0x55) -#define PPSMC_DPMStateHistoryStop ((uint16_t)0x56) -#define PPSMC_CACHistoryStart ((uint16_t)0x57) -#define PPSMC_CACHistoryStop ((uint16_t)0x58) -#define PPSMC_TDPClampingActive ((uint16_t)0x59) -#define PPSMC_TDPClampingInactive ((uint16_t)0x5A) -#define PPSMC_StartFanControl ((uint16_t)0x5B) -#define PPSMC_StopFanControl ((uint16_t)0x5C) -#define PPSMC_NoDisplay ((uint16_t)0x5D) -#define PPSMC_HasDisplay ((uint16_t)0x5E) -#define PPSMC_MSG_UVDPowerOFF ((uint16_t)0x60) -#define PPSMC_MSG_UVDPowerON ((uint16_t)0x61) -#define PPSMC_MSG_EnableULV ((uint16_t)0x62) -#define PPSMC_MSG_DisableULV ((uint16_t)0x63) -#define PPSMC_MSG_EnterULV ((uint16_t)0x64) -#define PPSMC_MSG_ExitULV ((uint16_t)0x65) -#define PPSMC_PowerShiftActive ((uint16_t)0x6A) -#define PPSMC_PowerShiftInactive ((uint16_t)0x6B) -#define PPSMC_OCPActive ((uint16_t)0x6C) -#define PPSMC_OCPInactive ((uint16_t)0x6D) -#define PPSMC_CACLongTermAvgEnable ((uint16_t)0x6E) -#define PPSMC_CACLongTermAvgDisable ((uint16_t)0x6F) -#define PPSMC_MSG_InferredStateSweep_Start ((uint16_t)0x70) -#define PPSMC_MSG_InferredStateSweep_Stop ((uint16_t)0x71) -#define PPSMC_MSG_SwitchToLowestInfState ((uint16_t)0x72) -#define PPSMC_MSG_SwitchToNonInfState ((uint16_t)0x73) -#define PPSMC_MSG_AllStateSweep_Start ((uint16_t)0x74) -#define PPSMC_MSG_AllStateSweep_Stop ((uint16_t)0x75) -#define PPSMC_MSG_SwitchNextLowerInfState ((uint16_t)0x76) -#define PPSMC_MSG_SwitchNextHigherInfState ((uint16_t)0x77) -#define PPSMC_MSG_MclkRetrainingTest ((uint16_t)0x78) -#define PPSMC_MSG_ForceTDPClamping ((uint16_t)0x79) -#define PPSMC_MSG_CollectCAC_PowerCorreln ((uint16_t)0x7A) -#define PPSMC_MSG_CollectCAC_WeightCalib ((uint16_t)0x7B) -#define PPSMC_MSG_CollectCAC_SQonly ((uint16_t)0x7C) -#define PPSMC_MSG_CollectCAC_TemperaturePwr ((uint16_t)0x7D) -#define PPSMC_MSG_ExtremitiesTest_Start ((uint16_t)0x7E) -#define PPSMC_MSG_ExtremitiesTest_Stop ((uint16_t)0x7F) -#define PPSMC_FlushDataCache ((uint16_t)0x80) -#define PPSMC_FlushInstrCache ((uint16_t)0x81) -#define PPSMC_MSG_SetEnabledLevels ((uint16_t)0x82) -#define PPSMC_MSG_SetForcedLevels ((uint16_t)0x83) -#define PPSMC_MSG_ResetToDefaults ((uint16_t)0x84) -#define PPSMC_MSG_SetForcedLevelsAndJump ((uint16_t)0x85) -#define PPSMC_MSG_SetCACHistoryMode ((uint16_t)0x86) -#define PPSMC_MSG_EnableDTE ((uint16_t)0x87) -#define PPSMC_MSG_DisableDTE ((uint16_t)0x88) -#define PPSMC_MSG_SmcSpaceSetAddress ((uint16_t)0x89) -#define PPSMC_MSG_SmcSpaceWriteDWordInc ((uint16_t)0x8A) -#define PPSMC_MSG_SmcSpaceWriteWordInc ((uint16_t)0x8B) -#define PPSMC_MSG_SmcSpaceWriteByteInc ((uint16_t)0x8C) - -#define PPSMC_MSG_BREAK ((uint16_t)0xF8) - -#define PPSMC_MSG_Test ((uint16_t)0x100) -#define PPSMC_MSG_DRV_DRAM_ADDR_HI ((uint16_t)0x250) -#define PPSMC_MSG_DRV_DRAM_ADDR_LO ((uint16_t)0x251) -#define PPSMC_MSG_SMU_DRAM_ADDR_HI ((uint16_t)0x252) -#define PPSMC_MSG_SMU_DRAM_ADDR_LO ((uint16_t)0x253) -#define PPSMC_MSG_LoadUcodes ((uint16_t)0x254) - -typedef uint16_t PPSMC_Msg; - -#define PPSMC_EVENT_STATUS_THERMAL 0x00000001 -#define PPSMC_EVENT_STATUS_REGULATORHOT 0x00000002 -#define PPSMC_EVENT_STATUS_DC 0x00000004 -#define PPSMC_EVENT_STATUS_GPIO17 0x00000008 - -#pragma pack(pop) - -#endif diff --git a/drivers/gpu/drm/amd/amdgpu/fiji_smc.c b/drivers/gpu/drm/amd/amdgpu/fiji_smc.c index bda1249eb871..b336c918d6a7 100644 --- a/drivers/gpu/drm/amd/amdgpu/fiji_smc.c +++ b/drivers/gpu/drm/amd/amdgpu/fiji_smc.c @@ -25,7 +25,7 @@ #include "drmP.h" #include "amdgpu.h" #include "fiji_ppsmc.h" -#include "fiji_smumgr.h" +#include "fiji_smum.h" #include "smu_ucode_xfer_vi.h" #include "amdgpu_ucode.h" @@ -272,6 +272,12 @@ static int fiji_smu_upload_firmware_image(struct amdgpu_device *adev) if (!adev->pm.fw) return -EINVAL; + /* Skip SMC ucode loading on SR-IOV capable boards. + * vbios does this for us in asic_init in that case. + */ + if (adev->virtualization.supports_sr_iov) + return 0; + hdr = (const struct smc_firmware_header_v1_0 *)adev->pm.fw->data; amdgpu_ucode_print_smc_hdr(&hdr->header); diff --git a/drivers/gpu/drm/amd/amdgpu/fiji_smumgr.h b/drivers/gpu/drm/amd/amdgpu/fiji_smum.h index 1cef03deeac3..1cef03deeac3 100644 --- a/drivers/gpu/drm/amd/amdgpu/fiji_smumgr.h +++ b/drivers/gpu/drm/amd/amdgpu/fiji_smum.h diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c index e992bf2ff66c..bb8709066fd8 100644 --- a/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c @@ -31,8 +31,6 @@ #include "amdgpu_ucode.h" #include "clearstate_ci.h" -#include "uvd/uvd_4_2_d.h" - #include "dce/dce_8_0_d.h" #include "dce/dce_8_0_sh_mask.h" @@ -1006,9 +1004,15 @@ out: */ static void gfx_v7_0_tiling_mode_table_init(struct amdgpu_device *adev) { - const u32 num_tile_mode_states = 32; - const u32 num_secondary_tile_mode_states = 16; - u32 reg_offset, gb_tile_moden, split_equal_to_row_size; + const u32 num_tile_mode_states = + ARRAY_SIZE(adev->gfx.config.tile_mode_array); + const u32 num_secondary_tile_mode_states = + ARRAY_SIZE(adev->gfx.config.macrotile_mode_array); + u32 reg_offset, split_equal_to_row_size; + uint32_t *tile, *macrotile; + + tile = adev->gfx.config.tile_mode_array; + macrotile = adev->gfx.config.macrotile_mode_array; switch (adev->gfx.config.mem_row_size_in_kb) { case 1: @@ -1023,832 +1027,531 @@ static void gfx_v7_0_tiling_mode_table_init(struct amdgpu_device *adev) break; } + for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++) + tile[reg_offset] = 0; + for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++) + macrotile[reg_offset] = 0; + switch (adev->asic_type) { case CHIP_BONAIRE: - for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++) { - switch (reg_offset) { - case 0: - gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P4_16x16) | - TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) | - MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); - break; - case 1: - gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P4_16x16) | - TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) | - MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); - break; - case 2: - gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P4_16x16) | - TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | - MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); - break; - case 3: - gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P4_16x16) | - TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) | - MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); - break; - case 4: - gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P4_16x16) | - MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) | - TILE_SPLIT(split_equal_to_row_size)); - break; - case 5: - gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P4_16x16) | - MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); - break; - case 6: - gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P4_16x16) | - MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) | - TILE_SPLIT(split_equal_to_row_size)); - break; - case 7: - gb_tile_moden = (TILE_SPLIT(split_equal_to_row_size)); - break; - - case 8: - gb_tile_moden = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) | - PIPE_CONFIG(ADDR_SURF_P4_16x16)); - break; - case 9: - gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P4_16x16) | - MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING)); - break; - case 10: - gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P4_16x16) | - MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); - break; - case 11: - gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P4_16x16) | - MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); - break; - case 12: - gb_tile_moden = (TILE_SPLIT(split_equal_to_row_size)); - break; - case 13: - gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P4_16x16) | - MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING)); - break; - case 14: - gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P4_16x16) | - MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); - break; - case 15: - gb_tile_moden = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P4_16x16) | - MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); - break; - case 16: - gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P4_16x16) | - MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); - break; - case 17: - gb_tile_moden = (TILE_SPLIT(split_equal_to_row_size)); - break; - case 18: - gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THICK) | - PIPE_CONFIG(ADDR_SURF_P4_16x16) | - MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); - break; - case 19: - gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THICK) | - PIPE_CONFIG(ADDR_SURF_P4_16x16) | - MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING)); - break; - case 20: - gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THICK) | - PIPE_CONFIG(ADDR_SURF_P4_16x16) | - MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); - break; - case 21: - gb_tile_moden = (ARRAY_MODE(ARRAY_3D_TILED_THICK) | - PIPE_CONFIG(ADDR_SURF_P4_16x16) | - MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); - break; - case 22: - gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) | - PIPE_CONFIG(ADDR_SURF_P4_16x16) | - MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); - break; - case 23: - gb_tile_moden = (TILE_SPLIT(split_equal_to_row_size)); - break; - case 24: - gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THICK) | - PIPE_CONFIG(ADDR_SURF_P4_16x16) | - MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); - break; - case 25: - gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) | - PIPE_CONFIG(ADDR_SURF_P4_16x16) | - MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); - break; - case 26: - gb_tile_moden = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) | - PIPE_CONFIG(ADDR_SURF_P4_16x16) | - MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); - break; - case 27: - gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P4_16x16) | - MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING)); - break; - case 28: - gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P4_16x16) | - MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); - break; - case 29: - gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P4_16x16) | - MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); - break; - case 30: - gb_tile_moden = (TILE_SPLIT(split_equal_to_row_size)); - break; - default: - gb_tile_moden = 0; - break; - } - adev->gfx.config.tile_mode_array[reg_offset] = gb_tile_moden; - WREG32(mmGB_TILE_MODE0 + reg_offset, gb_tile_moden); - } - for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++) { - switch (reg_offset) { - case 0: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | - NUM_BANKS(ADDR_SURF_16_BANK)); - break; - case 1: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | - NUM_BANKS(ADDR_SURF_16_BANK)); - break; - case 2: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | - NUM_BANKS(ADDR_SURF_16_BANK)); - break; - case 3: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | - NUM_BANKS(ADDR_SURF_16_BANK)); - break; - case 4: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | - NUM_BANKS(ADDR_SURF_16_BANK)); - break; - case 5: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | - NUM_BANKS(ADDR_SURF_8_BANK)); - break; - case 6: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | - NUM_BANKS(ADDR_SURF_4_BANK)); - break; - case 8: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | - NUM_BANKS(ADDR_SURF_16_BANK)); - break; - case 9: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | - NUM_BANKS(ADDR_SURF_16_BANK)); - break; - case 10: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | - NUM_BANKS(ADDR_SURF_16_BANK)); - break; - case 11: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | - NUM_BANKS(ADDR_SURF_16_BANK)); - break; - case 12: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | - NUM_BANKS(ADDR_SURF_16_BANK)); - break; - case 13: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | - NUM_BANKS(ADDR_SURF_8_BANK)); - break; - case 14: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | - NUM_BANKS(ADDR_SURF_4_BANK)); - break; - default: - gb_tile_moden = 0; - break; - } - adev->gfx.config.macrotile_mode_array[reg_offset] = gb_tile_moden; - WREG32(mmGB_MACROTILE_MODE0 + reg_offset, gb_tile_moden); - } + tile[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P4_16x16) | + TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); + tile[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P4_16x16) | + TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); + tile[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P4_16x16) | + TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); + tile[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P4_16x16) | + TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); + tile[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P4_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) | + TILE_SPLIT(split_equal_to_row_size)); + tile[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P4_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); + tile[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P4_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) | + TILE_SPLIT(split_equal_to_row_size)); + tile[7] = (TILE_SPLIT(split_equal_to_row_size)); + tile[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) | + PIPE_CONFIG(ADDR_SURF_P4_16x16)); + tile[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P4_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING)); + tile[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P4_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); + tile[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P4_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); + tile[12] = (TILE_SPLIT(split_equal_to_row_size)); + tile[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P4_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING)); + tile[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P4_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); + tile[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P4_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); + tile[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P4_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); + tile[17] = (TILE_SPLIT(split_equal_to_row_size)); + tile[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) | + PIPE_CONFIG(ADDR_SURF_P4_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); + tile[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) | + PIPE_CONFIG(ADDR_SURF_P4_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING)); + tile[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) | + PIPE_CONFIG(ADDR_SURF_P4_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); + tile[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) | + PIPE_CONFIG(ADDR_SURF_P4_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); + tile[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) | + PIPE_CONFIG(ADDR_SURF_P4_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); + tile[23] = (TILE_SPLIT(split_equal_to_row_size)); + tile[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) | + PIPE_CONFIG(ADDR_SURF_P4_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); + tile[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) | + PIPE_CONFIG(ADDR_SURF_P4_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); + tile[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) | + PIPE_CONFIG(ADDR_SURF_P4_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); + tile[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P4_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING)); + tile[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P4_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); + tile[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P4_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); + tile[30] = (TILE_SPLIT(split_equal_to_row_size)); + + macrotile[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | + NUM_BANKS(ADDR_SURF_16_BANK)); + macrotile[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | + NUM_BANKS(ADDR_SURF_16_BANK)); + macrotile[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | + NUM_BANKS(ADDR_SURF_16_BANK)); + macrotile[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | + NUM_BANKS(ADDR_SURF_16_BANK)); + macrotile[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | + NUM_BANKS(ADDR_SURF_16_BANK)); + macrotile[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | + NUM_BANKS(ADDR_SURF_8_BANK)); + macrotile[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | + NUM_BANKS(ADDR_SURF_4_BANK)); + macrotile[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | + NUM_BANKS(ADDR_SURF_16_BANK)); + macrotile[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | + NUM_BANKS(ADDR_SURF_16_BANK)); + macrotile[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | + NUM_BANKS(ADDR_SURF_16_BANK)); + macrotile[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | + NUM_BANKS(ADDR_SURF_16_BANK)); + macrotile[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | + NUM_BANKS(ADDR_SURF_16_BANK)); + macrotile[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | + NUM_BANKS(ADDR_SURF_8_BANK)); + macrotile[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | + NUM_BANKS(ADDR_SURF_4_BANK)); + + for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++) + WREG32(mmGB_TILE_MODE0 + reg_offset, tile[reg_offset]); + for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++) + if (reg_offset != 7) + WREG32(mmGB_MACROTILE_MODE0 + reg_offset, macrotile[reg_offset]); break; case CHIP_HAWAII: - for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++) { - switch (reg_offset) { - case 0: - gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | - TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) | - MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); - break; - case 1: - gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | - TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) | - MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); - break; - case 2: - gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | - TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | - MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); - break; - case 3: - gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | - TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) | - MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); - break; - case 4: - gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | - MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) | - TILE_SPLIT(split_equal_to_row_size)); - break; - case 5: - gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | - MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) | - TILE_SPLIT(split_equal_to_row_size)); - break; - case 6: - gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | - MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) | - TILE_SPLIT(split_equal_to_row_size)); - break; - case 7: - gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P4_16x16) | - MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) | - TILE_SPLIT(split_equal_to_row_size)); - break; - - case 8: - gb_tile_moden = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) | - PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16)); - break; - case 9: - gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | - MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING)); - break; - case 10: - gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | - MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); - break; - case 11: - gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | - MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); - break; - case 12: - gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P4_16x16) | - MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); - break; - case 13: - gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | - MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING)); - break; - case 14: - gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | - MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); - break; - case 15: - gb_tile_moden = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | - MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); - break; - case 16: - gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | - MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); - break; - case 17: - gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P4_16x16) | - MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); - break; - case 18: - gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THICK) | - PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | - MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); - break; - case 19: - gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THICK) | - PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | - MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING)); - break; - case 20: - gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THICK) | - PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | - MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); - break; - case 21: - gb_tile_moden = (ARRAY_MODE(ARRAY_3D_TILED_THICK) | - PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | - MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); - break; - case 22: - gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) | - PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | - MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); - break; - case 23: - gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) | - PIPE_CONFIG(ADDR_SURF_P4_16x16) | - MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); - break; - case 24: - gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THICK) | - PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | - MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); - break; - case 25: - gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) | - PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | - MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); - break; - case 26: - gb_tile_moden = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) | - PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | - MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); - break; - case 27: - gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | - MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING)); - break; - case 28: - gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | - MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); - break; - case 29: - gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | - MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); - break; - case 30: - gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P4_16x16) | - MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); - break; - default: - gb_tile_moden = 0; - break; - } - adev->gfx.config.tile_mode_array[reg_offset] = gb_tile_moden; - WREG32(mmGB_TILE_MODE0 + reg_offset, gb_tile_moden); - } - for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++) { - switch (reg_offset) { - case 0: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | - NUM_BANKS(ADDR_SURF_16_BANK)); - break; - case 1: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | - NUM_BANKS(ADDR_SURF_16_BANK)); - break; - case 2: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | - NUM_BANKS(ADDR_SURF_16_BANK)); - break; - case 3: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | - NUM_BANKS(ADDR_SURF_16_BANK)); - break; - case 4: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | - NUM_BANKS(ADDR_SURF_8_BANK)); - break; - case 5: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | - NUM_BANKS(ADDR_SURF_4_BANK)); - break; - case 6: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | - NUM_BANKS(ADDR_SURF_4_BANK)); - break; - case 8: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | - NUM_BANKS(ADDR_SURF_16_BANK)); - break; - case 9: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | - NUM_BANKS(ADDR_SURF_16_BANK)); - break; - case 10: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | - NUM_BANKS(ADDR_SURF_16_BANK)); - break; - case 11: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | - NUM_BANKS(ADDR_SURF_8_BANK)); - break; - case 12: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | - NUM_BANKS(ADDR_SURF_16_BANK)); - break; - case 13: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | - NUM_BANKS(ADDR_SURF_8_BANK)); - break; - case 14: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | - NUM_BANKS(ADDR_SURF_4_BANK)); - break; - default: - gb_tile_moden = 0; - break; - } - adev->gfx.config.macrotile_mode_array[reg_offset] = gb_tile_moden; - WREG32(mmGB_MACROTILE_MODE0 + reg_offset, gb_tile_moden); - } + tile[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | + TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); + tile[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | + TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); + tile[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | + TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); + tile[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | + TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); + tile[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) | + TILE_SPLIT(split_equal_to_row_size)); + tile[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) | + TILE_SPLIT(split_equal_to_row_size)); + tile[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) | + TILE_SPLIT(split_equal_to_row_size)); + tile[7] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P4_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) | + TILE_SPLIT(split_equal_to_row_size)); + tile[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) | + PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16)); + tile[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING)); + tile[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); + tile[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); + tile[12] = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P4_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); + tile[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING)); + tile[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); + tile[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); + tile[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); + tile[17] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P4_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); + tile[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) | + PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); + tile[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) | + PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING)); + tile[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) | + PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); + tile[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) | + PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); + tile[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) | + PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); + tile[23] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) | + PIPE_CONFIG(ADDR_SURF_P4_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); + tile[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) | + PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); + tile[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) | + PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); + tile[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) | + PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); + tile[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING)); + tile[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); + tile[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); + tile[30] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P4_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); + + macrotile[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | + NUM_BANKS(ADDR_SURF_16_BANK)); + macrotile[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | + NUM_BANKS(ADDR_SURF_16_BANK)); + macrotile[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | + NUM_BANKS(ADDR_SURF_16_BANK)); + macrotile[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | + NUM_BANKS(ADDR_SURF_16_BANK)); + macrotile[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | + NUM_BANKS(ADDR_SURF_8_BANK)); + macrotile[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | + NUM_BANKS(ADDR_SURF_4_BANK)); + macrotile[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | + NUM_BANKS(ADDR_SURF_4_BANK)); + macrotile[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | + NUM_BANKS(ADDR_SURF_16_BANK)); + macrotile[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | + NUM_BANKS(ADDR_SURF_16_BANK)); + macrotile[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | + NUM_BANKS(ADDR_SURF_16_BANK)); + macrotile[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | + NUM_BANKS(ADDR_SURF_8_BANK)); + macrotile[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | + NUM_BANKS(ADDR_SURF_16_BANK)); + macrotile[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | + NUM_BANKS(ADDR_SURF_8_BANK)); + macrotile[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | + NUM_BANKS(ADDR_SURF_4_BANK)); + + for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++) + WREG32(mmGB_TILE_MODE0 + reg_offset, tile[reg_offset]); + for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++) + if (reg_offset != 7) + WREG32(mmGB_MACROTILE_MODE0 + reg_offset, macrotile[reg_offset]); break; case CHIP_KABINI: case CHIP_KAVERI: case CHIP_MULLINS: default: - for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++) { - switch (reg_offset) { - case 0: - gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P2) | - TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) | - MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); - break; - case 1: - gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P2) | - TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) | - MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); - break; - case 2: - gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P2) | - TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | - MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); - break; - case 3: - gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P2) | - TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) | - MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); - break; - case 4: - gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P2) | - MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) | - TILE_SPLIT(split_equal_to_row_size)); - break; - case 5: - gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P2) | - MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); - break; - case 6: - gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P2) | - MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) | - TILE_SPLIT(split_equal_to_row_size)); - break; - case 7: - gb_tile_moden = (TILE_SPLIT(split_equal_to_row_size)); - break; - - case 8: - gb_tile_moden = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) | - PIPE_CONFIG(ADDR_SURF_P2)); - break; - case 9: - gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P2) | - MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING)); - break; - case 10: - gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P2) | - MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); - break; - case 11: - gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P2) | - MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); - break; - case 12: - gb_tile_moden = (TILE_SPLIT(split_equal_to_row_size)); - break; - case 13: - gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P2) | - MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING)); - break; - case 14: - gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P2) | - MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); - break; - case 15: - gb_tile_moden = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P2) | - MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); - break; - case 16: - gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P2) | - MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); - break; - case 17: - gb_tile_moden = (TILE_SPLIT(split_equal_to_row_size)); - break; - case 18: - gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THICK) | - PIPE_CONFIG(ADDR_SURF_P2) | - MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); - break; - case 19: - gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THICK) | - PIPE_CONFIG(ADDR_SURF_P2) | - MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING)); - break; - case 20: - gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THICK) | - PIPE_CONFIG(ADDR_SURF_P2) | - MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); - break; - case 21: - gb_tile_moden = (ARRAY_MODE(ARRAY_3D_TILED_THICK) | - PIPE_CONFIG(ADDR_SURF_P2) | - MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); - break; - case 22: - gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) | - PIPE_CONFIG(ADDR_SURF_P2) | - MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); - break; - case 23: - gb_tile_moden = (TILE_SPLIT(split_equal_to_row_size)); - break; - case 24: - gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THICK) | - PIPE_CONFIG(ADDR_SURF_P2) | - MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); - break; - case 25: - gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) | - PIPE_CONFIG(ADDR_SURF_P2) | - MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); - break; - case 26: - gb_tile_moden = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) | - PIPE_CONFIG(ADDR_SURF_P2) | - MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); - break; - case 27: - gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P2) | - MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING)); - break; - case 28: - gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P2) | - MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); - break; - case 29: - gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P2) | - MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); - break; - case 30: - gb_tile_moden = (TILE_SPLIT(split_equal_to_row_size)); - break; - default: - gb_tile_moden = 0; - break; - } - adev->gfx.config.tile_mode_array[reg_offset] = gb_tile_moden; - WREG32(mmGB_TILE_MODE0 + reg_offset, gb_tile_moden); - } - for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++) { - switch (reg_offset) { - case 0: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | - NUM_BANKS(ADDR_SURF_8_BANK)); - break; - case 1: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | - NUM_BANKS(ADDR_SURF_8_BANK)); - break; - case 2: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | - NUM_BANKS(ADDR_SURF_8_BANK)); - break; - case 3: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | - NUM_BANKS(ADDR_SURF_8_BANK)); - break; - case 4: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | - NUM_BANKS(ADDR_SURF_8_BANK)); - break; - case 5: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | - NUM_BANKS(ADDR_SURF_8_BANK)); - break; - case 6: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | - NUM_BANKS(ADDR_SURF_8_BANK)); - break; - case 8: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | - NUM_BANKS(ADDR_SURF_16_BANK)); - break; - case 9: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | - NUM_BANKS(ADDR_SURF_16_BANK)); - break; - case 10: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | - NUM_BANKS(ADDR_SURF_16_BANK)); - break; - case 11: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | - NUM_BANKS(ADDR_SURF_16_BANK)); - break; - case 12: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | - NUM_BANKS(ADDR_SURF_16_BANK)); - break; - case 13: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | - NUM_BANKS(ADDR_SURF_16_BANK)); - break; - case 14: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | - NUM_BANKS(ADDR_SURF_8_BANK)); - break; - default: - gb_tile_moden = 0; - break; - } - adev->gfx.config.macrotile_mode_array[reg_offset] = gb_tile_moden; - WREG32(mmGB_MACROTILE_MODE0 + reg_offset, gb_tile_moden); - } + tile[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P2) | + TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); + tile[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P2) | + TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); + tile[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P2) | + TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); + tile[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P2) | + TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); + tile[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P2) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) | + TILE_SPLIT(split_equal_to_row_size)); + tile[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P2) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); + tile[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P2) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) | + TILE_SPLIT(split_equal_to_row_size)); + tile[7] = (TILE_SPLIT(split_equal_to_row_size)); + tile[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) | + PIPE_CONFIG(ADDR_SURF_P2)); + tile[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P2) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING)); + tile[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P2) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); + tile[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P2) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); + tile[12] = (TILE_SPLIT(split_equal_to_row_size)); + tile[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P2) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING)); + tile[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P2) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); + tile[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P2) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); + tile[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P2) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); + tile[17] = (TILE_SPLIT(split_equal_to_row_size)); + tile[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) | + PIPE_CONFIG(ADDR_SURF_P2) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); + tile[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) | + PIPE_CONFIG(ADDR_SURF_P2) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING)); + tile[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) | + PIPE_CONFIG(ADDR_SURF_P2) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); + tile[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) | + PIPE_CONFIG(ADDR_SURF_P2) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); + tile[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) | + PIPE_CONFIG(ADDR_SURF_P2) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); + tile[23] = (TILE_SPLIT(split_equal_to_row_size)); + tile[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) | + PIPE_CONFIG(ADDR_SURF_P2) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); + tile[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) | + PIPE_CONFIG(ADDR_SURF_P2) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); + tile[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) | + PIPE_CONFIG(ADDR_SURF_P2) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); + tile[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P2) | + MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING)); + tile[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P2) | + MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); + tile[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P2) | + MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); + tile[30] = (TILE_SPLIT(split_equal_to_row_size)); + + macrotile[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | + NUM_BANKS(ADDR_SURF_8_BANK)); + macrotile[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | + NUM_BANKS(ADDR_SURF_8_BANK)); + macrotile[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | + NUM_BANKS(ADDR_SURF_8_BANK)); + macrotile[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | + NUM_BANKS(ADDR_SURF_8_BANK)); + macrotile[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | + NUM_BANKS(ADDR_SURF_8_BANK)); + macrotile[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | + NUM_BANKS(ADDR_SURF_8_BANK)); + macrotile[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | + NUM_BANKS(ADDR_SURF_8_BANK)); + macrotile[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | + NUM_BANKS(ADDR_SURF_16_BANK)); + macrotile[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | + NUM_BANKS(ADDR_SURF_16_BANK)); + macrotile[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | + NUM_BANKS(ADDR_SURF_16_BANK)); + macrotile[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | + NUM_BANKS(ADDR_SURF_16_BANK)); + macrotile[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | + NUM_BANKS(ADDR_SURF_16_BANK)); + macrotile[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | + NUM_BANKS(ADDR_SURF_16_BANK)); + macrotile[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | + NUM_BANKS(ADDR_SURF_8_BANK)); + + for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++) + WREG32(mmGB_TILE_MODE0 + reg_offset, tile[reg_offset]); + for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++) + if (reg_offset != 7) + WREG32(mmGB_MACROTILE_MODE0 + reg_offset, macrotile[reg_offset]); break; } } @@ -1893,45 +1596,31 @@ void gfx_v7_0_select_se_sh(struct amdgpu_device *adev, u32 se_num, u32 sh_num) */ static u32 gfx_v7_0_create_bitmask(u32 bit_width) { - u32 i, mask = 0; - - for (i = 0; i < bit_width; i++) { - mask <<= 1; - mask |= 1; - } - return mask; + return (u32)((1ULL << bit_width) - 1); } /** - * gfx_v7_0_get_rb_disabled - computes the mask of disabled RBs + * gfx_v7_0_get_rb_active_bitmap - computes the mask of enabled RBs * * @adev: amdgpu_device pointer - * @max_rb_num: max RBs (render backends) for the asic - * @se_num: number of SEs (shader engines) for the asic - * @sh_per_se: number of SH blocks per SE for the asic * - * Calculates the bitmask of disabled RBs (CIK). - * Returns the disabled RB bitmask. + * Calculates the bitmask of enabled RBs (CIK). + * Returns the enabled RB bitmask. */ -static u32 gfx_v7_0_get_rb_disabled(struct amdgpu_device *adev, - u32 max_rb_num_per_se, - u32 sh_per_se) +static u32 gfx_v7_0_get_rb_active_bitmap(struct amdgpu_device *adev) { u32 data, mask; data = RREG32(mmCC_RB_BACKEND_DISABLE); - if (data & 1) - data &= CC_RB_BACKEND_DISABLE__BACKEND_DISABLE_MASK; - else - data = 0; - data |= RREG32(mmGC_USER_RB_BACKEND_DISABLE); + data &= CC_RB_BACKEND_DISABLE__BACKEND_DISABLE_MASK; data >>= GC_USER_RB_BACKEND_DISABLE__BACKEND_DISABLE__SHIFT; - mask = gfx_v7_0_create_bitmask(max_rb_num_per_se / sh_per_se); + mask = gfx_v7_0_create_bitmask(adev->gfx.config.max_backends_per_se / + adev->gfx.config.max_sh_per_se); - return data & mask; + return (~data) & mask; } /** @@ -1940,73 +1629,31 @@ static u32 gfx_v7_0_get_rb_disabled(struct amdgpu_device *adev, * @adev: amdgpu_device pointer * @se_num: number of SEs (shader engines) for the asic * @sh_per_se: number of SH blocks per SE for the asic - * @max_rb_num: max RBs (render backends) for the asic * * Configures per-SE/SH RB registers (CIK). */ -static void gfx_v7_0_setup_rb(struct amdgpu_device *adev, - u32 se_num, u32 sh_per_se, - u32 max_rb_num_per_se) +static void gfx_v7_0_setup_rb(struct amdgpu_device *adev) { int i, j; - u32 data, mask; - u32 disabled_rbs = 0; - u32 enabled_rbs = 0; + u32 data; + u32 active_rbs = 0; + u32 rb_bitmap_width_per_sh = adev->gfx.config.max_backends_per_se / + adev->gfx.config.max_sh_per_se; mutex_lock(&adev->grbm_idx_mutex); - for (i = 0; i < se_num; i++) { - for (j = 0; j < sh_per_se; j++) { + for (i = 0; i < adev->gfx.config.max_shader_engines; i++) { + for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) { gfx_v7_0_select_se_sh(adev, i, j); - data = gfx_v7_0_get_rb_disabled(adev, max_rb_num_per_se, sh_per_se); - if (adev->asic_type == CHIP_HAWAII) - disabled_rbs |= data << ((i * sh_per_se + j) * HAWAII_RB_BITMAP_WIDTH_PER_SH); - else - disabled_rbs |= data << ((i * sh_per_se + j) * CIK_RB_BITMAP_WIDTH_PER_SH); + data = gfx_v7_0_get_rb_active_bitmap(adev); + active_rbs |= data << ((i * adev->gfx.config.max_sh_per_se + j) * + rb_bitmap_width_per_sh); } } gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff); mutex_unlock(&adev->grbm_idx_mutex); - mask = 1; - for (i = 0; i < max_rb_num_per_se * se_num; i++) { - if (!(disabled_rbs & mask)) - enabled_rbs |= mask; - mask <<= 1; - } - - adev->gfx.config.backend_enable_mask = enabled_rbs; - - mutex_lock(&adev->grbm_idx_mutex); - for (i = 0; i < se_num; i++) { - gfx_v7_0_select_se_sh(adev, i, 0xffffffff); - data = 0; - for (j = 0; j < sh_per_se; j++) { - switch (enabled_rbs & 3) { - case 0: - if (j == 0) - data |= (RASTER_CONFIG_RB_MAP_3 << - PA_SC_RASTER_CONFIG__PKR_MAP__SHIFT); - else - data |= (RASTER_CONFIG_RB_MAP_0 << - PA_SC_RASTER_CONFIG__PKR_MAP__SHIFT); - break; - case 1: - data |= (RASTER_CONFIG_RB_MAP_0 << (i * sh_per_se + j) * 2); - break; - case 2: - data |= (RASTER_CONFIG_RB_MAP_3 << (i * sh_per_se + j) * 2); - break; - case 3: - default: - data |= (RASTER_CONFIG_RB_MAP_2 << (i * sh_per_se + j) * 2); - break; - } - enabled_rbs >>= 2; - } - WREG32(mmPA_SC_RASTER_CONFIG, data); - } - gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff); - mutex_unlock(&adev->grbm_idx_mutex); + adev->gfx.config.backend_enable_mask = active_rbs; + adev->gfx.config.num_rbs = hweight32(active_rbs); } /** @@ -2059,192 +1706,23 @@ static void gmc_v7_0_init_compute_vmid(struct amdgpu_device *adev) */ static void gfx_v7_0_gpu_init(struct amdgpu_device *adev) { - u32 gb_addr_config; - u32 mc_shared_chmap, mc_arb_ramcfg; - u32 dimm00_addr_map, dimm01_addr_map, dimm10_addr_map, dimm11_addr_map; - u32 sh_mem_cfg; - u32 tmp; + u32 tmp, sh_mem_cfg; int i; - switch (adev->asic_type) { - case CHIP_BONAIRE: - adev->gfx.config.max_shader_engines = 2; - adev->gfx.config.max_tile_pipes = 4; - adev->gfx.config.max_cu_per_sh = 7; - adev->gfx.config.max_sh_per_se = 1; - adev->gfx.config.max_backends_per_se = 2; - adev->gfx.config.max_texture_channel_caches = 4; - adev->gfx.config.max_gprs = 256; - adev->gfx.config.max_gs_threads = 32; - adev->gfx.config.max_hw_contexts = 8; - - adev->gfx.config.sc_prim_fifo_size_frontend = 0x20; - adev->gfx.config.sc_prim_fifo_size_backend = 0x100; - adev->gfx.config.sc_hiz_tile_fifo_size = 0x30; - adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130; - gb_addr_config = BONAIRE_GB_ADDR_CONFIG_GOLDEN; - break; - case CHIP_HAWAII: - adev->gfx.config.max_shader_engines = 4; - adev->gfx.config.max_tile_pipes = 16; - adev->gfx.config.max_cu_per_sh = 11; - adev->gfx.config.max_sh_per_se = 1; - adev->gfx.config.max_backends_per_se = 4; - adev->gfx.config.max_texture_channel_caches = 16; - adev->gfx.config.max_gprs = 256; - adev->gfx.config.max_gs_threads = 32; - adev->gfx.config.max_hw_contexts = 8; - - adev->gfx.config.sc_prim_fifo_size_frontend = 0x20; - adev->gfx.config.sc_prim_fifo_size_backend = 0x100; - adev->gfx.config.sc_hiz_tile_fifo_size = 0x30; - adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130; - gb_addr_config = HAWAII_GB_ADDR_CONFIG_GOLDEN; - break; - case CHIP_KAVERI: - adev->gfx.config.max_shader_engines = 1; - adev->gfx.config.max_tile_pipes = 4; - if ((adev->pdev->device == 0x1304) || - (adev->pdev->device == 0x1305) || - (adev->pdev->device == 0x130C) || - (adev->pdev->device == 0x130F) || - (adev->pdev->device == 0x1310) || - (adev->pdev->device == 0x1311) || - (adev->pdev->device == 0x131C)) { - adev->gfx.config.max_cu_per_sh = 8; - adev->gfx.config.max_backends_per_se = 2; - } else if ((adev->pdev->device == 0x1309) || - (adev->pdev->device == 0x130A) || - (adev->pdev->device == 0x130D) || - (adev->pdev->device == 0x1313) || - (adev->pdev->device == 0x131D)) { - adev->gfx.config.max_cu_per_sh = 6; - adev->gfx.config.max_backends_per_se = 2; - } else if ((adev->pdev->device == 0x1306) || - (adev->pdev->device == 0x1307) || - (adev->pdev->device == 0x130B) || - (adev->pdev->device == 0x130E) || - (adev->pdev->device == 0x1315) || - (adev->pdev->device == 0x131B)) { - adev->gfx.config.max_cu_per_sh = 4; - adev->gfx.config.max_backends_per_se = 1; - } else { - adev->gfx.config.max_cu_per_sh = 3; - adev->gfx.config.max_backends_per_se = 1; - } - adev->gfx.config.max_sh_per_se = 1; - adev->gfx.config.max_texture_channel_caches = 4; - adev->gfx.config.max_gprs = 256; - adev->gfx.config.max_gs_threads = 16; - adev->gfx.config.max_hw_contexts = 8; - - adev->gfx.config.sc_prim_fifo_size_frontend = 0x20; - adev->gfx.config.sc_prim_fifo_size_backend = 0x100; - adev->gfx.config.sc_hiz_tile_fifo_size = 0x30; - adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130; - gb_addr_config = BONAIRE_GB_ADDR_CONFIG_GOLDEN; - break; - case CHIP_KABINI: - case CHIP_MULLINS: - default: - adev->gfx.config.max_shader_engines = 1; - adev->gfx.config.max_tile_pipes = 2; - adev->gfx.config.max_cu_per_sh = 2; - adev->gfx.config.max_sh_per_se = 1; - adev->gfx.config.max_backends_per_se = 1; - adev->gfx.config.max_texture_channel_caches = 2; - adev->gfx.config.max_gprs = 256; - adev->gfx.config.max_gs_threads = 16; - adev->gfx.config.max_hw_contexts = 8; - - adev->gfx.config.sc_prim_fifo_size_frontend = 0x20; - adev->gfx.config.sc_prim_fifo_size_backend = 0x100; - adev->gfx.config.sc_hiz_tile_fifo_size = 0x30; - adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130; - gb_addr_config = BONAIRE_GB_ADDR_CONFIG_GOLDEN; - break; - } - WREG32(mmGRBM_CNTL, (0xff << GRBM_CNTL__READ_TIMEOUT__SHIFT)); - mc_shared_chmap = RREG32(mmMC_SHARED_CHMAP); - adev->gfx.config.mc_arb_ramcfg = RREG32(mmMC_ARB_RAMCFG); - mc_arb_ramcfg = adev->gfx.config.mc_arb_ramcfg; - - adev->gfx.config.num_tile_pipes = adev->gfx.config.max_tile_pipes; - adev->gfx.config.mem_max_burst_length_bytes = 256; - if (adev->flags & AMD_IS_APU) { - /* Get memory bank mapping mode. */ - tmp = RREG32(mmMC_FUS_DRAM0_BANK_ADDR_MAPPING); - dimm00_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM0_BANK_ADDR_MAPPING, DIMM0ADDRMAP); - dimm01_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM0_BANK_ADDR_MAPPING, DIMM1ADDRMAP); - - tmp = RREG32(mmMC_FUS_DRAM1_BANK_ADDR_MAPPING); - dimm10_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM1_BANK_ADDR_MAPPING, DIMM0ADDRMAP); - dimm11_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM1_BANK_ADDR_MAPPING, DIMM1ADDRMAP); - - /* Validate settings in case only one DIMM installed. */ - if ((dimm00_addr_map == 0) || (dimm00_addr_map == 3) || (dimm00_addr_map == 4) || (dimm00_addr_map > 12)) - dimm00_addr_map = 0; - if ((dimm01_addr_map == 0) || (dimm01_addr_map == 3) || (dimm01_addr_map == 4) || (dimm01_addr_map > 12)) - dimm01_addr_map = 0; - if ((dimm10_addr_map == 0) || (dimm10_addr_map == 3) || (dimm10_addr_map == 4) || (dimm10_addr_map > 12)) - dimm10_addr_map = 0; - if ((dimm11_addr_map == 0) || (dimm11_addr_map == 3) || (dimm11_addr_map == 4) || (dimm11_addr_map > 12)) - dimm11_addr_map = 0; - - /* If DIMM Addr map is 8GB, ROW size should be 2KB. Otherwise 1KB. */ - /* If ROW size(DIMM1) != ROW size(DMIMM0), ROW size should be larger one. */ - if ((dimm00_addr_map == 11) || (dimm01_addr_map == 11) || (dimm10_addr_map == 11) || (dimm11_addr_map == 11)) - adev->gfx.config.mem_row_size_in_kb = 2; - else - adev->gfx.config.mem_row_size_in_kb = 1; - } else { - tmp = (mc_arb_ramcfg & MC_ARB_RAMCFG__NOOFCOLS_MASK) >> MC_ARB_RAMCFG__NOOFCOLS__SHIFT; - adev->gfx.config.mem_row_size_in_kb = (4 * (1 << (8 + tmp))) / 1024; - if (adev->gfx.config.mem_row_size_in_kb > 4) - adev->gfx.config.mem_row_size_in_kb = 4; - } - /* XXX use MC settings? */ - adev->gfx.config.shader_engine_tile_size = 32; - adev->gfx.config.num_gpus = 1; - adev->gfx.config.multi_gpu_tile_size = 64; - - /* fix up row size */ - gb_addr_config &= ~GB_ADDR_CONFIG__ROW_SIZE_MASK; - switch (adev->gfx.config.mem_row_size_in_kb) { - case 1: - default: - gb_addr_config |= (0 << GB_ADDR_CONFIG__ROW_SIZE__SHIFT); - break; - case 2: - gb_addr_config |= (1 << GB_ADDR_CONFIG__ROW_SIZE__SHIFT); - break; - case 4: - gb_addr_config |= (2 << GB_ADDR_CONFIG__ROW_SIZE__SHIFT); - break; - } - adev->gfx.config.gb_addr_config = gb_addr_config; - - WREG32(mmGB_ADDR_CONFIG, gb_addr_config); - WREG32(mmHDP_ADDR_CONFIG, gb_addr_config); - WREG32(mmDMIF_ADDR_CALC, gb_addr_config); - WREG32(mmSDMA0_TILING_CONFIG + SDMA0_REGISTER_OFFSET, gb_addr_config & 0x70); - WREG32(mmSDMA0_TILING_CONFIG + SDMA1_REGISTER_OFFSET, gb_addr_config & 0x70); - WREG32(mmUVD_UDEC_ADDR_CONFIG, gb_addr_config); - WREG32(mmUVD_UDEC_DB_ADDR_CONFIG, gb_addr_config); - WREG32(mmUVD_UDEC_DBW_ADDR_CONFIG, gb_addr_config); + WREG32(mmGB_ADDR_CONFIG, adev->gfx.config.gb_addr_config); + WREG32(mmHDP_ADDR_CONFIG, adev->gfx.config.gb_addr_config); + WREG32(mmDMIF_ADDR_CALC, adev->gfx.config.gb_addr_config); gfx_v7_0_tiling_mode_table_init(adev); - gfx_v7_0_setup_rb(adev, adev->gfx.config.max_shader_engines, - adev->gfx.config.max_sh_per_se, - adev->gfx.config.max_backends_per_se); + gfx_v7_0_setup_rb(adev); /* set HW defaults for 3D engine */ WREG32(mmCP_MEQ_THRESHOLDS, - (0x30 << CP_MEQ_THRESHOLDS__MEQ1_START__SHIFT) | - (0x60 << CP_MEQ_THRESHOLDS__MEQ2_START__SHIFT)); + (0x30 << CP_MEQ_THRESHOLDS__MEQ1_START__SHIFT) | + (0x60 << CP_MEQ_THRESHOLDS__MEQ2_START__SHIFT)); mutex_lock(&adev->grbm_idx_mutex); /* @@ -2255,7 +1733,7 @@ static void gfx_v7_0_gpu_init(struct amdgpu_device *adev) /* XXX SH_MEM regs */ /* where to put LDS, scratch, GPUVM in FSA64 space */ - sh_mem_cfg = REG_SET_FIELD(0, SH_MEM_CONFIG, ALIGNMENT_MODE, + sh_mem_cfg = REG_SET_FIELD(0, SH_MEM_CONFIG, ALIGNMENT_MODE, SH_MEM_ALIGNMENT_MODE_UNALIGNED); mutex_lock(&adev->srbm_mutex); @@ -2379,7 +1857,7 @@ static int gfx_v7_0_ring_test_ring(struct amdgpu_ring *ring) return r; } WREG32(scratch, 0xCAFEDEAD); - r = amdgpu_ring_lock(ring, 3); + r = amdgpu_ring_alloc(ring, 3); if (r) { DRM_ERROR("amdgpu: cp failed to lock ring %d (%d).\n", ring->idx, r); amdgpu_gfx_scratch_free(adev, scratch); @@ -2388,7 +1866,7 @@ static int gfx_v7_0_ring_test_ring(struct amdgpu_ring *ring) amdgpu_ring_write(ring, PACKET3(PACKET3_SET_UCONFIG_REG, 1)); amdgpu_ring_write(ring, (scratch - PACKET3_SET_UCONFIG_REG_START)); amdgpu_ring_write(ring, 0xDEADBEEF); - amdgpu_ring_unlock_commit(ring); + amdgpu_ring_commit(ring); for (i = 0; i < adev->usec_timeout; i++) { tmp = RREG32(scratch); @@ -2447,6 +1925,25 @@ static void gfx_v7_0_ring_emit_hdp_flush(struct amdgpu_ring *ring) } /** + * gfx_v7_0_ring_emit_hdp_invalidate - emit an hdp invalidate on the cp + * + * @adev: amdgpu_device pointer + * @ridx: amdgpu ring index + * + * Emits an hdp invalidate on the cp. + */ +static void gfx_v7_0_ring_emit_hdp_invalidate(struct amdgpu_ring *ring) +{ + amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3)); + amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) | + WRITE_DATA_DST_SEL(0) | + WR_CONFIRM)); + amdgpu_ring_write(ring, mmHDP_DEBUG0); + amdgpu_ring_write(ring, 0); + amdgpu_ring_write(ring, 1); +} + +/** * gfx_v7_0_ring_emit_fence_gfx - emit a fence on the gfx ring * * @adev: amdgpu_device pointer @@ -2516,36 +2013,6 @@ static void gfx_v7_0_ring_emit_fence_compute(struct amdgpu_ring *ring, amdgpu_ring_write(ring, upper_32_bits(seq)); } -/** - * gfx_v7_0_ring_emit_semaphore - emit a semaphore on the CP ring - * - * @ring: amdgpu ring buffer object - * @semaphore: amdgpu semaphore object - * @emit_wait: Is this a sempahore wait? - * - * Emits a semaphore signal/wait packet to the CP ring and prevents the PFP - * from running ahead of semaphore waits. - */ -static bool gfx_v7_0_ring_emit_semaphore(struct amdgpu_ring *ring, - struct amdgpu_semaphore *semaphore, - bool emit_wait) -{ - uint64_t addr = semaphore->gpu_addr; - unsigned sel = emit_wait ? PACKET3_SEM_SEL_WAIT : PACKET3_SEM_SEL_SIGNAL; - - amdgpu_ring_write(ring, PACKET3(PACKET3_MEM_SEMAPHORE, 1)); - amdgpu_ring_write(ring, addr & 0xffffffff); - amdgpu_ring_write(ring, (upper_32_bits(addr) & 0xffff) | sel); - - if (emit_wait && (ring->type == AMDGPU_RING_TYPE_GFX)) { - /* Prevent the PFP from running ahead of the semaphore wait */ - amdgpu_ring_write(ring, PACKET3(PACKET3_PFP_SYNC_ME, 0)); - amdgpu_ring_write(ring, 0x0); - } - - return true; -} - /* * IB stuff */ @@ -2593,8 +2060,7 @@ static void gfx_v7_0_ring_emit_ib_gfx(struct amdgpu_ring *ring, else header = PACKET3(PACKET3_INDIRECT_BUFFER, 2); - control |= ib->length_dw | - (ib->vm ? (ib->vm->ids[ring->idx].id << 24) : 0); + control |= ib->length_dw | (ib->vm_id << 24); amdgpu_ring_write(ring, header); amdgpu_ring_write(ring, @@ -2622,8 +2088,7 @@ static void gfx_v7_0_ring_emit_ib_compute(struct amdgpu_ring *ring, header = PACKET3(PACKET3_INDIRECT_BUFFER, 2); - control |= ib->length_dw | - (ib->vm ? (ib->vm->ids[ring->idx].id << 24) : 0); + control |= ib->length_dw | (ib->vm_id << 24); amdgpu_ring_write(ring, header); amdgpu_ring_write(ring, @@ -2661,7 +2126,7 @@ static int gfx_v7_0_ring_test_ib(struct amdgpu_ring *ring) } WREG32(scratch, 0xCAFEDEAD); memset(&ib, 0, sizeof(ib)); - r = amdgpu_ib_get(ring, NULL, 256, &ib); + r = amdgpu_ib_get(adev, NULL, 256, &ib); if (r) { DRM_ERROR("amdgpu: failed to get ib (%d).\n", r); goto err1; @@ -2671,9 +2136,7 @@ static int gfx_v7_0_ring_test_ib(struct amdgpu_ring *ring) ib.ptr[2] = 0xDEADBEEF; ib.length_dw = 3; - r = amdgpu_sched_ib_submit_kernel_helper(adev, ring, &ib, 1, NULL, - AMDGPU_FENCE_OWNER_UNDEFINED, - &f); + r = amdgpu_ib_schedule(ring, 1, &ib, NULL, &f); if (r) goto err2; @@ -2700,7 +2163,8 @@ static int gfx_v7_0_ring_test_ib(struct amdgpu_ring *ring) err2: fence_put(f); - amdgpu_ib_free(adev, &ib); + amdgpu_ib_free(adev, &ib, NULL); + fence_put(f); err1: amdgpu_gfx_scratch_free(adev, scratch); return r; @@ -2842,7 +2306,7 @@ static int gfx_v7_0_cp_gfx_start(struct amdgpu_device *adev) gfx_v7_0_cp_gfx_enable(adev, true); - r = amdgpu_ring_lock(ring, gfx_v7_0_get_csb_size(adev) + 8); + r = amdgpu_ring_alloc(ring, gfx_v7_0_get_csb_size(adev) + 8); if (r) { DRM_ERROR("amdgpu: cp failed to lock ring (%d).\n", r); return r; @@ -2911,7 +2375,7 @@ static int gfx_v7_0_cp_gfx_start(struct amdgpu_device *adev) amdgpu_ring_write(ring, 0x0000000e); /* VGT_VERTEX_REUSE_BLOCK_CNTL */ amdgpu_ring_write(ring, 0x00000010); /* VGT_OUT_DEALLOC_CNTL */ - amdgpu_ring_unlock_commit(ring); + amdgpu_ring_commit(ring); return 0; } @@ -2989,21 +2453,14 @@ static int gfx_v7_0_cp_gfx_resume(struct amdgpu_device *adev) static u32 gfx_v7_0_ring_get_rptr_gfx(struct amdgpu_ring *ring) { - u32 rptr; - - rptr = ring->adev->wb.wb[ring->rptr_offs]; - - return rptr; + return ring->adev->wb.wb[ring->rptr_offs]; } static u32 gfx_v7_0_ring_get_wptr_gfx(struct amdgpu_ring *ring) { struct amdgpu_device *adev = ring->adev; - u32 wptr; - - wptr = RREG32(mmCP_RB0_WPTR); - return wptr; + return RREG32(mmCP_RB0_WPTR); } static void gfx_v7_0_ring_set_wptr_gfx(struct amdgpu_ring *ring) @@ -3016,21 +2473,13 @@ static void gfx_v7_0_ring_set_wptr_gfx(struct amdgpu_ring *ring) static u32 gfx_v7_0_ring_get_rptr_compute(struct amdgpu_ring *ring) { - u32 rptr; - - rptr = ring->adev->wb.wb[ring->rptr_offs]; - - return rptr; + return ring->adev->wb.wb[ring->rptr_offs]; } static u32 gfx_v7_0_ring_get_wptr_compute(struct amdgpu_ring *ring) { - u32 wptr; - /* XXX check if swapping is necessary on BE */ - wptr = ring->adev->wb.wb[ring->wptr_offs]; - - return wptr; + return ring->adev->wb.wb[ring->wptr_offs]; } static void gfx_v7_0_ring_set_wptr_compute(struct amdgpu_ring *ring) @@ -3126,21 +2575,6 @@ static int gfx_v7_0_cp_compute_load_microcode(struct amdgpu_device *adev) } /** - * gfx_v7_0_cp_compute_start - start the compute queues - * - * @adev: amdgpu_device pointer - * - * Enable the compute queues. - * Returns 0 for success, error for failure. - */ -static int gfx_v7_0_cp_compute_start(struct amdgpu_device *adev) -{ - gfx_v7_0_cp_compute_enable(adev, true); - - return 0; -} - -/** * gfx_v7_0_cp_compute_fini - stop the compute queues * * @adev: amdgpu_device pointer @@ -3330,9 +2764,7 @@ static int gfx_v7_0_cp_compute_resume(struct amdgpu_device *adev) u32 *buf; struct bonaire_mqd *mqd; - r = gfx_v7_0_cp_compute_start(adev); - if (r) - return r; + gfx_v7_0_cp_compute_enable(adev, true); /* fix up chicken bits */ tmp = RREG32(mmCP_CPF_DEBUG); @@ -3610,6 +3042,26 @@ static int gfx_v7_0_cp_resume(struct amdgpu_device *adev) return 0; } +/** + * gfx_v7_0_ring_emit_vm_flush - cik vm flush using the CP + * + * @ring: the ring to emmit the commands to + * + * Sync the command pipeline with the PFP. E.g. wait for everything + * to be completed. + */ +static void gfx_v7_0_ring_emit_pipeline_sync(struct amdgpu_ring *ring) +{ + int usepfp = (ring->type == AMDGPU_RING_TYPE_GFX); + if (usepfp) { + /* synce CE with ME to prevent CE fetch CEIB before context switch done */ + amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0)); + amdgpu_ring_write(ring, 0); + amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0)); + amdgpu_ring_write(ring, 0); + } +} + /* * vm * VMID 0 is the physical GPU addresses as used by the kernel. @@ -3628,13 +3080,18 @@ static void gfx_v7_0_ring_emit_vm_flush(struct amdgpu_ring *ring, unsigned vm_id, uint64_t pd_addr) { int usepfp = (ring->type == AMDGPU_RING_TYPE_GFX); - if (usepfp) { - /* synce CE with ME to prevent CE fetch CEIB before context switch done */ - amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0)); - amdgpu_ring_write(ring, 0); - amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0)); - amdgpu_ring_write(ring, 0); - } + uint32_t seq = ring->fence_drv.sync_seq; + uint64_t addr = ring->fence_drv.gpu_addr; + + amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5)); + amdgpu_ring_write(ring, (WAIT_REG_MEM_MEM_SPACE(1) | /* memory */ + WAIT_REG_MEM_FUNCTION(3) | /* equal */ + WAIT_REG_MEM_ENGINE(usepfp))); /* pfp or me */ + amdgpu_ring_write(ring, addr & 0xfffffffc); + amdgpu_ring_write(ring, upper_32_bits(addr) & 0xffffffff); + amdgpu_ring_write(ring, seq); + amdgpu_ring_write(ring, 0xffffffff); + amdgpu_ring_write(ring, 4); /* poll interval */ amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3)); amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(usepfp) | @@ -4109,7 +3566,7 @@ static void gfx_v7_0_enable_cgcg(struct amdgpu_device *adev, bool enable) orig = data = RREG32(mmRLC_CGCG_CGLS_CTRL); - if (enable && (adev->cg_flags & AMDGPU_CG_SUPPORT_GFX_CGCG)) { + if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGCG)) { gfx_v7_0_enable_gui_idle_interrupt(adev, true); tmp = gfx_v7_0_halt_rlc(adev); @@ -4147,9 +3604,9 @@ static void gfx_v7_0_enable_mgcg(struct amdgpu_device *adev, bool enable) { u32 data, orig, tmp = 0; - if (enable && (adev->cg_flags & AMDGPU_CG_SUPPORT_GFX_MGCG)) { - if (adev->cg_flags & AMDGPU_CG_SUPPORT_GFX_MGLS) { - if (adev->cg_flags & AMDGPU_CG_SUPPORT_GFX_CP_LS) { + if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGCG)) { + if (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGLS) { + if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CP_LS) { orig = data = RREG32(mmCP_MEM_SLP_CNTL); data |= CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK; if (orig != data) @@ -4176,14 +3633,14 @@ static void gfx_v7_0_enable_mgcg(struct amdgpu_device *adev, bool enable) gfx_v7_0_update_rlc(adev, tmp); - if (adev->cg_flags & AMDGPU_CG_SUPPORT_GFX_CGTS) { + if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGTS) { orig = data = RREG32(mmCGTS_SM_CTRL_REG); data &= ~CGTS_SM_CTRL_REG__SM_MODE_MASK; data |= (0x2 << CGTS_SM_CTRL_REG__SM_MODE__SHIFT); data |= CGTS_SM_CTRL_REG__SM_MODE_ENABLE_MASK; data &= ~CGTS_SM_CTRL_REG__OVERRIDE_MASK; - if ((adev->cg_flags & AMDGPU_CG_SUPPORT_GFX_MGLS) && - (adev->cg_flags & AMDGPU_CG_SUPPORT_GFX_CGTS_LS)) + if ((adev->cg_flags & AMD_CG_SUPPORT_GFX_MGLS) && + (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGTS_LS)) data &= ~CGTS_SM_CTRL_REG__LS_OVERRIDE_MASK; data &= ~CGTS_SM_CTRL_REG__ON_MONITOR_ADD_MASK; data |= CGTS_SM_CTRL_REG__ON_MONITOR_ADD_EN_MASK; @@ -4249,7 +3706,7 @@ static void gfx_v7_0_enable_sclk_slowdown_on_pu(struct amdgpu_device *adev, u32 data, orig; orig = data = RREG32(mmRLC_PG_CNTL); - if (enable && (adev->pg_flags & AMDGPU_PG_SUPPORT_RLC_SMU_HS)) + if (enable && (adev->pg_flags & AMD_PG_SUPPORT_RLC_SMU_HS)) data |= RLC_PG_CNTL__SMU_CLK_SLOWDOWN_ON_PU_ENABLE_MASK; else data &= ~RLC_PG_CNTL__SMU_CLK_SLOWDOWN_ON_PU_ENABLE_MASK; @@ -4263,7 +3720,7 @@ static void gfx_v7_0_enable_sclk_slowdown_on_pd(struct amdgpu_device *adev, u32 data, orig; orig = data = RREG32(mmRLC_PG_CNTL); - if (enable && (adev->pg_flags & AMDGPU_PG_SUPPORT_RLC_SMU_HS)) + if (enable && (adev->pg_flags & AMD_PG_SUPPORT_RLC_SMU_HS)) data |= RLC_PG_CNTL__SMU_CLK_SLOWDOWN_ON_PD_ENABLE_MASK; else data &= ~RLC_PG_CNTL__SMU_CLK_SLOWDOWN_ON_PD_ENABLE_MASK; @@ -4276,7 +3733,7 @@ static void gfx_v7_0_enable_cp_pg(struct amdgpu_device *adev, bool enable) u32 data, orig; orig = data = RREG32(mmRLC_PG_CNTL); - if (enable && (adev->pg_flags & AMDGPU_PG_SUPPORT_CP)) + if (enable && (adev->pg_flags & AMD_PG_SUPPORT_CP)) data &= ~0x8000; else data |= 0x8000; @@ -4289,7 +3746,7 @@ static void gfx_v7_0_enable_gds_pg(struct amdgpu_device *adev, bool enable) u32 data, orig; orig = data = RREG32(mmRLC_PG_CNTL); - if (enable && (adev->pg_flags & AMDGPU_PG_SUPPORT_GDS)) + if (enable && (adev->pg_flags & AMD_PG_SUPPORT_GDS)) data &= ~0x2000; else data |= 0x2000; @@ -4370,7 +3827,7 @@ static void gfx_v7_0_enable_gfx_cgpg(struct amdgpu_device *adev, { u32 data, orig; - if (enable && (adev->pg_flags & AMDGPU_PG_SUPPORT_GFX_PG)) { + if (enable && (adev->pg_flags & AMD_PG_SUPPORT_GFX_PG)) { orig = data = RREG32(mmRLC_PG_CNTL); data |= RLC_PG_CNTL__GFX_POWER_GATING_ENABLE_MASK; if (orig != data) @@ -4395,28 +3852,19 @@ static void gfx_v7_0_enable_gfx_cgpg(struct amdgpu_device *adev, } } -static u32 gfx_v7_0_get_cu_active_bitmap(struct amdgpu_device *adev, - u32 se, u32 sh) +static u32 gfx_v7_0_get_cu_active_bitmap(struct amdgpu_device *adev) { - u32 mask = 0, tmp, tmp1; - int i; + u32 data, mask; - gfx_v7_0_select_se_sh(adev, se, sh); - tmp = RREG32(mmCC_GC_SHADER_ARRAY_CONFIG); - tmp1 = RREG32(mmGC_USER_SHADER_ARRAY_CONFIG); - gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff); + data = RREG32(mmCC_GC_SHADER_ARRAY_CONFIG); + data |= RREG32(mmGC_USER_SHADER_ARRAY_CONFIG); - tmp &= 0xffff0000; + data &= CC_GC_SHADER_ARRAY_CONFIG__INACTIVE_CUS_MASK; + data >>= CC_GC_SHADER_ARRAY_CONFIG__INACTIVE_CUS__SHIFT; - tmp |= tmp1; - tmp >>= 16; + mask = gfx_v7_0_create_bitmask(adev->gfx.config.max_cu_per_sh); - for (i = 0; i < adev->gfx.config.max_cu_per_sh; i ++) { - mask <<= 1; - mask |= 1; - } - - return (~tmp) & mask; + return (~data) & mask; } static void gfx_v7_0_init_ao_cu_mask(struct amdgpu_device *adev) @@ -4442,7 +3890,7 @@ static void gfx_v7_0_enable_gfx_static_mgpg(struct amdgpu_device *adev, u32 data, orig; orig = data = RREG32(mmRLC_PG_CNTL); - if (enable && (adev->pg_flags & AMDGPU_PG_SUPPORT_GFX_SMG)) + if (enable && (adev->pg_flags & AMD_PG_SUPPORT_GFX_SMG)) data |= RLC_PG_CNTL__STATIC_PER_CU_PG_ENABLE_MASK; else data &= ~RLC_PG_CNTL__STATIC_PER_CU_PG_ENABLE_MASK; @@ -4456,7 +3904,7 @@ static void gfx_v7_0_enable_gfx_dynamic_mgpg(struct amdgpu_device *adev, u32 data, orig; orig = data = RREG32(mmRLC_PG_CNTL); - if (enable && (adev->pg_flags & AMDGPU_PG_SUPPORT_GFX_DMG)) + if (enable && (adev->pg_flags & AMD_PG_SUPPORT_GFX_DMG)) data |= RLC_PG_CNTL__DYN_PER_CU_PG_ENABLE_MASK; else data &= ~RLC_PG_CNTL__DYN_PER_CU_PG_ENABLE_MASK; @@ -4623,15 +4071,15 @@ static void gfx_v7_0_get_csb_buffer(struct amdgpu_device *adev, static void gfx_v7_0_init_pg(struct amdgpu_device *adev) { - if (adev->pg_flags & (AMDGPU_PG_SUPPORT_GFX_PG | - AMDGPU_PG_SUPPORT_GFX_SMG | - AMDGPU_PG_SUPPORT_GFX_DMG | - AMDGPU_PG_SUPPORT_CP | - AMDGPU_PG_SUPPORT_GDS | - AMDGPU_PG_SUPPORT_RLC_SMU_HS)) { + if (adev->pg_flags & (AMD_PG_SUPPORT_GFX_PG | + AMD_PG_SUPPORT_GFX_SMG | + AMD_PG_SUPPORT_GFX_DMG | + AMD_PG_SUPPORT_CP | + AMD_PG_SUPPORT_GDS | + AMD_PG_SUPPORT_RLC_SMU_HS)) { gfx_v7_0_enable_sclk_slowdown_on_pu(adev, true); gfx_v7_0_enable_sclk_slowdown_on_pd(adev, true); - if (adev->pg_flags & AMDGPU_PG_SUPPORT_GFX_PG) { + if (adev->pg_flags & AMD_PG_SUPPORT_GFX_PG) { gfx_v7_0_init_gfx_cgpg(adev); gfx_v7_0_enable_cp_pg(adev, true); gfx_v7_0_enable_gds_pg(adev, true); @@ -4643,14 +4091,14 @@ static void gfx_v7_0_init_pg(struct amdgpu_device *adev) static void gfx_v7_0_fini_pg(struct amdgpu_device *adev) { - if (adev->pg_flags & (AMDGPU_PG_SUPPORT_GFX_PG | - AMDGPU_PG_SUPPORT_GFX_SMG | - AMDGPU_PG_SUPPORT_GFX_DMG | - AMDGPU_PG_SUPPORT_CP | - AMDGPU_PG_SUPPORT_GDS | - AMDGPU_PG_SUPPORT_RLC_SMU_HS)) { + if (adev->pg_flags & (AMD_PG_SUPPORT_GFX_PG | + AMD_PG_SUPPORT_GFX_SMG | + AMD_PG_SUPPORT_GFX_DMG | + AMD_PG_SUPPORT_CP | + AMD_PG_SUPPORT_GDS | + AMD_PG_SUPPORT_RLC_SMU_HS)) { gfx_v7_0_update_gfx_pg(adev, false); - if (adev->pg_flags & AMDGPU_PG_SUPPORT_GFX_PG) { + if (adev->pg_flags & AMD_PG_SUPPORT_GFX_PG) { gfx_v7_0_enable_cp_pg(adev, false); gfx_v7_0_enable_gds_pg(adev, false); } @@ -4738,6 +4186,188 @@ static int gfx_v7_0_early_init(void *handle) return 0; } +static int gfx_v7_0_late_init(void *handle) +{ + struct amdgpu_device *adev = (struct amdgpu_device *)handle; + int r; + + r = amdgpu_irq_get(adev, &adev->gfx.priv_reg_irq, 0); + if (r) + return r; + + r = amdgpu_irq_get(adev, &adev->gfx.priv_inst_irq, 0); + if (r) + return r; + + return 0; +} + +static void gfx_v7_0_gpu_early_init(struct amdgpu_device *adev) +{ + u32 gb_addr_config; + u32 mc_shared_chmap, mc_arb_ramcfg; + u32 dimm00_addr_map, dimm01_addr_map, dimm10_addr_map, dimm11_addr_map; + u32 tmp; + + switch (adev->asic_type) { + case CHIP_BONAIRE: + adev->gfx.config.max_shader_engines = 2; + adev->gfx.config.max_tile_pipes = 4; + adev->gfx.config.max_cu_per_sh = 7; + adev->gfx.config.max_sh_per_se = 1; + adev->gfx.config.max_backends_per_se = 2; + adev->gfx.config.max_texture_channel_caches = 4; + adev->gfx.config.max_gprs = 256; + adev->gfx.config.max_gs_threads = 32; + adev->gfx.config.max_hw_contexts = 8; + + adev->gfx.config.sc_prim_fifo_size_frontend = 0x20; + adev->gfx.config.sc_prim_fifo_size_backend = 0x100; + adev->gfx.config.sc_hiz_tile_fifo_size = 0x30; + adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130; + gb_addr_config = BONAIRE_GB_ADDR_CONFIG_GOLDEN; + break; + case CHIP_HAWAII: + adev->gfx.config.max_shader_engines = 4; + adev->gfx.config.max_tile_pipes = 16; + adev->gfx.config.max_cu_per_sh = 11; + adev->gfx.config.max_sh_per_se = 1; + adev->gfx.config.max_backends_per_se = 4; + adev->gfx.config.max_texture_channel_caches = 16; + adev->gfx.config.max_gprs = 256; + adev->gfx.config.max_gs_threads = 32; + adev->gfx.config.max_hw_contexts = 8; + + adev->gfx.config.sc_prim_fifo_size_frontend = 0x20; + adev->gfx.config.sc_prim_fifo_size_backend = 0x100; + adev->gfx.config.sc_hiz_tile_fifo_size = 0x30; + adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130; + gb_addr_config = HAWAII_GB_ADDR_CONFIG_GOLDEN; + break; + case CHIP_KAVERI: + adev->gfx.config.max_shader_engines = 1; + adev->gfx.config.max_tile_pipes = 4; + if ((adev->pdev->device == 0x1304) || + (adev->pdev->device == 0x1305) || + (adev->pdev->device == 0x130C) || + (adev->pdev->device == 0x130F) || + (adev->pdev->device == 0x1310) || + (adev->pdev->device == 0x1311) || + (adev->pdev->device == 0x131C)) { + adev->gfx.config.max_cu_per_sh = 8; + adev->gfx.config.max_backends_per_se = 2; + } else if ((adev->pdev->device == 0x1309) || + (adev->pdev->device == 0x130A) || + (adev->pdev->device == 0x130D) || + (adev->pdev->device == 0x1313) || + (adev->pdev->device == 0x131D)) { + adev->gfx.config.max_cu_per_sh = 6; + adev->gfx.config.max_backends_per_se = 2; + } else if ((adev->pdev->device == 0x1306) || + (adev->pdev->device == 0x1307) || + (adev->pdev->device == 0x130B) || + (adev->pdev->device == 0x130E) || + (adev->pdev->device == 0x1315) || + (adev->pdev->device == 0x131B)) { + adev->gfx.config.max_cu_per_sh = 4; + adev->gfx.config.max_backends_per_se = 1; + } else { + adev->gfx.config.max_cu_per_sh = 3; + adev->gfx.config.max_backends_per_se = 1; + } + adev->gfx.config.max_sh_per_se = 1; + adev->gfx.config.max_texture_channel_caches = 4; + adev->gfx.config.max_gprs = 256; + adev->gfx.config.max_gs_threads = 16; + adev->gfx.config.max_hw_contexts = 8; + + adev->gfx.config.sc_prim_fifo_size_frontend = 0x20; + adev->gfx.config.sc_prim_fifo_size_backend = 0x100; + adev->gfx.config.sc_hiz_tile_fifo_size = 0x30; + adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130; + gb_addr_config = BONAIRE_GB_ADDR_CONFIG_GOLDEN; + break; + case CHIP_KABINI: + case CHIP_MULLINS: + default: + adev->gfx.config.max_shader_engines = 1; + adev->gfx.config.max_tile_pipes = 2; + adev->gfx.config.max_cu_per_sh = 2; + adev->gfx.config.max_sh_per_se = 1; + adev->gfx.config.max_backends_per_se = 1; + adev->gfx.config.max_texture_channel_caches = 2; + adev->gfx.config.max_gprs = 256; + adev->gfx.config.max_gs_threads = 16; + adev->gfx.config.max_hw_contexts = 8; + + adev->gfx.config.sc_prim_fifo_size_frontend = 0x20; + adev->gfx.config.sc_prim_fifo_size_backend = 0x100; + adev->gfx.config.sc_hiz_tile_fifo_size = 0x30; + adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130; + gb_addr_config = BONAIRE_GB_ADDR_CONFIG_GOLDEN; + break; + } + + mc_shared_chmap = RREG32(mmMC_SHARED_CHMAP); + adev->gfx.config.mc_arb_ramcfg = RREG32(mmMC_ARB_RAMCFG); + mc_arb_ramcfg = adev->gfx.config.mc_arb_ramcfg; + + adev->gfx.config.num_tile_pipes = adev->gfx.config.max_tile_pipes; + adev->gfx.config.mem_max_burst_length_bytes = 256; + if (adev->flags & AMD_IS_APU) { + /* Get memory bank mapping mode. */ + tmp = RREG32(mmMC_FUS_DRAM0_BANK_ADDR_MAPPING); + dimm00_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM0_BANK_ADDR_MAPPING, DIMM0ADDRMAP); + dimm01_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM0_BANK_ADDR_MAPPING, DIMM1ADDRMAP); + + tmp = RREG32(mmMC_FUS_DRAM1_BANK_ADDR_MAPPING); + dimm10_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM1_BANK_ADDR_MAPPING, DIMM0ADDRMAP); + dimm11_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM1_BANK_ADDR_MAPPING, DIMM1ADDRMAP); + + /* Validate settings in case only one DIMM installed. */ + if ((dimm00_addr_map == 0) || (dimm00_addr_map == 3) || (dimm00_addr_map == 4) || (dimm00_addr_map > 12)) + dimm00_addr_map = 0; + if ((dimm01_addr_map == 0) || (dimm01_addr_map == 3) || (dimm01_addr_map == 4) || (dimm01_addr_map > 12)) + dimm01_addr_map = 0; + if ((dimm10_addr_map == 0) || (dimm10_addr_map == 3) || (dimm10_addr_map == 4) || (dimm10_addr_map > 12)) + dimm10_addr_map = 0; + if ((dimm11_addr_map == 0) || (dimm11_addr_map == 3) || (dimm11_addr_map == 4) || (dimm11_addr_map > 12)) + dimm11_addr_map = 0; + + /* If DIMM Addr map is 8GB, ROW size should be 2KB. Otherwise 1KB. */ + /* If ROW size(DIMM1) != ROW size(DMIMM0), ROW size should be larger one. */ + if ((dimm00_addr_map == 11) || (dimm01_addr_map == 11) || (dimm10_addr_map == 11) || (dimm11_addr_map == 11)) + adev->gfx.config.mem_row_size_in_kb = 2; + else + adev->gfx.config.mem_row_size_in_kb = 1; + } else { + tmp = (mc_arb_ramcfg & MC_ARB_RAMCFG__NOOFCOLS_MASK) >> MC_ARB_RAMCFG__NOOFCOLS__SHIFT; + adev->gfx.config.mem_row_size_in_kb = (4 * (1 << (8 + tmp))) / 1024; + if (adev->gfx.config.mem_row_size_in_kb > 4) + adev->gfx.config.mem_row_size_in_kb = 4; + } + /* XXX use MC settings? */ + adev->gfx.config.shader_engine_tile_size = 32; + adev->gfx.config.num_gpus = 1; + adev->gfx.config.multi_gpu_tile_size = 64; + + /* fix up row size */ + gb_addr_config &= ~GB_ADDR_CONFIG__ROW_SIZE_MASK; + switch (adev->gfx.config.mem_row_size_in_kb) { + case 1: + default: + gb_addr_config |= (0 << GB_ADDR_CONFIG__ROW_SIZE__SHIFT); + break; + case 2: + gb_addr_config |= (1 << GB_ADDR_CONFIG__ROW_SIZE__SHIFT); + break; + case 4: + gb_addr_config |= (2 << GB_ADDR_CONFIG__ROW_SIZE__SHIFT); + break; + } + adev->gfx.config.gb_addr_config = gb_addr_config; +} + static int gfx_v7_0_sw_init(void *handle) { struct amdgpu_ring *ring; @@ -4841,6 +4471,10 @@ static int gfx_v7_0_sw_init(void *handle) if (r) return r; + adev->gfx.ce_ram_size = 0x8000; + + gfx_v7_0_gpu_early_init(adev); + return r; } @@ -4881,8 +4515,6 @@ static int gfx_v7_0_hw_init(void *handle) if (r) return r; - adev->gfx.ce_ram_size = 0x8000; - return r; } @@ -4890,6 +4522,8 @@ static int gfx_v7_0_hw_fini(void *handle) { struct amdgpu_device *adev = (struct amdgpu_device *)handle; + amdgpu_irq_put(adev, &adev->gfx.priv_reg_irq, 0); + amdgpu_irq_put(adev, &adev->gfx.priv_inst_irq, 0); gfx_v7_0_cp_enable(adev, false); gfx_v7_0_rlc_stop(adev); gfx_v7_0_fini_pg(adev); @@ -4997,16 +4631,6 @@ static void gfx_v7_0_print_status(void *handle) RREG32(mmHDP_ADDR_CONFIG)); dev_info(adev->dev, " DMIF_ADDR_CALC=0x%08X\n", RREG32(mmDMIF_ADDR_CALC)); - dev_info(adev->dev, " SDMA0_TILING_CONFIG=0x%08X\n", - RREG32(mmSDMA0_TILING_CONFIG + SDMA0_REGISTER_OFFSET)); - dev_info(adev->dev, " SDMA1_TILING_CONFIG=0x%08X\n", - RREG32(mmSDMA0_TILING_CONFIG + SDMA1_REGISTER_OFFSET)); - dev_info(adev->dev, " UVD_UDEC_ADDR_CONFIG=0x%08X\n", - RREG32(mmUVD_UDEC_ADDR_CONFIG)); - dev_info(adev->dev, " UVD_UDEC_DB_ADDR_CONFIG=0x%08X\n", - RREG32(mmUVD_UDEC_DB_ADDR_CONFIG)); - dev_info(adev->dev, " UVD_UDEC_DBW_ADDR_CONFIG=0x%08X\n", - RREG32(mmUVD_UDEC_DBW_ADDR_CONFIG)); dev_info(adev->dev, " CP_MEQ_THRESHOLDS=0x%08X\n", RREG32(mmCP_MEQ_THRESHOLDS)); @@ -5509,14 +5133,14 @@ static int gfx_v7_0_set_powergating_state(void *handle, if (state == AMD_PG_STATE_GATE) gate = true; - if (adev->pg_flags & (AMDGPU_PG_SUPPORT_GFX_PG | - AMDGPU_PG_SUPPORT_GFX_SMG | - AMDGPU_PG_SUPPORT_GFX_DMG | - AMDGPU_PG_SUPPORT_CP | - AMDGPU_PG_SUPPORT_GDS | - AMDGPU_PG_SUPPORT_RLC_SMU_HS)) { + if (adev->pg_flags & (AMD_PG_SUPPORT_GFX_PG | + AMD_PG_SUPPORT_GFX_SMG | + AMD_PG_SUPPORT_GFX_DMG | + AMD_PG_SUPPORT_CP | + AMD_PG_SUPPORT_GDS | + AMD_PG_SUPPORT_RLC_SMU_HS)) { gfx_v7_0_update_gfx_pg(adev, gate); - if (adev->pg_flags & AMDGPU_PG_SUPPORT_GFX_PG) { + if (adev->pg_flags & AMD_PG_SUPPORT_GFX_PG) { gfx_v7_0_enable_cp_pg(adev, gate); gfx_v7_0_enable_gds_pg(adev, gate); } @@ -5527,7 +5151,7 @@ static int gfx_v7_0_set_powergating_state(void *handle, const struct amd_ip_funcs gfx_v7_0_ip_funcs = { .early_init = gfx_v7_0_early_init, - .late_init = NULL, + .late_init = gfx_v7_0_late_init, .sw_init = gfx_v7_0_sw_init, .sw_fini = gfx_v7_0_sw_fini, .hw_init = gfx_v7_0_hw_init, @@ -5542,24 +5166,6 @@ const struct amd_ip_funcs gfx_v7_0_ip_funcs = { .set_powergating_state = gfx_v7_0_set_powergating_state, }; -/** - * gfx_v7_0_ring_is_lockup - check if the 3D engine is locked up - * - * @adev: amdgpu_device pointer - * @ring: amdgpu_ring structure holding ring information - * - * Check if the 3D engine is locked up (CIK). - * Returns true if the engine is locked, false if not. - */ -static bool gfx_v7_0_ring_is_lockup(struct amdgpu_ring *ring) -{ - if (gfx_v7_0_is_idle(ring->adev)) { - amdgpu_ring_lockup_update(ring); - return false; - } - return amdgpu_ring_test_lockup(ring); -} - static const struct amdgpu_ring_funcs gfx_v7_0_ring_funcs_gfx = { .get_rptr = gfx_v7_0_ring_get_rptr_gfx, .get_wptr = gfx_v7_0_ring_get_wptr_gfx, @@ -5567,14 +5173,15 @@ static const struct amdgpu_ring_funcs gfx_v7_0_ring_funcs_gfx = { .parse_cs = NULL, .emit_ib = gfx_v7_0_ring_emit_ib_gfx, .emit_fence = gfx_v7_0_ring_emit_fence_gfx, - .emit_semaphore = gfx_v7_0_ring_emit_semaphore, + .emit_pipeline_sync = gfx_v7_0_ring_emit_pipeline_sync, .emit_vm_flush = gfx_v7_0_ring_emit_vm_flush, .emit_gds_switch = gfx_v7_0_ring_emit_gds_switch, .emit_hdp_flush = gfx_v7_0_ring_emit_hdp_flush, + .emit_hdp_invalidate = gfx_v7_0_ring_emit_hdp_invalidate, .test_ring = gfx_v7_0_ring_test_ring, .test_ib = gfx_v7_0_ring_test_ib, - .is_lockup = gfx_v7_0_ring_is_lockup, .insert_nop = amdgpu_ring_insert_nop, + .pad_ib = amdgpu_ring_generic_pad_ib, }; static const struct amdgpu_ring_funcs gfx_v7_0_ring_funcs_compute = { @@ -5584,14 +5191,15 @@ static const struct amdgpu_ring_funcs gfx_v7_0_ring_funcs_compute = { .parse_cs = NULL, .emit_ib = gfx_v7_0_ring_emit_ib_compute, .emit_fence = gfx_v7_0_ring_emit_fence_compute, - .emit_semaphore = gfx_v7_0_ring_emit_semaphore, + .emit_pipeline_sync = gfx_v7_0_ring_emit_pipeline_sync, .emit_vm_flush = gfx_v7_0_ring_emit_vm_flush, .emit_gds_switch = gfx_v7_0_ring_emit_gds_switch, .emit_hdp_flush = gfx_v7_0_ring_emit_hdp_flush, + .emit_hdp_invalidate = gfx_v7_0_ring_emit_hdp_invalidate, .test_ring = gfx_v7_0_ring_test_ring, .test_ib = gfx_v7_0_ring_test_ib, - .is_lockup = gfx_v7_0_ring_is_lockup, .insert_nop = amdgpu_ring_insert_nop, + .pad_ib = amdgpu_ring_generic_pad_ib, }; static void gfx_v7_0_set_ring_funcs(struct amdgpu_device *adev) @@ -5661,7 +5269,7 @@ static void gfx_v7_0_set_gds_init(struct amdgpu_device *adev) int gfx_v7_0_get_cu_info(struct amdgpu_device *adev, - struct amdgpu_cu_info *cu_info) + struct amdgpu_cu_info *cu_info) { int i, j, k, counter, active_cu_number = 0; u32 mask, bitmap, ao_bitmap, ao_cu_mask = 0; @@ -5669,16 +5277,19 @@ int gfx_v7_0_get_cu_info(struct amdgpu_device *adev, if (!adev || !cu_info) return -EINVAL; + memset(cu_info, 0, sizeof(*cu_info)); + mutex_lock(&adev->grbm_idx_mutex); for (i = 0; i < adev->gfx.config.max_shader_engines; i++) { for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) { mask = 1; ao_bitmap = 0; counter = 0; - bitmap = gfx_v7_0_get_cu_active_bitmap(adev, i, j); + gfx_v7_0_select_se_sh(adev, i, j); + bitmap = gfx_v7_0_get_cu_active_bitmap(adev); cu_info->bitmap[i][j] = bitmap; - for (k = 0; k < adev->gfx.config.max_cu_per_sh; k ++) { + for (k = 0; k < 16; k ++) { if (bitmap & mask) { if (counter < 2) ao_bitmap |= mask; @@ -5690,9 +5301,11 @@ int gfx_v7_0_get_cu_info(struct amdgpu_device *adev, ao_cu_mask |= (ao_bitmap << (i * 16 + j * 8)); } } + gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff); + mutex_unlock(&adev->grbm_idx_mutex); cu_info->number = active_cu_number; cu_info->ao_cu_mask = ao_cu_mask; - mutex_unlock(&adev->grbm_idx_mutex); + return 0; } diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c index cb4f68f53f24..f0c7b3596480 100644 --- a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c @@ -43,9 +43,6 @@ #include "gca/gfx_8_0_sh_mask.h" #include "gca/gfx_8_0_enum.h" -#include "uvd/uvd_5_0_d.h" -#include "uvd/uvd_5_0_sh_mask.h" - #include "dce/dce_10_0_d.h" #include "dce/dce_10_0_sh_mask.h" @@ -66,6 +63,27 @@ #define MACRO_TILE_ASPECT(x) ((x) << GB_MACROTILE_MODE0__MACRO_TILE_ASPECT__SHIFT) #define NUM_BANKS(x) ((x) << GB_MACROTILE_MODE0__NUM_BANKS__SHIFT) +#define RLC_CGTT_MGCG_OVERRIDE__CPF_MASK 0x00000001L +#define RLC_CGTT_MGCG_OVERRIDE__RLC_MASK 0x00000002L +#define RLC_CGTT_MGCG_OVERRIDE__MGCG_MASK 0x00000004L +#define RLC_CGTT_MGCG_OVERRIDE__CGCG_MASK 0x00000008L +#define RLC_CGTT_MGCG_OVERRIDE__CGLS_MASK 0x00000010L +#define RLC_CGTT_MGCG_OVERRIDE__GRBM_MASK 0x00000020L + +/* BPM SERDES CMD */ +#define SET_BPM_SERDES_CMD 1 +#define CLE_BPM_SERDES_CMD 0 + +/* BPM Register Address*/ +enum { + BPM_REG_CGLS_EN = 0, /* Enable/Disable CGLS */ + BPM_REG_CGLS_ON, /* ON/OFF CGLS: shall be controlled by RLC FW */ + BPM_REG_CGCG_OVERRIDE, /* Set/Clear CGCG Override */ + BPM_REG_MGCG_OVERRIDE, /* Set/Clear MGCG Override */ + BPM_REG_FGCG_OVERRIDE, /* Set/Clear FGCG Override */ + BPM_REG_FGCG_MAX +}; + MODULE_FIRMWARE("amdgpu/carrizo_ce.bin"); MODULE_FIRMWARE("amdgpu/carrizo_pfp.bin"); MODULE_FIRMWARE("amdgpu/carrizo_me.bin"); @@ -73,6 +91,12 @@ MODULE_FIRMWARE("amdgpu/carrizo_mec.bin"); MODULE_FIRMWARE("amdgpu/carrizo_mec2.bin"); MODULE_FIRMWARE("amdgpu/carrizo_rlc.bin"); +MODULE_FIRMWARE("amdgpu/stoney_ce.bin"); +MODULE_FIRMWARE("amdgpu/stoney_pfp.bin"); +MODULE_FIRMWARE("amdgpu/stoney_me.bin"); +MODULE_FIRMWARE("amdgpu/stoney_mec.bin"); +MODULE_FIRMWARE("amdgpu/stoney_rlc.bin"); + MODULE_FIRMWARE("amdgpu/tonga_ce.bin"); MODULE_FIRMWARE("amdgpu/tonga_pfp.bin"); MODULE_FIRMWARE("amdgpu/tonga_me.bin"); @@ -84,7 +108,6 @@ MODULE_FIRMWARE("amdgpu/topaz_ce.bin"); MODULE_FIRMWARE("amdgpu/topaz_pfp.bin"); MODULE_FIRMWARE("amdgpu/topaz_me.bin"); MODULE_FIRMWARE("amdgpu/topaz_mec.bin"); -MODULE_FIRMWARE("amdgpu/topaz_mec2.bin"); MODULE_FIRMWARE("amdgpu/topaz_rlc.bin"); MODULE_FIRMWARE("amdgpu/fiji_ce.bin"); @@ -229,11 +252,13 @@ static const u32 fiji_golden_common_all[] = mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000, mmPA_SC_RASTER_CONFIG, 0xffffffff, 0x3a00161a, mmPA_SC_RASTER_CONFIG_1, 0xffffffff, 0x0000002e, - mmGB_ADDR_CONFIG, 0xffffffff, 0x12011003, + mmGB_ADDR_CONFIG, 0xffffffff, 0x22011003, mmSPI_RESOURCE_RESERVE_CU_0, 0xffffffff, 0x00000800, mmSPI_RESOURCE_RESERVE_CU_1, 0xffffffff, 0x00000800, mmSPI_RESOURCE_RESERVE_EN_CU_0, 0xffffffff, 0x00007FBF, - mmSPI_RESOURCE_RESERVE_EN_CU_1, 0xffffffff, 0x00007FAF + mmSPI_RESOURCE_RESERVE_EN_CU_1, 0xffffffff, 0x00007FAF, + mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000, + mmSPI_CONFIG_CNTL_1, 0x0000000f, 0x00000009, }; static const u32 golden_settings_fiji_a10[] = @@ -241,18 +266,19 @@ static const u32 golden_settings_fiji_a10[] = mmCB_HW_CONTROL_3, 0x000001ff, 0x00000040, mmDB_DEBUG2, 0xf00fffff, 0x00000400, mmPA_SC_ENHANCE, 0xffffffff, 0x20000001, - mmPA_SC_FIFO_DEPTH_CNTL, 0x000003ff, 0x00000100, mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000, + mmRLC_CGCG_CGLS_CTRL, 0x00000003, 0x0001003c, + mmSQ_RANDOM_WAVE_PRI, 0x001fffff, 0x000006fd, mmTA_CNTL_AUX, 0x000f000f, 0x000b0000, - mmTCC_CTRL, 0x00100000, 0xf30fff7f, + mmTCC_CTRL, 0x00100000, 0xf31fff7f, + mmTCC_EXE_DISABLE, 0x00000002, 0x00000002, mmTCP_ADDR_CONFIG, 0x000003ff, 0x000000ff, - mmTCP_CHAN_STEER_HI, 0xffffffff, 0x7d6cf5e4, - mmTCP_CHAN_STEER_LO, 0xffffffff, 0x3928b1a0, + mmVGT_RESET_DEBUG, 0x00000004, 0x00000004, }; static const u32 fiji_mgcg_cgcg_init[] = { - mmRLC_CGTT_MGCG_OVERRIDE, 0xffffffff, 0xffffffc0, + mmRLC_CGTT_MGCG_OVERRIDE, 0xffffffff, 0xffffffff, mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000, mmCB_CGTT_SCLK_CTRL, 0xffffffff, 0x00000100, mmCGTT_BCI_CLK_CTRL, 0xffffffff, 0x00000100, @@ -493,6 +519,42 @@ static const u32 cz_mgcg_cgcg_init[] = mmCP_MEM_SLP_CNTL, 0x00000001, 0x00000001, }; +static const u32 stoney_golden_settings_a11[] = +{ + mmDB_DEBUG2, 0xf00fffff, 0x00000400, + mmGB_GPU_ID, 0x0000000f, 0x00000000, + mmPA_SC_ENHANCE, 0xffffffff, 0x20000001, + mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000, + mmRLC_CGCG_CGLS_CTRL, 0x00000003, 0x0001003c, + mmTA_CNTL_AUX, 0x000f000f, 0x000b0000, + mmTCC_CTRL, 0x00100000, 0xf31fff7f, + mmTCC_EXE_DISABLE, 0x00000002, 0x00000002, + mmTCP_ADDR_CONFIG, 0x0000000f, 0x000000f1, + mmTCP_CHAN_STEER_LO, 0xffffffff, 0x10101010, +}; + +static const u32 stoney_golden_common_all[] = +{ + mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000, + mmPA_SC_RASTER_CONFIG, 0xffffffff, 0x00000000, + mmPA_SC_RASTER_CONFIG_1, 0xffffffff, 0x00000000, + mmGB_ADDR_CONFIG, 0xffffffff, 0x12010001, + mmSPI_RESOURCE_RESERVE_CU_0, 0xffffffff, 0x00000800, + mmSPI_RESOURCE_RESERVE_CU_1, 0xffffffff, 0x00000800, + mmSPI_RESOURCE_RESERVE_EN_CU_0, 0xffffffff, 0x00007FBF, + mmSPI_RESOURCE_RESERVE_EN_CU_1, 0xffffffff, 0x00007FAF, +}; + +static const u32 stoney_mgcg_cgcg_init[] = +{ + mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000, + mmRLC_CGCG_CGLS_CTRL, 0xffffffff, 0x0020003f, + mmCP_MEM_SLP_CNTL, 0xffffffff, 0x00020201, + mmRLC_MEM_SLP_CNTL, 0xffffffff, 0x00020201, + mmCGTS_SM_CTRL_REG, 0xffffffff, 0x96940200, + mmATC_MISC_CG, 0xffffffff, 0x000c0200, +}; + static void gfx_v8_0_set_ring_funcs(struct amdgpu_device *adev); static void gfx_v8_0_set_irq_funcs(struct amdgpu_device *adev); static void gfx_v8_0_set_gds_init(struct amdgpu_device *adev); @@ -545,6 +607,17 @@ static void gfx_v8_0_init_golden_registers(struct amdgpu_device *adev) cz_golden_common_all, (const u32)ARRAY_SIZE(cz_golden_common_all)); break; + case CHIP_STONEY: + amdgpu_program_register_sequence(adev, + stoney_mgcg_cgcg_init, + (const u32)ARRAY_SIZE(stoney_mgcg_cgcg_init)); + amdgpu_program_register_sequence(adev, + stoney_golden_settings_a11, + (const u32)ARRAY_SIZE(stoney_golden_settings_a11)); + amdgpu_program_register_sequence(adev, + stoney_golden_common_all, + (const u32)ARRAY_SIZE(stoney_golden_common_all)); + break; default: break; } @@ -576,7 +649,7 @@ static int gfx_v8_0_ring_test_ring(struct amdgpu_ring *ring) return r; } WREG32(scratch, 0xCAFEDEAD); - r = amdgpu_ring_lock(ring, 3); + r = amdgpu_ring_alloc(ring, 3); if (r) { DRM_ERROR("amdgpu: cp failed to lock ring %d (%d).\n", ring->idx, r); @@ -586,7 +659,7 @@ static int gfx_v8_0_ring_test_ring(struct amdgpu_ring *ring) amdgpu_ring_write(ring, PACKET3(PACKET3_SET_UCONFIG_REG, 1)); amdgpu_ring_write(ring, (scratch - PACKET3_SET_UCONFIG_REG_START)); amdgpu_ring_write(ring, 0xDEADBEEF); - amdgpu_ring_unlock_commit(ring); + amdgpu_ring_commit(ring); for (i = 0; i < adev->usec_timeout; i++) { tmp = RREG32(scratch); @@ -623,7 +696,7 @@ static int gfx_v8_0_ring_test_ib(struct amdgpu_ring *ring) } WREG32(scratch, 0xCAFEDEAD); memset(&ib, 0, sizeof(ib)); - r = amdgpu_ib_get(ring, NULL, 256, &ib); + r = amdgpu_ib_get(adev, NULL, 256, &ib); if (r) { DRM_ERROR("amdgpu: failed to get ib (%d).\n", r); goto err1; @@ -633,9 +706,7 @@ static int gfx_v8_0_ring_test_ib(struct amdgpu_ring *ring) ib.ptr[2] = 0xDEADBEEF; ib.length_dw = 3; - r = amdgpu_sched_ib_submit_kernel_helper(adev, ring, &ib, 1, NULL, - AMDGPU_FENCE_OWNER_UNDEFINED, - &f); + r = amdgpu_ib_schedule(ring, 1, &ib, NULL, &f); if (r) goto err2; @@ -661,7 +732,8 @@ static int gfx_v8_0_ring_test_ib(struct amdgpu_ring *ring) } err2: fence_put(f); - amdgpu_ib_free(adev, &ib); + amdgpu_ib_free(adev, &ib, NULL); + fence_put(f); err1: amdgpu_gfx_scratch_free(adev, scratch); return r; @@ -691,6 +763,9 @@ static int gfx_v8_0_init_microcode(struct amdgpu_device *adev) case CHIP_FIJI: chip_name = "fiji"; break; + case CHIP_STONEY: + chip_name = "stoney"; + break; default: BUG(); } @@ -748,21 +823,24 @@ static int gfx_v8_0_init_microcode(struct amdgpu_device *adev) adev->gfx.mec_fw_version = le32_to_cpu(cp_hdr->header.ucode_version); adev->gfx.mec_feature_version = le32_to_cpu(cp_hdr->ucode_feature_version); - snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mec2.bin", chip_name); - err = request_firmware(&adev->gfx.mec2_fw, fw_name, adev->dev); - if (!err) { - err = amdgpu_ucode_validate(adev->gfx.mec2_fw); - if (err) - goto out; - cp_hdr = (const struct gfx_firmware_header_v1_0 *) - adev->gfx.mec2_fw->data; - adev->gfx.mec2_fw_version = le32_to_cpu( - cp_hdr->header.ucode_version); - adev->gfx.mec2_feature_version = le32_to_cpu( - cp_hdr->ucode_feature_version); - } else { - err = 0; - adev->gfx.mec2_fw = NULL; + if ((adev->asic_type != CHIP_STONEY) && + (adev->asic_type != CHIP_TOPAZ)) { + snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mec2.bin", chip_name); + err = request_firmware(&adev->gfx.mec2_fw, fw_name, adev->dev); + if (!err) { + err = amdgpu_ucode_validate(adev->gfx.mec2_fw); + if (err) + goto out; + cp_hdr = (const struct gfx_firmware_header_v1_0 *) + adev->gfx.mec2_fw->data; + adev->gfx.mec2_fw_version = + le32_to_cpu(cp_hdr->header.ucode_version); + adev->gfx.mec2_feature_version = + le32_to_cpu(cp_hdr->ucode_feature_version); + } else { + err = 0; + adev->gfx.mec2_fw = NULL; + } } if (adev->firmware.smu_load) { @@ -903,6 +981,547 @@ static int gfx_v8_0_mec_init(struct amdgpu_device *adev) return 0; } +static const u32 vgpr_init_compute_shader[] = +{ + 0x7e000209, 0x7e020208, + 0x7e040207, 0x7e060206, + 0x7e080205, 0x7e0a0204, + 0x7e0c0203, 0x7e0e0202, + 0x7e100201, 0x7e120200, + 0x7e140209, 0x7e160208, + 0x7e180207, 0x7e1a0206, + 0x7e1c0205, 0x7e1e0204, + 0x7e200203, 0x7e220202, + 0x7e240201, 0x7e260200, + 0x7e280209, 0x7e2a0208, + 0x7e2c0207, 0x7e2e0206, + 0x7e300205, 0x7e320204, + 0x7e340203, 0x7e360202, + 0x7e380201, 0x7e3a0200, + 0x7e3c0209, 0x7e3e0208, + 0x7e400207, 0x7e420206, + 0x7e440205, 0x7e460204, + 0x7e480203, 0x7e4a0202, + 0x7e4c0201, 0x7e4e0200, + 0x7e500209, 0x7e520208, + 0x7e540207, 0x7e560206, + 0x7e580205, 0x7e5a0204, + 0x7e5c0203, 0x7e5e0202, + 0x7e600201, 0x7e620200, + 0x7e640209, 0x7e660208, + 0x7e680207, 0x7e6a0206, + 0x7e6c0205, 0x7e6e0204, + 0x7e700203, 0x7e720202, + 0x7e740201, 0x7e760200, + 0x7e780209, 0x7e7a0208, + 0x7e7c0207, 0x7e7e0206, + 0xbf8a0000, 0xbf810000, +}; + +static const u32 sgpr_init_compute_shader[] = +{ + 0xbe8a0100, 0xbe8c0102, + 0xbe8e0104, 0xbe900106, + 0xbe920108, 0xbe940100, + 0xbe960102, 0xbe980104, + 0xbe9a0106, 0xbe9c0108, + 0xbe9e0100, 0xbea00102, + 0xbea20104, 0xbea40106, + 0xbea60108, 0xbea80100, + 0xbeaa0102, 0xbeac0104, + 0xbeae0106, 0xbeb00108, + 0xbeb20100, 0xbeb40102, + 0xbeb60104, 0xbeb80106, + 0xbeba0108, 0xbebc0100, + 0xbebe0102, 0xbec00104, + 0xbec20106, 0xbec40108, + 0xbec60100, 0xbec80102, + 0xbee60004, 0xbee70005, + 0xbeea0006, 0xbeeb0007, + 0xbee80008, 0xbee90009, + 0xbefc0000, 0xbf8a0000, + 0xbf810000, 0x00000000, +}; + +static const u32 vgpr_init_regs[] = +{ + mmCOMPUTE_STATIC_THREAD_MGMT_SE0, 0xffffffff, + mmCOMPUTE_RESOURCE_LIMITS, 0, + mmCOMPUTE_NUM_THREAD_X, 256*4, + mmCOMPUTE_NUM_THREAD_Y, 1, + mmCOMPUTE_NUM_THREAD_Z, 1, + mmCOMPUTE_PGM_RSRC2, 20, + mmCOMPUTE_USER_DATA_0, 0xedcedc00, + mmCOMPUTE_USER_DATA_1, 0xedcedc01, + mmCOMPUTE_USER_DATA_2, 0xedcedc02, + mmCOMPUTE_USER_DATA_3, 0xedcedc03, + mmCOMPUTE_USER_DATA_4, 0xedcedc04, + mmCOMPUTE_USER_DATA_5, 0xedcedc05, + mmCOMPUTE_USER_DATA_6, 0xedcedc06, + mmCOMPUTE_USER_DATA_7, 0xedcedc07, + mmCOMPUTE_USER_DATA_8, 0xedcedc08, + mmCOMPUTE_USER_DATA_9, 0xedcedc09, +}; + +static const u32 sgpr1_init_regs[] = +{ + mmCOMPUTE_STATIC_THREAD_MGMT_SE0, 0x0f, + mmCOMPUTE_RESOURCE_LIMITS, 0x1000000, + mmCOMPUTE_NUM_THREAD_X, 256*5, + mmCOMPUTE_NUM_THREAD_Y, 1, + mmCOMPUTE_NUM_THREAD_Z, 1, + mmCOMPUTE_PGM_RSRC2, 20, + mmCOMPUTE_USER_DATA_0, 0xedcedc00, + mmCOMPUTE_USER_DATA_1, 0xedcedc01, + mmCOMPUTE_USER_DATA_2, 0xedcedc02, + mmCOMPUTE_USER_DATA_3, 0xedcedc03, + mmCOMPUTE_USER_DATA_4, 0xedcedc04, + mmCOMPUTE_USER_DATA_5, 0xedcedc05, + mmCOMPUTE_USER_DATA_6, 0xedcedc06, + mmCOMPUTE_USER_DATA_7, 0xedcedc07, + mmCOMPUTE_USER_DATA_8, 0xedcedc08, + mmCOMPUTE_USER_DATA_9, 0xedcedc09, +}; + +static const u32 sgpr2_init_regs[] = +{ + mmCOMPUTE_STATIC_THREAD_MGMT_SE0, 0xf0, + mmCOMPUTE_RESOURCE_LIMITS, 0x1000000, + mmCOMPUTE_NUM_THREAD_X, 256*5, + mmCOMPUTE_NUM_THREAD_Y, 1, + mmCOMPUTE_NUM_THREAD_Z, 1, + mmCOMPUTE_PGM_RSRC2, 20, + mmCOMPUTE_USER_DATA_0, 0xedcedc00, + mmCOMPUTE_USER_DATA_1, 0xedcedc01, + mmCOMPUTE_USER_DATA_2, 0xedcedc02, + mmCOMPUTE_USER_DATA_3, 0xedcedc03, + mmCOMPUTE_USER_DATA_4, 0xedcedc04, + mmCOMPUTE_USER_DATA_5, 0xedcedc05, + mmCOMPUTE_USER_DATA_6, 0xedcedc06, + mmCOMPUTE_USER_DATA_7, 0xedcedc07, + mmCOMPUTE_USER_DATA_8, 0xedcedc08, + mmCOMPUTE_USER_DATA_9, 0xedcedc09, +}; + +static const u32 sec_ded_counter_registers[] = +{ + mmCPC_EDC_ATC_CNT, + mmCPC_EDC_SCRATCH_CNT, + mmCPC_EDC_UCODE_CNT, + mmCPF_EDC_ATC_CNT, + mmCPF_EDC_ROQ_CNT, + mmCPF_EDC_TAG_CNT, + mmCPG_EDC_ATC_CNT, + mmCPG_EDC_DMA_CNT, + mmCPG_EDC_TAG_CNT, + mmDC_EDC_CSINVOC_CNT, + mmDC_EDC_RESTORE_CNT, + mmDC_EDC_STATE_CNT, + mmGDS_EDC_CNT, + mmGDS_EDC_GRBM_CNT, + mmGDS_EDC_OA_DED, + mmSPI_EDC_CNT, + mmSQC_ATC_EDC_GATCL1_CNT, + mmSQC_EDC_CNT, + mmSQ_EDC_DED_CNT, + mmSQ_EDC_INFO, + mmSQ_EDC_SEC_CNT, + mmTCC_EDC_CNT, + mmTCP_ATC_EDC_GATCL1_CNT, + mmTCP_EDC_CNT, + mmTD_EDC_CNT +}; + +static int gfx_v8_0_do_edc_gpr_workarounds(struct amdgpu_device *adev) +{ + struct amdgpu_ring *ring = &adev->gfx.compute_ring[0]; + struct amdgpu_ib ib; + struct fence *f = NULL; + int r, i; + u32 tmp; + unsigned total_size, vgpr_offset, sgpr_offset; + u64 gpu_addr; + + /* only supported on CZ */ + if (adev->asic_type != CHIP_CARRIZO) + return 0; + + /* bail if the compute ring is not ready */ + if (!ring->ready) + return 0; + + tmp = RREG32(mmGB_EDC_MODE); + WREG32(mmGB_EDC_MODE, 0); + + total_size = + (((ARRAY_SIZE(vgpr_init_regs) / 2) * 3) + 4 + 5 + 2) * 4; + total_size += + (((ARRAY_SIZE(sgpr1_init_regs) / 2) * 3) + 4 + 5 + 2) * 4; + total_size += + (((ARRAY_SIZE(sgpr2_init_regs) / 2) * 3) + 4 + 5 + 2) * 4; + total_size = ALIGN(total_size, 256); + vgpr_offset = total_size; + total_size += ALIGN(sizeof(vgpr_init_compute_shader), 256); + sgpr_offset = total_size; + total_size += sizeof(sgpr_init_compute_shader); + + /* allocate an indirect buffer to put the commands in */ + memset(&ib, 0, sizeof(ib)); + r = amdgpu_ib_get(adev, NULL, total_size, &ib); + if (r) { + DRM_ERROR("amdgpu: failed to get ib (%d).\n", r); + return r; + } + + /* load the compute shaders */ + for (i = 0; i < ARRAY_SIZE(vgpr_init_compute_shader); i++) + ib.ptr[i + (vgpr_offset / 4)] = vgpr_init_compute_shader[i]; + + for (i = 0; i < ARRAY_SIZE(sgpr_init_compute_shader); i++) + ib.ptr[i + (sgpr_offset / 4)] = sgpr_init_compute_shader[i]; + + /* init the ib length to 0 */ + ib.length_dw = 0; + + /* VGPR */ + /* write the register state for the compute dispatch */ + for (i = 0; i < ARRAY_SIZE(vgpr_init_regs); i += 2) { + ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 1); + ib.ptr[ib.length_dw++] = vgpr_init_regs[i] - PACKET3_SET_SH_REG_START; + ib.ptr[ib.length_dw++] = vgpr_init_regs[i + 1]; + } + /* write the shader start address: mmCOMPUTE_PGM_LO, mmCOMPUTE_PGM_HI */ + gpu_addr = (ib.gpu_addr + (u64)vgpr_offset) >> 8; + ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 2); + ib.ptr[ib.length_dw++] = mmCOMPUTE_PGM_LO - PACKET3_SET_SH_REG_START; + ib.ptr[ib.length_dw++] = lower_32_bits(gpu_addr); + ib.ptr[ib.length_dw++] = upper_32_bits(gpu_addr); + + /* write dispatch packet */ + ib.ptr[ib.length_dw++] = PACKET3(PACKET3_DISPATCH_DIRECT, 3); + ib.ptr[ib.length_dw++] = 8; /* x */ + ib.ptr[ib.length_dw++] = 1; /* y */ + ib.ptr[ib.length_dw++] = 1; /* z */ + ib.ptr[ib.length_dw++] = + REG_SET_FIELD(0, COMPUTE_DISPATCH_INITIATOR, COMPUTE_SHADER_EN, 1); + + /* write CS partial flush packet */ + ib.ptr[ib.length_dw++] = PACKET3(PACKET3_EVENT_WRITE, 0); + ib.ptr[ib.length_dw++] = EVENT_TYPE(7) | EVENT_INDEX(4); + + /* SGPR1 */ + /* write the register state for the compute dispatch */ + for (i = 0; i < ARRAY_SIZE(sgpr1_init_regs); i += 2) { + ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 1); + ib.ptr[ib.length_dw++] = sgpr1_init_regs[i] - PACKET3_SET_SH_REG_START; + ib.ptr[ib.length_dw++] = sgpr1_init_regs[i + 1]; + } + /* write the shader start address: mmCOMPUTE_PGM_LO, mmCOMPUTE_PGM_HI */ + gpu_addr = (ib.gpu_addr + (u64)sgpr_offset) >> 8; + ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 2); + ib.ptr[ib.length_dw++] = mmCOMPUTE_PGM_LO - PACKET3_SET_SH_REG_START; + ib.ptr[ib.length_dw++] = lower_32_bits(gpu_addr); + ib.ptr[ib.length_dw++] = upper_32_bits(gpu_addr); + + /* write dispatch packet */ + ib.ptr[ib.length_dw++] = PACKET3(PACKET3_DISPATCH_DIRECT, 3); + ib.ptr[ib.length_dw++] = 8; /* x */ + ib.ptr[ib.length_dw++] = 1; /* y */ + ib.ptr[ib.length_dw++] = 1; /* z */ + ib.ptr[ib.length_dw++] = + REG_SET_FIELD(0, COMPUTE_DISPATCH_INITIATOR, COMPUTE_SHADER_EN, 1); + + /* write CS partial flush packet */ + ib.ptr[ib.length_dw++] = PACKET3(PACKET3_EVENT_WRITE, 0); + ib.ptr[ib.length_dw++] = EVENT_TYPE(7) | EVENT_INDEX(4); + + /* SGPR2 */ + /* write the register state for the compute dispatch */ + for (i = 0; i < ARRAY_SIZE(sgpr2_init_regs); i += 2) { + ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 1); + ib.ptr[ib.length_dw++] = sgpr2_init_regs[i] - PACKET3_SET_SH_REG_START; + ib.ptr[ib.length_dw++] = sgpr2_init_regs[i + 1]; + } + /* write the shader start address: mmCOMPUTE_PGM_LO, mmCOMPUTE_PGM_HI */ + gpu_addr = (ib.gpu_addr + (u64)sgpr_offset) >> 8; + ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 2); + ib.ptr[ib.length_dw++] = mmCOMPUTE_PGM_LO - PACKET3_SET_SH_REG_START; + ib.ptr[ib.length_dw++] = lower_32_bits(gpu_addr); + ib.ptr[ib.length_dw++] = upper_32_bits(gpu_addr); + + /* write dispatch packet */ + ib.ptr[ib.length_dw++] = PACKET3(PACKET3_DISPATCH_DIRECT, 3); + ib.ptr[ib.length_dw++] = 8; /* x */ + ib.ptr[ib.length_dw++] = 1; /* y */ + ib.ptr[ib.length_dw++] = 1; /* z */ + ib.ptr[ib.length_dw++] = + REG_SET_FIELD(0, COMPUTE_DISPATCH_INITIATOR, COMPUTE_SHADER_EN, 1); + + /* write CS partial flush packet */ + ib.ptr[ib.length_dw++] = PACKET3(PACKET3_EVENT_WRITE, 0); + ib.ptr[ib.length_dw++] = EVENT_TYPE(7) | EVENT_INDEX(4); + + /* shedule the ib on the ring */ + r = amdgpu_ib_schedule(ring, 1, &ib, NULL, &f); + if (r) { + DRM_ERROR("amdgpu: ib submit failed (%d).\n", r); + goto fail; + } + + /* wait for the GPU to finish processing the IB */ + r = fence_wait(f, false); + if (r) { + DRM_ERROR("amdgpu: fence wait failed (%d).\n", r); + goto fail; + } + + tmp = REG_SET_FIELD(tmp, GB_EDC_MODE, DED_MODE, 2); + tmp = REG_SET_FIELD(tmp, GB_EDC_MODE, PROP_FED, 1); + WREG32(mmGB_EDC_MODE, tmp); + + tmp = RREG32(mmCC_GC_EDC_CONFIG); + tmp = REG_SET_FIELD(tmp, CC_GC_EDC_CONFIG, DIS_EDC, 0) | 1; + WREG32(mmCC_GC_EDC_CONFIG, tmp); + + + /* read back registers to clear the counters */ + for (i = 0; i < ARRAY_SIZE(sec_ded_counter_registers); i++) + RREG32(sec_ded_counter_registers[i]); + +fail: + fence_put(f); + amdgpu_ib_free(adev, &ib, NULL); + fence_put(f); + + return r; +} + +static void gfx_v8_0_gpu_early_init(struct amdgpu_device *adev) +{ + u32 gb_addr_config; + u32 mc_shared_chmap, mc_arb_ramcfg; + u32 dimm00_addr_map, dimm01_addr_map, dimm10_addr_map, dimm11_addr_map; + u32 tmp; + + switch (adev->asic_type) { + case CHIP_TOPAZ: + adev->gfx.config.max_shader_engines = 1; + adev->gfx.config.max_tile_pipes = 2; + adev->gfx.config.max_cu_per_sh = 6; + adev->gfx.config.max_sh_per_se = 1; + adev->gfx.config.max_backends_per_se = 2; + adev->gfx.config.max_texture_channel_caches = 2; + adev->gfx.config.max_gprs = 256; + adev->gfx.config.max_gs_threads = 32; + adev->gfx.config.max_hw_contexts = 8; + + adev->gfx.config.sc_prim_fifo_size_frontend = 0x20; + adev->gfx.config.sc_prim_fifo_size_backend = 0x100; + adev->gfx.config.sc_hiz_tile_fifo_size = 0x30; + adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130; + gb_addr_config = TOPAZ_GB_ADDR_CONFIG_GOLDEN; + break; + case CHIP_FIJI: + adev->gfx.config.max_shader_engines = 4; + adev->gfx.config.max_tile_pipes = 16; + adev->gfx.config.max_cu_per_sh = 16; + adev->gfx.config.max_sh_per_se = 1; + adev->gfx.config.max_backends_per_se = 4; + adev->gfx.config.max_texture_channel_caches = 16; + adev->gfx.config.max_gprs = 256; + adev->gfx.config.max_gs_threads = 32; + adev->gfx.config.max_hw_contexts = 8; + + adev->gfx.config.sc_prim_fifo_size_frontend = 0x20; + adev->gfx.config.sc_prim_fifo_size_backend = 0x100; + adev->gfx.config.sc_hiz_tile_fifo_size = 0x30; + adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130; + gb_addr_config = TONGA_GB_ADDR_CONFIG_GOLDEN; + break; + case CHIP_TONGA: + adev->gfx.config.max_shader_engines = 4; + adev->gfx.config.max_tile_pipes = 8; + adev->gfx.config.max_cu_per_sh = 8; + adev->gfx.config.max_sh_per_se = 1; + adev->gfx.config.max_backends_per_se = 2; + adev->gfx.config.max_texture_channel_caches = 8; + adev->gfx.config.max_gprs = 256; + adev->gfx.config.max_gs_threads = 32; + adev->gfx.config.max_hw_contexts = 8; + + adev->gfx.config.sc_prim_fifo_size_frontend = 0x20; + adev->gfx.config.sc_prim_fifo_size_backend = 0x100; + adev->gfx.config.sc_hiz_tile_fifo_size = 0x30; + adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130; + gb_addr_config = TONGA_GB_ADDR_CONFIG_GOLDEN; + break; + case CHIP_CARRIZO: + adev->gfx.config.max_shader_engines = 1; + adev->gfx.config.max_tile_pipes = 2; + adev->gfx.config.max_sh_per_se = 1; + adev->gfx.config.max_backends_per_se = 2; + + switch (adev->pdev->revision) { + case 0xc4: + case 0x84: + case 0xc8: + case 0xcc: + case 0xe1: + case 0xe3: + /* B10 */ + adev->gfx.config.max_cu_per_sh = 8; + break; + case 0xc5: + case 0x81: + case 0x85: + case 0xc9: + case 0xcd: + case 0xe2: + case 0xe4: + /* B8 */ + adev->gfx.config.max_cu_per_sh = 6; + break; + case 0xc6: + case 0xca: + case 0xce: + case 0x88: + /* B6 */ + adev->gfx.config.max_cu_per_sh = 6; + break; + case 0xc7: + case 0x87: + case 0xcb: + case 0xe5: + case 0x89: + default: + /* B4 */ + adev->gfx.config.max_cu_per_sh = 4; + break; + } + + adev->gfx.config.max_texture_channel_caches = 2; + adev->gfx.config.max_gprs = 256; + adev->gfx.config.max_gs_threads = 32; + adev->gfx.config.max_hw_contexts = 8; + + adev->gfx.config.sc_prim_fifo_size_frontend = 0x20; + adev->gfx.config.sc_prim_fifo_size_backend = 0x100; + adev->gfx.config.sc_hiz_tile_fifo_size = 0x30; + adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130; + gb_addr_config = CARRIZO_GB_ADDR_CONFIG_GOLDEN; + break; + case CHIP_STONEY: + adev->gfx.config.max_shader_engines = 1; + adev->gfx.config.max_tile_pipes = 2; + adev->gfx.config.max_sh_per_se = 1; + adev->gfx.config.max_backends_per_se = 1; + + switch (adev->pdev->revision) { + case 0xc0: + case 0xc1: + case 0xc2: + case 0xc4: + case 0xc8: + case 0xc9: + adev->gfx.config.max_cu_per_sh = 3; + break; + case 0xd0: + case 0xd1: + case 0xd2: + default: + adev->gfx.config.max_cu_per_sh = 2; + break; + } + + adev->gfx.config.max_texture_channel_caches = 2; + adev->gfx.config.max_gprs = 256; + adev->gfx.config.max_gs_threads = 16; + adev->gfx.config.max_hw_contexts = 8; + + adev->gfx.config.sc_prim_fifo_size_frontend = 0x20; + adev->gfx.config.sc_prim_fifo_size_backend = 0x100; + adev->gfx.config.sc_hiz_tile_fifo_size = 0x30; + adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130; + gb_addr_config = CARRIZO_GB_ADDR_CONFIG_GOLDEN; + break; + default: + adev->gfx.config.max_shader_engines = 2; + adev->gfx.config.max_tile_pipes = 4; + adev->gfx.config.max_cu_per_sh = 2; + adev->gfx.config.max_sh_per_se = 1; + adev->gfx.config.max_backends_per_se = 2; + adev->gfx.config.max_texture_channel_caches = 4; + adev->gfx.config.max_gprs = 256; + adev->gfx.config.max_gs_threads = 32; + adev->gfx.config.max_hw_contexts = 8; + + adev->gfx.config.sc_prim_fifo_size_frontend = 0x20; + adev->gfx.config.sc_prim_fifo_size_backend = 0x100; + adev->gfx.config.sc_hiz_tile_fifo_size = 0x30; + adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130; + gb_addr_config = TONGA_GB_ADDR_CONFIG_GOLDEN; + break; + } + + mc_shared_chmap = RREG32(mmMC_SHARED_CHMAP); + adev->gfx.config.mc_arb_ramcfg = RREG32(mmMC_ARB_RAMCFG); + mc_arb_ramcfg = adev->gfx.config.mc_arb_ramcfg; + + adev->gfx.config.num_tile_pipes = adev->gfx.config.max_tile_pipes; + adev->gfx.config.mem_max_burst_length_bytes = 256; + if (adev->flags & AMD_IS_APU) { + /* Get memory bank mapping mode. */ + tmp = RREG32(mmMC_FUS_DRAM0_BANK_ADDR_MAPPING); + dimm00_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM0_BANK_ADDR_MAPPING, DIMM0ADDRMAP); + dimm01_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM0_BANK_ADDR_MAPPING, DIMM1ADDRMAP); + + tmp = RREG32(mmMC_FUS_DRAM1_BANK_ADDR_MAPPING); + dimm10_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM1_BANK_ADDR_MAPPING, DIMM0ADDRMAP); + dimm11_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM1_BANK_ADDR_MAPPING, DIMM1ADDRMAP); + + /* Validate settings in case only one DIMM installed. */ + if ((dimm00_addr_map == 0) || (dimm00_addr_map == 3) || (dimm00_addr_map == 4) || (dimm00_addr_map > 12)) + dimm00_addr_map = 0; + if ((dimm01_addr_map == 0) || (dimm01_addr_map == 3) || (dimm01_addr_map == 4) || (dimm01_addr_map > 12)) + dimm01_addr_map = 0; + if ((dimm10_addr_map == 0) || (dimm10_addr_map == 3) || (dimm10_addr_map == 4) || (dimm10_addr_map > 12)) + dimm10_addr_map = 0; + if ((dimm11_addr_map == 0) || (dimm11_addr_map == 3) || (dimm11_addr_map == 4) || (dimm11_addr_map > 12)) + dimm11_addr_map = 0; + + /* If DIMM Addr map is 8GB, ROW size should be 2KB. Otherwise 1KB. */ + /* If ROW size(DIMM1) != ROW size(DMIMM0), ROW size should be larger one. */ + if ((dimm00_addr_map == 11) || (dimm01_addr_map == 11) || (dimm10_addr_map == 11) || (dimm11_addr_map == 11)) + adev->gfx.config.mem_row_size_in_kb = 2; + else + adev->gfx.config.mem_row_size_in_kb = 1; + } else { + tmp = REG_GET_FIELD(mc_arb_ramcfg, MC_ARB_RAMCFG, NOOFCOLS); + adev->gfx.config.mem_row_size_in_kb = (4 * (1 << (8 + tmp))) / 1024; + if (adev->gfx.config.mem_row_size_in_kb > 4) + adev->gfx.config.mem_row_size_in_kb = 4; + } + + adev->gfx.config.shader_engine_tile_size = 32; + adev->gfx.config.num_gpus = 1; + adev->gfx.config.multi_gpu_tile_size = 64; + + /* fix up row size */ + switch (adev->gfx.config.mem_row_size_in_kb) { + case 1: + default: + gb_addr_config = REG_SET_FIELD(gb_addr_config, GB_ADDR_CONFIG, ROW_SIZE, 0); + break; + case 2: + gb_addr_config = REG_SET_FIELD(gb_addr_config, GB_ADDR_CONFIG, ROW_SIZE, 1); + break; + case 4: + gb_addr_config = REG_SET_FIELD(gb_addr_config, GB_ADDR_CONFIG, ROW_SIZE, 2); + break; + } + adev->gfx.config.gb_addr_config = gb_addr_config; +} + static int gfx_v8_0_sw_init(void *handle) { int i, r; @@ -1010,6 +1629,8 @@ static int gfx_v8_0_sw_init(void *handle) adev->gfx.ce_ram_size = 0x8000; + gfx_v8_0_gpu_early_init(adev); + return 0; } @@ -1034,861 +1655,918 @@ static int gfx_v8_0_sw_fini(void *handle) static void gfx_v8_0_tiling_mode_table_init(struct amdgpu_device *adev) { - const u32 num_tile_mode_states = 32; - const u32 num_secondary_tile_mode_states = 16; - u32 reg_offset, gb_tile_moden, split_equal_to_row_size; + uint32_t *modearray, *mod2array; + const u32 num_tile_mode_states = ARRAY_SIZE(adev->gfx.config.tile_mode_array); + const u32 num_secondary_tile_mode_states = ARRAY_SIZE(adev->gfx.config.macrotile_mode_array); + u32 reg_offset; - switch (adev->gfx.config.mem_row_size_in_kb) { - case 1: - split_equal_to_row_size = ADDR_SURF_TILE_SPLIT_1KB; - break; - case 2: - default: - split_equal_to_row_size = ADDR_SURF_TILE_SPLIT_2KB; - break; - case 4: - split_equal_to_row_size = ADDR_SURF_TILE_SPLIT_4KB; - break; - } + modearray = adev->gfx.config.tile_mode_array; + mod2array = adev->gfx.config.macrotile_mode_array; + + for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++) + modearray[reg_offset] = 0; + + for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++) + mod2array[reg_offset] = 0; switch (adev->asic_type) { case CHIP_TOPAZ: - for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++) { - switch (reg_offset) { - case 0: - gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P2) | - TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) | - MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); - break; - case 1: - gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P2) | - TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) | - MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); - break; - case 2: - gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P2) | - TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | - MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); - break; - case 3: - gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P2) | - TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) | - MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); - break; - case 4: - gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P2) | - TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) | - MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); - break; - case 5: - gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P2) | - TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) | - MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); - break; - case 6: - gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P2) | - TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) | - MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); - break; - case 8: - gb_tile_moden = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) | - PIPE_CONFIG(ADDR_SURF_P2)); - break; - case 9: - gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P2) | - MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); - break; - case 10: - gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P2) | - MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); - break; - case 11: - gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P2) | - MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); - break; - case 13: - gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P2) | - MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); - break; - case 14: - gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P2) | - MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); - break; - case 15: - gb_tile_moden = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P2) | - MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); - break; - case 16: - gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P2) | - MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); - break; - case 18: - gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THICK) | - PIPE_CONFIG(ADDR_SURF_P2) | - MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); - break; - case 19: - gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THICK) | - PIPE_CONFIG(ADDR_SURF_P2) | - MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); - break; - case 20: - gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THICK) | - PIPE_CONFIG(ADDR_SURF_P2) | - MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); - break; - case 21: - gb_tile_moden = (ARRAY_MODE(ARRAY_3D_TILED_THICK) | - PIPE_CONFIG(ADDR_SURF_P2) | - MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); - break; - case 22: - gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) | - PIPE_CONFIG(ADDR_SURF_P2) | - MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); - break; - case 24: - gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THICK) | - PIPE_CONFIG(ADDR_SURF_P2) | - MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); - break; - case 25: - gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) | - PIPE_CONFIG(ADDR_SURF_P2) | - MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); - break; - case 26: - gb_tile_moden = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) | - PIPE_CONFIG(ADDR_SURF_P2) | - MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); - break; - case 27: - gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P2) | - MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); - break; - case 28: - gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P2) | - MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); - break; - case 29: - gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P2) | - MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); - break; - case 7: - case 12: - case 17: - case 23: - /* unused idx */ - continue; - default: - gb_tile_moden = 0; - break; - }; - adev->gfx.config.tile_mode_array[reg_offset] = gb_tile_moden; - WREG32(mmGB_TILE_MODE0 + reg_offset, gb_tile_moden); - } - for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++) { - switch (reg_offset) { - case 0: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | - NUM_BANKS(ADDR_SURF_8_BANK)); - break; - case 1: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | - NUM_BANKS(ADDR_SURF_8_BANK)); - break; - case 2: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | - NUM_BANKS(ADDR_SURF_8_BANK)); - break; - case 3: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | - NUM_BANKS(ADDR_SURF_8_BANK)); - break; - case 4: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | - NUM_BANKS(ADDR_SURF_8_BANK)); - break; - case 5: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | - NUM_BANKS(ADDR_SURF_8_BANK)); - break; - case 6: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | - NUM_BANKS(ADDR_SURF_8_BANK)); - break; - case 8: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | - NUM_BANKS(ADDR_SURF_16_BANK)); - break; - case 9: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | - NUM_BANKS(ADDR_SURF_16_BANK)); - break; - case 10: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | - NUM_BANKS(ADDR_SURF_16_BANK)); - break; - case 11: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | - NUM_BANKS(ADDR_SURF_16_BANK)); - break; - case 12: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | - NUM_BANKS(ADDR_SURF_16_BANK)); - break; - case 13: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | - NUM_BANKS(ADDR_SURF_16_BANK)); - break; - case 14: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | - NUM_BANKS(ADDR_SURF_8_BANK)); - break; - case 7: - /* unused idx */ - continue; - default: - gb_tile_moden = 0; - break; - }; - adev->gfx.config.macrotile_mode_array[reg_offset] = gb_tile_moden; - WREG32(mmGB_MACROTILE_MODE0 + reg_offset, gb_tile_moden); - } + modearray[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P2) | + TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); + modearray[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P2) | + TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); + modearray[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P2) | + TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); + modearray[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P2) | + TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); + modearray[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P2) | + TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); + modearray[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P2) | + TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); + modearray[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P2) | + TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); + modearray[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) | + PIPE_CONFIG(ADDR_SURF_P2)); + modearray[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P2) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); + modearray[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P2) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); + modearray[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P2) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); + modearray[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P2) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); + modearray[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P2) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); + modearray[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P2) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); + modearray[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P2) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); + modearray[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) | + PIPE_CONFIG(ADDR_SURF_P2) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); + modearray[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) | + PIPE_CONFIG(ADDR_SURF_P2) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); + modearray[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) | + PIPE_CONFIG(ADDR_SURF_P2) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); + modearray[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) | + PIPE_CONFIG(ADDR_SURF_P2) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); + modearray[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) | + PIPE_CONFIG(ADDR_SURF_P2) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); + modearray[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) | + PIPE_CONFIG(ADDR_SURF_P2) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); + modearray[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) | + PIPE_CONFIG(ADDR_SURF_P2) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); + modearray[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) | + PIPE_CONFIG(ADDR_SURF_P2) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); + modearray[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P2) | + MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); + modearray[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P2) | + MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); + modearray[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P2) | + MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); + + mod2array[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | + NUM_BANKS(ADDR_SURF_8_BANK)); + mod2array[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | + NUM_BANKS(ADDR_SURF_8_BANK)); + mod2array[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | + NUM_BANKS(ADDR_SURF_8_BANK)); + mod2array[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | + NUM_BANKS(ADDR_SURF_8_BANK)); + mod2array[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | + NUM_BANKS(ADDR_SURF_8_BANK)); + mod2array[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | + NUM_BANKS(ADDR_SURF_8_BANK)); + mod2array[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | + NUM_BANKS(ADDR_SURF_8_BANK)); + mod2array[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | + NUM_BANKS(ADDR_SURF_16_BANK)); + mod2array[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | + NUM_BANKS(ADDR_SURF_16_BANK)); + mod2array[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | + NUM_BANKS(ADDR_SURF_16_BANK)); + mod2array[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | + NUM_BANKS(ADDR_SURF_16_BANK)); + mod2array[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | + NUM_BANKS(ADDR_SURF_16_BANK)); + mod2array[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | + NUM_BANKS(ADDR_SURF_16_BANK)); + mod2array[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | + NUM_BANKS(ADDR_SURF_8_BANK)); + + for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++) + if (reg_offset != 7 && reg_offset != 12 && reg_offset != 17 && + reg_offset != 23) + WREG32(mmGB_TILE_MODE0 + reg_offset, modearray[reg_offset]); + + for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++) + if (reg_offset != 7) + WREG32(mmGB_MACROTILE_MODE0 + reg_offset, mod2array[reg_offset]); + + break; case CHIP_FIJI: + modearray[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | + TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); + modearray[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | + TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); + modearray[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | + TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); + modearray[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | + TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); + modearray[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | + TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); + modearray[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | + TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); + modearray[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | + TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); + modearray[7] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P4_16x16) | + TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); + modearray[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) | + PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16)); + modearray[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); + modearray[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); + modearray[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); + modearray[12] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P4_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); + modearray[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); + modearray[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); + modearray[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); + modearray[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); + modearray[17] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P4_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); + modearray[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) | + PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); + modearray[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) | + PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); + modearray[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) | + PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); + modearray[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) | + PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); + modearray[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) | + PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); + modearray[23] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) | + PIPE_CONFIG(ADDR_SURF_P4_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); + modearray[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) | + PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); + modearray[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) | + PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); + modearray[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) | + PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); + modearray[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); + modearray[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); + modearray[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); + modearray[30] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P4_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); + + mod2array[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | + NUM_BANKS(ADDR_SURF_8_BANK)); + mod2array[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | + NUM_BANKS(ADDR_SURF_8_BANK)); + mod2array[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | + NUM_BANKS(ADDR_SURF_8_BANK)); + mod2array[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | + NUM_BANKS(ADDR_SURF_8_BANK)); + mod2array[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | + NUM_BANKS(ADDR_SURF_8_BANK)); + mod2array[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | + NUM_BANKS(ADDR_SURF_8_BANK)); + mod2array[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | + NUM_BANKS(ADDR_SURF_8_BANK)); + mod2array[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | + NUM_BANKS(ADDR_SURF_8_BANK)); + mod2array[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | + NUM_BANKS(ADDR_SURF_8_BANK)); + mod2array[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | + NUM_BANKS(ADDR_SURF_8_BANK)); + mod2array[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | + NUM_BANKS(ADDR_SURF_8_BANK)); + mod2array[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | + NUM_BANKS(ADDR_SURF_8_BANK)); + mod2array[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | + NUM_BANKS(ADDR_SURF_8_BANK)); + mod2array[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | + NUM_BANKS(ADDR_SURF_4_BANK)); + + for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++) + WREG32(mmGB_TILE_MODE0 + reg_offset, modearray[reg_offset]); + + for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++) + if (reg_offset != 7) + WREG32(mmGB_MACROTILE_MODE0 + reg_offset, mod2array[reg_offset]); + + break; case CHIP_TONGA: - for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++) { - switch (reg_offset) { - case 0: - gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | - TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) | - MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); - break; - case 1: - gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | - TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) | - MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); - break; - case 2: - gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | - TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | - MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); - break; - case 3: - gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | - TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) | - MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); - break; - case 4: - gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | - TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) | - MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); - break; - case 5: - gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | - TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) | - MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); - break; - case 6: - gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | - TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) | - MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); - break; - case 7: - gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P4_16x16) | - TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) | - MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); - break; - case 8: - gb_tile_moden = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) | - PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16)); - break; - case 9: - gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | - MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); - break; - case 10: - gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | - MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); - break; - case 11: - gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | - MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); - break; - case 12: - gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P4_16x16) | - MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); - break; - case 13: - gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | - MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); - break; - case 14: - gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | - MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); - break; - case 15: - gb_tile_moden = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | - MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); - break; - case 16: - gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | - MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); - break; - case 17: - gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P4_16x16) | - MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); - break; - case 18: - gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THICK) | - PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | - MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); - break; - case 19: - gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THICK) | - PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | - MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); - break; - case 20: - gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THICK) | - PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | - MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); - break; - case 21: - gb_tile_moden = (ARRAY_MODE(ARRAY_3D_TILED_THICK) | - PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | - MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); - break; - case 22: - gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) | - PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | - MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); - break; - case 23: - gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) | - PIPE_CONFIG(ADDR_SURF_P4_16x16) | - MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); - break; - case 24: - gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THICK) | - PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | - MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); - break; - case 25: - gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) | - PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | - MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); - break; - case 26: - gb_tile_moden = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) | - PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | - MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); - break; - case 27: - gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | - MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); - break; - case 28: - gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | - MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); - break; - case 29: - gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | - MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); - break; - case 30: - gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P4_16x16) | - MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); - break; - default: - gb_tile_moden = 0; - break; - }; - adev->gfx.config.tile_mode_array[reg_offset] = gb_tile_moden; - WREG32(mmGB_TILE_MODE0 + reg_offset, gb_tile_moden); - } - for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++) { - switch (reg_offset) { - case 0: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | - NUM_BANKS(ADDR_SURF_16_BANK)); - break; - case 1: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | - NUM_BANKS(ADDR_SURF_16_BANK)); - break; - case 2: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | - NUM_BANKS(ADDR_SURF_16_BANK)); - break; - case 3: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | - NUM_BANKS(ADDR_SURF_16_BANK)); - break; - case 4: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | - NUM_BANKS(ADDR_SURF_16_BANK)); - break; - case 5: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | - NUM_BANKS(ADDR_SURF_16_BANK)); - break; - case 6: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | - NUM_BANKS(ADDR_SURF_16_BANK)); - break; - case 8: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | - NUM_BANKS(ADDR_SURF_16_BANK)); - break; - case 9: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | - NUM_BANKS(ADDR_SURF_16_BANK)); - break; - case 10: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | - NUM_BANKS(ADDR_SURF_16_BANK)); - break; - case 11: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | - NUM_BANKS(ADDR_SURF_16_BANK)); - break; - case 12: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | - NUM_BANKS(ADDR_SURF_8_BANK)); - break; - case 13: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | - NUM_BANKS(ADDR_SURF_4_BANK)); - break; - case 14: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | - NUM_BANKS(ADDR_SURF_4_BANK)); - break; - case 7: - /* unused idx */ - continue; - default: - gb_tile_moden = 0; - break; - }; - adev->gfx.config.macrotile_mode_array[reg_offset] = gb_tile_moden; - WREG32(mmGB_MACROTILE_MODE0 + reg_offset, gb_tile_moden); - } + modearray[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | + TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); + modearray[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | + TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); + modearray[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | + TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); + modearray[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | + TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); + modearray[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | + TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); + modearray[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | + TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); + modearray[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | + TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); + modearray[7] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P4_16x16) | + TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); + modearray[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) | + PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16)); + modearray[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); + modearray[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); + modearray[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); + modearray[12] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P4_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); + modearray[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); + modearray[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); + modearray[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); + modearray[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); + modearray[17] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P4_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); + modearray[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) | + PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); + modearray[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) | + PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); + modearray[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) | + PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); + modearray[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) | + PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); + modearray[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) | + PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); + modearray[23] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) | + PIPE_CONFIG(ADDR_SURF_P4_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); + modearray[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) | + PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); + modearray[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) | + PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); + modearray[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) | + PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); + modearray[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); + modearray[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); + modearray[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); + modearray[30] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P4_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); + + mod2array[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | + NUM_BANKS(ADDR_SURF_16_BANK)); + mod2array[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | + NUM_BANKS(ADDR_SURF_16_BANK)); + mod2array[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | + NUM_BANKS(ADDR_SURF_16_BANK)); + mod2array[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | + NUM_BANKS(ADDR_SURF_16_BANK)); + mod2array[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | + NUM_BANKS(ADDR_SURF_16_BANK)); + mod2array[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | + NUM_BANKS(ADDR_SURF_16_BANK)); + mod2array[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | + NUM_BANKS(ADDR_SURF_16_BANK)); + mod2array[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | + NUM_BANKS(ADDR_SURF_16_BANK)); + mod2array[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | + NUM_BANKS(ADDR_SURF_16_BANK)); + mod2array[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | + NUM_BANKS(ADDR_SURF_16_BANK)); + mod2array[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | + NUM_BANKS(ADDR_SURF_16_BANK)); + mod2array[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | + NUM_BANKS(ADDR_SURF_8_BANK)); + mod2array[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | + NUM_BANKS(ADDR_SURF_4_BANK)); + mod2array[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | + NUM_BANKS(ADDR_SURF_4_BANK)); + + for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++) + WREG32(mmGB_TILE_MODE0 + reg_offset, modearray[reg_offset]); + + for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++) + if (reg_offset != 7) + WREG32(mmGB_MACROTILE_MODE0 + reg_offset, mod2array[reg_offset]); + + break; + case CHIP_STONEY: + modearray[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P2) | + TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); + modearray[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P2) | + TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); + modearray[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P2) | + TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); + modearray[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P2) | + TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); + modearray[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P2) | + TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); + modearray[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P2) | + TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); + modearray[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P2) | + TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); + modearray[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) | + PIPE_CONFIG(ADDR_SURF_P2)); + modearray[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P2) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); + modearray[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P2) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); + modearray[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P2) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); + modearray[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P2) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); + modearray[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P2) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); + modearray[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P2) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); + modearray[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P2) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); + modearray[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) | + PIPE_CONFIG(ADDR_SURF_P2) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); + modearray[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) | + PIPE_CONFIG(ADDR_SURF_P2) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); + modearray[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) | + PIPE_CONFIG(ADDR_SURF_P2) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); + modearray[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) | + PIPE_CONFIG(ADDR_SURF_P2) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); + modearray[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) | + PIPE_CONFIG(ADDR_SURF_P2) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); + modearray[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) | + PIPE_CONFIG(ADDR_SURF_P2) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); + modearray[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) | + PIPE_CONFIG(ADDR_SURF_P2) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); + modearray[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) | + PIPE_CONFIG(ADDR_SURF_P2) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); + modearray[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P2) | + MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); + modearray[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P2) | + MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); + modearray[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P2) | + MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); + + mod2array[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | + NUM_BANKS(ADDR_SURF_8_BANK)); + mod2array[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | + NUM_BANKS(ADDR_SURF_8_BANK)); + mod2array[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | + NUM_BANKS(ADDR_SURF_8_BANK)); + mod2array[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | + NUM_BANKS(ADDR_SURF_8_BANK)); + mod2array[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | + NUM_BANKS(ADDR_SURF_8_BANK)); + mod2array[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | + NUM_BANKS(ADDR_SURF_8_BANK)); + mod2array[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | + NUM_BANKS(ADDR_SURF_8_BANK)); + mod2array[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | + NUM_BANKS(ADDR_SURF_16_BANK)); + mod2array[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | + NUM_BANKS(ADDR_SURF_16_BANK)); + mod2array[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | + NUM_BANKS(ADDR_SURF_16_BANK)); + mod2array[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | + NUM_BANKS(ADDR_SURF_16_BANK)); + mod2array[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | + NUM_BANKS(ADDR_SURF_16_BANK)); + mod2array[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | + NUM_BANKS(ADDR_SURF_16_BANK)); + mod2array[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | + NUM_BANKS(ADDR_SURF_8_BANK)); + + for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++) + if (reg_offset != 7 && reg_offset != 12 && reg_offset != 17 && + reg_offset != 23) + WREG32(mmGB_TILE_MODE0 + reg_offset, modearray[reg_offset]); + + for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++) + if (reg_offset != 7) + WREG32(mmGB_MACROTILE_MODE0 + reg_offset, mod2array[reg_offset]); + break; - case CHIP_CARRIZO: default: - for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++) { - switch (reg_offset) { - case 0: - gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P2) | - TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) | - MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); - break; - case 1: - gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P2) | - TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) | - MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); - break; - case 2: - gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P2) | - TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | - MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); - break; - case 3: - gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P2) | - TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) | - MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); - break; - case 4: - gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P2) | - TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) | - MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); - break; - case 5: - gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P2) | - TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) | - MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); - break; - case 6: - gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P2) | - TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) | - MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); - break; - case 8: - gb_tile_moden = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) | - PIPE_CONFIG(ADDR_SURF_P2)); - break; - case 9: - gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P2) | - MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); - break; - case 10: - gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P2) | - MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); - break; - case 11: - gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P2) | - MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); - break; - case 13: - gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P2) | - MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); - break; - case 14: - gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P2) | - MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); - break; - case 15: - gb_tile_moden = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P2) | - MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); - break; - case 16: - gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P2) | - MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); - break; - case 18: - gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THICK) | - PIPE_CONFIG(ADDR_SURF_P2) | - MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); - break; - case 19: - gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THICK) | - PIPE_CONFIG(ADDR_SURF_P2) | - MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); - break; - case 20: - gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THICK) | - PIPE_CONFIG(ADDR_SURF_P2) | - MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); - break; - case 21: - gb_tile_moden = (ARRAY_MODE(ARRAY_3D_TILED_THICK) | - PIPE_CONFIG(ADDR_SURF_P2) | - MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); - break; - case 22: - gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) | - PIPE_CONFIG(ADDR_SURF_P2) | - MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); - break; - case 24: - gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THICK) | - PIPE_CONFIG(ADDR_SURF_P2) | - MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); - break; - case 25: - gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) | - PIPE_CONFIG(ADDR_SURF_P2) | - MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); - break; - case 26: - gb_tile_moden = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) | - PIPE_CONFIG(ADDR_SURF_P2) | - MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); - break; - case 27: - gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P2) | - MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); - break; - case 28: - gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P2) | - MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); - break; - case 29: - gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | - PIPE_CONFIG(ADDR_SURF_P2) | - MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | - SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); - break; - case 7: - case 12: - case 17: - case 23: - /* unused idx */ - continue; - default: - gb_tile_moden = 0; - break; - }; - adev->gfx.config.tile_mode_array[reg_offset] = gb_tile_moden; - WREG32(mmGB_TILE_MODE0 + reg_offset, gb_tile_moden); - } - for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++) { - switch (reg_offset) { - case 0: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | - NUM_BANKS(ADDR_SURF_8_BANK)); - break; - case 1: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | - NUM_BANKS(ADDR_SURF_8_BANK)); - break; - case 2: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | - NUM_BANKS(ADDR_SURF_8_BANK)); - break; - case 3: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | - NUM_BANKS(ADDR_SURF_8_BANK)); - break; - case 4: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | - NUM_BANKS(ADDR_SURF_8_BANK)); - break; - case 5: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | - NUM_BANKS(ADDR_SURF_8_BANK)); - break; - case 6: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | - NUM_BANKS(ADDR_SURF_8_BANK)); - break; - case 8: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | - NUM_BANKS(ADDR_SURF_16_BANK)); - break; - case 9: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | - NUM_BANKS(ADDR_SURF_16_BANK)); - break; - case 10: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | - NUM_BANKS(ADDR_SURF_16_BANK)); - break; - case 11: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | - NUM_BANKS(ADDR_SURF_16_BANK)); - break; - case 12: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | - NUM_BANKS(ADDR_SURF_16_BANK)); - break; - case 13: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | - NUM_BANKS(ADDR_SURF_16_BANK)); - break; - case 14: - gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | - BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | - MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | - NUM_BANKS(ADDR_SURF_8_BANK)); - break; - case 7: - /* unused idx */ - continue; - default: - gb_tile_moden = 0; - break; - }; - adev->gfx.config.macrotile_mode_array[reg_offset] = gb_tile_moden; - WREG32(mmGB_MACROTILE_MODE0 + reg_offset, gb_tile_moden); - } - } -} + dev_warn(adev->dev, + "Unknown chip type (%d) in function gfx_v8_0_tiling_mode_table_init() falling through to CHIP_CARRIZO\n", + adev->asic_type); -static u32 gfx_v8_0_create_bitmask(u32 bit_width) -{ - u32 i, mask = 0; + case CHIP_CARRIZO: + modearray[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P2) | + TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); + modearray[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P2) | + TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); + modearray[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P2) | + TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); + modearray[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P2) | + TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); + modearray[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P2) | + TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); + modearray[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P2) | + TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); + modearray[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P2) | + TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); + modearray[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) | + PIPE_CONFIG(ADDR_SURF_P2)); + modearray[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P2) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); + modearray[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P2) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); + modearray[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P2) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); + modearray[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P2) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); + modearray[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P2) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); + modearray[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P2) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); + modearray[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P2) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); + modearray[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) | + PIPE_CONFIG(ADDR_SURF_P2) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); + modearray[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) | + PIPE_CONFIG(ADDR_SURF_P2) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); + modearray[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) | + PIPE_CONFIG(ADDR_SURF_P2) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); + modearray[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) | + PIPE_CONFIG(ADDR_SURF_P2) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); + modearray[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) | + PIPE_CONFIG(ADDR_SURF_P2) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); + modearray[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) | + PIPE_CONFIG(ADDR_SURF_P2) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); + modearray[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) | + PIPE_CONFIG(ADDR_SURF_P2) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); + modearray[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) | + PIPE_CONFIG(ADDR_SURF_P2) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); + modearray[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P2) | + MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); + modearray[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P2) | + MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); + modearray[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P2) | + MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); + + mod2array[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | + NUM_BANKS(ADDR_SURF_8_BANK)); + mod2array[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | + NUM_BANKS(ADDR_SURF_8_BANK)); + mod2array[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | + NUM_BANKS(ADDR_SURF_8_BANK)); + mod2array[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | + NUM_BANKS(ADDR_SURF_8_BANK)); + mod2array[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | + NUM_BANKS(ADDR_SURF_8_BANK)); + mod2array[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | + NUM_BANKS(ADDR_SURF_8_BANK)); + mod2array[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | + NUM_BANKS(ADDR_SURF_8_BANK)); + mod2array[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | + NUM_BANKS(ADDR_SURF_16_BANK)); + mod2array[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | + NUM_BANKS(ADDR_SURF_16_BANK)); + mod2array[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | + NUM_BANKS(ADDR_SURF_16_BANK)); + mod2array[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | + NUM_BANKS(ADDR_SURF_16_BANK)); + mod2array[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | + NUM_BANKS(ADDR_SURF_16_BANK)); + mod2array[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | + NUM_BANKS(ADDR_SURF_16_BANK)); + mod2array[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | + NUM_BANKS(ADDR_SURF_8_BANK)); + + for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++) + if (reg_offset != 7 && reg_offset != 12 && reg_offset != 17 && + reg_offset != 23) + WREG32(mmGB_TILE_MODE0 + reg_offset, modearray[reg_offset]); + + for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++) + if (reg_offset != 7) + WREG32(mmGB_MACROTILE_MODE0 + reg_offset, mod2array[reg_offset]); - for (i = 0; i < bit_width; i++) { - mask <<= 1; - mask |= 1; + break; } - return mask; } void gfx_v8_0_select_se_sh(struct amdgpu_device *adev, u32 se_num, u32 sh_num) @@ -1911,89 +2589,49 @@ void gfx_v8_0_select_se_sh(struct amdgpu_device *adev, u32 se_num, u32 sh_num) WREG32(mmGRBM_GFX_INDEX, data); } -static u32 gfx_v8_0_get_rb_disabled(struct amdgpu_device *adev, - u32 max_rb_num_per_se, - u32 sh_per_se) +static u32 gfx_v8_0_create_bitmask(u32 bit_width) +{ + return (u32)((1ULL << bit_width) - 1); +} + +static u32 gfx_v8_0_get_rb_active_bitmap(struct amdgpu_device *adev) { u32 data, mask; data = RREG32(mmCC_RB_BACKEND_DISABLE); - data &= CC_RB_BACKEND_DISABLE__BACKEND_DISABLE_MASK; - data |= RREG32(mmGC_USER_RB_BACKEND_DISABLE); + data &= CC_RB_BACKEND_DISABLE__BACKEND_DISABLE_MASK; data >>= GC_USER_RB_BACKEND_DISABLE__BACKEND_DISABLE__SHIFT; - mask = gfx_v8_0_create_bitmask(max_rb_num_per_se / sh_per_se); + mask = gfx_v8_0_create_bitmask(adev->gfx.config.max_backends_per_se / + adev->gfx.config.max_sh_per_se); - return data & mask; + return (~data) & mask; } -static void gfx_v8_0_setup_rb(struct amdgpu_device *adev, - u32 se_num, u32 sh_per_se, - u32 max_rb_num_per_se) +static void gfx_v8_0_setup_rb(struct amdgpu_device *adev) { int i, j; - u32 data, mask; - u32 disabled_rbs = 0; - u32 enabled_rbs = 0; + u32 data; + u32 active_rbs = 0; + u32 rb_bitmap_width_per_sh = adev->gfx.config.max_backends_per_se / + adev->gfx.config.max_sh_per_se; mutex_lock(&adev->grbm_idx_mutex); - for (i = 0; i < se_num; i++) { - for (j = 0; j < sh_per_se; j++) { + for (i = 0; i < adev->gfx.config.max_shader_engines; i++) { + for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) { gfx_v8_0_select_se_sh(adev, i, j); - data = gfx_v8_0_get_rb_disabled(adev, - max_rb_num_per_se, sh_per_se); - disabled_rbs |= data << ((i * sh_per_se + j) * - RB_BITMAP_WIDTH_PER_SH); + data = gfx_v8_0_get_rb_active_bitmap(adev); + active_rbs |= data << ((i * adev->gfx.config.max_sh_per_se + j) * + rb_bitmap_width_per_sh); } } gfx_v8_0_select_se_sh(adev, 0xffffffff, 0xffffffff); mutex_unlock(&adev->grbm_idx_mutex); - mask = 1; - for (i = 0; i < max_rb_num_per_se * se_num; i++) { - if (!(disabled_rbs & mask)) - enabled_rbs |= mask; - mask <<= 1; - } - - adev->gfx.config.backend_enable_mask = enabled_rbs; - - mutex_lock(&adev->grbm_idx_mutex); - for (i = 0; i < se_num; i++) { - gfx_v8_0_select_se_sh(adev, i, 0xffffffff); - data = 0; - for (j = 0; j < sh_per_se; j++) { - switch (enabled_rbs & 3) { - case 0: - if (j == 0) - data |= (RASTER_CONFIG_RB_MAP_3 << - PA_SC_RASTER_CONFIG__PKR_MAP__SHIFT); - else - data |= (RASTER_CONFIG_RB_MAP_0 << - PA_SC_RASTER_CONFIG__PKR_MAP__SHIFT); - break; - case 1: - data |= (RASTER_CONFIG_RB_MAP_0 << - (i * sh_per_se + j) * 2); - break; - case 2: - data |= (RASTER_CONFIG_RB_MAP_3 << - (i * sh_per_se + j) * 2); - break; - case 3: - default: - data |= (RASTER_CONFIG_RB_MAP_2 << - (i * sh_per_se + j) * 2); - break; - } - enabled_rbs >>= 2; - } - WREG32(mmPA_SC_RASTER_CONFIG, data); - } - gfx_v8_0_select_se_sh(adev, 0xffffffff, 0xffffffff); - mutex_unlock(&adev->grbm_idx_mutex); + adev->gfx.config.backend_enable_mask = active_rbs; + adev->gfx.config.num_rbs = hweight32(active_rbs); } /** @@ -2043,209 +2681,20 @@ static void gfx_v8_0_init_compute_vmid(struct amdgpu_device *adev) static void gfx_v8_0_gpu_init(struct amdgpu_device *adev) { - u32 gb_addr_config; - u32 mc_shared_chmap, mc_arb_ramcfg; - u32 dimm00_addr_map, dimm01_addr_map, dimm10_addr_map, dimm11_addr_map; u32 tmp; int i; - switch (adev->asic_type) { - case CHIP_TOPAZ: - adev->gfx.config.max_shader_engines = 1; - adev->gfx.config.max_tile_pipes = 2; - adev->gfx.config.max_cu_per_sh = 6; - adev->gfx.config.max_sh_per_se = 1; - adev->gfx.config.max_backends_per_se = 2; - adev->gfx.config.max_texture_channel_caches = 2; - adev->gfx.config.max_gprs = 256; - adev->gfx.config.max_gs_threads = 32; - adev->gfx.config.max_hw_contexts = 8; - - adev->gfx.config.sc_prim_fifo_size_frontend = 0x20; - adev->gfx.config.sc_prim_fifo_size_backend = 0x100; - adev->gfx.config.sc_hiz_tile_fifo_size = 0x30; - adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130; - gb_addr_config = TOPAZ_GB_ADDR_CONFIG_GOLDEN; - break; - case CHIP_FIJI: - adev->gfx.config.max_shader_engines = 4; - adev->gfx.config.max_tile_pipes = 16; - adev->gfx.config.max_cu_per_sh = 16; - adev->gfx.config.max_sh_per_se = 1; - adev->gfx.config.max_backends_per_se = 4; - adev->gfx.config.max_texture_channel_caches = 8; - adev->gfx.config.max_gprs = 256; - adev->gfx.config.max_gs_threads = 32; - adev->gfx.config.max_hw_contexts = 8; - - adev->gfx.config.sc_prim_fifo_size_frontend = 0x20; - adev->gfx.config.sc_prim_fifo_size_backend = 0x100; - adev->gfx.config.sc_hiz_tile_fifo_size = 0x30; - adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130; - gb_addr_config = TONGA_GB_ADDR_CONFIG_GOLDEN; - break; - case CHIP_TONGA: - adev->gfx.config.max_shader_engines = 4; - adev->gfx.config.max_tile_pipes = 8; - adev->gfx.config.max_cu_per_sh = 8; - adev->gfx.config.max_sh_per_se = 1; - adev->gfx.config.max_backends_per_se = 2; - adev->gfx.config.max_texture_channel_caches = 8; - adev->gfx.config.max_gprs = 256; - adev->gfx.config.max_gs_threads = 32; - adev->gfx.config.max_hw_contexts = 8; - - adev->gfx.config.sc_prim_fifo_size_frontend = 0x20; - adev->gfx.config.sc_prim_fifo_size_backend = 0x100; - adev->gfx.config.sc_hiz_tile_fifo_size = 0x30; - adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130; - gb_addr_config = TONGA_GB_ADDR_CONFIG_GOLDEN; - break; - case CHIP_CARRIZO: - adev->gfx.config.max_shader_engines = 1; - adev->gfx.config.max_tile_pipes = 2; - adev->gfx.config.max_sh_per_se = 1; - adev->gfx.config.max_backends_per_se = 2; - - switch (adev->pdev->revision) { - case 0xc4: - case 0x84: - case 0xc8: - case 0xcc: - /* B10 */ - adev->gfx.config.max_cu_per_sh = 8; - break; - case 0xc5: - case 0x81: - case 0x85: - case 0xc9: - case 0xcd: - /* B8 */ - adev->gfx.config.max_cu_per_sh = 6; - break; - case 0xc6: - case 0xca: - case 0xce: - /* B6 */ - adev->gfx.config.max_cu_per_sh = 6; - break; - case 0xc7: - case 0x87: - case 0xcb: - default: - /* B4 */ - adev->gfx.config.max_cu_per_sh = 4; - break; - } - - adev->gfx.config.max_texture_channel_caches = 2; - adev->gfx.config.max_gprs = 256; - adev->gfx.config.max_gs_threads = 32; - adev->gfx.config.max_hw_contexts = 8; - - adev->gfx.config.sc_prim_fifo_size_frontend = 0x20; - adev->gfx.config.sc_prim_fifo_size_backend = 0x100; - adev->gfx.config.sc_hiz_tile_fifo_size = 0x30; - adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130; - gb_addr_config = CARRIZO_GB_ADDR_CONFIG_GOLDEN; - break; - default: - adev->gfx.config.max_shader_engines = 2; - adev->gfx.config.max_tile_pipes = 4; - adev->gfx.config.max_cu_per_sh = 2; - adev->gfx.config.max_sh_per_se = 1; - adev->gfx.config.max_backends_per_se = 2; - adev->gfx.config.max_texture_channel_caches = 4; - adev->gfx.config.max_gprs = 256; - adev->gfx.config.max_gs_threads = 32; - adev->gfx.config.max_hw_contexts = 8; - - adev->gfx.config.sc_prim_fifo_size_frontend = 0x20; - adev->gfx.config.sc_prim_fifo_size_backend = 0x100; - adev->gfx.config.sc_hiz_tile_fifo_size = 0x30; - adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130; - gb_addr_config = TONGA_GB_ADDR_CONFIG_GOLDEN; - break; - } - tmp = RREG32(mmGRBM_CNTL); tmp = REG_SET_FIELD(tmp, GRBM_CNTL, READ_TIMEOUT, 0xff); WREG32(mmGRBM_CNTL, tmp); - mc_shared_chmap = RREG32(mmMC_SHARED_CHMAP); - adev->gfx.config.mc_arb_ramcfg = RREG32(mmMC_ARB_RAMCFG); - mc_arb_ramcfg = adev->gfx.config.mc_arb_ramcfg; - - adev->gfx.config.num_tile_pipes = adev->gfx.config.max_tile_pipes; - adev->gfx.config.mem_max_burst_length_bytes = 256; - if (adev->flags & AMD_IS_APU) { - /* Get memory bank mapping mode. */ - tmp = RREG32(mmMC_FUS_DRAM0_BANK_ADDR_MAPPING); - dimm00_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM0_BANK_ADDR_MAPPING, DIMM0ADDRMAP); - dimm01_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM0_BANK_ADDR_MAPPING, DIMM1ADDRMAP); - - tmp = RREG32(mmMC_FUS_DRAM1_BANK_ADDR_MAPPING); - dimm10_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM1_BANK_ADDR_MAPPING, DIMM0ADDRMAP); - dimm11_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM1_BANK_ADDR_MAPPING, DIMM1ADDRMAP); - - /* Validate settings in case only one DIMM installed. */ - if ((dimm00_addr_map == 0) || (dimm00_addr_map == 3) || (dimm00_addr_map == 4) || (dimm00_addr_map > 12)) - dimm00_addr_map = 0; - if ((dimm01_addr_map == 0) || (dimm01_addr_map == 3) || (dimm01_addr_map == 4) || (dimm01_addr_map > 12)) - dimm01_addr_map = 0; - if ((dimm10_addr_map == 0) || (dimm10_addr_map == 3) || (dimm10_addr_map == 4) || (dimm10_addr_map > 12)) - dimm10_addr_map = 0; - if ((dimm11_addr_map == 0) || (dimm11_addr_map == 3) || (dimm11_addr_map == 4) || (dimm11_addr_map > 12)) - dimm11_addr_map = 0; - - /* If DIMM Addr map is 8GB, ROW size should be 2KB. Otherwise 1KB. */ - /* If ROW size(DIMM1) != ROW size(DMIMM0), ROW size should be larger one. */ - if ((dimm00_addr_map == 11) || (dimm01_addr_map == 11) || (dimm10_addr_map == 11) || (dimm11_addr_map == 11)) - adev->gfx.config.mem_row_size_in_kb = 2; - else - adev->gfx.config.mem_row_size_in_kb = 1; - } else { - tmp = REG_GET_FIELD(mc_arb_ramcfg, MC_ARB_RAMCFG, NOOFCOLS); - adev->gfx.config.mem_row_size_in_kb = (4 * (1 << (8 + tmp))) / 1024; - if (adev->gfx.config.mem_row_size_in_kb > 4) - adev->gfx.config.mem_row_size_in_kb = 4; - } - - adev->gfx.config.shader_engine_tile_size = 32; - adev->gfx.config.num_gpus = 1; - adev->gfx.config.multi_gpu_tile_size = 64; - - /* fix up row size */ - switch (adev->gfx.config.mem_row_size_in_kb) { - case 1: - default: - gb_addr_config = REG_SET_FIELD(gb_addr_config, GB_ADDR_CONFIG, ROW_SIZE, 0); - break; - case 2: - gb_addr_config = REG_SET_FIELD(gb_addr_config, GB_ADDR_CONFIG, ROW_SIZE, 1); - break; - case 4: - gb_addr_config = REG_SET_FIELD(gb_addr_config, GB_ADDR_CONFIG, ROW_SIZE, 2); - break; - } - adev->gfx.config.gb_addr_config = gb_addr_config; - - WREG32(mmGB_ADDR_CONFIG, gb_addr_config); - WREG32(mmHDP_ADDR_CONFIG, gb_addr_config); - WREG32(mmDMIF_ADDR_CALC, gb_addr_config); - WREG32(mmSDMA0_TILING_CONFIG + SDMA0_REGISTER_OFFSET, - gb_addr_config & 0x70); - WREG32(mmSDMA0_TILING_CONFIG + SDMA1_REGISTER_OFFSET, - gb_addr_config & 0x70); - WREG32(mmUVD_UDEC_ADDR_CONFIG, gb_addr_config); - WREG32(mmUVD_UDEC_DB_ADDR_CONFIG, gb_addr_config); - WREG32(mmUVD_UDEC_DBW_ADDR_CONFIG, gb_addr_config); + WREG32(mmGB_ADDR_CONFIG, adev->gfx.config.gb_addr_config); + WREG32(mmHDP_ADDR_CONFIG, adev->gfx.config.gb_addr_config); + WREG32(mmDMIF_ADDR_CALC, adev->gfx.config.gb_addr_config); gfx_v8_0_tiling_mode_table_init(adev); - gfx_v8_0_setup_rb(adev, adev->gfx.config.max_shader_engines, - adev->gfx.config.max_sh_per_se, - adev->gfx.config.max_backends_per_se); + gfx_v8_0_setup_rb(adev); /* XXX SH_MEM regs */ /* where to put LDS, scratch, GPUVM in FSA64 space */ @@ -2256,13 +2705,13 @@ static void gfx_v8_0_gpu_init(struct amdgpu_device *adev) if (i == 0) { tmp = REG_SET_FIELD(0, SH_MEM_CONFIG, DEFAULT_MTYPE, MTYPE_UC); tmp = REG_SET_FIELD(tmp, SH_MEM_CONFIG, APE1_MTYPE, MTYPE_UC); - tmp = REG_SET_FIELD(tmp, SH_MEM_CONFIG, ALIGNMENT_MODE, + tmp = REG_SET_FIELD(tmp, SH_MEM_CONFIG, ALIGNMENT_MODE, SH_MEM_ALIGNMENT_MODE_UNALIGNED); WREG32(mmSH_MEM_CONFIG, tmp); } else { tmp = REG_SET_FIELD(0, SH_MEM_CONFIG, DEFAULT_MTYPE, MTYPE_NC); tmp = REG_SET_FIELD(tmp, SH_MEM_CONFIG, APE1_MTYPE, MTYPE_NC); - tmp = REG_SET_FIELD(tmp, SH_MEM_CONFIG, ALIGNMENT_MODE, + tmp = REG_SET_FIELD(tmp, SH_MEM_CONFIG, ALIGNMENT_MODE, SH_MEM_ALIGNMENT_MODE_UNALIGNED); WREG32(mmSH_MEM_CONFIG, tmp); } @@ -2331,17 +2780,11 @@ static void gfx_v8_0_enable_gui_idle_interrupt(struct amdgpu_device *adev, { u32 tmp = RREG32(mmCP_INT_CNTL_RING0); - if (enable) { - tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CNTX_BUSY_INT_ENABLE, 1); - tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CNTX_EMPTY_INT_ENABLE, 1); - tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CMP_BUSY_INT_ENABLE, 1); - tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, GFX_IDLE_INT_ENABLE, 1); - } else { - tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CNTX_BUSY_INT_ENABLE, 0); - tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CNTX_EMPTY_INT_ENABLE, 0); - tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CMP_BUSY_INT_ENABLE, 0); - tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, GFX_IDLE_INT_ENABLE, 0); - } + tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CNTX_BUSY_INT_ENABLE, enable ? 1 : 0); + tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CNTX_EMPTY_INT_ENABLE, enable ? 1 : 0); + tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CMP_BUSY_INT_ENABLE, enable ? 1 : 0); + tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, GFX_IDLE_INT_ENABLE, enable ? 1 : 0); + WREG32(mmCP_INT_CNTL_RING0, tmp); } @@ -2377,7 +2820,7 @@ static void gfx_v8_0_rlc_start(struct amdgpu_device *adev) WREG32(mmRLC_CNTL, tmp); /* carrizo do enable cp interrupt after cp inited */ - if (adev->asic_type != CHIP_CARRIZO) + if (!(adev->flags & AMD_IS_APU)) gfx_v8_0_enable_gui_idle_interrupt(adev, true); udelay(50); @@ -2421,16 +2864,18 @@ static int gfx_v8_0_rlc_resume(struct amdgpu_device *adev) gfx_v8_0_rlc_reset(adev); - if (!adev->firmware.smu_load) { - /* legacy rlc firmware loading */ - r = gfx_v8_0_rlc_load_microcode(adev); - if (r) - return r; - } else { - r = adev->smu.smumgr_funcs->check_fw_load_finish(adev, - AMDGPU_UCODE_ID_RLC_G); - if (r) - return -EINVAL; + if (!adev->pp_enabled) { + if (!adev->firmware.smu_load) { + /* legacy rlc firmware loading */ + r = gfx_v8_0_rlc_load_microcode(adev); + if (r) + return r; + } else { + r = adev->smu.smumgr_funcs->check_fw_load_finish(adev, + AMDGPU_UCODE_ID_RLC_G); + if (r) + return -EINVAL; + } } gfx_v8_0_rlc_start(adev); @@ -2558,7 +3003,7 @@ static int gfx_v8_0_cp_gfx_start(struct amdgpu_device *adev) gfx_v8_0_cp_gfx_enable(adev, true); - r = amdgpu_ring_lock(ring, gfx_v8_0_get_csb_size(adev) + 4); + r = amdgpu_ring_alloc(ring, gfx_v8_0_get_csb_size(adev) + 4); if (r) { DRM_ERROR("amdgpu: cp failed to lock ring (%d).\n", r); return r; @@ -2590,15 +3035,22 @@ static int gfx_v8_0_cp_gfx_start(struct amdgpu_device *adev) amdgpu_ring_write(ring, mmPA_SC_RASTER_CONFIG - PACKET3_SET_CONTEXT_REG_START); switch (adev->asic_type) { case CHIP_TONGA: - case CHIP_FIJI: amdgpu_ring_write(ring, 0x16000012); amdgpu_ring_write(ring, 0x0000002A); break; + case CHIP_FIJI: + amdgpu_ring_write(ring, 0x3a00161a); + amdgpu_ring_write(ring, 0x0000002e); + break; case CHIP_TOPAZ: case CHIP_CARRIZO: amdgpu_ring_write(ring, 0x00000002); amdgpu_ring_write(ring, 0x00000000); break; + case CHIP_STONEY: + amdgpu_ring_write(ring, 0x00000000); + amdgpu_ring_write(ring, 0x00000000); + break; default: BUG(); } @@ -2615,7 +3067,7 @@ static int gfx_v8_0_cp_gfx_start(struct amdgpu_device *adev) amdgpu_ring_write(ring, 0x8000); amdgpu_ring_write(ring, 0x8000); - amdgpu_ring_unlock_commit(ring); + amdgpu_ring_commit(ring); return 0; } @@ -2715,13 +3167,6 @@ static void gfx_v8_0_cp_compute_enable(struct amdgpu_device *adev, bool enable) udelay(50); } -static int gfx_v8_0_cp_compute_start(struct amdgpu_device *adev) -{ - gfx_v8_0_cp_compute_enable(adev, true); - - return 0; -} - static int gfx_v8_0_cp_compute_load_microcode(struct amdgpu_device *adev) { const struct gfx_firmware_header_v1_0 *mec_hdr; @@ -3233,7 +3678,8 @@ static int gfx_v8_0_cp_compute_resume(struct amdgpu_device *adev) /* enable the doorbell if requested */ if (use_doorbell) { if ((adev->asic_type == CHIP_CARRIZO) || - (adev->asic_type == CHIP_FIJI)) { + (adev->asic_type == CHIP_FIJI) || + (adev->asic_type == CHIP_STONEY)) { WREG32(mmCP_MEC_DOORBELL_RANGE_LOWER, AMDGPU_DOORBELL_KIQ << 2); WREG32(mmCP_MEC_DOORBELL_RANGE_UPPER, @@ -3267,6 +3713,11 @@ static int gfx_v8_0_cp_compute_resume(struct amdgpu_device *adev) tmp = REG_SET_FIELD(tmp, CP_HQD_PERSISTENT_STATE, PRELOAD_SIZE, 0x53); WREG32(mmCP_HQD_PERSISTENT_STATE, tmp); mqd->cp_hqd_persistent_state = tmp; + if (adev->asic_type == CHIP_STONEY) { + tmp = RREG32(mmCP_ME1_PIPE3_INT_CNTL); + tmp = REG_SET_FIELD(tmp, CP_ME1_PIPE3_INT_CNTL, GENERIC2_INT_ENABLE, 1); + WREG32(mmCP_ME1_PIPE3_INT_CNTL, tmp); + } /* activate the queue */ mqd->cp_hqd_active = 1; @@ -3285,9 +3736,7 @@ static int gfx_v8_0_cp_compute_resume(struct amdgpu_device *adev) WREG32(mmCP_PQ_STATUS, tmp); } - r = gfx_v8_0_cp_compute_start(adev); - if (r) - return r; + gfx_v8_0_cp_compute_enable(adev, true); for (i = 0; i < adev->gfx.num_compute_rings; i++) { struct amdgpu_ring *ring = &adev->gfx.compute_ring[i]; @@ -3305,38 +3754,46 @@ static int gfx_v8_0_cp_resume(struct amdgpu_device *adev) { int r; - if (adev->asic_type != CHIP_CARRIZO) + if (!(adev->flags & AMD_IS_APU)) gfx_v8_0_enable_gui_idle_interrupt(adev, false); - if (!adev->firmware.smu_load) { - /* legacy firmware loading */ - r = gfx_v8_0_cp_gfx_load_microcode(adev); - if (r) - return r; - - r = gfx_v8_0_cp_compute_load_microcode(adev); - if (r) - return r; - } else { - r = adev->smu.smumgr_funcs->check_fw_load_finish(adev, - AMDGPU_UCODE_ID_CP_CE); - if (r) - return -EINVAL; - - r = adev->smu.smumgr_funcs->check_fw_load_finish(adev, - AMDGPU_UCODE_ID_CP_PFP); - if (r) - return -EINVAL; - - r = adev->smu.smumgr_funcs->check_fw_load_finish(adev, - AMDGPU_UCODE_ID_CP_ME); - if (r) - return -EINVAL; + if (!adev->pp_enabled) { + if (!adev->firmware.smu_load) { + /* legacy firmware loading */ + r = gfx_v8_0_cp_gfx_load_microcode(adev); + if (r) + return r; - r = adev->smu.smumgr_funcs->check_fw_load_finish(adev, - AMDGPU_UCODE_ID_CP_MEC1); - if (r) - return -EINVAL; + r = gfx_v8_0_cp_compute_load_microcode(adev); + if (r) + return r; + } else { + r = adev->smu.smumgr_funcs->check_fw_load_finish(adev, + AMDGPU_UCODE_ID_CP_CE); + if (r) + return -EINVAL; + + r = adev->smu.smumgr_funcs->check_fw_load_finish(adev, + AMDGPU_UCODE_ID_CP_PFP); + if (r) + return -EINVAL; + + r = adev->smu.smumgr_funcs->check_fw_load_finish(adev, + AMDGPU_UCODE_ID_CP_ME); + if (r) + return -EINVAL; + + if (adev->asic_type == CHIP_TOPAZ) { + r = gfx_v8_0_cp_compute_load_microcode(adev); + if (r) + return r; + } else { + r = adev->smu.smumgr_funcs->check_fw_load_finish(adev, + AMDGPU_UCODE_ID_CP_MEC1); + if (r) + return -EINVAL; + } + } } r = gfx_v8_0_cp_gfx_resume(adev); @@ -3382,6 +3839,8 @@ static int gfx_v8_0_hw_fini(void *handle) { struct amdgpu_device *adev = (struct amdgpu_device *)handle; + amdgpu_irq_put(adev, &adev->gfx.priv_reg_irq, 0); + amdgpu_irq_put(adev, &adev->gfx.priv_inst_irq, 0); gfx_v8_0_cp_enable(adev, false); gfx_v8_0_rlc_stop(adev); gfx_v8_0_cp_compute_fini(adev); @@ -3489,16 +3948,6 @@ static void gfx_v8_0_print_status(void *handle) RREG32(mmHDP_ADDR_CONFIG)); dev_info(adev->dev, " DMIF_ADDR_CALC=0x%08X\n", RREG32(mmDMIF_ADDR_CALC)); - dev_info(adev->dev, " SDMA0_TILING_CONFIG=0x%08X\n", - RREG32(mmSDMA0_TILING_CONFIG + SDMA0_REGISTER_OFFSET)); - dev_info(adev->dev, " SDMA1_TILING_CONFIG=0x%08X\n", - RREG32(mmSDMA0_TILING_CONFIG + SDMA1_REGISTER_OFFSET)); - dev_info(adev->dev, " UVD_UDEC_ADDR_CONFIG=0x%08X\n", - RREG32(mmUVD_UDEC_ADDR_CONFIG)); - dev_info(adev->dev, " UVD_UDEC_DB_ADDR_CONFIG=0x%08X\n", - RREG32(mmUVD_UDEC_DB_ADDR_CONFIG)); - dev_info(adev->dev, " UVD_UDEC_DBW_ADDR_CONFIG=0x%08X\n", - RREG32(mmUVD_UDEC_DBW_ADDR_CONFIG)); dev_info(adev->dev, " CP_MEQ_THRESHOLDS=0x%08X\n", RREG32(mmCP_MEQ_THRESHOLDS)); @@ -3667,7 +4116,18 @@ static int gfx_v8_0_soft_reset(void *handle) gfx_v8_0_cp_gfx_enable(adev, false); /* Disable MEC parsing/prefetching */ - /* XXX todo */ + gfx_v8_0_cp_compute_enable(adev, false); + + if (grbm_soft_reset || srbm_soft_reset) { + tmp = RREG32(mmGMCON_DEBUG); + tmp = REG_SET_FIELD(tmp, + GMCON_DEBUG, GFX_STALL, 1); + tmp = REG_SET_FIELD(tmp, + GMCON_DEBUG, GFX_CLEAR, 1); + WREG32(mmGMCON_DEBUG, tmp); + + udelay(50); + } if (grbm_soft_reset) { tmp = RREG32(mmGRBM_SOFT_RESET); @@ -3696,6 +4156,16 @@ static int gfx_v8_0_soft_reset(void *handle) WREG32(mmSRBM_SOFT_RESET, tmp); tmp = RREG32(mmSRBM_SOFT_RESET); } + + if (grbm_soft_reset || srbm_soft_reset) { + tmp = RREG32(mmGMCON_DEBUG); + tmp = REG_SET_FIELD(tmp, + GMCON_DEBUG, GFX_STALL, 0); + tmp = REG_SET_FIELD(tmp, + GMCON_DEBUG, GFX_CLEAR, 0); + WREG32(mmGMCON_DEBUG, tmp); + } + /* Wait a little for things to settle down */ udelay(50); gfx_v8_0_print_status((void *)adev); @@ -3784,15 +4254,269 @@ static int gfx_v8_0_early_init(void *handle) return 0; } +static int gfx_v8_0_late_init(void *handle) +{ + struct amdgpu_device *adev = (struct amdgpu_device *)handle; + int r; + + r = amdgpu_irq_get(adev, &adev->gfx.priv_reg_irq, 0); + if (r) + return r; + + r = amdgpu_irq_get(adev, &adev->gfx.priv_inst_irq, 0); + if (r) + return r; + + /* requires IBs so do in late init after IB pool is initialized */ + r = gfx_v8_0_do_edc_gpr_workarounds(adev); + if (r) + return r; + + return 0; +} + static int gfx_v8_0_set_powergating_state(void *handle, enum amd_powergating_state state) { return 0; } +static void fiji_send_serdes_cmd(struct amdgpu_device *adev, + uint32_t reg_addr, uint32_t cmd) +{ + uint32_t data; + + gfx_v8_0_select_se_sh(adev, 0xffffffff, 0xffffffff); + + WREG32(mmRLC_SERDES_WR_CU_MASTER_MASK, 0xffffffff); + WREG32(mmRLC_SERDES_WR_NONCU_MASTER_MASK, 0xffffffff); + + data = RREG32(mmRLC_SERDES_WR_CTRL); + data &= ~(RLC_SERDES_WR_CTRL__WRITE_COMMAND_MASK | + RLC_SERDES_WR_CTRL__READ_COMMAND_MASK | + RLC_SERDES_WR_CTRL__P1_SELECT_MASK | + RLC_SERDES_WR_CTRL__P2_SELECT_MASK | + RLC_SERDES_WR_CTRL__RDDATA_RESET_MASK | + RLC_SERDES_WR_CTRL__POWER_DOWN_MASK | + RLC_SERDES_WR_CTRL__POWER_UP_MASK | + RLC_SERDES_WR_CTRL__SHORT_FORMAT_MASK | + RLC_SERDES_WR_CTRL__BPM_DATA_MASK | + RLC_SERDES_WR_CTRL__REG_ADDR_MASK | + RLC_SERDES_WR_CTRL__SRBM_OVERRIDE_MASK); + data |= (RLC_SERDES_WR_CTRL__RSVD_BPM_ADDR_MASK | + (cmd << RLC_SERDES_WR_CTRL__BPM_DATA__SHIFT) | + (reg_addr << RLC_SERDES_WR_CTRL__REG_ADDR__SHIFT) | + (0xff << RLC_SERDES_WR_CTRL__BPM_ADDR__SHIFT)); + + WREG32(mmRLC_SERDES_WR_CTRL, data); +} + +static void fiji_update_medium_grain_clock_gating(struct amdgpu_device *adev, + bool enable) +{ + uint32_t temp, data; + + /* It is disabled by HW by default */ + if (enable) { + /* 1 - RLC memory Light sleep */ + temp = data = RREG32(mmRLC_MEM_SLP_CNTL); + data |= RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK; + if (temp != data) + WREG32(mmRLC_MEM_SLP_CNTL, data); + + /* 2 - CP memory Light sleep */ + temp = data = RREG32(mmCP_MEM_SLP_CNTL); + data |= CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK; + if (temp != data) + WREG32(mmCP_MEM_SLP_CNTL, data); + + /* 3 - RLC_CGTT_MGCG_OVERRIDE */ + temp = data = RREG32(mmRLC_CGTT_MGCG_OVERRIDE); + data &= ~(RLC_CGTT_MGCG_OVERRIDE__CPF_MASK | + RLC_CGTT_MGCG_OVERRIDE__RLC_MASK | + RLC_CGTT_MGCG_OVERRIDE__MGCG_MASK | + RLC_CGTT_MGCG_OVERRIDE__GRBM_MASK); + + if (temp != data) + WREG32(mmRLC_CGTT_MGCG_OVERRIDE, data); + + /* 4 - wait for RLC_SERDES_CU_MASTER & RLC_SERDES_NONCU_MASTER idle */ + gfx_v8_0_wait_for_rlc_serdes(adev); + + /* 5 - clear mgcg override */ + fiji_send_serdes_cmd(adev, BPM_REG_MGCG_OVERRIDE, CLE_BPM_SERDES_CMD); + + /* 6 - Enable CGTS(Tree Shade) MGCG /MGLS */ + temp = data = RREG32(mmCGTS_SM_CTRL_REG); + data &= ~(CGTS_SM_CTRL_REG__SM_MODE_MASK); + data |= (0x2 << CGTS_SM_CTRL_REG__SM_MODE__SHIFT); + data |= CGTS_SM_CTRL_REG__SM_MODE_ENABLE_MASK; + data &= ~CGTS_SM_CTRL_REG__OVERRIDE_MASK; + data &= ~CGTS_SM_CTRL_REG__LS_OVERRIDE_MASK; + data |= CGTS_SM_CTRL_REG__ON_MONITOR_ADD_EN_MASK; + data |= (0x96 << CGTS_SM_CTRL_REG__ON_MONITOR_ADD__SHIFT); + if (temp != data) + WREG32(mmCGTS_SM_CTRL_REG, data); + udelay(50); + + /* 7 - wait for RLC_SERDES_CU_MASTER & RLC_SERDES_NONCU_MASTER idle */ + gfx_v8_0_wait_for_rlc_serdes(adev); + } else { + /* 1 - MGCG_OVERRIDE[0] for CP and MGCG_OVERRIDE[1] for RLC */ + temp = data = RREG32(mmRLC_CGTT_MGCG_OVERRIDE); + data |= (RLC_CGTT_MGCG_OVERRIDE__CPF_MASK | + RLC_CGTT_MGCG_OVERRIDE__RLC_MASK | + RLC_CGTT_MGCG_OVERRIDE__MGCG_MASK | + RLC_CGTT_MGCG_OVERRIDE__GRBM_MASK); + if (temp != data) + WREG32(mmRLC_CGTT_MGCG_OVERRIDE, data); + + /* 2 - disable MGLS in RLC */ + data = RREG32(mmRLC_MEM_SLP_CNTL); + if (data & RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK) { + data &= ~RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK; + WREG32(mmRLC_MEM_SLP_CNTL, data); + } + + /* 3 - disable MGLS in CP */ + data = RREG32(mmCP_MEM_SLP_CNTL); + if (data & CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK) { + data &= ~CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK; + WREG32(mmCP_MEM_SLP_CNTL, data); + } + + /* 4 - Disable CGTS(Tree Shade) MGCG and MGLS */ + temp = data = RREG32(mmCGTS_SM_CTRL_REG); + data |= (CGTS_SM_CTRL_REG__OVERRIDE_MASK | + CGTS_SM_CTRL_REG__LS_OVERRIDE_MASK); + if (temp != data) + WREG32(mmCGTS_SM_CTRL_REG, data); + + /* 5 - wait for RLC_SERDES_CU_MASTER & RLC_SERDES_NONCU_MASTER idle */ + gfx_v8_0_wait_for_rlc_serdes(adev); + + /* 6 - set mgcg override */ + fiji_send_serdes_cmd(adev, BPM_REG_MGCG_OVERRIDE, SET_BPM_SERDES_CMD); + + udelay(50); + + /* 7- wait for RLC_SERDES_CU_MASTER & RLC_SERDES_NONCU_MASTER idle */ + gfx_v8_0_wait_for_rlc_serdes(adev); + } +} + +static void fiji_update_coarse_grain_clock_gating(struct amdgpu_device *adev, + bool enable) +{ + uint32_t temp, temp1, data, data1; + + temp = data = RREG32(mmRLC_CGCG_CGLS_CTRL); + + if (enable) { + /* 1 enable cntx_empty_int_enable/cntx_busy_int_enable/ + * Cmp_busy/GFX_Idle interrupts + */ + gfx_v8_0_enable_gui_idle_interrupt(adev, true); + + temp1 = data1 = RREG32(mmRLC_CGTT_MGCG_OVERRIDE); + data1 &= ~RLC_CGTT_MGCG_OVERRIDE__CGCG_MASK; + if (temp1 != data1) + WREG32(mmRLC_CGTT_MGCG_OVERRIDE, data1); + + /* 2 wait for RLC_SERDES_CU_MASTER & RLC_SERDES_NONCU_MASTER idle */ + gfx_v8_0_wait_for_rlc_serdes(adev); + + /* 3 - clear cgcg override */ + fiji_send_serdes_cmd(adev, BPM_REG_CGCG_OVERRIDE, CLE_BPM_SERDES_CMD); + + /* wait for RLC_SERDES_CU_MASTER & RLC_SERDES_NONCU_MASTER idle */ + gfx_v8_0_wait_for_rlc_serdes(adev); + + /* 4 - write cmd to set CGLS */ + fiji_send_serdes_cmd(adev, BPM_REG_CGLS_EN, SET_BPM_SERDES_CMD); + + /* 5 - enable cgcg */ + data |= RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK; + + /* enable cgls*/ + data |= RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK; + + temp1 = data1 = RREG32(mmRLC_CGTT_MGCG_OVERRIDE); + data1 &= ~RLC_CGTT_MGCG_OVERRIDE__CGLS_MASK; + + if (temp1 != data1) + WREG32(mmRLC_CGTT_MGCG_OVERRIDE, data1); + + if (temp != data) + WREG32(mmRLC_CGCG_CGLS_CTRL, data); + } else { + /* disable cntx_empty_int_enable & GFX Idle interrupt */ + gfx_v8_0_enable_gui_idle_interrupt(adev, false); + + /* TEST CGCG */ + temp1 = data1 = RREG32(mmRLC_CGTT_MGCG_OVERRIDE); + data1 |= (RLC_CGTT_MGCG_OVERRIDE__CGCG_MASK | + RLC_CGTT_MGCG_OVERRIDE__CGLS_MASK); + if (temp1 != data1) + WREG32(mmRLC_CGTT_MGCG_OVERRIDE, data1); + + /* read gfx register to wake up cgcg */ + RREG32(mmCB_CGTT_SCLK_CTRL); + RREG32(mmCB_CGTT_SCLK_CTRL); + RREG32(mmCB_CGTT_SCLK_CTRL); + RREG32(mmCB_CGTT_SCLK_CTRL); + + /* wait for RLC_SERDES_CU_MASTER & RLC_SERDES_NONCU_MASTER idle */ + gfx_v8_0_wait_for_rlc_serdes(adev); + + /* write cmd to Set CGCG Overrride */ + fiji_send_serdes_cmd(adev, BPM_REG_CGCG_OVERRIDE, SET_BPM_SERDES_CMD); + + /* wait for RLC_SERDES_CU_MASTER & RLC_SERDES_NONCU_MASTER idle */ + gfx_v8_0_wait_for_rlc_serdes(adev); + + /* write cmd to Clear CGLS */ + fiji_send_serdes_cmd(adev, BPM_REG_CGLS_EN, CLE_BPM_SERDES_CMD); + + /* disable cgcg, cgls should be disabled too. */ + data &= ~(RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK | + RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK); + if (temp != data) + WREG32(mmRLC_CGCG_CGLS_CTRL, data); + } +} +static int fiji_update_gfx_clock_gating(struct amdgpu_device *adev, + bool enable) +{ + if (enable) { + /* CGCG/CGLS should be enabled after MGCG/MGLS/TS(CG/LS) + * === MGCG + MGLS + TS(CG/LS) === + */ + fiji_update_medium_grain_clock_gating(adev, enable); + fiji_update_coarse_grain_clock_gating(adev, enable); + } else { + /* CGCG/CGLS should be disabled before MGCG/MGLS/TS(CG/LS) + * === CGCG + CGLS === + */ + fiji_update_coarse_grain_clock_gating(adev, enable); + fiji_update_medium_grain_clock_gating(adev, enable); + } + return 0; +} + static int gfx_v8_0_set_clockgating_state(void *handle, enum amd_clockgating_state state) { + struct amdgpu_device *adev = (struct amdgpu_device *)handle; + + switch (adev->asic_type) { + case CHIP_FIJI: + fiji_update_gfx_clock_gating(adev, + state == AMD_CG_STATE_GATE ? true : false); + break; + default: + break; + } return 0; } @@ -3865,6 +4589,18 @@ static void gfx_v8_0_ring_emit_hdp_flush(struct amdgpu_ring *ring) amdgpu_ring_write(ring, 0x20); /* poll interval */ } +static void gfx_v8_0_ring_emit_hdp_invalidate(struct amdgpu_ring *ring) +{ + amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3)); + amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) | + WRITE_DATA_DST_SEL(0) | + WR_CONFIRM)); + amdgpu_ring_write(ring, mmHDP_DEBUG0); + amdgpu_ring_write(ring, 0); + amdgpu_ring_write(ring, 1); + +} + static void gfx_v8_0_ring_emit_ib_gfx(struct amdgpu_ring *ring, struct amdgpu_ib *ib) { @@ -3897,8 +4633,7 @@ static void gfx_v8_0_ring_emit_ib_gfx(struct amdgpu_ring *ring, else header = PACKET3(PACKET3_INDIRECT_BUFFER, 2); - control |= ib->length_dw | - (ib->vm ? (ib->vm->ids[ring->idx].id << 24) : 0); + control |= ib->length_dw | (ib->vm_id << 24); amdgpu_ring_write(ring, header); amdgpu_ring_write(ring, @@ -3927,8 +4662,7 @@ static void gfx_v8_0_ring_emit_ib_compute(struct amdgpu_ring *ring, header = PACKET3(PACKET3_INDIRECT_BUFFER, 2); - control |= ib->length_dw | - (ib->vm ? (ib->vm->ids[ring->idx].id << 24) : 0); + control |= ib->length_dw | (ib->vm_id << 24); amdgpu_ring_write(ring, header); amdgpu_ring_write(ring, @@ -3953,61 +4687,23 @@ static void gfx_v8_0_ring_emit_fence_gfx(struct amdgpu_ring *ring, u64 addr, EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) | EVENT_INDEX(5))); amdgpu_ring_write(ring, addr & 0xfffffffc); - amdgpu_ring_write(ring, (upper_32_bits(addr) & 0xffff) | + amdgpu_ring_write(ring, (upper_32_bits(addr) & 0xffff) | DATA_SEL(write64bit ? 2 : 1) | INT_SEL(int_sel ? 2 : 0)); amdgpu_ring_write(ring, lower_32_bits(seq)); amdgpu_ring_write(ring, upper_32_bits(seq)); } -/** - * gfx_v8_0_ring_emit_semaphore - emit a semaphore on the CP ring - * - * @ring: amdgpu ring buffer object - * @semaphore: amdgpu semaphore object - * @emit_wait: Is this a sempahore wait? - * - * Emits a semaphore signal/wait packet to the CP ring and prevents the PFP - * from running ahead of semaphore waits. - */ -static bool gfx_v8_0_ring_emit_semaphore(struct amdgpu_ring *ring, - struct amdgpu_semaphore *semaphore, - bool emit_wait) -{ - uint64_t addr = semaphore->gpu_addr; - unsigned sel = emit_wait ? PACKET3_SEM_SEL_WAIT : PACKET3_SEM_SEL_SIGNAL; - - if (ring->adev->asic_type == CHIP_TOPAZ || - ring->adev->asic_type == CHIP_TONGA || - ring->adev->asic_type == CHIP_FIJI) - /* we got a hw semaphore bug in VI TONGA, return false to switch back to sw fence wait */ - return false; - else { - amdgpu_ring_write(ring, PACKET3(PACKET3_MEM_SEMAPHORE, 2)); - amdgpu_ring_write(ring, lower_32_bits(addr)); - amdgpu_ring_write(ring, upper_32_bits(addr)); - amdgpu_ring_write(ring, sel); - } - - if (emit_wait && (ring->type == AMDGPU_RING_TYPE_GFX)) { - /* Prevent the PFP from running ahead of the semaphore wait */ - amdgpu_ring_write(ring, PACKET3(PACKET3_PFP_SYNC_ME, 0)); - amdgpu_ring_write(ring, 0x0); - } - - return true; -} - -static void gfx_v8_0_ring_emit_vm_flush(struct amdgpu_ring *ring, - unsigned vm_id, uint64_t pd_addr) +static void gfx_v8_0_ring_emit_pipeline_sync(struct amdgpu_ring *ring) { int usepfp = (ring->type == AMDGPU_RING_TYPE_GFX); - uint32_t seq = ring->fence_drv.sync_seq[ring->idx]; + uint32_t seq = ring->fence_drv.sync_seq; uint64_t addr = ring->fence_drv.gpu_addr; amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5)); amdgpu_ring_write(ring, (WAIT_REG_MEM_MEM_SPACE(1) | /* memory */ - WAIT_REG_MEM_FUNCTION(3))); /* equal */ + WAIT_REG_MEM_FUNCTION(3) | /* equal */ + WAIT_REG_MEM_ENGINE(usepfp))); /* pfp or me */ amdgpu_ring_write(ring, addr & 0xfffffffc); amdgpu_ring_write(ring, upper_32_bits(addr) & 0xffffffff); amdgpu_ring_write(ring, seq); @@ -4021,6 +4717,12 @@ static void gfx_v8_0_ring_emit_vm_flush(struct amdgpu_ring *ring, amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0)); amdgpu_ring_write(ring, 0); } +} + +static void gfx_v8_0_ring_emit_vm_flush(struct amdgpu_ring *ring, + unsigned vm_id, uint64_t pd_addr) +{ + int usepfp = (ring->type == AMDGPU_RING_TYPE_GFX); amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3)); amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(usepfp) | @@ -4068,15 +4770,6 @@ static void gfx_v8_0_ring_emit_vm_flush(struct amdgpu_ring *ring, } } -static bool gfx_v8_0_ring_is_lockup(struct amdgpu_ring *ring) -{ - if (gfx_v8_0_is_idle(ring->adev)) { - amdgpu_ring_lockup_update(ring); - return false; - } - return amdgpu_ring_test_lockup(ring); -} - static u32 gfx_v8_0_ring_get_rptr_compute(struct amdgpu_ring *ring) { return ring->adev->wb.wb[ring->rptr_offs]; @@ -4107,6 +4800,7 @@ static void gfx_v8_0_ring_emit_fence_compute(struct amdgpu_ring *ring, amdgpu_ring_write(ring, PACKET3(PACKET3_RELEASE_MEM, 5)); amdgpu_ring_write(ring, (EOP_TCL1_ACTION_EN | EOP_TC_ACTION_EN | + EOP_TC_WB_ACTION_EN | EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) | EVENT_INDEX(5))); amdgpu_ring_write(ring, DATA_SEL(write64bit ? 2 : 1) | INT_SEL(int_sel ? 2 : 0)); @@ -4201,7 +4895,7 @@ static int gfx_v8_0_set_priv_reg_fault_state(struct amdgpu_device *adev, case AMDGPU_IRQ_STATE_ENABLE: cp_int_cntl = RREG32(mmCP_INT_CNTL_RING0); cp_int_cntl = REG_SET_FIELD(cp_int_cntl, CP_INT_CNTL_RING0, - PRIV_REG_INT_ENABLE, 0); + PRIV_REG_INT_ENABLE, 1); WREG32(mmCP_INT_CNTL_RING0, cp_int_cntl); break; default: @@ -4329,7 +5023,7 @@ static int gfx_v8_0_priv_inst_irq(struct amdgpu_device *adev, const struct amd_ip_funcs gfx_v8_0_ip_funcs = { .early_init = gfx_v8_0_early_init, - .late_init = NULL, + .late_init = gfx_v8_0_late_init, .sw_init = gfx_v8_0_sw_init, .sw_fini = gfx_v8_0_sw_fini, .hw_init = gfx_v8_0_hw_init, @@ -4351,14 +5045,15 @@ static const struct amdgpu_ring_funcs gfx_v8_0_ring_funcs_gfx = { .parse_cs = NULL, .emit_ib = gfx_v8_0_ring_emit_ib_gfx, .emit_fence = gfx_v8_0_ring_emit_fence_gfx, - .emit_semaphore = gfx_v8_0_ring_emit_semaphore, + .emit_pipeline_sync = gfx_v8_0_ring_emit_pipeline_sync, .emit_vm_flush = gfx_v8_0_ring_emit_vm_flush, .emit_gds_switch = gfx_v8_0_ring_emit_gds_switch, .emit_hdp_flush = gfx_v8_0_ring_emit_hdp_flush, + .emit_hdp_invalidate = gfx_v8_0_ring_emit_hdp_invalidate, .test_ring = gfx_v8_0_ring_test_ring, .test_ib = gfx_v8_0_ring_test_ib, - .is_lockup = gfx_v8_0_ring_is_lockup, .insert_nop = amdgpu_ring_insert_nop, + .pad_ib = amdgpu_ring_generic_pad_ib, }; static const struct amdgpu_ring_funcs gfx_v8_0_ring_funcs_compute = { @@ -4368,14 +5063,15 @@ static const struct amdgpu_ring_funcs gfx_v8_0_ring_funcs_compute = { .parse_cs = NULL, .emit_ib = gfx_v8_0_ring_emit_ib_compute, .emit_fence = gfx_v8_0_ring_emit_fence_compute, - .emit_semaphore = gfx_v8_0_ring_emit_semaphore, + .emit_pipeline_sync = gfx_v8_0_ring_emit_pipeline_sync, .emit_vm_flush = gfx_v8_0_ring_emit_vm_flush, .emit_gds_switch = gfx_v8_0_ring_emit_gds_switch, .emit_hdp_flush = gfx_v8_0_ring_emit_hdp_flush, + .emit_hdp_invalidate = gfx_v8_0_ring_emit_hdp_invalidate, .test_ring = gfx_v8_0_ring_test_ring, .test_ib = gfx_v8_0_ring_test_ib, - .is_lockup = gfx_v8_0_ring_is_lockup, .insert_nop = amdgpu_ring_insert_nop, + .pad_ib = amdgpu_ring_generic_pad_ib, }; static void gfx_v8_0_set_ring_funcs(struct amdgpu_device *adev) @@ -4444,32 +5140,23 @@ static void gfx_v8_0_set_gds_init(struct amdgpu_device *adev) } } -static u32 gfx_v8_0_get_cu_active_bitmap(struct amdgpu_device *adev, - u32 se, u32 sh) +static u32 gfx_v8_0_get_cu_active_bitmap(struct amdgpu_device *adev) { - u32 mask = 0, tmp, tmp1; - int i; - - gfx_v8_0_select_se_sh(adev, se, sh); - tmp = RREG32(mmCC_GC_SHADER_ARRAY_CONFIG); - tmp1 = RREG32(mmGC_USER_SHADER_ARRAY_CONFIG); - gfx_v8_0_select_se_sh(adev, 0xffffffff, 0xffffffff); + u32 data, mask; - tmp &= 0xffff0000; + data = RREG32(mmCC_GC_SHADER_ARRAY_CONFIG); + data |= RREG32(mmGC_USER_SHADER_ARRAY_CONFIG); - tmp |= tmp1; - tmp >>= 16; + data &= CC_GC_SHADER_ARRAY_CONFIG__INACTIVE_CUS_MASK; + data >>= CC_GC_SHADER_ARRAY_CONFIG__INACTIVE_CUS__SHIFT; - for (i = 0; i < adev->gfx.config.max_cu_per_sh; i ++) { - mask <<= 1; - mask |= 1; - } + mask = gfx_v8_0_create_bitmask(adev->gfx.config.max_cu_per_sh); - return (~tmp) & mask; + return (~data) & mask; } int gfx_v8_0_get_cu_info(struct amdgpu_device *adev, - struct amdgpu_cu_info *cu_info) + struct amdgpu_cu_info *cu_info) { int i, j, k, counter, active_cu_number = 0; u32 mask, bitmap, ao_bitmap, ao_cu_mask = 0; @@ -4477,16 +5164,19 @@ int gfx_v8_0_get_cu_info(struct amdgpu_device *adev, if (!adev || !cu_info) return -EINVAL; + memset(cu_info, 0, sizeof(*cu_info)); + mutex_lock(&adev->grbm_idx_mutex); for (i = 0; i < adev->gfx.config.max_shader_engines; i++) { for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) { mask = 1; ao_bitmap = 0; counter = 0; - bitmap = gfx_v8_0_get_cu_active_bitmap(adev, i, j); + gfx_v8_0_select_se_sh(adev, i, j); + bitmap = gfx_v8_0_get_cu_active_bitmap(adev); cu_info->bitmap[i][j] = bitmap; - for (k = 0; k < adev->gfx.config.max_cu_per_sh; k ++) { + for (k = 0; k < 16; k ++) { if (bitmap & mask) { if (counter < 2) ao_bitmap |= mask; @@ -4498,9 +5188,11 @@ int gfx_v8_0_get_cu_info(struct amdgpu_device *adev, ao_cu_mask |= (ao_bitmap << (i * 16 + j * 8)); } } + gfx_v8_0_select_se_sh(adev, 0xffffffff, 0xffffffff); + mutex_unlock(&adev->grbm_idx_mutex); cu_info->number = active_cu_number; cu_info->ao_cu_mask = ao_cu_mask; - mutex_unlock(&adev->grbm_idx_mutex); + return 0; } diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c b/drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c index fab5471d25d7..82ce7d943884 100644 --- a/drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c @@ -40,11 +40,41 @@ static void gmc_v7_0_set_gart_funcs(struct amdgpu_device *adev); static void gmc_v7_0_set_irq_funcs(struct amdgpu_device *adev); -MODULE_FIRMWARE("radeon/boniare_mc.bin"); +MODULE_FIRMWARE("radeon/bonaire_mc.bin"); MODULE_FIRMWARE("radeon/hawaii_mc.bin"); +MODULE_FIRMWARE("amdgpu/topaz_mc.bin"); + +static const u32 golden_settings_iceland_a11[] = +{ + mmVM_PRT_APERTURE0_LOW_ADDR, 0x0fffffff, 0x0fffffff, + mmVM_PRT_APERTURE1_LOW_ADDR, 0x0fffffff, 0x0fffffff, + mmVM_PRT_APERTURE2_LOW_ADDR, 0x0fffffff, 0x0fffffff, + mmVM_PRT_APERTURE3_LOW_ADDR, 0x0fffffff, 0x0fffffff +}; + +static const u32 iceland_mgcg_cgcg_init[] = +{ + mmMC_MEM_POWER_LS, 0xffffffff, 0x00000104 +}; + +static void gmc_v7_0_init_golden_registers(struct amdgpu_device *adev) +{ + switch (adev->asic_type) { + case CHIP_TOPAZ: + amdgpu_program_register_sequence(adev, + iceland_mgcg_cgcg_init, + (const u32)ARRAY_SIZE(iceland_mgcg_cgcg_init)); + amdgpu_program_register_sequence(adev, + golden_settings_iceland_a11, + (const u32)ARRAY_SIZE(golden_settings_iceland_a11)); + break; + default: + break; + } +} /** - * gmc8_mc_wait_for_idle - wait for MC idle callback. + * gmc7_mc_wait_for_idle - wait for MC idle callback. * * @adev: amdgpu_device pointer * @@ -132,13 +162,20 @@ static int gmc_v7_0_init_microcode(struct amdgpu_device *adev) case CHIP_HAWAII: chip_name = "hawaii"; break; + case CHIP_TOPAZ: + chip_name = "topaz"; + break; case CHIP_KAVERI: case CHIP_KABINI: return 0; default: BUG(); } - snprintf(fw_name, sizeof(fw_name), "radeon/%s_mc.bin", chip_name); + if (adev->asic_type == CHIP_TOPAZ) + snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mc.bin", chip_name); + else + snprintf(fw_name, sizeof(fw_name), "radeon/%s_mc.bin", chip_name); + err = request_firmware(&adev->mc.fw, fw_name, adev->dev); if (err) goto out; @@ -302,7 +339,7 @@ static void gmc_v7_0_mc_program(struct amdgpu_device *adev) WREG32(mmBIF_FB_EN, BIF_FB_EN__FB_READ_EN_MASK | BIF_FB_EN__FB_WRITE_EN_MASK); tmp = RREG32(mmHDP_MISC_CNTL); - tmp = REG_SET_FIELD(tmp, HDP_MISC_CNTL, FLUSH_INVALIDATE_CACHE, 1); + tmp = REG_SET_FIELD(tmp, HDP_MISC_CNTL, FLUSH_INVALIDATE_CACHE, 0); WREG32(mmHDP_MISC_CNTL, tmp); tmp = RREG32(mmHDP_HOST_PATH_CNTL); @@ -370,6 +407,10 @@ static int gmc_v7_0_mc_init(struct amdgpu_device *adev) adev->mc.real_vram_size = RREG32(mmCONFIG_MEMSIZE) * 1024ULL * 1024ULL; adev->mc.visible_vram_size = adev->mc.aper_size; + /* In case the PCI BAR is larger than the actual amount of vram */ + if (adev->mc.visible_vram_size > adev->mc.real_vram_size) + adev->mc.visible_vram_size = adev->mc.real_vram_size; + /* unless the user had overridden it, set the gart * size equal to the 1024 or vram, whichever is larger. */ @@ -436,6 +477,33 @@ static int gmc_v7_0_gart_set_pte_pde(struct amdgpu_device *adev, } /** + * gmc_v8_0_set_fault_enable_default - update VM fault handling + * + * @adev: amdgpu_device pointer + * @value: true redirects VM faults to the default page + */ +static void gmc_v7_0_set_fault_enable_default(struct amdgpu_device *adev, + bool value) +{ + u32 tmp; + + tmp = RREG32(mmVM_CONTEXT1_CNTL); + tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, + RANGE_PROTECTION_FAULT_ENABLE_DEFAULT, value); + tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, + DUMMY_PAGE_PROTECTION_FAULT_ENABLE_DEFAULT, value); + tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, + PDE0_PROTECTION_FAULT_ENABLE_DEFAULT, value); + tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, + VALID_PROTECTION_FAULT_ENABLE_DEFAULT, value); + tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, + READ_PROTECTION_FAULT_ENABLE_DEFAULT, value); + tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, + WRITE_PROTECTION_FAULT_ENABLE_DEFAULT, value); + WREG32(mmVM_CONTEXT1_CNTL, tmp); +} + +/** * gmc_v7_0_gart_enable - gart enable * * @adev: amdgpu_device pointer @@ -474,6 +542,7 @@ static int gmc_v7_0_gart_enable(struct amdgpu_device *adev) tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_L2_PDE0_CACHE_LRU_UPDATE_BY_WRITE, 1); tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, EFFECTIVE_L2_QUEUE_SIZE, 7); tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, CONTEXT1_IDENTITY_ACCESS_MODE, 1); + tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_DEFAULT_PAGE_OUT_TO_SYSTEM_MEMORY, 1); WREG32(mmVM_L2_CNTL, tmp); tmp = REG_SET_FIELD(0, VM_L2_CNTL2, INVALIDATE_ALL_L1_TLBS, 1); tmp = REG_SET_FIELD(tmp, VM_L2_CNTL2, INVALIDATE_L2_CACHE, 1); @@ -485,7 +554,7 @@ static int gmc_v7_0_gart_enable(struct amdgpu_device *adev) WREG32(mmVM_L2_CNTL3, tmp); /* setup context0 */ WREG32(mmVM_CONTEXT0_PAGE_TABLE_START_ADDR, adev->mc.gtt_start >> 12); - WREG32(mmVM_CONTEXT0_PAGE_TABLE_END_ADDR, (adev->mc.gtt_end >> 12) - 1); + WREG32(mmVM_CONTEXT0_PAGE_TABLE_END_ADDR, adev->mc.gtt_end >> 12); WREG32(mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR, adev->gart.table_addr >> 12); WREG32(mmVM_CONTEXT0_PROTECTION_FAULT_DEFAULT_ADDR, (u32)(adev->dummy_page.addr >> 12)); @@ -523,15 +592,13 @@ static int gmc_v7_0_gart_enable(struct amdgpu_device *adev) tmp = RREG32(mmVM_CONTEXT1_CNTL); tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, ENABLE_CONTEXT, 1); tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, PAGE_TABLE_DEPTH, 1); - tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, RANGE_PROTECTION_FAULT_ENABLE_DEFAULT, 1); - tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, DUMMY_PAGE_PROTECTION_FAULT_ENABLE_DEFAULT, 1); - tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, PDE0_PROTECTION_FAULT_ENABLE_DEFAULT, 1); - tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, VALID_PROTECTION_FAULT_ENABLE_DEFAULT, 1); - tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, READ_PROTECTION_FAULT_ENABLE_DEFAULT, 1); - tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, WRITE_PROTECTION_FAULT_ENABLE_DEFAULT, 1); tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, PAGE_TABLE_BLOCK_SIZE, amdgpu_vm_block_size - 9); WREG32(mmVM_CONTEXT1_CNTL, tmp); + if (amdgpu_vm_fault_stop == AMDGPU_VM_FAULT_STOP_ALWAYS) + gmc_v7_0_set_fault_enable_default(adev, false); + else + gmc_v7_0_set_fault_enable_default(adev, true); if (adev->asic_type == CHIP_KAVERI) { tmp = RREG32(mmCHUB_CONTROL); @@ -627,7 +694,8 @@ static int gmc_v7_0_vm_init(struct amdgpu_device *adev) * amdgpu graphics/compute will use VMIDs 1-7 * amdkfd will use VMIDs 8-15 */ - adev->vm_manager.nvm = AMDGPU_NUM_OF_VMIDS; + adev->vm_manager.num_ids = AMDGPU_NUM_OF_VMIDS; + amdgpu_vm_manager_init(adev); /* base offset of vram pages */ if (adev->flags & AMD_IS_APU) { @@ -725,7 +793,7 @@ static void gmc_v7_0_enable_mc_ls(struct amdgpu_device *adev, for (i = 0; i < ARRAY_SIZE(mc_cg_registers); i++) { orig = data = RREG32(mc_cg_registers[i]); - if (enable && (adev->cg_flags & AMDGPU_CG_SUPPORT_MC_LS)) + if (enable && (adev->cg_flags & AMD_CG_SUPPORT_MC_LS)) data |= mc_cg_ls_en[i]; else data &= ~mc_cg_ls_en[i]; @@ -742,7 +810,7 @@ static void gmc_v7_0_enable_mc_mgcg(struct amdgpu_device *adev, for (i = 0; i < ARRAY_SIZE(mc_cg_registers); i++) { orig = data = RREG32(mc_cg_registers[i]); - if (enable && (adev->cg_flags & AMDGPU_CG_SUPPORT_MC_MGCG)) + if (enable && (adev->cg_flags & AMD_CG_SUPPORT_MC_MGCG)) data |= mc_cg_en[i]; else data &= ~mc_cg_en[i]; @@ -758,7 +826,7 @@ static void gmc_v7_0_enable_bif_mgls(struct amdgpu_device *adev, orig = data = RREG32_PCIE(ixPCIE_CNTL2); - if (enable && (adev->cg_flags & AMDGPU_CG_SUPPORT_BIF_LS)) { + if (enable && (adev->cg_flags & AMD_CG_SUPPORT_BIF_LS)) { data = REG_SET_FIELD(data, PCIE_CNTL2, SLV_MEM_LS_EN, 1); data = REG_SET_FIELD(data, PCIE_CNTL2, MST_MEM_LS_EN, 1); data = REG_SET_FIELD(data, PCIE_CNTL2, REPLAY_MEM_LS_EN, 1); @@ -781,7 +849,7 @@ static void gmc_v7_0_enable_hdp_mgcg(struct amdgpu_device *adev, orig = data = RREG32(mmHDP_HOST_PATH_CNTL); - if (enable && (adev->cg_flags & AMDGPU_CG_SUPPORT_HDP_MGCG)) + if (enable && (adev->cg_flags & AMD_CG_SUPPORT_HDP_MGCG)) data = REG_SET_FIELD(data, HDP_HOST_PATH_CNTL, CLOCK_GATING_DIS, 0); else data = REG_SET_FIELD(data, HDP_HOST_PATH_CNTL, CLOCK_GATING_DIS, 1); @@ -797,7 +865,7 @@ static void gmc_v7_0_enable_hdp_ls(struct amdgpu_device *adev, orig = data = RREG32(mmHDP_MEM_POWER_LS); - if (enable && (adev->cg_flags & AMDGPU_CG_SUPPORT_HDP_LS)) + if (enable && (adev->cg_flags & AMD_CG_SUPPORT_HDP_LS)) data = REG_SET_FIELD(data, HDP_MEM_POWER_LS, LS_ENABLE, 1); else data = REG_SET_FIELD(data, HDP_MEM_POWER_LS, LS_ENABLE, 0); @@ -859,10 +927,6 @@ static int gmc_v7_0_sw_init(void *handle) int dma_bits; struct amdgpu_device *adev = (struct amdgpu_device *)handle; - r = amdgpu_gem_init(adev); - if (r) - return r; - r = amdgpu_irq_add_id(adev, 146, &adev->mc.vm_fault); if (r) return r; @@ -935,17 +999,15 @@ static int gmc_v7_0_sw_init(void *handle) static int gmc_v7_0_sw_fini(void *handle) { - int i; struct amdgpu_device *adev = (struct amdgpu_device *)handle; if (adev->vm_manager.enabled) { - for (i = 0; i < AMDGPU_NUM_VM; ++i) - amdgpu_fence_unref(&adev->vm_manager.active[i]); + amdgpu_vm_manager_fini(adev); gmc_v7_0_vm_fini(adev); adev->vm_manager.enabled = false; } gmc_v7_0_gart_fini(adev); - amdgpu_gem_fini(adev); + amdgpu_gem_force_release(adev); amdgpu_bo_fini(adev); return 0; @@ -956,6 +1018,8 @@ static int gmc_v7_0_hw_init(void *handle) int r; struct amdgpu_device *adev = (struct amdgpu_device *)handle; + gmc_v7_0_init_golden_registers(adev); + gmc_v7_0_mc_program(adev); if (!(adev->flags & AMD_IS_APU)) { @@ -985,12 +1049,9 @@ static int gmc_v7_0_hw_fini(void *handle) static int gmc_v7_0_suspend(void *handle) { - int i; struct amdgpu_device *adev = (struct amdgpu_device *)handle; if (adev->vm_manager.enabled) { - for (i = 0; i < AMDGPU_NUM_VM; ++i) - amdgpu_fence_unref(&adev->vm_manager.active[i]); gmc_v7_0_vm_fini(adev); adev->vm_manager.enabled = false; } @@ -1268,6 +1329,9 @@ static int gmc_v7_0_process_interrupt(struct amdgpu_device *adev, if (!addr && !status) return 0; + if (amdgpu_vm_fault_stop == AMDGPU_VM_FAULT_STOP_FIRST) + gmc_v7_0_set_fault_enable_default(adev, false); + dev_err(adev->dev, "GPU fault detected: %d 0x%08x\n", entry->src_id, entry->src_data); dev_err(adev->dev, " VM_CONTEXT1_PROTECTION_FAULT_ADDR 0x%08X\n", diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c b/drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c index 7bc9e9fcf3d2..29bd7b57dc91 100644 --- a/drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c @@ -42,9 +42,7 @@ static void gmc_v8_0_set_gart_funcs(struct amdgpu_device *adev); static void gmc_v8_0_set_irq_funcs(struct amdgpu_device *adev); -MODULE_FIRMWARE("amdgpu/topaz_mc.bin"); MODULE_FIRMWARE("amdgpu/tonga_mc.bin"); -MODULE_FIRMWARE("amdgpu/fiji_mc.bin"); static const u32 golden_settings_tonga_a11[] = { @@ -75,35 +73,20 @@ static const u32 fiji_mgcg_cgcg_init[] = mmMC_MEM_POWER_LS, 0xffffffff, 0x00000104 }; -static const u32 golden_settings_iceland_a11[] = -{ - mmVM_PRT_APERTURE0_LOW_ADDR, 0x0fffffff, 0x0fffffff, - mmVM_PRT_APERTURE1_LOW_ADDR, 0x0fffffff, 0x0fffffff, - mmVM_PRT_APERTURE2_LOW_ADDR, 0x0fffffff, 0x0fffffff, - mmVM_PRT_APERTURE3_LOW_ADDR, 0x0fffffff, 0x0fffffff -}; - -static const u32 iceland_mgcg_cgcg_init[] = +static const u32 cz_mgcg_cgcg_init[] = { mmMC_MEM_POWER_LS, 0xffffffff, 0x00000104 }; -static const u32 cz_mgcg_cgcg_init[] = +static const u32 stoney_mgcg_cgcg_init[] = { mmMC_MEM_POWER_LS, 0xffffffff, 0x00000104 }; + static void gmc_v8_0_init_golden_registers(struct amdgpu_device *adev) { switch (adev->asic_type) { - case CHIP_TOPAZ: - amdgpu_program_register_sequence(adev, - iceland_mgcg_cgcg_init, - (const u32)ARRAY_SIZE(iceland_mgcg_cgcg_init)); - amdgpu_program_register_sequence(adev, - golden_settings_iceland_a11, - (const u32)ARRAY_SIZE(golden_settings_iceland_a11)); - break; case CHIP_FIJI: amdgpu_program_register_sequence(adev, fiji_mgcg_cgcg_init, @@ -125,6 +108,11 @@ static void gmc_v8_0_init_golden_registers(struct amdgpu_device *adev) cz_mgcg_cgcg_init, (const u32)ARRAY_SIZE(cz_mgcg_cgcg_init)); break; + case CHIP_STONEY: + amdgpu_program_register_sequence(adev, + stoney_mgcg_cgcg_init, + (const u32)ARRAY_SIZE(stoney_mgcg_cgcg_init)); + break; default: break; } @@ -218,16 +206,12 @@ static int gmc_v8_0_init_microcode(struct amdgpu_device *adev) DRM_DEBUG("\n"); switch (adev->asic_type) { - case CHIP_TOPAZ: - chip_name = "topaz"; - break; case CHIP_TONGA: chip_name = "tonga"; break; case CHIP_FIJI: - chip_name = "fiji"; - break; case CHIP_CARRIZO: + case CHIP_STONEY: return 0; default: BUG(); } @@ -268,6 +252,12 @@ static int gmc_v8_0_mc_load_microcode(struct amdgpu_device *adev) if (!adev->mc.fw) return -EINVAL; + /* Skip MC ucode loading on SR-IOV capable boards. + * vbios does this for us in asic_init in that case. + */ + if (adev->virtualization.supports_sr_iov) + return 0; + hdr = (const struct mc_firmware_header_v1_0 *)adev->mc.fw->data; amdgpu_ucode_print_mc_hdr(&hdr->header); @@ -396,7 +386,7 @@ static void gmc_v8_0_mc_program(struct amdgpu_device *adev) WREG32(mmBIF_FB_EN, BIF_FB_EN__FB_READ_EN_MASK | BIF_FB_EN__FB_WRITE_EN_MASK); tmp = RREG32(mmHDP_MISC_CNTL); - tmp = REG_SET_FIELD(tmp, HDP_MISC_CNTL, FLUSH_INVALIDATE_CACHE, 1); + tmp = REG_SET_FIELD(tmp, HDP_MISC_CNTL, FLUSH_INVALIDATE_CACHE, 0); WREG32(mmHDP_MISC_CNTL, tmp); tmp = RREG32(mmHDP_HOST_PATH_CNTL); @@ -464,6 +454,10 @@ static int gmc_v8_0_mc_init(struct amdgpu_device *adev) adev->mc.real_vram_size = RREG32(mmCONFIG_MEMSIZE) * 1024ULL * 1024ULL; adev->mc.visible_vram_size = adev->mc.aper_size; + /* In case the PCI BAR is larger than the actual amount of vram */ + if (adev->mc.visible_vram_size > adev->mc.real_vram_size) + adev->mc.visible_vram_size = adev->mc.real_vram_size; + /* unless the user had overridden it, set the gart * size equal to the 1024 or vram, whichever is larger. */ @@ -550,6 +544,35 @@ static int gmc_v8_0_gart_set_pte_pde(struct amdgpu_device *adev, } /** + * gmc_v8_0_set_fault_enable_default - update VM fault handling + * + * @adev: amdgpu_device pointer + * @value: true redirects VM faults to the default page + */ +static void gmc_v8_0_set_fault_enable_default(struct amdgpu_device *adev, + bool value) +{ + u32 tmp; + + tmp = RREG32(mmVM_CONTEXT1_CNTL); + tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, + RANGE_PROTECTION_FAULT_ENABLE_DEFAULT, value); + tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, + DUMMY_PAGE_PROTECTION_FAULT_ENABLE_DEFAULT, value); + tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, + PDE0_PROTECTION_FAULT_ENABLE_DEFAULT, value); + tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, + VALID_PROTECTION_FAULT_ENABLE_DEFAULT, value); + tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, + READ_PROTECTION_FAULT_ENABLE_DEFAULT, value); + tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, + WRITE_PROTECTION_FAULT_ENABLE_DEFAULT, value); + tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, + EXECUTE_PROTECTION_FAULT_ENABLE_DEFAULT, value); + WREG32(mmVM_CONTEXT1_CNTL, tmp); +} + +/** * gmc_v8_0_gart_enable - gart enable * * @adev: amdgpu_device pointer @@ -588,6 +611,7 @@ static int gmc_v8_0_gart_enable(struct amdgpu_device *adev) tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_L2_PDE0_CACHE_LRU_UPDATE_BY_WRITE, 1); tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, EFFECTIVE_L2_QUEUE_SIZE, 7); tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, CONTEXT1_IDENTITY_ACCESS_MODE, 1); + tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_DEFAULT_PAGE_OUT_TO_SYSTEM_MEMORY, 1); WREG32(mmVM_L2_CNTL, tmp); tmp = RREG32(mmVM_L2_CNTL2); tmp = REG_SET_FIELD(tmp, VM_L2_CNTL2, INVALIDATE_ALL_L1_TLBS, 1); @@ -615,7 +639,7 @@ static int gmc_v8_0_gart_enable(struct amdgpu_device *adev) WREG32(mmVM_L2_CNTL4, tmp); /* setup context0 */ WREG32(mmVM_CONTEXT0_PAGE_TABLE_START_ADDR, adev->mc.gtt_start >> 12); - WREG32(mmVM_CONTEXT0_PAGE_TABLE_END_ADDR, (adev->mc.gtt_end >> 12) - 1); + WREG32(mmVM_CONTEXT0_PAGE_TABLE_END_ADDR, adev->mc.gtt_end >> 12); WREG32(mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR, adev->gart.table_addr >> 12); WREG32(mmVM_CONTEXT0_PROTECTION_FAULT_DEFAULT_ADDR, (u32)(adev->dummy_page.addr >> 12)); @@ -663,6 +687,10 @@ static int gmc_v8_0_gart_enable(struct amdgpu_device *adev) tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, PAGE_TABLE_BLOCK_SIZE, amdgpu_vm_block_size - 9); WREG32(mmVM_CONTEXT1_CNTL, tmp); + if (amdgpu_vm_fault_stop == AMDGPU_VM_FAULT_STOP_ALWAYS) + gmc_v8_0_set_fault_enable_default(adev, false); + else + gmc_v8_0_set_fault_enable_default(adev, true); gmc_v8_0_gart_flush_gpu_tlb(adev, 0); DRM_INFO("PCIE GART of %uM enabled (table at 0x%016llX).\n", @@ -752,7 +780,8 @@ static int gmc_v8_0_vm_init(struct amdgpu_device *adev) * amdgpu graphics/compute will use VMIDs 1-7 * amdkfd will use VMIDs 8-15 */ - adev->vm_manager.nvm = AMDGPU_NUM_OF_VMIDS; + adev->vm_manager.num_ids = AMDGPU_NUM_OF_VMIDS; + amdgpu_vm_manager_init(adev); /* base offset of vram pages */ if (adev->flags & AMD_IS_APU) { @@ -858,10 +887,6 @@ static int gmc_v8_0_sw_init(void *handle) int dma_bits; struct amdgpu_device *adev = (struct amdgpu_device *)handle; - r = amdgpu_gem_init(adev); - if (r) - return r; - r = amdgpu_irq_add_id(adev, 146, &adev->mc.vm_fault); if (r) return r; @@ -934,17 +959,15 @@ static int gmc_v8_0_sw_init(void *handle) static int gmc_v8_0_sw_fini(void *handle) { - int i; struct amdgpu_device *adev = (struct amdgpu_device *)handle; if (adev->vm_manager.enabled) { - for (i = 0; i < AMDGPU_NUM_VM; ++i) - amdgpu_fence_unref(&adev->vm_manager.active[i]); + amdgpu_vm_manager_fini(adev); gmc_v8_0_vm_fini(adev); adev->vm_manager.enabled = false; } gmc_v8_0_gart_fini(adev); - amdgpu_gem_fini(adev); + amdgpu_gem_force_release(adev); amdgpu_bo_fini(adev); return 0; @@ -959,7 +982,7 @@ static int gmc_v8_0_hw_init(void *handle) gmc_v8_0_mc_program(adev); - if (!(adev->flags & AMD_IS_APU)) { + if (adev->asic_type == CHIP_TONGA) { r = gmc_v8_0_mc_load_microcode(adev); if (r) { DRM_ERROR("Failed to load MC firmware!\n"); @@ -986,12 +1009,9 @@ static int gmc_v8_0_hw_fini(void *handle) static int gmc_v8_0_suspend(void *handle) { - int i; struct amdgpu_device *adev = (struct amdgpu_device *)handle; if (adev->vm_manager.enabled) { - for (i = 0; i < AMDGPU_NUM_VM; ++i) - amdgpu_fence_unref(&adev->vm_manager.active[i]); gmc_v8_0_vm_fini(adev); adev->vm_manager.enabled = false; } @@ -1268,6 +1288,9 @@ static int gmc_v8_0_process_interrupt(struct amdgpu_device *adev, if (!addr && !status) return 0; + if (amdgpu_vm_fault_stop == AMDGPU_VM_FAULT_STOP_FIRST) + gmc_v8_0_set_fault_enable_default(adev, false); + dev_err(adev->dev, "GPU fault detected: %d 0x%08x\n", entry->src_id, entry->src_data); dev_err(adev->dev, " VM_CONTEXT1_PROTECTION_FAULT_ADDR 0x%08X\n", @@ -1279,9 +1302,181 @@ static int gmc_v8_0_process_interrupt(struct amdgpu_device *adev, return 0; } +static void fiji_update_mc_medium_grain_clock_gating(struct amdgpu_device *adev, + bool enable) +{ + uint32_t data; + + if (enable) { + data = RREG32(mmMC_HUB_MISC_HUB_CG); + data |= MC_HUB_MISC_HUB_CG__ENABLE_MASK; + WREG32(mmMC_HUB_MISC_HUB_CG, data); + + data = RREG32(mmMC_HUB_MISC_SIP_CG); + data |= MC_HUB_MISC_SIP_CG__ENABLE_MASK; + WREG32(mmMC_HUB_MISC_SIP_CG, data); + + data = RREG32(mmMC_HUB_MISC_VM_CG); + data |= MC_HUB_MISC_VM_CG__ENABLE_MASK; + WREG32(mmMC_HUB_MISC_VM_CG, data); + + data = RREG32(mmMC_XPB_CLK_GAT); + data |= MC_XPB_CLK_GAT__ENABLE_MASK; + WREG32(mmMC_XPB_CLK_GAT, data); + + data = RREG32(mmATC_MISC_CG); + data |= ATC_MISC_CG__ENABLE_MASK; + WREG32(mmATC_MISC_CG, data); + + data = RREG32(mmMC_CITF_MISC_WR_CG); + data |= MC_CITF_MISC_WR_CG__ENABLE_MASK; + WREG32(mmMC_CITF_MISC_WR_CG, data); + + data = RREG32(mmMC_CITF_MISC_RD_CG); + data |= MC_CITF_MISC_RD_CG__ENABLE_MASK; + WREG32(mmMC_CITF_MISC_RD_CG, data); + + data = RREG32(mmMC_CITF_MISC_VM_CG); + data |= MC_CITF_MISC_VM_CG__ENABLE_MASK; + WREG32(mmMC_CITF_MISC_VM_CG, data); + + data = RREG32(mmVM_L2_CG); + data |= VM_L2_CG__ENABLE_MASK; + WREG32(mmVM_L2_CG, data); + } else { + data = RREG32(mmMC_HUB_MISC_HUB_CG); + data &= ~MC_HUB_MISC_HUB_CG__ENABLE_MASK; + WREG32(mmMC_HUB_MISC_HUB_CG, data); + + data = RREG32(mmMC_HUB_MISC_SIP_CG); + data &= ~MC_HUB_MISC_SIP_CG__ENABLE_MASK; + WREG32(mmMC_HUB_MISC_SIP_CG, data); + + data = RREG32(mmMC_HUB_MISC_VM_CG); + data &= ~MC_HUB_MISC_VM_CG__ENABLE_MASK; + WREG32(mmMC_HUB_MISC_VM_CG, data); + + data = RREG32(mmMC_XPB_CLK_GAT); + data &= ~MC_XPB_CLK_GAT__ENABLE_MASK; + WREG32(mmMC_XPB_CLK_GAT, data); + + data = RREG32(mmATC_MISC_CG); + data &= ~ATC_MISC_CG__ENABLE_MASK; + WREG32(mmATC_MISC_CG, data); + + data = RREG32(mmMC_CITF_MISC_WR_CG); + data &= ~MC_CITF_MISC_WR_CG__ENABLE_MASK; + WREG32(mmMC_CITF_MISC_WR_CG, data); + + data = RREG32(mmMC_CITF_MISC_RD_CG); + data &= ~MC_CITF_MISC_RD_CG__ENABLE_MASK; + WREG32(mmMC_CITF_MISC_RD_CG, data); + + data = RREG32(mmMC_CITF_MISC_VM_CG); + data &= ~MC_CITF_MISC_VM_CG__ENABLE_MASK; + WREG32(mmMC_CITF_MISC_VM_CG, data); + + data = RREG32(mmVM_L2_CG); + data &= ~VM_L2_CG__ENABLE_MASK; + WREG32(mmVM_L2_CG, data); + } +} + +static void fiji_update_mc_light_sleep(struct amdgpu_device *adev, + bool enable) +{ + uint32_t data; + + if (enable) { + data = RREG32(mmMC_HUB_MISC_HUB_CG); + data |= MC_HUB_MISC_HUB_CG__MEM_LS_ENABLE_MASK; + WREG32(mmMC_HUB_MISC_HUB_CG, data); + + data = RREG32(mmMC_HUB_MISC_SIP_CG); + data |= MC_HUB_MISC_SIP_CG__MEM_LS_ENABLE_MASK; + WREG32(mmMC_HUB_MISC_SIP_CG, data); + + data = RREG32(mmMC_HUB_MISC_VM_CG); + data |= MC_HUB_MISC_VM_CG__MEM_LS_ENABLE_MASK; + WREG32(mmMC_HUB_MISC_VM_CG, data); + + data = RREG32(mmMC_XPB_CLK_GAT); + data |= MC_XPB_CLK_GAT__MEM_LS_ENABLE_MASK; + WREG32(mmMC_XPB_CLK_GAT, data); + + data = RREG32(mmATC_MISC_CG); + data |= ATC_MISC_CG__MEM_LS_ENABLE_MASK; + WREG32(mmATC_MISC_CG, data); + + data = RREG32(mmMC_CITF_MISC_WR_CG); + data |= MC_CITF_MISC_WR_CG__MEM_LS_ENABLE_MASK; + WREG32(mmMC_CITF_MISC_WR_CG, data); + + data = RREG32(mmMC_CITF_MISC_RD_CG); + data |= MC_CITF_MISC_RD_CG__MEM_LS_ENABLE_MASK; + WREG32(mmMC_CITF_MISC_RD_CG, data); + + data = RREG32(mmMC_CITF_MISC_VM_CG); + data |= MC_CITF_MISC_VM_CG__MEM_LS_ENABLE_MASK; + WREG32(mmMC_CITF_MISC_VM_CG, data); + + data = RREG32(mmVM_L2_CG); + data |= VM_L2_CG__MEM_LS_ENABLE_MASK; + WREG32(mmVM_L2_CG, data); + } else { + data = RREG32(mmMC_HUB_MISC_HUB_CG); + data &= ~MC_HUB_MISC_HUB_CG__MEM_LS_ENABLE_MASK; + WREG32(mmMC_HUB_MISC_HUB_CG, data); + + data = RREG32(mmMC_HUB_MISC_SIP_CG); + data &= ~MC_HUB_MISC_SIP_CG__MEM_LS_ENABLE_MASK; + WREG32(mmMC_HUB_MISC_SIP_CG, data); + + data = RREG32(mmMC_HUB_MISC_VM_CG); + data &= ~MC_HUB_MISC_VM_CG__MEM_LS_ENABLE_MASK; + WREG32(mmMC_HUB_MISC_VM_CG, data); + + data = RREG32(mmMC_XPB_CLK_GAT); + data &= ~MC_XPB_CLK_GAT__MEM_LS_ENABLE_MASK; + WREG32(mmMC_XPB_CLK_GAT, data); + + data = RREG32(mmATC_MISC_CG); + data &= ~ATC_MISC_CG__MEM_LS_ENABLE_MASK; + WREG32(mmATC_MISC_CG, data); + + data = RREG32(mmMC_CITF_MISC_WR_CG); + data &= ~MC_CITF_MISC_WR_CG__MEM_LS_ENABLE_MASK; + WREG32(mmMC_CITF_MISC_WR_CG, data); + + data = RREG32(mmMC_CITF_MISC_RD_CG); + data &= ~MC_CITF_MISC_RD_CG__MEM_LS_ENABLE_MASK; + WREG32(mmMC_CITF_MISC_RD_CG, data); + + data = RREG32(mmMC_CITF_MISC_VM_CG); + data &= ~MC_CITF_MISC_VM_CG__MEM_LS_ENABLE_MASK; + WREG32(mmMC_CITF_MISC_VM_CG, data); + + data = RREG32(mmVM_L2_CG); + data &= ~VM_L2_CG__MEM_LS_ENABLE_MASK; + WREG32(mmVM_L2_CG, data); + } +} + static int gmc_v8_0_set_clockgating_state(void *handle, enum amd_clockgating_state state) { + struct amdgpu_device *adev = (struct amdgpu_device *)handle; + + switch (adev->asic_type) { + case CHIP_FIJI: + fiji_update_mc_medium_grain_clock_gating(adev, + state == AMD_CG_STATE_GATE ? true : false); + fiji_update_mc_light_sleep(adev, + state == AMD_CG_STATE_GATE ? true : false); + break; + default: + break; + } return 0; } diff --git a/drivers/gpu/drm/amd/amdgpu/iceland_ih.c b/drivers/gpu/drm/amd/amdgpu/iceland_ih.c index 779532d350ff..679e7394a495 100644 --- a/drivers/gpu/drm/amd/amdgpu/iceland_ih.c +++ b/drivers/gpu/drm/amd/amdgpu/iceland_ih.c @@ -253,8 +253,14 @@ static void iceland_ih_set_rptr(struct amdgpu_device *adev) static int iceland_ih_early_init(void *handle) { struct amdgpu_device *adev = (struct amdgpu_device *)handle; + int ret; + + ret = amdgpu_irq_add_domain(adev); + if (ret) + return ret; iceland_ih_set_interrupt_funcs(adev); + return 0; } @@ -278,6 +284,7 @@ static int iceland_ih_sw_fini(void *handle) amdgpu_irq_fini(adev); amdgpu_ih_ring_fini(adev); + amdgpu_irq_remove_domain(adev); return 0; } diff --git a/drivers/gpu/drm/amd/amdgpu/iceland_smc.c b/drivers/gpu/drm/amd/amdgpu/iceland_smc.c index 966d4b2ed9da..52ee08193295 100644 --- a/drivers/gpu/drm/amd/amdgpu/iceland_smc.c +++ b/drivers/gpu/drm/amd/amdgpu/iceland_smc.c @@ -279,6 +279,12 @@ static int iceland_smu_upload_firmware_image(struct amdgpu_device *adev) if (!adev->pm.fw) return -EINVAL; + /* Skip SMC ucode loading on SR-IOV capable boards. + * vbios does this for us in asic_init in that case. + */ + if (adev->virtualization.supports_sr_iov) + return 0; + hdr = (const struct smc_firmware_header_v1_0 *)adev->pm.fw->data; amdgpu_ucode_print_smc_hdr(&hdr->header); @@ -432,7 +438,7 @@ static uint32_t iceland_smu_get_mask_for_fw_type(uint32_t fw_type) case AMDGPU_UCODE_ID_CP_ME: return UCODE_ID_CP_ME_MASK; case AMDGPU_UCODE_ID_CP_MEC1: - return UCODE_ID_CP_MEC_MASK | UCODE_ID_CP_MEC_JT1_MASK | UCODE_ID_CP_MEC_JT2_MASK; + return UCODE_ID_CP_MEC_MASK | UCODE_ID_CP_MEC_JT1_MASK; case AMDGPU_UCODE_ID_CP_MEC2: return UCODE_ID_CP_MEC_MASK; case AMDGPU_UCODE_ID_RLC_G: @@ -522,12 +528,6 @@ static int iceland_smu_request_load_fw(struct amdgpu_device *adev) return -EINVAL; } - if (iceland_smu_populate_single_firmware_entry(adev, UCODE_ID_CP_MEC_JT2, - &toc->entry[toc->num_entries++])) { - DRM_ERROR("Failed to get firmware entry for MEC_JT2\n"); - return -EINVAL; - } - if (iceland_smu_populate_single_firmware_entry(adev, UCODE_ID_SDMA0, &toc->entry[toc->num_entries++])) { DRM_ERROR("Failed to get firmware entry for SDMA0\n"); @@ -550,8 +550,8 @@ static int iceland_smu_request_load_fw(struct amdgpu_device *adev) UCODE_ID_CP_ME_MASK | UCODE_ID_CP_PFP_MASK | UCODE_ID_CP_MEC_MASK | - UCODE_ID_CP_MEC_JT1_MASK | - UCODE_ID_CP_MEC_JT2_MASK; + UCODE_ID_CP_MEC_JT1_MASK; + if (iceland_send_msg_to_smc_with_parameter_without_waiting(adev, PPSMC_MSG_LoadUcodes, fw_to_load)) { DRM_ERROR("Fail to request SMU load ucode\n"); diff --git a/drivers/gpu/drm/amd/amdgpu/kv_dpm.c b/drivers/gpu/drm/amd/amdgpu/kv_dpm.c index 7e9154c7f1db..654d76723bc3 100644 --- a/drivers/gpu/drm/amd/amdgpu/kv_dpm.c +++ b/drivers/gpu/drm/amd/amdgpu/kv_dpm.c @@ -2859,11 +2859,11 @@ static int kv_dpm_init(struct amdgpu_device *adev) pi->voltage_drop_t = 0; pi->caps_sclk_throttle_low_notification = false; pi->caps_fps = false; /* true? */ - pi->caps_uvd_pg = (adev->pg_flags & AMDGPU_PG_SUPPORT_UVD) ? true : false; + pi->caps_uvd_pg = (adev->pg_flags & AMD_PG_SUPPORT_UVD) ? true : false; pi->caps_uvd_dpm = true; - pi->caps_vce_pg = (adev->pg_flags & AMDGPU_PG_SUPPORT_VCE) ? true : false; - pi->caps_samu_pg = (adev->pg_flags & AMDGPU_PG_SUPPORT_SAMU) ? true : false; - pi->caps_acp_pg = (adev->pg_flags & AMDGPU_PG_SUPPORT_ACP) ? true : false; + pi->caps_vce_pg = (adev->pg_flags & AMD_PG_SUPPORT_VCE) ? true : false; + pi->caps_samu_pg = (adev->pg_flags & AMD_PG_SUPPORT_SAMU) ? true : false; + pi->caps_acp_pg = (adev->pg_flags & AMD_PG_SUPPORT_ACP) ? true : false; pi->caps_stable_p_state = false; ret = kv_parse_sys_info_table(adev); diff --git a/drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c b/drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c index 14e87234171a..6e0a86a563f3 100644 --- a/drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c +++ b/drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c @@ -32,8 +32,8 @@ #include "oss/oss_2_4_d.h" #include "oss/oss_2_4_sh_mask.h" -#include "gmc/gmc_8_1_d.h" -#include "gmc/gmc_8_1_sh_mask.h" +#include "gmc/gmc_7_1_d.h" +#include "gmc/gmc_7_1_sh_mask.h" #include "gca/gfx_8_0_d.h" #include "gca/gfx_8_0_enum.h" @@ -118,7 +118,7 @@ static int sdma_v2_4_init_microcode(struct amdgpu_device *adev) { const char *chip_name; char fw_name[30]; - int err, i; + int err = 0, i; struct amdgpu_firmware_info *info = NULL; const struct common_firmware_header *header = NULL; const struct sdma_firmware_header_v1_0 *hdr; @@ -132,27 +132,27 @@ static int sdma_v2_4_init_microcode(struct amdgpu_device *adev) default: BUG(); } - for (i = 0; i < SDMA_MAX_INSTANCE; i++) { + for (i = 0; i < adev->sdma.num_instances; i++) { if (i == 0) snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_sdma.bin", chip_name); else snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_sdma1.bin", chip_name); - err = request_firmware(&adev->sdma[i].fw, fw_name, adev->dev); + err = request_firmware(&adev->sdma.instance[i].fw, fw_name, adev->dev); if (err) goto out; - err = amdgpu_ucode_validate(adev->sdma[i].fw); + err = amdgpu_ucode_validate(adev->sdma.instance[i].fw); if (err) goto out; - hdr = (const struct sdma_firmware_header_v1_0 *)adev->sdma[i].fw->data; - adev->sdma[i].fw_version = le32_to_cpu(hdr->header.ucode_version); - adev->sdma[i].feature_version = le32_to_cpu(hdr->ucode_feature_version); - if (adev->sdma[i].feature_version >= 20) - adev->sdma[i].burst_nop = true; + hdr = (const struct sdma_firmware_header_v1_0 *)adev->sdma.instance[i].fw->data; + adev->sdma.instance[i].fw_version = le32_to_cpu(hdr->header.ucode_version); + adev->sdma.instance[i].feature_version = le32_to_cpu(hdr->ucode_feature_version); + if (adev->sdma.instance[i].feature_version >= 20) + adev->sdma.instance[i].burst_nop = true; if (adev->firmware.smu_load) { info = &adev->firmware.ucode[AMDGPU_UCODE_ID_SDMA0 + i]; info->ucode_id = AMDGPU_UCODE_ID_SDMA0 + i; - info->fw = adev->sdma[i].fw; + info->fw = adev->sdma.instance[i].fw; header = (const struct common_firmware_header *)info->fw->data; adev->firmware.fw_size += ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE); @@ -164,9 +164,9 @@ out: printk(KERN_ERR "sdma_v2_4: Failed to load firmware \"%s\"\n", fw_name); - for (i = 0; i < SDMA_MAX_INSTANCE; i++) { - release_firmware(adev->sdma[i].fw); - adev->sdma[i].fw = NULL; + for (i = 0; i < adev->sdma.num_instances; i++) { + release_firmware(adev->sdma.instance[i].fw); + adev->sdma.instance[i].fw = NULL; } } return err; @@ -199,7 +199,7 @@ static uint32_t sdma_v2_4_ring_get_rptr(struct amdgpu_ring *ring) static uint32_t sdma_v2_4_ring_get_wptr(struct amdgpu_ring *ring) { struct amdgpu_device *adev = ring->adev; - int me = (ring == &ring->adev->sdma[0].ring) ? 0 : 1; + int me = (ring == &ring->adev->sdma.instance[0].ring) ? 0 : 1; u32 wptr = RREG32(mmSDMA0_GFX_RB_WPTR + sdma_offsets[me]) >> 2; return wptr; @@ -215,14 +215,14 @@ static uint32_t sdma_v2_4_ring_get_wptr(struct amdgpu_ring *ring) static void sdma_v2_4_ring_set_wptr(struct amdgpu_ring *ring) { struct amdgpu_device *adev = ring->adev; - int me = (ring == &ring->adev->sdma[0].ring) ? 0 : 1; + int me = (ring == &ring->adev->sdma.instance[0].ring) ? 0 : 1; WREG32(mmSDMA0_GFX_RB_WPTR + sdma_offsets[me], ring->wptr << 2); } static void sdma_v2_4_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count) { - struct amdgpu_sdma *sdma = amdgpu_get_sdma_instance(ring); + struct amdgpu_sdma_instance *sdma = amdgpu_get_sdma_instance(ring); int i; for (i = 0; i < count; i++) @@ -244,7 +244,7 @@ static void sdma_v2_4_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count) static void sdma_v2_4_ring_emit_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib) { - u32 vmid = (ib->vm ? ib->vm->ids[ring->idx].id : 0) & 0xf; + u32 vmid = ib->vm_id & 0xf; u32 next_rptr = ring->wptr + 5; while ((next_rptr & 7) != 2) @@ -284,7 +284,7 @@ static void sdma_v2_4_ring_emit_hdp_flush(struct amdgpu_ring *ring) { u32 ref_and_mask = 0; - if (ring == &ring->adev->sdma[0].ring) + if (ring == &ring->adev->sdma.instance[0].ring) ref_and_mask = REG_SET_FIELD(ref_and_mask, GPU_HDP_FLUSH_DONE, SDMA0, 1); else ref_and_mask = REG_SET_FIELD(ref_and_mask, GPU_HDP_FLUSH_DONE, SDMA1, 1); @@ -300,6 +300,13 @@ static void sdma_v2_4_ring_emit_hdp_flush(struct amdgpu_ring *ring) SDMA_PKT_POLL_REGMEM_DW5_INTERVAL(10)); /* retry count, poll interval */ } +static void sdma_v2_4_ring_emit_hdp_invalidate(struct amdgpu_ring *ring) +{ + amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_SRBM_WRITE) | + SDMA_PKT_SRBM_WRITE_HEADER_BYTE_EN(0xf)); + amdgpu_ring_write(ring, mmHDP_DEBUG0); + amdgpu_ring_write(ring, 1); +} /** * sdma_v2_4_ring_emit_fence - emit a fence on the DMA ring * @@ -335,31 +342,6 @@ static void sdma_v2_4_ring_emit_fence(struct amdgpu_ring *ring, u64 addr, u64 se } /** - * sdma_v2_4_ring_emit_semaphore - emit a semaphore on the dma ring - * - * @ring: amdgpu_ring structure holding ring information - * @semaphore: amdgpu semaphore object - * @emit_wait: wait or signal semaphore - * - * Add a DMA semaphore packet to the ring wait on or signal - * other rings (VI). - */ -static bool sdma_v2_4_ring_emit_semaphore(struct amdgpu_ring *ring, - struct amdgpu_semaphore *semaphore, - bool emit_wait) -{ - u64 addr = semaphore->gpu_addr; - u32 sig = emit_wait ? 0 : 1; - - amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_SEM) | - SDMA_PKT_SEMAPHORE_HEADER_SIGNAL(sig)); - amdgpu_ring_write(ring, lower_32_bits(addr) & 0xfffffff8); - amdgpu_ring_write(ring, upper_32_bits(addr)); - - return true; -} - -/** * sdma_v2_4_gfx_stop - stop the gfx async dma engines * * @adev: amdgpu_device pointer @@ -368,8 +350,8 @@ static bool sdma_v2_4_ring_emit_semaphore(struct amdgpu_ring *ring, */ static void sdma_v2_4_gfx_stop(struct amdgpu_device *adev) { - struct amdgpu_ring *sdma0 = &adev->sdma[0].ring; - struct amdgpu_ring *sdma1 = &adev->sdma[1].ring; + struct amdgpu_ring *sdma0 = &adev->sdma.instance[0].ring; + struct amdgpu_ring *sdma1 = &adev->sdma.instance[1].ring; u32 rb_cntl, ib_cntl; int i; @@ -377,7 +359,7 @@ static void sdma_v2_4_gfx_stop(struct amdgpu_device *adev) (adev->mman.buffer_funcs_ring == sdma1)) amdgpu_ttm_set_active_vram_size(adev, adev->mc.visible_vram_size); - for (i = 0; i < SDMA_MAX_INSTANCE; i++) { + for (i = 0; i < adev->sdma.num_instances; i++) { rb_cntl = RREG32(mmSDMA0_GFX_RB_CNTL + sdma_offsets[i]); rb_cntl = REG_SET_FIELD(rb_cntl, SDMA0_GFX_RB_CNTL, RB_ENABLE, 0); WREG32(mmSDMA0_GFX_RB_CNTL + sdma_offsets[i], rb_cntl); @@ -419,7 +401,7 @@ static void sdma_v2_4_enable(struct amdgpu_device *adev, bool enable) sdma_v2_4_rlc_stop(adev); } - for (i = 0; i < SDMA_MAX_INSTANCE; i++) { + for (i = 0; i < adev->sdma.num_instances; i++) { f32_cntl = RREG32(mmSDMA0_F32_CNTL + sdma_offsets[i]); if (enable) f32_cntl = REG_SET_FIELD(f32_cntl, SDMA0_F32_CNTL, HALT, 0); @@ -445,8 +427,8 @@ static int sdma_v2_4_gfx_resume(struct amdgpu_device *adev) u32 wb_offset; int i, j, r; - for (i = 0; i < SDMA_MAX_INSTANCE; i++) { - ring = &adev->sdma[i].ring; + for (i = 0; i < adev->sdma.num_instances; i++) { + ring = &adev->sdma.instance[i].ring; wb_offset = (ring->rptr_offs * 4); mutex_lock(&adev->srbm_mutex); @@ -459,6 +441,9 @@ static int sdma_v2_4_gfx_resume(struct amdgpu_device *adev) vi_srbm_select(adev, 0, 0, 0, 0); mutex_unlock(&adev->srbm_mutex); + WREG32(mmSDMA0_TILING_CONFIG + sdma_offsets[i], + adev->gfx.config.gb_addr_config & 0x70); + WREG32(mmSDMA0_SEM_WAIT_FAIL_TIMER_CNTL + sdma_offsets[i], 0); /* Set ring buffer size in dwords */ @@ -545,29 +530,23 @@ static int sdma_v2_4_load_microcode(struct amdgpu_device *adev) const __le32 *fw_data; u32 fw_size; int i, j; - bool smc_loads_fw = false; /* XXX fix me */ - - if (!adev->sdma[0].fw || !adev->sdma[1].fw) - return -EINVAL; /* halt the MEs */ sdma_v2_4_enable(adev, false); - if (smc_loads_fw) { - /* XXX query SMC for fw load complete */ - } else { - for (i = 0; i < SDMA_MAX_INSTANCE; i++) { - hdr = (const struct sdma_firmware_header_v1_0 *)adev->sdma[i].fw->data; - amdgpu_ucode_print_sdma_hdr(&hdr->header); - fw_size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4; - fw_data = (const __le32 *) - (adev->sdma[i].fw->data + - le32_to_cpu(hdr->header.ucode_array_offset_bytes)); - WREG32(mmSDMA0_UCODE_ADDR + sdma_offsets[i], 0); - for (j = 0; j < fw_size; j++) - WREG32(mmSDMA0_UCODE_DATA + sdma_offsets[i], le32_to_cpup(fw_data++)); - WREG32(mmSDMA0_UCODE_ADDR + sdma_offsets[i], adev->sdma[i].fw_version); - } + for (i = 0; i < adev->sdma.num_instances; i++) { + if (!adev->sdma.instance[i].fw) + return -EINVAL; + hdr = (const struct sdma_firmware_header_v1_0 *)adev->sdma.instance[i].fw->data; + amdgpu_ucode_print_sdma_hdr(&hdr->header); + fw_size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4; + fw_data = (const __le32 *) + (adev->sdma.instance[i].fw->data + + le32_to_cpu(hdr->header.ucode_array_offset_bytes)); + WREG32(mmSDMA0_UCODE_ADDR + sdma_offsets[i], 0); + for (j = 0; j < fw_size; j++) + WREG32(mmSDMA0_UCODE_DATA + sdma_offsets[i], le32_to_cpup(fw_data++)); + WREG32(mmSDMA0_UCODE_ADDR + sdma_offsets[i], adev->sdma.instance[i].fw_version); } return 0; @@ -642,7 +621,7 @@ static int sdma_v2_4_ring_test_ring(struct amdgpu_ring *ring) tmp = 0xCAFEDEAD; adev->wb.wb[index] = cpu_to_le32(tmp); - r = amdgpu_ring_lock(ring, 5); + r = amdgpu_ring_alloc(ring, 5); if (r) { DRM_ERROR("amdgpu: dma failed to lock ring %d (%d).\n", ring->idx, r); amdgpu_wb_free(adev, index); @@ -655,7 +634,7 @@ static int sdma_v2_4_ring_test_ring(struct amdgpu_ring *ring) amdgpu_ring_write(ring, upper_32_bits(gpu_addr)); amdgpu_ring_write(ring, SDMA_PKT_WRITE_UNTILED_DW_3_COUNT(1)); amdgpu_ring_write(ring, 0xDEADBEEF); - amdgpu_ring_unlock_commit(ring); + amdgpu_ring_commit(ring); for (i = 0; i < adev->usec_timeout; i++) { tmp = le32_to_cpu(adev->wb.wb[index]); @@ -705,7 +684,7 @@ static int sdma_v2_4_ring_test_ib(struct amdgpu_ring *ring) tmp = 0xCAFEDEAD; adev->wb.wb[index] = cpu_to_le32(tmp); memset(&ib, 0, sizeof(ib)); - r = amdgpu_ib_get(ring, NULL, 256, &ib); + r = amdgpu_ib_get(adev, NULL, 256, &ib); if (r) { DRM_ERROR("amdgpu: failed to get ib (%d).\n", r); goto err0; @@ -722,9 +701,7 @@ static int sdma_v2_4_ring_test_ib(struct amdgpu_ring *ring) ib.ptr[7] = SDMA_PKT_HEADER_OP(SDMA_OP_NOP); ib.length_dw = 8; - r = amdgpu_sched_ib_submit_kernel_helper(adev, ring, &ib, 1, NULL, - AMDGPU_FENCE_OWNER_UNDEFINED, - &f); + r = amdgpu_ib_schedule(ring, 1, &ib, NULL, &f); if (r) goto err1; @@ -750,7 +727,8 @@ static int sdma_v2_4_ring_test_ib(struct amdgpu_ring *ring) err1: fence_put(f); - amdgpu_ib_free(adev, &ib); + amdgpu_ib_free(adev, &ib, NULL); + fence_put(f); err0: amdgpu_wb_free(adev, index); return r; @@ -803,7 +781,7 @@ static void sdma_v2_4_vm_copy_pte(struct amdgpu_ib *ib, * Update PTEs by writing them manually using sDMA (CIK). */ static void sdma_v2_4_vm_write_pte(struct amdgpu_ib *ib, - uint64_t pe, + const dma_addr_t *pages_addr, uint64_t pe, uint64_t addr, unsigned count, uint32_t incr, uint32_t flags) { @@ -822,14 +800,7 @@ static void sdma_v2_4_vm_write_pte(struct amdgpu_ib *ib, ib->ptr[ib->length_dw++] = upper_32_bits(pe); ib->ptr[ib->length_dw++] = ndw; for (; ndw > 0; ndw -= 2, --count, pe += 8) { - if (flags & AMDGPU_PTE_SYSTEM) { - value = amdgpu_vm_map_gart(ib->ring->adev, addr); - value &= 0xFFFFFFFFFFFFF000ULL; - } else if (flags & AMDGPU_PTE_VALID) { - value = addr; - } else { - value = 0; - } + value = amdgpu_vm_map_gart(pages_addr, addr); addr += incr; value |= flags; ib->ptr[ib->length_dw++] = value; @@ -887,14 +858,14 @@ static void sdma_v2_4_vm_set_pte_pde(struct amdgpu_ib *ib, } /** - * sdma_v2_4_vm_pad_ib - pad the IB to the required number of dw + * sdma_v2_4_ring_pad_ib - pad the IB to the required number of dw * * @ib: indirect buffer to fill with padding * */ -static void sdma_v2_4_vm_pad_ib(struct amdgpu_ib *ib) +static void sdma_v2_4_ring_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib) { - struct amdgpu_sdma *sdma = amdgpu_get_sdma_instance(ib->ring); + struct amdgpu_sdma_instance *sdma = amdgpu_get_sdma_instance(ring); u32 pad_count; int i; @@ -910,6 +881,31 @@ static void sdma_v2_4_vm_pad_ib(struct amdgpu_ib *ib) } /** + * sdma_v2_4_ring_emit_pipeline_sync - sync the pipeline + * + * @ring: amdgpu_ring pointer + * + * Make sure all previous operations are completed (CIK). + */ +static void sdma_v2_4_ring_emit_pipeline_sync(struct amdgpu_ring *ring) +{ + uint32_t seq = ring->fence_drv.sync_seq; + uint64_t addr = ring->fence_drv.gpu_addr; + + /* wait for idle */ + amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_POLL_REGMEM) | + SDMA_PKT_POLL_REGMEM_HEADER_HDP_FLUSH(0) | + SDMA_PKT_POLL_REGMEM_HEADER_FUNC(3) | /* equal */ + SDMA_PKT_POLL_REGMEM_HEADER_MEM_POLL(1)); + amdgpu_ring_write(ring, addr & 0xfffffffc); + amdgpu_ring_write(ring, upper_32_bits(addr) & 0xffffffff); + amdgpu_ring_write(ring, seq); /* reference */ + amdgpu_ring_write(ring, 0xfffffff); /* mask */ + amdgpu_ring_write(ring, SDMA_PKT_POLL_REGMEM_DW5_RETRY_COUNT(0xfff) | + SDMA_PKT_POLL_REGMEM_DW5_INTERVAL(4)); /* retry count, poll interval */ +} + +/** * sdma_v2_4_ring_emit_vm_flush - cik vm flush using sDMA * * @ring: amdgpu_ring pointer @@ -952,6 +948,8 @@ static int sdma_v2_4_early_init(void *handle) { struct amdgpu_device *adev = (struct amdgpu_device *)handle; + adev->sdma.num_instances = SDMA_MAX_INSTANCE; + sdma_v2_4_set_ring_funcs(adev); sdma_v2_4_set_buffer_funcs(adev); sdma_v2_4_set_vm_pte_funcs(adev); @@ -963,21 +961,21 @@ static int sdma_v2_4_early_init(void *handle) static int sdma_v2_4_sw_init(void *handle) { struct amdgpu_ring *ring; - int r; + int r, i; struct amdgpu_device *adev = (struct amdgpu_device *)handle; /* SDMA trap event */ - r = amdgpu_irq_add_id(adev, 224, &adev->sdma_trap_irq); + r = amdgpu_irq_add_id(adev, 224, &adev->sdma.trap_irq); if (r) return r; /* SDMA Privileged inst */ - r = amdgpu_irq_add_id(adev, 241, &adev->sdma_illegal_inst_irq); + r = amdgpu_irq_add_id(adev, 241, &adev->sdma.illegal_inst_irq); if (r) return r; /* SDMA Privileged inst */ - r = amdgpu_irq_add_id(adev, 247, &adev->sdma_illegal_inst_irq); + r = amdgpu_irq_add_id(adev, 247, &adev->sdma.illegal_inst_irq); if (r) return r; @@ -987,31 +985,20 @@ static int sdma_v2_4_sw_init(void *handle) return r; } - ring = &adev->sdma[0].ring; - ring->ring_obj = NULL; - ring->use_doorbell = false; - - ring = &adev->sdma[1].ring; - ring->ring_obj = NULL; - ring->use_doorbell = false; - - ring = &adev->sdma[0].ring; - sprintf(ring->name, "sdma0"); - r = amdgpu_ring_init(adev, ring, 256 * 1024, - SDMA_PKT_NOP_HEADER_OP(SDMA_OP_NOP), 0xf, - &adev->sdma_trap_irq, AMDGPU_SDMA_IRQ_TRAP0, - AMDGPU_RING_TYPE_SDMA); - if (r) - return r; - - ring = &adev->sdma[1].ring; - sprintf(ring->name, "sdma1"); - r = amdgpu_ring_init(adev, ring, 256 * 1024, - SDMA_PKT_NOP_HEADER_OP(SDMA_OP_NOP), 0xf, - &adev->sdma_trap_irq, AMDGPU_SDMA_IRQ_TRAP1, - AMDGPU_RING_TYPE_SDMA); - if (r) - return r; + for (i = 0; i < adev->sdma.num_instances; i++) { + ring = &adev->sdma.instance[i].ring; + ring->ring_obj = NULL; + ring->use_doorbell = false; + sprintf(ring->name, "sdma%d", i); + r = amdgpu_ring_init(adev, ring, 256 * 1024, + SDMA_PKT_NOP_HEADER_OP(SDMA_OP_NOP), 0xf, + &adev->sdma.trap_irq, + (i == 0) ? + AMDGPU_SDMA_IRQ_TRAP0 : AMDGPU_SDMA_IRQ_TRAP1, + AMDGPU_RING_TYPE_SDMA); + if (r) + return r; + } return r; } @@ -1019,9 +1006,10 @@ static int sdma_v2_4_sw_init(void *handle) static int sdma_v2_4_sw_fini(void *handle) { struct amdgpu_device *adev = (struct amdgpu_device *)handle; + int i; - amdgpu_ring_fini(&adev->sdma[0].ring); - amdgpu_ring_fini(&adev->sdma[1].ring); + for (i = 0; i < adev->sdma.num_instances; i++) + amdgpu_ring_fini(&adev->sdma.instance[i].ring); return 0; } @@ -1100,7 +1088,7 @@ static void sdma_v2_4_print_status(void *handle) dev_info(adev->dev, "VI SDMA registers\n"); dev_info(adev->dev, " SRBM_STATUS2=0x%08X\n", RREG32(mmSRBM_STATUS2)); - for (i = 0; i < SDMA_MAX_INSTANCE; i++) { + for (i = 0; i < adev->sdma.num_instances; i++) { dev_info(adev->dev, " SDMA%d_STATUS_REG=0x%08X\n", i, RREG32(mmSDMA0_STATUS_REG + sdma_offsets[i])); dev_info(adev->dev, " SDMA%d_F32_CNTL=0x%08X\n", @@ -1125,6 +1113,8 @@ static void sdma_v2_4_print_status(void *handle) i, RREG32(mmSDMA0_GFX_RB_BASE + sdma_offsets[i])); dev_info(adev->dev, " SDMA%d_GFX_RB_BASE_HI=0x%08X\n", i, RREG32(mmSDMA0_GFX_RB_BASE_HI + sdma_offsets[i])); + dev_info(adev->dev, " SDMA%d_TILING_CONFIG=0x%08X\n", + i, RREG32(mmSDMA0_TILING_CONFIG + sdma_offsets[i])); mutex_lock(&adev->srbm_mutex); for (j = 0; j < 16; j++) { vi_srbm_select(adev, 0, 0, 0, j); @@ -1243,7 +1233,7 @@ static int sdma_v2_4_process_trap_irq(struct amdgpu_device *adev, case 0: switch (queue_id) { case 0: - amdgpu_fence_process(&adev->sdma[0].ring); + amdgpu_fence_process(&adev->sdma.instance[0].ring); break; case 1: /* XXX compute */ @@ -1256,7 +1246,7 @@ static int sdma_v2_4_process_trap_irq(struct amdgpu_device *adev, case 1: switch (queue_id) { case 0: - amdgpu_fence_process(&adev->sdma[1].ring); + amdgpu_fence_process(&adev->sdma.instance[1].ring); break; case 1: /* XXX compute */ @@ -1309,24 +1299,6 @@ const struct amd_ip_funcs sdma_v2_4_ip_funcs = { .set_powergating_state = sdma_v2_4_set_powergating_state, }; -/** - * sdma_v2_4_ring_is_lockup - Check if the DMA engine is locked up - * - * @ring: amdgpu_ring structure holding ring information - * - * Check if the async DMA engine is locked up (VI). - * Returns true if the engine appears to be locked up, false if not. - */ -static bool sdma_v2_4_ring_is_lockup(struct amdgpu_ring *ring) -{ - - if (sdma_v2_4_is_idle(ring->adev)) { - amdgpu_ring_lockup_update(ring); - return false; - } - return amdgpu_ring_test_lockup(ring); -} - static const struct amdgpu_ring_funcs sdma_v2_4_ring_funcs = { .get_rptr = sdma_v2_4_ring_get_rptr, .get_wptr = sdma_v2_4_ring_get_wptr, @@ -1334,19 +1306,22 @@ static const struct amdgpu_ring_funcs sdma_v2_4_ring_funcs = { .parse_cs = NULL, .emit_ib = sdma_v2_4_ring_emit_ib, .emit_fence = sdma_v2_4_ring_emit_fence, - .emit_semaphore = sdma_v2_4_ring_emit_semaphore, + .emit_pipeline_sync = sdma_v2_4_ring_emit_pipeline_sync, .emit_vm_flush = sdma_v2_4_ring_emit_vm_flush, .emit_hdp_flush = sdma_v2_4_ring_emit_hdp_flush, + .emit_hdp_invalidate = sdma_v2_4_ring_emit_hdp_invalidate, .test_ring = sdma_v2_4_ring_test_ring, .test_ib = sdma_v2_4_ring_test_ib, - .is_lockup = sdma_v2_4_ring_is_lockup, .insert_nop = sdma_v2_4_ring_insert_nop, + .pad_ib = sdma_v2_4_ring_pad_ib, }; static void sdma_v2_4_set_ring_funcs(struct amdgpu_device *adev) { - adev->sdma[0].ring.funcs = &sdma_v2_4_ring_funcs; - adev->sdma[1].ring.funcs = &sdma_v2_4_ring_funcs; + int i; + + for (i = 0; i < adev->sdma.num_instances; i++) + adev->sdma.instance[i].ring.funcs = &sdma_v2_4_ring_funcs; } static const struct amdgpu_irq_src_funcs sdma_v2_4_trap_irq_funcs = { @@ -1360,9 +1335,9 @@ static const struct amdgpu_irq_src_funcs sdma_v2_4_illegal_inst_irq_funcs = { static void sdma_v2_4_set_irq_funcs(struct amdgpu_device *adev) { - adev->sdma_trap_irq.num_types = AMDGPU_SDMA_IRQ_LAST; - adev->sdma_trap_irq.funcs = &sdma_v2_4_trap_irq_funcs; - adev->sdma_illegal_inst_irq.funcs = &sdma_v2_4_illegal_inst_irq_funcs; + adev->sdma.trap_irq.num_types = AMDGPU_SDMA_IRQ_LAST; + adev->sdma.trap_irq.funcs = &sdma_v2_4_trap_irq_funcs; + adev->sdma.illegal_inst_irq.funcs = &sdma_v2_4_illegal_inst_irq_funcs; } /** @@ -1428,7 +1403,7 @@ static void sdma_v2_4_set_buffer_funcs(struct amdgpu_device *adev) { if (adev->mman.buffer_funcs == NULL) { adev->mman.buffer_funcs = &sdma_v2_4_buffer_funcs; - adev->mman.buffer_funcs_ring = &adev->sdma[0].ring; + adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].ring; } } @@ -1436,14 +1411,18 @@ static const struct amdgpu_vm_pte_funcs sdma_v2_4_vm_pte_funcs = { .copy_pte = sdma_v2_4_vm_copy_pte, .write_pte = sdma_v2_4_vm_write_pte, .set_pte_pde = sdma_v2_4_vm_set_pte_pde, - .pad_ib = sdma_v2_4_vm_pad_ib, }; static void sdma_v2_4_set_vm_pte_funcs(struct amdgpu_device *adev) { + unsigned i; + if (adev->vm_manager.vm_pte_funcs == NULL) { adev->vm_manager.vm_pte_funcs = &sdma_v2_4_vm_pte_funcs; - adev->vm_manager.vm_pte_funcs_ring = &adev->sdma[0].ring; - adev->vm_manager.vm_pte_funcs_ring->is_pte_ring = true; + for (i = 0; i < adev->sdma.num_instances; i++) + adev->vm_manager.vm_pte_rings[i] = + &adev->sdma.instance[i].ring; + + adev->vm_manager.vm_pte_num_rings = adev->sdma.num_instances; } } diff --git a/drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c b/drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c index 9bfe92df15f7..8c8ca98dd129 100644 --- a/drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c +++ b/drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c @@ -55,6 +55,7 @@ MODULE_FIRMWARE("amdgpu/carrizo_sdma.bin"); MODULE_FIRMWARE("amdgpu/carrizo_sdma1.bin"); MODULE_FIRMWARE("amdgpu/fiji_sdma.bin"); MODULE_FIRMWARE("amdgpu/fiji_sdma1.bin"); +MODULE_FIRMWARE("amdgpu/stoney_sdma.bin"); static const u32 sdma_offsets[SDMA_MAX_INSTANCE] = { @@ -122,6 +123,19 @@ static const u32 cz_mgcg_cgcg_init[] = mmSDMA1_CLK_CTRL, 0xff000ff0, 0x00000100 }; +static const u32 stoney_golden_settings_a11[] = +{ + mmSDMA0_GFX_IB_CNTL, 0x00000100, 0x00000100, + mmSDMA0_POWER_CNTL, 0x00000800, 0x0003c800, + mmSDMA0_RLC0_IB_CNTL, 0x00000100, 0x00000100, + mmSDMA0_RLC1_IB_CNTL, 0x00000100, 0x00000100, +}; + +static const u32 stoney_mgcg_cgcg_init[] = +{ + mmSDMA0_CLK_CTRL, 0xffffffff, 0x00000100, +}; + /* * sDMA - System DMA * Starting with CIK, the GPU has new asynchronous @@ -166,6 +180,14 @@ static void sdma_v3_0_init_golden_registers(struct amdgpu_device *adev) cz_golden_settings_a11, (const u32)ARRAY_SIZE(cz_golden_settings_a11)); break; + case CHIP_STONEY: + amdgpu_program_register_sequence(adev, + stoney_mgcg_cgcg_init, + (const u32)ARRAY_SIZE(stoney_mgcg_cgcg_init)); + amdgpu_program_register_sequence(adev, + stoney_golden_settings_a11, + (const u32)ARRAY_SIZE(stoney_golden_settings_a11)); + break; default: break; } @@ -184,7 +206,7 @@ static int sdma_v3_0_init_microcode(struct amdgpu_device *adev) { const char *chip_name; char fw_name[30]; - int err, i; + int err = 0, i; struct amdgpu_firmware_info *info = NULL; const struct common_firmware_header *header = NULL; const struct sdma_firmware_header_v1_0 *hdr; @@ -201,30 +223,33 @@ static int sdma_v3_0_init_microcode(struct amdgpu_device *adev) case CHIP_CARRIZO: chip_name = "carrizo"; break; + case CHIP_STONEY: + chip_name = "stoney"; + break; default: BUG(); } - for (i = 0; i < SDMA_MAX_INSTANCE; i++) { + for (i = 0; i < adev->sdma.num_instances; i++) { if (i == 0) snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_sdma.bin", chip_name); else snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_sdma1.bin", chip_name); - err = request_firmware(&adev->sdma[i].fw, fw_name, adev->dev); + err = request_firmware(&adev->sdma.instance[i].fw, fw_name, adev->dev); if (err) goto out; - err = amdgpu_ucode_validate(adev->sdma[i].fw); + err = amdgpu_ucode_validate(adev->sdma.instance[i].fw); if (err) goto out; - hdr = (const struct sdma_firmware_header_v1_0 *)adev->sdma[i].fw->data; - adev->sdma[i].fw_version = le32_to_cpu(hdr->header.ucode_version); - adev->sdma[i].feature_version = le32_to_cpu(hdr->ucode_feature_version); - if (adev->sdma[i].feature_version >= 20) - adev->sdma[i].burst_nop = true; + hdr = (const struct sdma_firmware_header_v1_0 *)adev->sdma.instance[i].fw->data; + adev->sdma.instance[i].fw_version = le32_to_cpu(hdr->header.ucode_version); + adev->sdma.instance[i].feature_version = le32_to_cpu(hdr->ucode_feature_version); + if (adev->sdma.instance[i].feature_version >= 20) + adev->sdma.instance[i].burst_nop = true; if (adev->firmware.smu_load) { info = &adev->firmware.ucode[AMDGPU_UCODE_ID_SDMA0 + i]; info->ucode_id = AMDGPU_UCODE_ID_SDMA0 + i; - info->fw = adev->sdma[i].fw; + info->fw = adev->sdma.instance[i].fw; header = (const struct common_firmware_header *)info->fw->data; adev->firmware.fw_size += ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE); @@ -235,9 +260,9 @@ out: printk(KERN_ERR "sdma_v3_0: Failed to load firmware \"%s\"\n", fw_name); - for (i = 0; i < SDMA_MAX_INSTANCE; i++) { - release_firmware(adev->sdma[i].fw); - adev->sdma[i].fw = NULL; + for (i = 0; i < adev->sdma.num_instances; i++) { + release_firmware(adev->sdma.instance[i].fw); + adev->sdma.instance[i].fw = NULL; } } return err; @@ -276,7 +301,7 @@ static uint32_t sdma_v3_0_ring_get_wptr(struct amdgpu_ring *ring) /* XXX check if swapping is necessary on BE */ wptr = ring->adev->wb.wb[ring->wptr_offs] >> 2; } else { - int me = (ring == &ring->adev->sdma[0].ring) ? 0 : 1; + int me = (ring == &ring->adev->sdma.instance[0].ring) ? 0 : 1; wptr = RREG32(mmSDMA0_GFX_RB_WPTR + sdma_offsets[me]) >> 2; } @@ -300,7 +325,7 @@ static void sdma_v3_0_ring_set_wptr(struct amdgpu_ring *ring) adev->wb.wb[ring->wptr_offs] = ring->wptr << 2; WDOORBELL32(ring->doorbell_index, ring->wptr << 2); } else { - int me = (ring == &ring->adev->sdma[0].ring) ? 0 : 1; + int me = (ring == &ring->adev->sdma.instance[0].ring) ? 0 : 1; WREG32(mmSDMA0_GFX_RB_WPTR + sdma_offsets[me], ring->wptr << 2); } @@ -308,7 +333,7 @@ static void sdma_v3_0_ring_set_wptr(struct amdgpu_ring *ring) static void sdma_v3_0_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count) { - struct amdgpu_sdma *sdma = amdgpu_get_sdma_instance(ring); + struct amdgpu_sdma_instance *sdma = amdgpu_get_sdma_instance(ring); int i; for (i = 0; i < count; i++) @@ -330,7 +355,7 @@ static void sdma_v3_0_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count) static void sdma_v3_0_ring_emit_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib) { - u32 vmid = (ib->vm ? ib->vm->ids[ring->idx].id : 0) & 0xf; + u32 vmid = ib->vm_id & 0xf; u32 next_rptr = ring->wptr + 5; while ((next_rptr & 7) != 2) @@ -369,7 +394,7 @@ static void sdma_v3_0_ring_emit_hdp_flush(struct amdgpu_ring *ring) { u32 ref_and_mask = 0; - if (ring == &ring->adev->sdma[0].ring) + if (ring == &ring->adev->sdma.instance[0].ring) ref_and_mask = REG_SET_FIELD(ref_and_mask, GPU_HDP_FLUSH_DONE, SDMA0, 1); else ref_and_mask = REG_SET_FIELD(ref_and_mask, GPU_HDP_FLUSH_DONE, SDMA1, 1); @@ -385,6 +410,14 @@ static void sdma_v3_0_ring_emit_hdp_flush(struct amdgpu_ring *ring) SDMA_PKT_POLL_REGMEM_DW5_INTERVAL(10)); /* retry count, poll interval */ } +static void sdma_v3_0_ring_emit_hdp_invalidate(struct amdgpu_ring *ring) +{ + amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_SRBM_WRITE) | + SDMA_PKT_SRBM_WRITE_HEADER_BYTE_EN(0xf)); + amdgpu_ring_write(ring, mmHDP_DEBUG0); + amdgpu_ring_write(ring, 1); +} + /** * sdma_v3_0_ring_emit_fence - emit a fence on the DMA ring * @@ -419,32 +452,6 @@ static void sdma_v3_0_ring_emit_fence(struct amdgpu_ring *ring, u64 addr, u64 se amdgpu_ring_write(ring, SDMA_PKT_TRAP_INT_CONTEXT_INT_CONTEXT(0)); } - -/** - * sdma_v3_0_ring_emit_semaphore - emit a semaphore on the dma ring - * - * @ring: amdgpu_ring structure holding ring information - * @semaphore: amdgpu semaphore object - * @emit_wait: wait or signal semaphore - * - * Add a DMA semaphore packet to the ring wait on or signal - * other rings (VI). - */ -static bool sdma_v3_0_ring_emit_semaphore(struct amdgpu_ring *ring, - struct amdgpu_semaphore *semaphore, - bool emit_wait) -{ - u64 addr = semaphore->gpu_addr; - u32 sig = emit_wait ? 0 : 1; - - amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_SEM) | - SDMA_PKT_SEMAPHORE_HEADER_SIGNAL(sig)); - amdgpu_ring_write(ring, lower_32_bits(addr) & 0xfffffff8); - amdgpu_ring_write(ring, upper_32_bits(addr)); - - return true; -} - /** * sdma_v3_0_gfx_stop - stop the gfx async dma engines * @@ -454,8 +461,8 @@ static bool sdma_v3_0_ring_emit_semaphore(struct amdgpu_ring *ring, */ static void sdma_v3_0_gfx_stop(struct amdgpu_device *adev) { - struct amdgpu_ring *sdma0 = &adev->sdma[0].ring; - struct amdgpu_ring *sdma1 = &adev->sdma[1].ring; + struct amdgpu_ring *sdma0 = &adev->sdma.instance[0].ring; + struct amdgpu_ring *sdma1 = &adev->sdma.instance[1].ring; u32 rb_cntl, ib_cntl; int i; @@ -463,7 +470,7 @@ static void sdma_v3_0_gfx_stop(struct amdgpu_device *adev) (adev->mman.buffer_funcs_ring == sdma1)) amdgpu_ttm_set_active_vram_size(adev, adev->mc.visible_vram_size); - for (i = 0; i < SDMA_MAX_INSTANCE; i++) { + for (i = 0; i < adev->sdma.num_instances; i++) { rb_cntl = RREG32(mmSDMA0_GFX_RB_CNTL + sdma_offsets[i]); rb_cntl = REG_SET_FIELD(rb_cntl, SDMA0_GFX_RB_CNTL, RB_ENABLE, 0); WREG32(mmSDMA0_GFX_RB_CNTL + sdma_offsets[i], rb_cntl); @@ -500,7 +507,7 @@ static void sdma_v3_0_ctx_switch_enable(struct amdgpu_device *adev, bool enable) u32 f32_cntl; int i; - for (i = 0; i < SDMA_MAX_INSTANCE; i++) { + for (i = 0; i < adev->sdma.num_instances; i++) { f32_cntl = RREG32(mmSDMA0_CNTL + sdma_offsets[i]); if (enable) f32_cntl = REG_SET_FIELD(f32_cntl, SDMA0_CNTL, @@ -530,7 +537,7 @@ static void sdma_v3_0_enable(struct amdgpu_device *adev, bool enable) sdma_v3_0_rlc_stop(adev); } - for (i = 0; i < SDMA_MAX_INSTANCE; i++) { + for (i = 0; i < adev->sdma.num_instances; i++) { f32_cntl = RREG32(mmSDMA0_F32_CNTL + sdma_offsets[i]); if (enable) f32_cntl = REG_SET_FIELD(f32_cntl, SDMA0_F32_CNTL, HALT, 0); @@ -557,8 +564,8 @@ static int sdma_v3_0_gfx_resume(struct amdgpu_device *adev) u32 doorbell; int i, j, r; - for (i = 0; i < SDMA_MAX_INSTANCE; i++) { - ring = &adev->sdma[i].ring; + for (i = 0; i < adev->sdma.num_instances; i++) { + ring = &adev->sdma.instance[i].ring; wb_offset = (ring->rptr_offs * 4); mutex_lock(&adev->srbm_mutex); @@ -571,6 +578,9 @@ static int sdma_v3_0_gfx_resume(struct amdgpu_device *adev) vi_srbm_select(adev, 0, 0, 0, 0); mutex_unlock(&adev->srbm_mutex); + WREG32(mmSDMA0_TILING_CONFIG + sdma_offsets[i], + adev->gfx.config.gb_addr_config & 0x70); + WREG32(mmSDMA0_SEM_WAIT_FAIL_TIMER_CNTL + sdma_offsets[i], 0); /* Set ring buffer size in dwords */ @@ -669,23 +679,22 @@ static int sdma_v3_0_load_microcode(struct amdgpu_device *adev) u32 fw_size; int i, j; - if (!adev->sdma[0].fw || !adev->sdma[1].fw) - return -EINVAL; - /* halt the MEs */ sdma_v3_0_enable(adev, false); - for (i = 0; i < SDMA_MAX_INSTANCE; i++) { - hdr = (const struct sdma_firmware_header_v1_0 *)adev->sdma[i].fw->data; + for (i = 0; i < adev->sdma.num_instances; i++) { + if (!adev->sdma.instance[i].fw) + return -EINVAL; + hdr = (const struct sdma_firmware_header_v1_0 *)adev->sdma.instance[i].fw->data; amdgpu_ucode_print_sdma_hdr(&hdr->header); fw_size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4; fw_data = (const __le32 *) - (adev->sdma[i].fw->data + + (adev->sdma.instance[i].fw->data + le32_to_cpu(hdr->header.ucode_array_offset_bytes)); WREG32(mmSDMA0_UCODE_ADDR + sdma_offsets[i], 0); for (j = 0; j < fw_size; j++) WREG32(mmSDMA0_UCODE_DATA + sdma_offsets[i], le32_to_cpup(fw_data++)); - WREG32(mmSDMA0_UCODE_ADDR + sdma_offsets[i], adev->sdma[i].fw_version); + WREG32(mmSDMA0_UCODE_ADDR + sdma_offsets[i], adev->sdma.instance[i].fw_version); } return 0; @@ -701,21 +710,23 @@ static int sdma_v3_0_load_microcode(struct amdgpu_device *adev) */ static int sdma_v3_0_start(struct amdgpu_device *adev) { - int r; + int r, i; - if (!adev->firmware.smu_load) { - r = sdma_v3_0_load_microcode(adev); - if (r) - return r; - } else { - r = adev->smu.smumgr_funcs->check_fw_load_finish(adev, - AMDGPU_UCODE_ID_SDMA0); - if (r) - return -EINVAL; - r = adev->smu.smumgr_funcs->check_fw_load_finish(adev, - AMDGPU_UCODE_ID_SDMA1); - if (r) - return -EINVAL; + if (!adev->pp_enabled) { + if (!adev->firmware.smu_load) { + r = sdma_v3_0_load_microcode(adev); + if (r) + return r; + } else { + for (i = 0; i < adev->sdma.num_instances; i++) { + r = adev->smu.smumgr_funcs->check_fw_load_finish(adev, + (i == 0) ? + AMDGPU_UCODE_ID_SDMA0 : + AMDGPU_UCODE_ID_SDMA1); + if (r) + return -EINVAL; + } + } } /* unhalt the MEs */ @@ -762,7 +773,7 @@ static int sdma_v3_0_ring_test_ring(struct amdgpu_ring *ring) tmp = 0xCAFEDEAD; adev->wb.wb[index] = cpu_to_le32(tmp); - r = amdgpu_ring_lock(ring, 5); + r = amdgpu_ring_alloc(ring, 5); if (r) { DRM_ERROR("amdgpu: dma failed to lock ring %d (%d).\n", ring->idx, r); amdgpu_wb_free(adev, index); @@ -775,7 +786,7 @@ static int sdma_v3_0_ring_test_ring(struct amdgpu_ring *ring) amdgpu_ring_write(ring, upper_32_bits(gpu_addr)); amdgpu_ring_write(ring, SDMA_PKT_WRITE_UNTILED_DW_3_COUNT(1)); amdgpu_ring_write(ring, 0xDEADBEEF); - amdgpu_ring_unlock_commit(ring); + amdgpu_ring_commit(ring); for (i = 0; i < adev->usec_timeout; i++) { tmp = le32_to_cpu(adev->wb.wb[index]); @@ -825,7 +836,7 @@ static int sdma_v3_0_ring_test_ib(struct amdgpu_ring *ring) tmp = 0xCAFEDEAD; adev->wb.wb[index] = cpu_to_le32(tmp); memset(&ib, 0, sizeof(ib)); - r = amdgpu_ib_get(ring, NULL, 256, &ib); + r = amdgpu_ib_get(adev, NULL, 256, &ib); if (r) { DRM_ERROR("amdgpu: failed to get ib (%d).\n", r); goto err0; @@ -842,9 +853,7 @@ static int sdma_v3_0_ring_test_ib(struct amdgpu_ring *ring) ib.ptr[7] = SDMA_PKT_NOP_HEADER_OP(SDMA_OP_NOP); ib.length_dw = 8; - r = amdgpu_sched_ib_submit_kernel_helper(adev, ring, &ib, 1, NULL, - AMDGPU_FENCE_OWNER_UNDEFINED, - &f); + r = amdgpu_ib_schedule(ring, 1, &ib, NULL, &f); if (r) goto err1; @@ -869,7 +878,8 @@ static int sdma_v3_0_ring_test_ib(struct amdgpu_ring *ring) } err1: fence_put(f); - amdgpu_ib_free(adev, &ib); + amdgpu_ib_free(adev, &ib, NULL); + fence_put(f); err0: amdgpu_wb_free(adev, index); return r; @@ -922,7 +932,7 @@ static void sdma_v3_0_vm_copy_pte(struct amdgpu_ib *ib, * Update PTEs by writing them manually using sDMA (CIK). */ static void sdma_v3_0_vm_write_pte(struct amdgpu_ib *ib, - uint64_t pe, + const dma_addr_t *pages_addr, uint64_t pe, uint64_t addr, unsigned count, uint32_t incr, uint32_t flags) { @@ -941,14 +951,7 @@ static void sdma_v3_0_vm_write_pte(struct amdgpu_ib *ib, ib->ptr[ib->length_dw++] = upper_32_bits(pe); ib->ptr[ib->length_dw++] = ndw; for (; ndw > 0; ndw -= 2, --count, pe += 8) { - if (flags & AMDGPU_PTE_SYSTEM) { - value = amdgpu_vm_map_gart(ib->ring->adev, addr); - value &= 0xFFFFFFFFFFFFF000ULL; - } else if (flags & AMDGPU_PTE_VALID) { - value = addr; - } else { - value = 0; - } + value = amdgpu_vm_map_gart(pages_addr, addr); addr += incr; value |= flags; ib->ptr[ib->length_dw++] = value; @@ -1006,14 +1009,14 @@ static void sdma_v3_0_vm_set_pte_pde(struct amdgpu_ib *ib, } /** - * sdma_v3_0_vm_pad_ib - pad the IB to the required number of dw + * sdma_v3_0_ring_pad_ib - pad the IB to the required number of dw * * @ib: indirect buffer to fill with padding * */ -static void sdma_v3_0_vm_pad_ib(struct amdgpu_ib *ib) +static void sdma_v3_0_ring_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib) { - struct amdgpu_sdma *sdma = amdgpu_get_sdma_instance(ib->ring); + struct amdgpu_sdma_instance *sdma = amdgpu_get_sdma_instance(ring); u32 pad_count; int i; @@ -1029,6 +1032,31 @@ static void sdma_v3_0_vm_pad_ib(struct amdgpu_ib *ib) } /** + * sdma_v3_0_ring_emit_pipeline_sync - sync the pipeline + * + * @ring: amdgpu_ring pointer + * + * Make sure all previous operations are completed (CIK). + */ +static void sdma_v3_0_ring_emit_pipeline_sync(struct amdgpu_ring *ring) +{ + uint32_t seq = ring->fence_drv.sync_seq; + uint64_t addr = ring->fence_drv.gpu_addr; + + /* wait for idle */ + amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_POLL_REGMEM) | + SDMA_PKT_POLL_REGMEM_HEADER_HDP_FLUSH(0) | + SDMA_PKT_POLL_REGMEM_HEADER_FUNC(3) | /* equal */ + SDMA_PKT_POLL_REGMEM_HEADER_MEM_POLL(1)); + amdgpu_ring_write(ring, addr & 0xfffffffc); + amdgpu_ring_write(ring, upper_32_bits(addr) & 0xffffffff); + amdgpu_ring_write(ring, seq); /* reference */ + amdgpu_ring_write(ring, 0xfffffff); /* mask */ + amdgpu_ring_write(ring, SDMA_PKT_POLL_REGMEM_DW5_RETRY_COUNT(0xfff) | + SDMA_PKT_POLL_REGMEM_DW5_INTERVAL(4)); /* retry count, poll interval */ +} + +/** * sdma_v3_0_ring_emit_vm_flush - cik vm flush using sDMA * * @ring: amdgpu_ring pointer @@ -1071,6 +1099,15 @@ static int sdma_v3_0_early_init(void *handle) { struct amdgpu_device *adev = (struct amdgpu_device *)handle; + switch (adev->asic_type) { + case CHIP_STONEY: + adev->sdma.num_instances = 1; + break; + default: + adev->sdma.num_instances = SDMA_MAX_INSTANCE; + break; + } + sdma_v3_0_set_ring_funcs(adev); sdma_v3_0_set_buffer_funcs(adev); sdma_v3_0_set_vm_pte_funcs(adev); @@ -1082,21 +1119,21 @@ static int sdma_v3_0_early_init(void *handle) static int sdma_v3_0_sw_init(void *handle) { struct amdgpu_ring *ring; - int r; + int r, i; struct amdgpu_device *adev = (struct amdgpu_device *)handle; /* SDMA trap event */ - r = amdgpu_irq_add_id(adev, 224, &adev->sdma_trap_irq); + r = amdgpu_irq_add_id(adev, 224, &adev->sdma.trap_irq); if (r) return r; /* SDMA Privileged inst */ - r = amdgpu_irq_add_id(adev, 241, &adev->sdma_illegal_inst_irq); + r = amdgpu_irq_add_id(adev, 241, &adev->sdma.illegal_inst_irq); if (r) return r; /* SDMA Privileged inst */ - r = amdgpu_irq_add_id(adev, 247, &adev->sdma_illegal_inst_irq); + r = amdgpu_irq_add_id(adev, 247, &adev->sdma.illegal_inst_irq); if (r) return r; @@ -1106,33 +1143,23 @@ static int sdma_v3_0_sw_init(void *handle) return r; } - ring = &adev->sdma[0].ring; - ring->ring_obj = NULL; - ring->use_doorbell = true; - ring->doorbell_index = AMDGPU_DOORBELL_sDMA_ENGINE0; - - ring = &adev->sdma[1].ring; - ring->ring_obj = NULL; - ring->use_doorbell = true; - ring->doorbell_index = AMDGPU_DOORBELL_sDMA_ENGINE1; - - ring = &adev->sdma[0].ring; - sprintf(ring->name, "sdma0"); - r = amdgpu_ring_init(adev, ring, 256 * 1024, - SDMA_PKT_NOP_HEADER_OP(SDMA_OP_NOP), 0xf, - &adev->sdma_trap_irq, AMDGPU_SDMA_IRQ_TRAP0, - AMDGPU_RING_TYPE_SDMA); - if (r) - return r; - - ring = &adev->sdma[1].ring; - sprintf(ring->name, "sdma1"); - r = amdgpu_ring_init(adev, ring, 256 * 1024, - SDMA_PKT_NOP_HEADER_OP(SDMA_OP_NOP), 0xf, - &adev->sdma_trap_irq, AMDGPU_SDMA_IRQ_TRAP1, - AMDGPU_RING_TYPE_SDMA); - if (r) - return r; + for (i = 0; i < adev->sdma.num_instances; i++) { + ring = &adev->sdma.instance[i].ring; + ring->ring_obj = NULL; + ring->use_doorbell = true; + ring->doorbell_index = (i == 0) ? + AMDGPU_DOORBELL_sDMA_ENGINE0 : AMDGPU_DOORBELL_sDMA_ENGINE1; + + sprintf(ring->name, "sdma%d", i); + r = amdgpu_ring_init(adev, ring, 256 * 1024, + SDMA_PKT_NOP_HEADER_OP(SDMA_OP_NOP), 0xf, + &adev->sdma.trap_irq, + (i == 0) ? + AMDGPU_SDMA_IRQ_TRAP0 : AMDGPU_SDMA_IRQ_TRAP1, + AMDGPU_RING_TYPE_SDMA); + if (r) + return r; + } return r; } @@ -1140,9 +1167,10 @@ static int sdma_v3_0_sw_init(void *handle) static int sdma_v3_0_sw_fini(void *handle) { struct amdgpu_device *adev = (struct amdgpu_device *)handle; + int i; - amdgpu_ring_fini(&adev->sdma[0].ring); - amdgpu_ring_fini(&adev->sdma[1].ring); + for (i = 0; i < adev->sdma.num_instances; i++) + amdgpu_ring_fini(&adev->sdma.instance[i].ring); return 0; } @@ -1222,7 +1250,7 @@ static void sdma_v3_0_print_status(void *handle) dev_info(adev->dev, "VI SDMA registers\n"); dev_info(adev->dev, " SRBM_STATUS2=0x%08X\n", RREG32(mmSRBM_STATUS2)); - for (i = 0; i < SDMA_MAX_INSTANCE; i++) { + for (i = 0; i < adev->sdma.num_instances; i++) { dev_info(adev->dev, " SDMA%d_STATUS_REG=0x%08X\n", i, RREG32(mmSDMA0_STATUS_REG + sdma_offsets[i])); dev_info(adev->dev, " SDMA%d_F32_CNTL=0x%08X\n", @@ -1249,6 +1277,8 @@ static void sdma_v3_0_print_status(void *handle) i, RREG32(mmSDMA0_GFX_RB_BASE_HI + sdma_offsets[i])); dev_info(adev->dev, " SDMA%d_GFX_DOORBELL=0x%08X\n", i, RREG32(mmSDMA0_GFX_DOORBELL + sdma_offsets[i])); + dev_info(adev->dev, " SDMA%d_TILING_CONFIG=0x%08X\n", + i, RREG32(mmSDMA0_TILING_CONFIG + sdma_offsets[i])); mutex_lock(&adev->srbm_mutex); for (j = 0; j < 16; j++) { vi_srbm_select(adev, 0, 0, 0, j); @@ -1367,7 +1397,7 @@ static int sdma_v3_0_process_trap_irq(struct amdgpu_device *adev, case 0: switch (queue_id) { case 0: - amdgpu_fence_process(&adev->sdma[0].ring); + amdgpu_fence_process(&adev->sdma.instance[0].ring); break; case 1: /* XXX compute */ @@ -1380,7 +1410,7 @@ static int sdma_v3_0_process_trap_irq(struct amdgpu_device *adev, case 1: switch (queue_id) { case 0: - amdgpu_fence_process(&adev->sdma[1].ring); + amdgpu_fence_process(&adev->sdma.instance[1].ring); break; case 1: /* XXX compute */ @@ -1403,9 +1433,114 @@ static int sdma_v3_0_process_illegal_inst_irq(struct amdgpu_device *adev, return 0; } +static void fiji_update_sdma_medium_grain_clock_gating( + struct amdgpu_device *adev, + bool enable) +{ + uint32_t temp, data; + + if (enable) { + temp = data = RREG32(mmSDMA0_CLK_CTRL); + data &= ~(SDMA0_CLK_CTRL__SOFT_OVERRIDE7_MASK | + SDMA0_CLK_CTRL__SOFT_OVERRIDE6_MASK | + SDMA0_CLK_CTRL__SOFT_OVERRIDE5_MASK | + SDMA0_CLK_CTRL__SOFT_OVERRIDE4_MASK | + SDMA0_CLK_CTRL__SOFT_OVERRIDE3_MASK | + SDMA0_CLK_CTRL__SOFT_OVERRIDE2_MASK | + SDMA0_CLK_CTRL__SOFT_OVERRIDE1_MASK | + SDMA0_CLK_CTRL__SOFT_OVERRIDE0_MASK); + if (data != temp) + WREG32(mmSDMA0_CLK_CTRL, data); + + temp = data = RREG32(mmSDMA1_CLK_CTRL); + data &= ~(SDMA1_CLK_CTRL__SOFT_OVERRIDE7_MASK | + SDMA1_CLK_CTRL__SOFT_OVERRIDE6_MASK | + SDMA1_CLK_CTRL__SOFT_OVERRIDE5_MASK | + SDMA1_CLK_CTRL__SOFT_OVERRIDE4_MASK | + SDMA1_CLK_CTRL__SOFT_OVERRIDE3_MASK | + SDMA1_CLK_CTRL__SOFT_OVERRIDE2_MASK | + SDMA1_CLK_CTRL__SOFT_OVERRIDE1_MASK | + SDMA1_CLK_CTRL__SOFT_OVERRIDE0_MASK); + + if (data != temp) + WREG32(mmSDMA1_CLK_CTRL, data); + } else { + temp = data = RREG32(mmSDMA0_CLK_CTRL); + data |= SDMA0_CLK_CTRL__SOFT_OVERRIDE7_MASK | + SDMA0_CLK_CTRL__SOFT_OVERRIDE6_MASK | + SDMA0_CLK_CTRL__SOFT_OVERRIDE5_MASK | + SDMA0_CLK_CTRL__SOFT_OVERRIDE4_MASK | + SDMA0_CLK_CTRL__SOFT_OVERRIDE3_MASK | + SDMA0_CLK_CTRL__SOFT_OVERRIDE2_MASK | + SDMA0_CLK_CTRL__SOFT_OVERRIDE1_MASK | + SDMA0_CLK_CTRL__SOFT_OVERRIDE0_MASK; + + if (data != temp) + WREG32(mmSDMA0_CLK_CTRL, data); + + temp = data = RREG32(mmSDMA1_CLK_CTRL); + data |= SDMA1_CLK_CTRL__SOFT_OVERRIDE7_MASK | + SDMA1_CLK_CTRL__SOFT_OVERRIDE6_MASK | + SDMA1_CLK_CTRL__SOFT_OVERRIDE5_MASK | + SDMA1_CLK_CTRL__SOFT_OVERRIDE4_MASK | + SDMA1_CLK_CTRL__SOFT_OVERRIDE3_MASK | + SDMA1_CLK_CTRL__SOFT_OVERRIDE2_MASK | + SDMA1_CLK_CTRL__SOFT_OVERRIDE1_MASK | + SDMA1_CLK_CTRL__SOFT_OVERRIDE0_MASK; + + if (data != temp) + WREG32(mmSDMA1_CLK_CTRL, data); + } +} + +static void fiji_update_sdma_medium_grain_light_sleep( + struct amdgpu_device *adev, + bool enable) +{ + uint32_t temp, data; + + if (enable) { + temp = data = RREG32(mmSDMA0_POWER_CNTL); + data |= SDMA0_POWER_CNTL__MEM_POWER_OVERRIDE_MASK; + + if (temp != data) + WREG32(mmSDMA0_POWER_CNTL, data); + + temp = data = RREG32(mmSDMA1_POWER_CNTL); + data |= SDMA1_POWER_CNTL__MEM_POWER_OVERRIDE_MASK; + + if (temp != data) + WREG32(mmSDMA1_POWER_CNTL, data); + } else { + temp = data = RREG32(mmSDMA0_POWER_CNTL); + data &= ~SDMA0_POWER_CNTL__MEM_POWER_OVERRIDE_MASK; + + if (temp != data) + WREG32(mmSDMA0_POWER_CNTL, data); + + temp = data = RREG32(mmSDMA1_POWER_CNTL); + data &= ~SDMA1_POWER_CNTL__MEM_POWER_OVERRIDE_MASK; + + if (temp != data) + WREG32(mmSDMA1_POWER_CNTL, data); + } +} + static int sdma_v3_0_set_clockgating_state(void *handle, enum amd_clockgating_state state) { + struct amdgpu_device *adev = (struct amdgpu_device *)handle; + + switch (adev->asic_type) { + case CHIP_FIJI: + fiji_update_sdma_medium_grain_clock_gating(adev, + state == AMD_CG_STATE_GATE ? true : false); + fiji_update_sdma_medium_grain_light_sleep(adev, + state == AMD_CG_STATE_GATE ? true : false); + break; + default: + break; + } return 0; } @@ -1432,24 +1567,6 @@ const struct amd_ip_funcs sdma_v3_0_ip_funcs = { .set_powergating_state = sdma_v3_0_set_powergating_state, }; -/** - * sdma_v3_0_ring_is_lockup - Check if the DMA engine is locked up - * - * @ring: amdgpu_ring structure holding ring information - * - * Check if the async DMA engine is locked up (VI). - * Returns true if the engine appears to be locked up, false if not. - */ -static bool sdma_v3_0_ring_is_lockup(struct amdgpu_ring *ring) -{ - - if (sdma_v3_0_is_idle(ring->adev)) { - amdgpu_ring_lockup_update(ring); - return false; - } - return amdgpu_ring_test_lockup(ring); -} - static const struct amdgpu_ring_funcs sdma_v3_0_ring_funcs = { .get_rptr = sdma_v3_0_ring_get_rptr, .get_wptr = sdma_v3_0_ring_get_wptr, @@ -1457,19 +1574,22 @@ static const struct amdgpu_ring_funcs sdma_v3_0_ring_funcs = { .parse_cs = NULL, .emit_ib = sdma_v3_0_ring_emit_ib, .emit_fence = sdma_v3_0_ring_emit_fence, - .emit_semaphore = sdma_v3_0_ring_emit_semaphore, + .emit_pipeline_sync = sdma_v3_0_ring_emit_pipeline_sync, .emit_vm_flush = sdma_v3_0_ring_emit_vm_flush, .emit_hdp_flush = sdma_v3_0_ring_emit_hdp_flush, + .emit_hdp_invalidate = sdma_v3_0_ring_emit_hdp_invalidate, .test_ring = sdma_v3_0_ring_test_ring, .test_ib = sdma_v3_0_ring_test_ib, - .is_lockup = sdma_v3_0_ring_is_lockup, .insert_nop = sdma_v3_0_ring_insert_nop, + .pad_ib = sdma_v3_0_ring_pad_ib, }; static void sdma_v3_0_set_ring_funcs(struct amdgpu_device *adev) { - adev->sdma[0].ring.funcs = &sdma_v3_0_ring_funcs; - adev->sdma[1].ring.funcs = &sdma_v3_0_ring_funcs; + int i; + + for (i = 0; i < adev->sdma.num_instances; i++) + adev->sdma.instance[i].ring.funcs = &sdma_v3_0_ring_funcs; } static const struct amdgpu_irq_src_funcs sdma_v3_0_trap_irq_funcs = { @@ -1483,9 +1603,9 @@ static const struct amdgpu_irq_src_funcs sdma_v3_0_illegal_inst_irq_funcs = { static void sdma_v3_0_set_irq_funcs(struct amdgpu_device *adev) { - adev->sdma_trap_irq.num_types = AMDGPU_SDMA_IRQ_LAST; - adev->sdma_trap_irq.funcs = &sdma_v3_0_trap_irq_funcs; - adev->sdma_illegal_inst_irq.funcs = &sdma_v3_0_illegal_inst_irq_funcs; + adev->sdma.trap_irq.num_types = AMDGPU_SDMA_IRQ_LAST; + adev->sdma.trap_irq.funcs = &sdma_v3_0_trap_irq_funcs; + adev->sdma.illegal_inst_irq.funcs = &sdma_v3_0_illegal_inst_irq_funcs; } /** @@ -1551,7 +1671,7 @@ static void sdma_v3_0_set_buffer_funcs(struct amdgpu_device *adev) { if (adev->mman.buffer_funcs == NULL) { adev->mman.buffer_funcs = &sdma_v3_0_buffer_funcs; - adev->mman.buffer_funcs_ring = &adev->sdma[0].ring; + adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].ring; } } @@ -1559,14 +1679,18 @@ static const struct amdgpu_vm_pte_funcs sdma_v3_0_vm_pte_funcs = { .copy_pte = sdma_v3_0_vm_copy_pte, .write_pte = sdma_v3_0_vm_write_pte, .set_pte_pde = sdma_v3_0_vm_set_pte_pde, - .pad_ib = sdma_v3_0_vm_pad_ib, }; static void sdma_v3_0_set_vm_pte_funcs(struct amdgpu_device *adev) { + unsigned i; + if (adev->vm_manager.vm_pte_funcs == NULL) { adev->vm_manager.vm_pte_funcs = &sdma_v3_0_vm_pte_funcs; - adev->vm_manager.vm_pte_funcs_ring = &adev->sdma[0].ring; - adev->vm_manager.vm_pte_funcs_ring->is_pte_ring = true; + for (i = 0; i < adev->sdma.num_instances; i++) + adev->vm_manager.vm_pte_rings[i] = + &adev->sdma.instance[i].ring; + + adev->vm_manager.vm_pte_num_rings = adev->sdma.num_instances; } } diff --git a/drivers/gpu/drm/amd/amdgpu/smu7.h b/drivers/gpu/drm/amd/amdgpu/smu7.h deleted file mode 100644 index 75a380a15292..000000000000 --- a/drivers/gpu/drm/amd/amdgpu/smu7.h +++ /dev/null @@ -1,170 +0,0 @@ -/* - * Copyright 2013 Advanced Micro Devices, Inc. - * - * Permission is hereby granted, free of charge, to any person obtaining a - * copy of this software and associated documentation files (the "Software"), - * to deal in the Software without restriction, including without limitation - * the rights to use, copy, modify, merge, publish, distribute, sublicense, - * and/or sell copies of the Software, and to permit persons to whom the - * Software is furnished to do so, subject to the following conditions: - * - * The above copyright notice and this permission notice shall be included in - * all copies or substantial portions of the Software. - * - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR - * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, - * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL - * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR - * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, - * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR - * OTHER DEALINGS IN THE SOFTWARE. - * - */ - -#ifndef SMU7_H -#define SMU7_H - -#pragma pack(push, 1) - -#define SMU7_CONTEXT_ID_SMC 1 -#define SMU7_CONTEXT_ID_VBIOS 2 - - -#define SMU7_CONTEXT_ID_SMC 1 -#define SMU7_CONTEXT_ID_VBIOS 2 - -#define SMU7_MAX_LEVELS_VDDC 8 -#define SMU7_MAX_LEVELS_VDDCI 4 -#define SMU7_MAX_LEVELS_MVDD 4 -#define SMU7_MAX_LEVELS_VDDNB 8 - -#define SMU7_MAX_LEVELS_GRAPHICS SMU__NUM_SCLK_DPM_STATE // SCLK + SQ DPM + ULV -#define SMU7_MAX_LEVELS_MEMORY SMU__NUM_MCLK_DPM_LEVELS // MCLK Levels DPM -#define SMU7_MAX_LEVELS_GIO SMU__NUM_LCLK_DPM_LEVELS // LCLK Levels -#define SMU7_MAX_LEVELS_LINK SMU__NUM_PCIE_DPM_LEVELS // PCIe speed and number of lanes. -#define SMU7_MAX_LEVELS_UVD 8 // VCLK/DCLK levels for UVD. -#define SMU7_MAX_LEVELS_VCE 8 // ECLK levels for VCE. -#define SMU7_MAX_LEVELS_ACP 8 // ACLK levels for ACP. -#define SMU7_MAX_LEVELS_SAMU 8 // SAMCLK levels for SAMU. -#define SMU7_MAX_ENTRIES_SMIO 32 // Number of entries in SMIO table. - -#define DPM_NO_LIMIT 0 -#define DPM_NO_UP 1 -#define DPM_GO_DOWN 2 -#define DPM_GO_UP 3 - -#define SMU7_FIRST_DPM_GRAPHICS_LEVEL 0 -#define SMU7_FIRST_DPM_MEMORY_LEVEL 0 - -#define GPIO_CLAMP_MODE_VRHOT 1 -#define GPIO_CLAMP_MODE_THERM 2 -#define GPIO_CLAMP_MODE_DC 4 - -#define SCRATCH_B_TARG_PCIE_INDEX_SHIFT 0 -#define SCRATCH_B_TARG_PCIE_INDEX_MASK (0x7<<SCRATCH_B_TARG_PCIE_INDEX_SHIFT) -#define SCRATCH_B_CURR_PCIE_INDEX_SHIFT 3 -#define SCRATCH_B_CURR_PCIE_INDEX_MASK (0x7<<SCRATCH_B_CURR_PCIE_INDEX_SHIFT) -#define SCRATCH_B_TARG_UVD_INDEX_SHIFT 6 -#define SCRATCH_B_TARG_UVD_INDEX_MASK (0x7<<SCRATCH_B_TARG_UVD_INDEX_SHIFT) -#define SCRATCH_B_CURR_UVD_INDEX_SHIFT 9 -#define SCRATCH_B_CURR_UVD_INDEX_MASK (0x7<<SCRATCH_B_CURR_UVD_INDEX_SHIFT) -#define SCRATCH_B_TARG_VCE_INDEX_SHIFT 12 -#define SCRATCH_B_TARG_VCE_INDEX_MASK (0x7<<SCRATCH_B_TARG_VCE_INDEX_SHIFT) -#define SCRATCH_B_CURR_VCE_INDEX_SHIFT 15 -#define SCRATCH_B_CURR_VCE_INDEX_MASK (0x7<<SCRATCH_B_CURR_VCE_INDEX_SHIFT) -#define SCRATCH_B_TARG_ACP_INDEX_SHIFT 18 -#define SCRATCH_B_TARG_ACP_INDEX_MASK (0x7<<SCRATCH_B_TARG_ACP_INDEX_SHIFT) -#define SCRATCH_B_CURR_ACP_INDEX_SHIFT 21 -#define SCRATCH_B_CURR_ACP_INDEX_MASK (0x7<<SCRATCH_B_CURR_ACP_INDEX_SHIFT) -#define SCRATCH_B_TARG_SAMU_INDEX_SHIFT 24 -#define SCRATCH_B_TARG_SAMU_INDEX_MASK (0x7<<SCRATCH_B_TARG_SAMU_INDEX_SHIFT) -#define SCRATCH_B_CURR_SAMU_INDEX_SHIFT 27 -#define SCRATCH_B_CURR_SAMU_INDEX_MASK (0x7<<SCRATCH_B_CURR_SAMU_INDEX_SHIFT) - - -struct SMU7_PIDController -{ - uint32_t Ki; - int32_t LFWindupUL; - int32_t LFWindupLL; - uint32_t StatePrecision; - uint32_t LfPrecision; - uint32_t LfOffset; - uint32_t MaxState; - uint32_t MaxLfFraction; - uint32_t StateShift; -}; - -typedef struct SMU7_PIDController SMU7_PIDController; - -// ------------------------------------------------------------------------------------------------------------------------- -#define SMU7_MAX_PCIE_LINK_SPEEDS 3 /* 0:Gen1 1:Gen2 2:Gen3 */ - -#define SMU7_SCLK_DPM_CONFIG_MASK 0x01 -#define SMU7_VOLTAGE_CONTROLLER_CONFIG_MASK 0x02 -#define SMU7_THERMAL_CONTROLLER_CONFIG_MASK 0x04 -#define SMU7_MCLK_DPM_CONFIG_MASK 0x08 -#define SMU7_UVD_DPM_CONFIG_MASK 0x10 -#define SMU7_VCE_DPM_CONFIG_MASK 0x20 -#define SMU7_ACP_DPM_CONFIG_MASK 0x40 -#define SMU7_SAMU_DPM_CONFIG_MASK 0x80 -#define SMU7_PCIEGEN_DPM_CONFIG_MASK 0x100 - -#define SMU7_ACP_MCLK_HANDSHAKE_DISABLE 0x00000001 -#define SMU7_ACP_SCLK_HANDSHAKE_DISABLE 0x00000002 -#define SMU7_UVD_MCLK_HANDSHAKE_DISABLE 0x00000100 -#define SMU7_UVD_SCLK_HANDSHAKE_DISABLE 0x00000200 -#define SMU7_VCE_MCLK_HANDSHAKE_DISABLE 0x00010000 -#define SMU7_VCE_SCLK_HANDSHAKE_DISABLE 0x00020000 - -struct SMU7_Firmware_Header -{ - uint32_t Digest[5]; - uint32_t Version; - uint32_t HeaderSize; - uint32_t Flags; - uint32_t EntryPoint; - uint32_t CodeSize; - uint32_t ImageSize; - - uint32_t Rtos; - uint32_t SoftRegisters; - uint32_t DpmTable; - uint32_t FanTable; - uint32_t CacConfigTable; - uint32_t CacStatusTable; - - uint32_t mcRegisterTable; - - uint32_t mcArbDramTimingTable; - - uint32_t PmFuseTable; - uint32_t Globals; - uint32_t Reserved[42]; - uint32_t Signature; -}; - -typedef struct SMU7_Firmware_Header SMU7_Firmware_Header; - -#define SMU7_FIRMWARE_HEADER_LOCATION 0x20000 - -enum DisplayConfig { - PowerDown = 1, - DP54x4, - DP54x2, - DP54x1, - DP27x4, - DP27x2, - DP27x1, - HDMI297, - HDMI162, - LVDS, - DP324x4, - DP324x2, - DP324x1 -}; - -#pragma pack(pop) - -#endif - diff --git a/drivers/gpu/drm/amd/amdgpu/smu7_discrete.h b/drivers/gpu/drm/amd/amdgpu/smu7_discrete.h deleted file mode 100644 index 0b0b404ff091..000000000000 --- a/drivers/gpu/drm/amd/amdgpu/smu7_discrete.h +++ /dev/null @@ -1,514 +0,0 @@ -/* - * Copyright 2013 Advanced Micro Devices, Inc. - * - * Permission is hereby granted, free of charge, to any person obtaining a - * copy of this software and associated documentation files (the "Software"), - * to deal in the Software without restriction, including without limitation - * the rights to use, copy, modify, merge, publish, distribute, sublicense, - * and/or sell copies of the Software, and to permit persons to whom the - * Software is furnished to do so, subject to the following conditions: - * - * The above copyright notice and this permission notice shall be included in - * all copies or substantial portions of the Software. - * - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR - * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, - * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL - * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR - * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, - * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR - * OTHER DEALINGS IN THE SOFTWARE. - * - */ - -#ifndef SMU7_DISCRETE_H -#define SMU7_DISCRETE_H - -#include "smu7.h" - -#pragma pack(push, 1) - -#define SMU7_DTE_ITERATIONS 5 -#define SMU7_DTE_SOURCES 3 -#define SMU7_DTE_SINKS 1 -#define SMU7_NUM_CPU_TES 0 -#define SMU7_NUM_GPU_TES 1 -#define SMU7_NUM_NON_TES 2 - -struct SMU7_SoftRegisters -{ - uint32_t RefClockFrequency; - uint32_t PmTimerP; - uint32_t FeatureEnables; - uint32_t PreVBlankGap; - uint32_t VBlankTimeout; - uint32_t TrainTimeGap; - - uint32_t MvddSwitchTime; - uint32_t LongestAcpiTrainTime; - uint32_t AcpiDelay; - uint32_t G5TrainTime; - uint32_t DelayMpllPwron; - uint32_t VoltageChangeTimeout; - uint32_t HandshakeDisables; - - uint8_t DisplayPhy1Config; - uint8_t DisplayPhy2Config; - uint8_t DisplayPhy3Config; - uint8_t DisplayPhy4Config; - - uint8_t DisplayPhy5Config; - uint8_t DisplayPhy6Config; - uint8_t DisplayPhy7Config; - uint8_t DisplayPhy8Config; - - uint32_t AverageGraphicsA; - uint32_t AverageMemoryA; - uint32_t AverageGioA; - - uint8_t SClkDpmEnabledLevels; - uint8_t MClkDpmEnabledLevels; - uint8_t LClkDpmEnabledLevels; - uint8_t PCIeDpmEnabledLevels; - - uint8_t UVDDpmEnabledLevels; - uint8_t SAMUDpmEnabledLevels; - uint8_t ACPDpmEnabledLevels; - uint8_t VCEDpmEnabledLevels; - - uint32_t DRAM_LOG_ADDR_H; - uint32_t DRAM_LOG_ADDR_L; - uint32_t DRAM_LOG_PHY_ADDR_H; - uint32_t DRAM_LOG_PHY_ADDR_L; - uint32_t DRAM_LOG_BUFF_SIZE; - uint32_t UlvEnterC; - uint32_t UlvTime; - uint32_t Reserved[3]; - -}; - -typedef struct SMU7_SoftRegisters SMU7_SoftRegisters; - -struct SMU7_Discrete_VoltageLevel -{ - uint16_t Voltage; - uint16_t StdVoltageHiSidd; - uint16_t StdVoltageLoSidd; - uint8_t Smio; - uint8_t padding; -}; - -typedef struct SMU7_Discrete_VoltageLevel SMU7_Discrete_VoltageLevel; - -struct SMU7_Discrete_GraphicsLevel -{ - uint32_t Flags; - uint32_t MinVddc; - uint32_t MinVddcPhases; - - uint32_t SclkFrequency; - - uint8_t padding1[2]; - uint16_t ActivityLevel; - - uint32_t CgSpllFuncCntl3; - uint32_t CgSpllFuncCntl4; - uint32_t SpllSpreadSpectrum; - uint32_t SpllSpreadSpectrum2; - uint32_t CcPwrDynRm; - uint32_t CcPwrDynRm1; - uint8_t SclkDid; - uint8_t DisplayWatermark; - uint8_t EnabledForActivity; - uint8_t EnabledForThrottle; - uint8_t UpH; - uint8_t DownH; - uint8_t VoltageDownH; - uint8_t PowerThrottle; - uint8_t DeepSleepDivId; - uint8_t padding[3]; -}; - -typedef struct SMU7_Discrete_GraphicsLevel SMU7_Discrete_GraphicsLevel; - -struct SMU7_Discrete_ACPILevel -{ - uint32_t Flags; - uint32_t MinVddc; - uint32_t MinVddcPhases; - uint32_t SclkFrequency; - uint8_t SclkDid; - uint8_t DisplayWatermark; - uint8_t DeepSleepDivId; - uint8_t padding; - uint32_t CgSpllFuncCntl; - uint32_t CgSpllFuncCntl2; - uint32_t CgSpllFuncCntl3; - uint32_t CgSpllFuncCntl4; - uint32_t SpllSpreadSpectrum; - uint32_t SpllSpreadSpectrum2; - uint32_t CcPwrDynRm; - uint32_t CcPwrDynRm1; -}; - -typedef struct SMU7_Discrete_ACPILevel SMU7_Discrete_ACPILevel; - -struct SMU7_Discrete_Ulv -{ - uint32_t CcPwrDynRm; - uint32_t CcPwrDynRm1; - uint16_t VddcOffset; - uint8_t VddcOffsetVid; - uint8_t VddcPhase; - uint32_t Reserved; -}; - -typedef struct SMU7_Discrete_Ulv SMU7_Discrete_Ulv; - -struct SMU7_Discrete_MemoryLevel -{ - uint32_t MinVddc; - uint32_t MinVddcPhases; - uint32_t MinVddci; - uint32_t MinMvdd; - - uint32_t MclkFrequency; - - uint8_t EdcReadEnable; - uint8_t EdcWriteEnable; - uint8_t RttEnable; - uint8_t StutterEnable; - - uint8_t StrobeEnable; - uint8_t StrobeRatio; - uint8_t EnabledForThrottle; - uint8_t EnabledForActivity; - - uint8_t UpH; - uint8_t DownH; - uint8_t VoltageDownH; - uint8_t padding; - - uint16_t ActivityLevel; - uint8_t DisplayWatermark; - uint8_t padding1; - - uint32_t MpllFuncCntl; - uint32_t MpllFuncCntl_1; - uint32_t MpllFuncCntl_2; - uint32_t MpllAdFuncCntl; - uint32_t MpllDqFuncCntl; - uint32_t MclkPwrmgtCntl; - uint32_t DllCntl; - uint32_t MpllSs1; - uint32_t MpllSs2; -}; - -typedef struct SMU7_Discrete_MemoryLevel SMU7_Discrete_MemoryLevel; - -struct SMU7_Discrete_LinkLevel -{ - uint8_t PcieGenSpeed; - uint8_t PcieLaneCount; - uint8_t EnabledForActivity; - uint8_t Padding; - uint32_t DownT; - uint32_t UpT; - uint32_t Reserved; -}; - -typedef struct SMU7_Discrete_LinkLevel SMU7_Discrete_LinkLevel; - - -struct SMU7_Discrete_MCArbDramTimingTableEntry -{ - uint32_t McArbDramTiming; - uint32_t McArbDramTiming2; - uint8_t McArbBurstTime; - uint8_t padding[3]; -}; - -typedef struct SMU7_Discrete_MCArbDramTimingTableEntry SMU7_Discrete_MCArbDramTimingTableEntry; - -struct SMU7_Discrete_MCArbDramTimingTable -{ - SMU7_Discrete_MCArbDramTimingTableEntry entries[SMU__NUM_SCLK_DPM_STATE][SMU__NUM_MCLK_DPM_LEVELS]; -}; - -typedef struct SMU7_Discrete_MCArbDramTimingTable SMU7_Discrete_MCArbDramTimingTable; - -struct SMU7_Discrete_UvdLevel -{ - uint32_t VclkFrequency; - uint32_t DclkFrequency; - uint16_t MinVddc; - uint8_t MinVddcPhases; - uint8_t VclkDivider; - uint8_t DclkDivider; - uint8_t padding[3]; -}; - -typedef struct SMU7_Discrete_UvdLevel SMU7_Discrete_UvdLevel; - -struct SMU7_Discrete_ExtClkLevel -{ - uint32_t Frequency; - uint16_t MinVoltage; - uint8_t MinPhases; - uint8_t Divider; -}; - -typedef struct SMU7_Discrete_ExtClkLevel SMU7_Discrete_ExtClkLevel; - -struct SMU7_Discrete_StateInfo -{ - uint32_t SclkFrequency; - uint32_t MclkFrequency; - uint32_t VclkFrequency; - uint32_t DclkFrequency; - uint32_t SamclkFrequency; - uint32_t AclkFrequency; - uint32_t EclkFrequency; - uint16_t MvddVoltage; - uint16_t padding16; - uint8_t DisplayWatermark; - uint8_t McArbIndex; - uint8_t McRegIndex; - uint8_t SeqIndex; - uint8_t SclkDid; - int8_t SclkIndex; - int8_t MclkIndex; - uint8_t PCIeGen; - -}; - -typedef struct SMU7_Discrete_StateInfo SMU7_Discrete_StateInfo; - - -struct SMU7_Discrete_DpmTable -{ - SMU7_PIDController GraphicsPIDController; - SMU7_PIDController MemoryPIDController; - SMU7_PIDController LinkPIDController; - - uint32_t SystemFlags; - - - uint32_t SmioMaskVddcVid; - uint32_t SmioMaskVddcPhase; - uint32_t SmioMaskVddciVid; - uint32_t SmioMaskMvddVid; - - uint32_t VddcLevelCount; - uint32_t VddciLevelCount; - uint32_t MvddLevelCount; - - SMU7_Discrete_VoltageLevel VddcLevel [SMU7_MAX_LEVELS_VDDC]; -// SMU7_Discrete_VoltageLevel VddcStandardReference [SMU7_MAX_LEVELS_VDDC]; - SMU7_Discrete_VoltageLevel VddciLevel [SMU7_MAX_LEVELS_VDDCI]; - SMU7_Discrete_VoltageLevel MvddLevel [SMU7_MAX_LEVELS_MVDD]; - - uint8_t GraphicsDpmLevelCount; - uint8_t MemoryDpmLevelCount; - uint8_t LinkLevelCount; - uint8_t UvdLevelCount; - uint8_t VceLevelCount; - uint8_t AcpLevelCount; - uint8_t SamuLevelCount; - uint8_t MasterDeepSleepControl; - uint32_t Reserved[5]; -// uint32_t SamuDefaultLevel; - - SMU7_Discrete_GraphicsLevel GraphicsLevel [SMU7_MAX_LEVELS_GRAPHICS]; - SMU7_Discrete_MemoryLevel MemoryACPILevel; - SMU7_Discrete_MemoryLevel MemoryLevel [SMU7_MAX_LEVELS_MEMORY]; - SMU7_Discrete_LinkLevel LinkLevel [SMU7_MAX_LEVELS_LINK]; - SMU7_Discrete_ACPILevel ACPILevel; - SMU7_Discrete_UvdLevel UvdLevel [SMU7_MAX_LEVELS_UVD]; - SMU7_Discrete_ExtClkLevel VceLevel [SMU7_MAX_LEVELS_VCE]; - SMU7_Discrete_ExtClkLevel AcpLevel [SMU7_MAX_LEVELS_ACP]; - SMU7_Discrete_ExtClkLevel SamuLevel [SMU7_MAX_LEVELS_SAMU]; - SMU7_Discrete_Ulv Ulv; - - uint32_t SclkStepSize; - uint32_t Smio [SMU7_MAX_ENTRIES_SMIO]; - - uint8_t UvdBootLevel; - uint8_t VceBootLevel; - uint8_t AcpBootLevel; - uint8_t SamuBootLevel; - - uint8_t UVDInterval; - uint8_t VCEInterval; - uint8_t ACPInterval; - uint8_t SAMUInterval; - - uint8_t GraphicsBootLevel; - uint8_t GraphicsVoltageChangeEnable; - uint8_t GraphicsThermThrottleEnable; - uint8_t GraphicsInterval; - - uint8_t VoltageInterval; - uint8_t ThermalInterval; - uint16_t TemperatureLimitHigh; - - uint16_t TemperatureLimitLow; - uint8_t MemoryBootLevel; - uint8_t MemoryVoltageChangeEnable; - - uint8_t MemoryInterval; - uint8_t MemoryThermThrottleEnable; - uint16_t VddcVddciDelta; - - uint16_t VoltageResponseTime; - uint16_t PhaseResponseTime; - - uint8_t PCIeBootLinkLevel; - uint8_t PCIeGenInterval; - uint8_t DTEInterval; - uint8_t DTEMode; - - uint8_t SVI2Enable; - uint8_t VRHotGpio; - uint8_t AcDcGpio; - uint8_t ThermGpio; - - uint16_t PPM_PkgPwrLimit; - uint16_t PPM_TemperatureLimit; - - uint16_t DefaultTdp; - uint16_t TargetTdp; - - uint16_t FpsHighT; - uint16_t FpsLowT; - - uint16_t BAPMTI_R [SMU7_DTE_ITERATIONS][SMU7_DTE_SOURCES][SMU7_DTE_SINKS]; - uint16_t BAPMTI_RC [SMU7_DTE_ITERATIONS][SMU7_DTE_SOURCES][SMU7_DTE_SINKS]; - - uint8_t DTEAmbientTempBase; - uint8_t DTETjOffset; - uint8_t GpuTjMax; - uint8_t GpuTjHyst; - - uint16_t BootVddc; - uint16_t BootVddci; - - uint16_t BootMVdd; - uint16_t padding; - - uint32_t BAPM_TEMP_GRADIENT; - - uint32_t LowSclkInterruptT; -}; - -typedef struct SMU7_Discrete_DpmTable SMU7_Discrete_DpmTable; - -#define SMU7_DISCRETE_MC_REGISTER_ARRAY_SIZE 16 -#define SMU7_DISCRETE_MC_REGISTER_ARRAY_SET_COUNT SMU7_MAX_LEVELS_MEMORY - -struct SMU7_Discrete_MCRegisterAddress -{ - uint16_t s0; - uint16_t s1; -}; - -typedef struct SMU7_Discrete_MCRegisterAddress SMU7_Discrete_MCRegisterAddress; - -struct SMU7_Discrete_MCRegisterSet -{ - uint32_t value[SMU7_DISCRETE_MC_REGISTER_ARRAY_SIZE]; -}; - -typedef struct SMU7_Discrete_MCRegisterSet SMU7_Discrete_MCRegisterSet; - -struct SMU7_Discrete_MCRegisters -{ - uint8_t last; - uint8_t reserved[3]; - SMU7_Discrete_MCRegisterAddress address[SMU7_DISCRETE_MC_REGISTER_ARRAY_SIZE]; - SMU7_Discrete_MCRegisterSet data[SMU7_DISCRETE_MC_REGISTER_ARRAY_SET_COUNT]; -}; - -typedef struct SMU7_Discrete_MCRegisters SMU7_Discrete_MCRegisters; - -struct SMU7_Discrete_FanTable -{ - uint16_t FdoMode; - int16_t TempMin; - int16_t TempMed; - int16_t TempMax; - int16_t Slope1; - int16_t Slope2; - int16_t FdoMin; - int16_t HystUp; - int16_t HystDown; - int16_t HystSlope; - int16_t TempRespLim; - int16_t TempCurr; - int16_t SlopeCurr; - int16_t PwmCurr; - uint32_t RefreshPeriod; - int16_t FdoMax; - uint8_t TempSrc; - int8_t Padding; -}; - -typedef struct SMU7_Discrete_FanTable SMU7_Discrete_FanTable; - - -struct SMU7_Discrete_PmFuses { - // dw0-dw1 - uint8_t BapmVddCVidHiSidd[8]; - - // dw2-dw3 - uint8_t BapmVddCVidLoSidd[8]; - - // dw4-dw5 - uint8_t VddCVid[8]; - - // dw6 - uint8_t SviLoadLineEn; - uint8_t SviLoadLineVddC; - uint8_t SviLoadLineTrimVddC; - uint8_t SviLoadLineOffsetVddC; - - // dw7 - uint16_t TDC_VDDC_PkgLimit; - uint8_t TDC_VDDC_ThrottleReleaseLimitPerc; - uint8_t TDC_MAWt; - - // dw8 - uint8_t TdcWaterfallCtl; - uint8_t LPMLTemperatureMin; - uint8_t LPMLTemperatureMax; - uint8_t Reserved; - - // dw9-dw10 - uint8_t BapmVddCVidHiSidd2[8]; - - // dw11-dw12 - int16_t FuzzyFan_ErrorSetDelta; - int16_t FuzzyFan_ErrorRateSetDelta; - int16_t FuzzyFan_PwmSetDelta; - uint16_t CalcMeasPowerBlend; - - // dw13-dw16 - uint8_t GnbLPML[16]; - - // dw17 - uint8_t GnbLPMLMaxVid; - uint8_t GnbLPMLMinVid; - uint8_t Reserved1[2]; - - // dw18 - uint16_t BapmVddCBaseLeakageHiSidd; - uint16_t BapmVddCBaseLeakageLoSidd; -}; - -typedef struct SMU7_Discrete_PmFuses SMU7_Discrete_PmFuses; - - -#pragma pack(pop) - -#endif - diff --git a/drivers/gpu/drm/amd/amdgpu/smu7_fusion.h b/drivers/gpu/drm/amd/amdgpu/smu7_fusion.h deleted file mode 100644 index 78ada9ffd508..000000000000 --- a/drivers/gpu/drm/amd/amdgpu/smu7_fusion.h +++ /dev/null @@ -1,300 +0,0 @@ -/* - * Copyright 2013 Advanced Micro Devices, Inc. - * - * Permission is hereby granted, free of charge, to any person obtaining a - * copy of this software and associated documentation files (the "Software"), - * to deal in the Software without restriction, including without limitation - * the rights to use, copy, modify, merge, publish, distribute, sublicense, - * and/or sell copies of the Software, and to permit persons to whom the - * Software is furnished to do so, subject to the following conditions: - * - * The above copyright notice and this permission notice shall be included in - * all copies or substantial portions of the Software. - * - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR - * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, - * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL - * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR - * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, - * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR - * OTHER DEALINGS IN THE SOFTWARE. - * - */ - -#ifndef SMU7_FUSION_H -#define SMU7_FUSION_H - -#include "smu7.h" - -#pragma pack(push, 1) - -#define SMU7_DTE_ITERATIONS 5 -#define SMU7_DTE_SOURCES 5 -#define SMU7_DTE_SINKS 3 -#define SMU7_NUM_CPU_TES 2 -#define SMU7_NUM_GPU_TES 1 -#define SMU7_NUM_NON_TES 2 - -// All 'soft registers' should be uint32_t. -struct SMU7_SoftRegisters -{ - uint32_t RefClockFrequency; - uint32_t PmTimerP; - uint32_t FeatureEnables; - uint32_t HandshakeDisables; - - uint8_t DisplayPhy1Config; - uint8_t DisplayPhy2Config; - uint8_t DisplayPhy3Config; - uint8_t DisplayPhy4Config; - - uint8_t DisplayPhy5Config; - uint8_t DisplayPhy6Config; - uint8_t DisplayPhy7Config; - uint8_t DisplayPhy8Config; - - uint32_t AverageGraphicsA; - uint32_t AverageMemoryA; - uint32_t AverageGioA; - - uint8_t SClkDpmEnabledLevels; - uint8_t MClkDpmEnabledLevels; - uint8_t LClkDpmEnabledLevels; - uint8_t PCIeDpmEnabledLevels; - - uint8_t UVDDpmEnabledLevels; - uint8_t SAMUDpmEnabledLevels; - uint8_t ACPDpmEnabledLevels; - uint8_t VCEDpmEnabledLevels; - - uint32_t DRAM_LOG_ADDR_H; - uint32_t DRAM_LOG_ADDR_L; - uint32_t DRAM_LOG_PHY_ADDR_H; - uint32_t DRAM_LOG_PHY_ADDR_L; - uint32_t DRAM_LOG_BUFF_SIZE; - uint32_t UlvEnterC; - uint32_t UlvTime; - uint32_t Reserved[3]; - -}; - -typedef struct SMU7_SoftRegisters SMU7_SoftRegisters; - -struct SMU7_Fusion_GraphicsLevel -{ - uint32_t MinVddNb; - - uint32_t SclkFrequency; - - uint8_t Vid; - uint8_t VidOffset; - uint16_t AT; - - uint8_t PowerThrottle; - uint8_t GnbSlow; - uint8_t ForceNbPs1; - uint8_t SclkDid; - - uint8_t DisplayWatermark; - uint8_t EnabledForActivity; - uint8_t EnabledForThrottle; - uint8_t UpH; - - uint8_t DownH; - uint8_t VoltageDownH; - uint8_t DeepSleepDivId; - - uint8_t ClkBypassCntl; - - uint32_t reserved; -}; - -typedef struct SMU7_Fusion_GraphicsLevel SMU7_Fusion_GraphicsLevel; - -struct SMU7_Fusion_GIOLevel -{ - uint8_t EnabledForActivity; - uint8_t LclkDid; - uint8_t Vid; - uint8_t VoltageDownH; - - uint32_t MinVddNb; - - uint16_t ResidencyCounter; - uint8_t UpH; - uint8_t DownH; - - uint32_t LclkFrequency; - - uint8_t ActivityLevel; - uint8_t EnabledForThrottle; - - uint8_t ClkBypassCntl; - - uint8_t padding; -}; - -typedef struct SMU7_Fusion_GIOLevel SMU7_Fusion_GIOLevel; - -// UVD VCLK/DCLK state (level) definition. -struct SMU7_Fusion_UvdLevel -{ - uint32_t VclkFrequency; - uint32_t DclkFrequency; - uint16_t MinVddNb; - uint8_t VclkDivider; - uint8_t DclkDivider; - - uint8_t VClkBypassCntl; - uint8_t DClkBypassCntl; - - uint8_t padding[2]; - -}; - -typedef struct SMU7_Fusion_UvdLevel SMU7_Fusion_UvdLevel; - -// Clocks for other external blocks (VCE, ACP, SAMU). -struct SMU7_Fusion_ExtClkLevel -{ - uint32_t Frequency; - uint16_t MinVoltage; - uint8_t Divider; - uint8_t ClkBypassCntl; - - uint32_t Reserved; -}; -typedef struct SMU7_Fusion_ExtClkLevel SMU7_Fusion_ExtClkLevel; - -struct SMU7_Fusion_ACPILevel -{ - uint32_t Flags; - uint32_t MinVddNb; - uint32_t SclkFrequency; - uint8_t SclkDid; - uint8_t GnbSlow; - uint8_t ForceNbPs1; - uint8_t DisplayWatermark; - uint8_t DeepSleepDivId; - uint8_t padding[3]; -}; - -typedef struct SMU7_Fusion_ACPILevel SMU7_Fusion_ACPILevel; - -struct SMU7_Fusion_NbDpm -{ - uint8_t DpmXNbPsHi; - uint8_t DpmXNbPsLo; - uint8_t Dpm0PgNbPsHi; - uint8_t Dpm0PgNbPsLo; - uint8_t EnablePsi1; - uint8_t SkipDPM0; - uint8_t SkipPG; - uint8_t Hysteresis; - uint8_t EnableDpmPstatePoll; - uint8_t padding[3]; -}; - -typedef struct SMU7_Fusion_NbDpm SMU7_Fusion_NbDpm; - -struct SMU7_Fusion_StateInfo -{ - uint32_t SclkFrequency; - uint32_t LclkFrequency; - uint32_t VclkFrequency; - uint32_t DclkFrequency; - uint32_t SamclkFrequency; - uint32_t AclkFrequency; - uint32_t EclkFrequency; - uint8_t DisplayWatermark; - uint8_t McArbIndex; - int8_t SclkIndex; - int8_t MclkIndex; -}; - -typedef struct SMU7_Fusion_StateInfo SMU7_Fusion_StateInfo; - -struct SMU7_Fusion_DpmTable -{ - uint32_t SystemFlags; - - SMU7_PIDController GraphicsPIDController; - SMU7_PIDController GioPIDController; - - uint8_t GraphicsDpmLevelCount; - uint8_t GIOLevelCount; - uint8_t UvdLevelCount; - uint8_t VceLevelCount; - - uint8_t AcpLevelCount; - uint8_t SamuLevelCount; - uint16_t FpsHighT; - - SMU7_Fusion_GraphicsLevel GraphicsLevel [SMU__NUM_SCLK_DPM_STATE]; - SMU7_Fusion_ACPILevel ACPILevel; - SMU7_Fusion_UvdLevel UvdLevel [SMU7_MAX_LEVELS_UVD]; - SMU7_Fusion_ExtClkLevel VceLevel [SMU7_MAX_LEVELS_VCE]; - SMU7_Fusion_ExtClkLevel AcpLevel [SMU7_MAX_LEVELS_ACP]; - SMU7_Fusion_ExtClkLevel SamuLevel [SMU7_MAX_LEVELS_SAMU]; - - uint8_t UvdBootLevel; - uint8_t VceBootLevel; - uint8_t AcpBootLevel; - uint8_t SamuBootLevel; - uint8_t UVDInterval; - uint8_t VCEInterval; - uint8_t ACPInterval; - uint8_t SAMUInterval; - - uint8_t GraphicsBootLevel; - uint8_t GraphicsInterval; - uint8_t GraphicsThermThrottleEnable; - uint8_t GraphicsVoltageChangeEnable; - - uint8_t GraphicsClkSlowEnable; - uint8_t GraphicsClkSlowDivider; - uint16_t FpsLowT; - - uint32_t DisplayCac; - uint32_t LowSclkInterruptT; - - uint32_t DRAM_LOG_ADDR_H; - uint32_t DRAM_LOG_ADDR_L; - uint32_t DRAM_LOG_PHY_ADDR_H; - uint32_t DRAM_LOG_PHY_ADDR_L; - uint32_t DRAM_LOG_BUFF_SIZE; - -}; - -struct SMU7_Fusion_GIODpmTable -{ - - SMU7_Fusion_GIOLevel GIOLevel [SMU7_MAX_LEVELS_GIO]; - - SMU7_PIDController GioPIDController; - - uint32_t GIOLevelCount; - - uint8_t Enable; - uint8_t GIOVoltageChangeEnable; - uint8_t GIOBootLevel; - uint8_t padding; - uint8_t padding1[2]; - uint8_t TargetState; - uint8_t CurrenttState; - uint8_t ThrottleOnHtc; - uint8_t ThermThrottleStatus; - uint8_t ThermThrottleTempSelect; - uint8_t ThermThrottleEnable; - uint16_t TemperatureLimitHigh; - uint16_t TemperatureLimitLow; - -}; - -typedef struct SMU7_Fusion_DpmTable SMU7_Fusion_DpmTable; -typedef struct SMU7_Fusion_GIODpmTable SMU7_Fusion_GIODpmTable; - -#pragma pack(pop) - -#endif - diff --git a/drivers/gpu/drm/amd/amdgpu/smu8_fusion.h b/drivers/gpu/drm/amd/amdgpu/smu8_fusion.h deleted file mode 100644 index 5c9cc3c0bbfa..000000000000 --- a/drivers/gpu/drm/amd/amdgpu/smu8_fusion.h +++ /dev/null @@ -1,127 +0,0 @@ -/* - * Copyright 2014 Advanced Micro Devices, Inc. - * - * Permission is hereby granted, free of charge, to any person obtaining a - * copy of this software and associated documentation files (the "Software"), - * to deal in the Software without restriction, including without limitation - * the rights to use, copy, modify, merge, publish, distribute, sublicense, - * and/or sell copies of the Software, and to permit persons to whom the - * Software is furnished to do so, subject to the following conditions: - * - * The above copyright notice and this permission notice shall be included in - * all copies or substantial portions of the Software. - * - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR - * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, - * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL - * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR - * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, - * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR - * OTHER DEALINGS IN THE SOFTWARE. - * - */ - -#ifndef SMU8_FUSION_H -#define SMU8_FUSION_H - -#include "smu8.h" - -#pragma pack(push, 1) - -#define SMU8_MAX_CUS 2 -#define SMU8_PSMS_PER_CU 4 -#define SMU8_CACS_PER_CU 4 - -struct SMU8_GfxCuPgScoreboard { - uint8_t Enabled; - uint8_t spare[3]; -}; - -struct SMU8_Port80MonitorTable { - uint32_t MmioAddress; - uint32_t MemoryBaseHi; - uint32_t MemoryBaseLo; - uint16_t MemoryBufferSize; - uint16_t MemoryPosition; - uint16_t PollingInterval; - uint8_t EnableCsrShadow; - uint8_t EnableDramShadow; -}; - -/* Clock Table Definitions */ -#define NUM_SCLK_LEVELS 8 -#define NUM_LCLK_LEVELS 8 -#define NUM_UVD_LEVELS 8 -#define NUM_ECLK_LEVELS 8 -#define NUM_ACLK_LEVELS 8 - -struct SMU8_Fusion_ClkLevel { - uint8_t GnbVid; - uint8_t GfxVid; - uint8_t DfsDid; - uint8_t DeepSleepDid; - uint32_t DfsBypass; - uint32_t Frequency; -}; - -struct SMU8_Fusion_SclkBreakdownTable { - struct SMU8_Fusion_ClkLevel ClkLevel[NUM_SCLK_LEVELS]; - struct SMU8_Fusion_ClkLevel DpmOffLevel; - /* SMU8_Fusion_ClkLevel PwrOffLevel; */ - uint32_t SclkValidMask; - uint32_t MaxSclkIndex; -}; - -struct SMU8_Fusion_LclkBreakdownTable { - struct SMU8_Fusion_ClkLevel ClkLevel[NUM_LCLK_LEVELS]; - struct SMU8_Fusion_ClkLevel DpmOffLevel; - /* SMU8_Fusion_ClkLevel PwrOffLevel; */ - uint32_t LclkValidMask; - uint32_t MaxLclkIndex; -}; - -struct SMU8_Fusion_EclkBreakdownTable { - struct SMU8_Fusion_ClkLevel ClkLevel[NUM_ECLK_LEVELS]; - struct SMU8_Fusion_ClkLevel DpmOffLevel; - struct SMU8_Fusion_ClkLevel PwrOffLevel; - uint32_t EclkValidMask; - uint32_t MaxEclkIndex; -}; - -struct SMU8_Fusion_VclkBreakdownTable { - struct SMU8_Fusion_ClkLevel ClkLevel[NUM_UVD_LEVELS]; - struct SMU8_Fusion_ClkLevel DpmOffLevel; - struct SMU8_Fusion_ClkLevel PwrOffLevel; - uint32_t VclkValidMask; - uint32_t MaxVclkIndex; -}; - -struct SMU8_Fusion_DclkBreakdownTable { - struct SMU8_Fusion_ClkLevel ClkLevel[NUM_UVD_LEVELS]; - struct SMU8_Fusion_ClkLevel DpmOffLevel; - struct SMU8_Fusion_ClkLevel PwrOffLevel; - uint32_t DclkValidMask; - uint32_t MaxDclkIndex; -}; - -struct SMU8_Fusion_AclkBreakdownTable { - struct SMU8_Fusion_ClkLevel ClkLevel[NUM_ACLK_LEVELS]; - struct SMU8_Fusion_ClkLevel DpmOffLevel; - struct SMU8_Fusion_ClkLevel PwrOffLevel; - uint32_t AclkValidMask; - uint32_t MaxAclkIndex; -}; - - -struct SMU8_Fusion_ClkTable { - struct SMU8_Fusion_SclkBreakdownTable SclkBreakdownTable; - struct SMU8_Fusion_LclkBreakdownTable LclkBreakdownTable; - struct SMU8_Fusion_EclkBreakdownTable EclkBreakdownTable; - struct SMU8_Fusion_VclkBreakdownTable VclkBreakdownTable; - struct SMU8_Fusion_DclkBreakdownTable DclkBreakdownTable; - struct SMU8_Fusion_AclkBreakdownTable AclkBreakdownTable; -}; - -#pragma pack(pop) - -#endif diff --git a/drivers/gpu/drm/amd/amdgpu/smu_ucode_xfer_cz.h b/drivers/gpu/drm/amd/amdgpu/smu_ucode_xfer_cz.h deleted file mode 100644 index f8ba071f39c8..000000000000 --- a/drivers/gpu/drm/amd/amdgpu/smu_ucode_xfer_cz.h +++ /dev/null @@ -1,147 +0,0 @@ -// CZ Ucode Loading Definitions -#ifndef SMU_UCODE_XFER_CZ_H -#define SMU_UCODE_XFER_CZ_H - -#define NUM_JOBLIST_ENTRIES 32 - -#define TASK_TYPE_NO_ACTION 0 -#define TASK_TYPE_UCODE_LOAD 1 -#define TASK_TYPE_UCODE_SAVE 2 -#define TASK_TYPE_REG_LOAD 3 -#define TASK_TYPE_REG_SAVE 4 -#define TASK_TYPE_INITIALIZE 5 - -#define TASK_ARG_REG_SMCIND 0 -#define TASK_ARG_REG_MMIO 1 -#define TASK_ARG_REG_FCH 2 -#define TASK_ARG_REG_UNB 3 - -#define TASK_ARG_INIT_MM_PWR_LOG 0 -#define TASK_ARG_INIT_CLK_TABLE 1 - -#define JOB_GFX_SAVE 0 -#define JOB_GFX_RESTORE 1 -#define JOB_FCH_SAVE 2 -#define JOB_FCH_RESTORE 3 -#define JOB_UNB_SAVE 4 -#define JOB_UNB_RESTORE 5 -#define JOB_GMC_SAVE 6 -#define JOB_GMC_RESTORE 7 -#define JOB_GNB_SAVE 8 -#define JOB_GNB_RESTORE 9 - -#define IGNORE_JOB 0xff -#define END_OF_TASK_LIST (uint16_t)0xffff - -// Size of DRAM regions (in bytes) requested by SMU: -#define SMU_DRAM_REQ_MM_PWR_LOG 48 - -#define UCODE_ID_SDMA0 0 -#define UCODE_ID_SDMA1 1 -#define UCODE_ID_CP_CE 2 -#define UCODE_ID_CP_PFP 3 -#define UCODE_ID_CP_ME 4 -#define UCODE_ID_CP_MEC_JT1 5 -#define UCODE_ID_CP_MEC_JT2 6 -#define UCODE_ID_GMCON_RENG 7 -#define UCODE_ID_RLC_G 8 -#define UCODE_ID_RLC_SCRATCH 9 -#define UCODE_ID_RLC_SRM_ARAM 10 -#define UCODE_ID_RLC_SRM_DRAM 11 -#define UCODE_ID_DMCU_ERAM 12 -#define UCODE_ID_DMCU_IRAM 13 - -#define UCODE_ID_SDMA0_MASK 0x00000001 -#define UCODE_ID_SDMA1_MASK 0x00000002 -#define UCODE_ID_CP_CE_MASK 0x00000004 -#define UCODE_ID_CP_PFP_MASK 0x00000008 -#define UCODE_ID_CP_ME_MASK 0x00000010 -#define UCODE_ID_CP_MEC_JT1_MASK 0x00000020 -#define UCODE_ID_CP_MEC_JT2_MASK 0x00000040 -#define UCODE_ID_GMCON_RENG_MASK 0x00000080 -#define UCODE_ID_RLC_G_MASK 0x00000100 -#define UCODE_ID_RLC_SCRATCH_MASK 0x00000200 -#define UCODE_ID_RLC_SRM_ARAM_MASK 0x00000400 -#define UCODE_ID_RLC_SRM_DRAM_MASK 0x00000800 -#define UCODE_ID_DMCU_ERAM_MASK 0x00001000 -#define UCODE_ID_DMCU_IRAM_MASK 0x00002000 - -#define UCODE_ID_SDMA0_SIZE_BYTE 10368 -#define UCODE_ID_SDMA1_SIZE_BYTE 10368 -#define UCODE_ID_CP_CE_SIZE_BYTE 8576 -#define UCODE_ID_CP_PFP_SIZE_BYTE 16768 -#define UCODE_ID_CP_ME_SIZE_BYTE 16768 -#define UCODE_ID_CP_MEC_JT1_SIZE_BYTE 384 -#define UCODE_ID_CP_MEC_JT2_SIZE_BYTE 384 -#define UCODE_ID_GMCON_RENG_SIZE_BYTE 4096 -#define UCODE_ID_RLC_G_SIZE_BYTE 2048 -#define UCODE_ID_RLC_SCRATCH_SIZE_BYTE 132 -#define UCODE_ID_RLC_SRM_ARAM_SIZE_BYTE 8192 -#define UCODE_ID_RLC_SRM_DRAM_SIZE_BYTE 4096 -#define UCODE_ID_DMCU_ERAM_SIZE_BYTE 24576 -#define UCODE_ID_DMCU_IRAM_SIZE_BYTE 1024 - -#define NUM_UCODES 14 - -typedef struct { - uint32_t high; - uint32_t low; -} data_64_t; - -struct SMU_Task { - uint8_t type; - uint8_t arg; - uint16_t next; - data_64_t addr; - uint32_t size_bytes; -}; -typedef struct SMU_Task SMU_Task; - -struct TOC { - uint8_t JobList[NUM_JOBLIST_ENTRIES]; - SMU_Task tasks[1]; -}; - -// META DATA COMMAND Definitions -#define METADATA_CMD_MODE0 0x00000103 -#define METADATA_CMD_MODE1 0x00000113 -#define METADATA_CMD_MODE2 0x00000123 -#define METADATA_CMD_MODE3 0x00000133 -#define METADATA_CMD_DELAY 0x00000203 -#define METADATA_CMD_CHNG_REGSPACE 0x00000303 -#define METADATA_PERFORM_ON_SAVE 0x00001000 -#define METADATA_PERFORM_ON_LOAD 0x00002000 -#define METADATA_CMD_ARG_MASK 0xFFFF0000 -#define METADATA_CMD_ARG_SHIFT 16 - -// Simple register addr/data fields -struct SMU_MetaData_Mode0 { - uint32_t register_address; - uint32_t register_data; -}; -typedef struct SMU_MetaData_Mode0 SMU_MetaData_Mode0; - -// Register addr/data with mask -struct SMU_MetaData_Mode1 { - uint32_t register_address; - uint32_t register_mask; - uint32_t register_data; -}; -typedef struct SMU_MetaData_Mode1 SMU_MetaData_Mode1; - -struct SMU_MetaData_Mode2 { - uint32_t register_address; - uint32_t register_mask; - uint32_t target_value; -}; -typedef struct SMU_MetaData_Mode2 SMU_MetaData_Mode2; - -// Always write data (even on a save operation) -struct SMU_MetaData_Mode3 { - uint32_t register_address; - uint32_t register_mask; - uint32_t register_data; -}; -typedef struct SMU_MetaData_Mode3 SMU_MetaData_Mode3; - -#endif diff --git a/drivers/gpu/drm/amd/amdgpu/tonga_dpm.c b/drivers/gpu/drm/amd/amdgpu/tonga_dpm.c index 204903897b4f..0497784b3652 100644 --- a/drivers/gpu/drm/amd/amdgpu/tonga_dpm.c +++ b/drivers/gpu/drm/amd/amdgpu/tonga_dpm.c @@ -24,7 +24,7 @@ #include <linux/firmware.h> #include "drmP.h" #include "amdgpu.h" -#include "tonga_smumgr.h" +#include "tonga_smum.h" MODULE_FIRMWARE("amdgpu/tonga_smc.bin"); @@ -122,25 +122,12 @@ static int tonga_dpm_hw_fini(void *handle) static int tonga_dpm_suspend(void *handle) { - return 0; + return tonga_dpm_hw_fini(handle); } static int tonga_dpm_resume(void *handle) { - int ret; - struct amdgpu_device *adev = (struct amdgpu_device *)handle; - - mutex_lock(&adev->pm.mutex); - - ret = tonga_smu_start(adev); - if (ret) { - DRM_ERROR("SMU start failed\n"); - goto fail; - } - -fail: - mutex_unlock(&adev->pm.mutex); - return ret; + return tonga_dpm_hw_init(handle); } static int tonga_dpm_set_clockgating_state(void *handle, diff --git a/drivers/gpu/drm/amd/amdgpu/tonga_ih.c b/drivers/gpu/drm/amd/amdgpu/tonga_ih.c index 743c372837aa..b6f7d7bff929 100644 --- a/drivers/gpu/drm/amd/amdgpu/tonga_ih.c +++ b/drivers/gpu/drm/amd/amdgpu/tonga_ih.c @@ -273,8 +273,14 @@ static void tonga_ih_set_rptr(struct amdgpu_device *adev) static int tonga_ih_early_init(void *handle) { struct amdgpu_device *adev = (struct amdgpu_device *)handle; + int ret; + + ret = amdgpu_irq_add_domain(adev); + if (ret) + return ret; tonga_ih_set_interrupt_funcs(adev); + return 0; } @@ -301,6 +307,7 @@ static int tonga_ih_sw_fini(void *handle) amdgpu_irq_fini(adev); amdgpu_ih_ring_fini(adev); + amdgpu_irq_add_domain(adev); return 0; } diff --git a/drivers/gpu/drm/amd/amdgpu/tonga_ppsmc.h b/drivers/gpu/drm/amd/amdgpu/tonga_ppsmc.h deleted file mode 100644 index 811781f69482..000000000000 --- a/drivers/gpu/drm/amd/amdgpu/tonga_ppsmc.h +++ /dev/null @@ -1,198 +0,0 @@ -/* - * Copyright 2014 Advanced Micro Devices, Inc. - * - * Permission is hereby granted, free of charge, to any person obtaining a - * copy of this software and associated documentation files (the "Software"), - * to deal in the Software without restriction, including without limitation - * the rights to use, copy, modify, merge, publish, distribute, sublicense, - * and/or sell copies of the Software, and to permit persons to whom the - * Software is furnished to do so, subject to the following conditions: - * - * The above copyright notice and this permission notice shall be included in - * all copies or substantial portions of the Software. - * - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR - * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, - * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL - * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR - * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, - * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR - * OTHER DEALINGS IN THE SOFTWARE. - * - */ - -#ifndef TONGA_PP_SMC_H -#define TONGA_PP_SMC_H - -#pragma pack(push, 1) - -#define PPSMC_SWSTATE_FLAG_DC 0x01 -#define PPSMC_SWSTATE_FLAG_UVD 0x02 -#define PPSMC_SWSTATE_FLAG_VCE 0x04 -#define PPSMC_SWSTATE_FLAG_PCIE_X1 0x08 - -#define PPSMC_THERMAL_PROTECT_TYPE_INTERNAL 0x00 -#define PPSMC_THERMAL_PROTECT_TYPE_EXTERNAL 0x01 -#define PPSMC_THERMAL_PROTECT_TYPE_NONE 0xff - -#define PPSMC_SYSTEMFLAG_GPIO_DC 0x01 -#define PPSMC_SYSTEMFLAG_STEPVDDC 0x02 -#define PPSMC_SYSTEMFLAG_GDDR5 0x04 - -#define PPSMC_SYSTEMFLAG_DISABLE_BABYSTEP 0x08 - -#define PPSMC_SYSTEMFLAG_REGULATOR_HOT 0x10 -#define PPSMC_SYSTEMFLAG_REGULATOR_HOT_ANALOG 0x20 -#define PPSMC_SYSTEMFLAG_12CHANNEL 0x40 - -#define PPSMC_EXTRAFLAGS_AC2DC_ACTION_MASK 0x07 -#define PPSMC_EXTRAFLAGS_AC2DC_DONT_WAIT_FOR_VBLANK 0x08 - -#define PPSMC_EXTRAFLAGS_AC2DC_ACTION_GOTODPMLOWSTATE 0x00 -#define PPSMC_EXTRAFLAGS_AC2DC_ACTION_GOTOINITIALSTATE 0x01 - -#define PPSMC_EXTRAFLAGS_AC2DC_GPIO5_POLARITY_HIGH 0x10 -#define PPSMC_EXTRAFLAGS_DRIVER_TO_GPIO17 0x20 -#define PPSMC_EXTRAFLAGS_PCC_TO_GPIO17 0x40 - -#define PPSMC_DPM2FLAGS_TDPCLMP 0x01 -#define PPSMC_DPM2FLAGS_PWRSHFT 0x02 -#define PPSMC_DPM2FLAGS_OCP 0x04 - -#define PPSMC_DISPLAY_WATERMARK_LOW 0 -#define PPSMC_DISPLAY_WATERMARK_HIGH 1 - -#define PPSMC_STATEFLAG_AUTO_PULSE_SKIP 0x01 -#define PPSMC_STATEFLAG_POWERBOOST 0x02 -#define PPSMC_STATEFLAG_PSKIP_ON_TDP_FAULT 0x04 -#define PPSMC_STATEFLAG_POWERSHIFT 0x08 -#define PPSMC_STATEFLAG_SLOW_READ_MARGIN 0x10 -#define PPSMC_STATEFLAG_DEEPSLEEP_THROTTLE 0x20 -#define PPSMC_STATEFLAG_DEEPSLEEP_BYPASS 0x40 - -#define FDO_MODE_HARDWARE 0 -#define FDO_MODE_PIECE_WISE_LINEAR 1 - -enum FAN_CONTROL { - FAN_CONTROL_FUZZY, - FAN_CONTROL_TABLE -}; - -#define PPSMC_Result_OK ((uint16_t)0x01) -#define PPSMC_Result_NoMore ((uint16_t)0x02) -#define PPSMC_Result_NotNow ((uint16_t)0x03) -#define PPSMC_Result_Failed ((uint16_t)0xFF) -#define PPSMC_Result_UnknownCmd ((uint16_t)0xFE) -#define PPSMC_Result_UnknownVT ((uint16_t)0xFD) - -typedef uint16_t PPSMC_Result; - -#define PPSMC_isERROR(x) ((uint16_t)0x80 & (x)) - -#define PPSMC_MSG_Halt ((uint16_t)0x10) -#define PPSMC_MSG_Resume ((uint16_t)0x11) -#define PPSMC_MSG_EnableDPMLevel ((uint16_t)0x12) -#define PPSMC_MSG_ZeroLevelsDisabled ((uint16_t)0x13) -#define PPSMC_MSG_OneLevelsDisabled ((uint16_t)0x14) -#define PPSMC_MSG_TwoLevelsDisabled ((uint16_t)0x15) -#define PPSMC_MSG_EnableThermalInterrupt ((uint16_t)0x16) -#define PPSMC_MSG_RunningOnAC ((uint16_t)0x17) -#define PPSMC_MSG_LevelUp ((uint16_t)0x18) -#define PPSMC_MSG_LevelDown ((uint16_t)0x19) -#define PPSMC_MSG_ResetDPMCounters ((uint16_t)0x1a) -#define PPSMC_MSG_SwitchToSwState ((uint16_t)0x20) -#define PPSMC_MSG_SwitchToSwStateLast ((uint16_t)0x3f) -#define PPSMC_MSG_SwitchToInitialState ((uint16_t)0x40) -#define PPSMC_MSG_NoForcedLevel ((uint16_t)0x41) -#define PPSMC_MSG_ForceHigh ((uint16_t)0x42) -#define PPSMC_MSG_ForceMediumOrHigh ((uint16_t)0x43) -#define PPSMC_MSG_SwitchToMinimumPower ((uint16_t)0x51) -#define PPSMC_MSG_ResumeFromMinimumPower ((uint16_t)0x52) -#define PPSMC_MSG_EnableCac ((uint16_t)0x53) -#define PPSMC_MSG_DisableCac ((uint16_t)0x54) -#define PPSMC_DPMStateHistoryStart ((uint16_t)0x55) -#define PPSMC_DPMStateHistoryStop ((uint16_t)0x56) -#define PPSMC_CACHistoryStart ((uint16_t)0x57) -#define PPSMC_CACHistoryStop ((uint16_t)0x58) -#define PPSMC_TDPClampingActive ((uint16_t)0x59) -#define PPSMC_TDPClampingInactive ((uint16_t)0x5A) -#define PPSMC_StartFanControl ((uint16_t)0x5B) -#define PPSMC_StopFanControl ((uint16_t)0x5C) -#define PPSMC_NoDisplay ((uint16_t)0x5D) -#define PPSMC_HasDisplay ((uint16_t)0x5E) -#define PPSMC_MSG_UVDPowerOFF ((uint16_t)0x60) -#define PPSMC_MSG_UVDPowerON ((uint16_t)0x61) -#define PPSMC_MSG_EnableULV ((uint16_t)0x62) -#define PPSMC_MSG_DisableULV ((uint16_t)0x63) -#define PPSMC_MSG_EnterULV ((uint16_t)0x64) -#define PPSMC_MSG_ExitULV ((uint16_t)0x65) -#define PPSMC_PowerShiftActive ((uint16_t)0x6A) -#define PPSMC_PowerShiftInactive ((uint16_t)0x6B) -#define PPSMC_OCPActive ((uint16_t)0x6C) -#define PPSMC_OCPInactive ((uint16_t)0x6D) -#define PPSMC_CACLongTermAvgEnable ((uint16_t)0x6E) -#define PPSMC_CACLongTermAvgDisable ((uint16_t)0x6F) -#define PPSMC_MSG_InferredStateSweep_Start ((uint16_t)0x70) -#define PPSMC_MSG_InferredStateSweep_Stop ((uint16_t)0x71) -#define PPSMC_MSG_SwitchToLowestInfState ((uint16_t)0x72) -#define PPSMC_MSG_SwitchToNonInfState ((uint16_t)0x73) -#define PPSMC_MSG_AllStateSweep_Start ((uint16_t)0x74) -#define PPSMC_MSG_AllStateSweep_Stop ((uint16_t)0x75) -#define PPSMC_MSG_SwitchNextLowerInfState ((uint16_t)0x76) -#define PPSMC_MSG_SwitchNextHigherInfState ((uint16_t)0x77) -#define PPSMC_MSG_MclkRetrainingTest ((uint16_t)0x78) -#define PPSMC_MSG_ForceTDPClamping ((uint16_t)0x79) -#define PPSMC_MSG_CollectCAC_PowerCorreln ((uint16_t)0x7A) -#define PPSMC_MSG_CollectCAC_WeightCalib ((uint16_t)0x7B) -#define PPSMC_MSG_CollectCAC_SQonly ((uint16_t)0x7C) -#define PPSMC_MSG_CollectCAC_TemperaturePwr ((uint16_t)0x7D) -#define PPSMC_MSG_ExtremitiesTest_Start ((uint16_t)0x7E) -#define PPSMC_MSG_ExtremitiesTest_Stop ((uint16_t)0x7F) -#define PPSMC_FlushDataCache ((uint16_t)0x80) -#define PPSMC_FlushInstrCache ((uint16_t)0x81) -#define PPSMC_MSG_SetEnabledLevels ((uint16_t)0x82) -#define PPSMC_MSG_SetForcedLevels ((uint16_t)0x83) -#define PPSMC_MSG_ResetToDefaults ((uint16_t)0x84) -#define PPSMC_MSG_SetForcedLevelsAndJump ((uint16_t)0x85) -#define PPSMC_MSG_SetCACHistoryMode ((uint16_t)0x86) -#define PPSMC_MSG_EnableDTE ((uint16_t)0x87) -#define PPSMC_MSG_DisableDTE ((uint16_t)0x88) -#define PPSMC_MSG_SmcSpaceSetAddress ((uint16_t)0x89) -#define PPSMC_MSG_SmcSpaceWriteDWordInc ((uint16_t)0x8A) -#define PPSMC_MSG_SmcSpaceWriteWordInc ((uint16_t)0x8B) -#define PPSMC_MSG_SmcSpaceWriteByteInc ((uint16_t)0x8C) -#define PPSMC_MSG_ChangeNearTDPLimit ((uint16_t)0x90) -#define PPSMC_MSG_ChangeSafePowerLimit ((uint16_t)0x91) -#define PPSMC_MSG_DPMStateSweepStart ((uint16_t)0x92) -#define PPSMC_MSG_DPMStateSweepStop ((uint16_t)0x93) -#define PPSMC_MSG_OVRDDisableSCLKDS ((uint16_t)0x94) -#define PPSMC_MSG_CancelDisableOVRDSCLKDS ((uint16_t)0x95) -#define PPSMC_MSG_ThrottleOVRDSCLKDS ((uint16_t)0x96) -#define PPSMC_MSG_CancelThrottleOVRDSCLKDS ((uint16_t)0x97) -#define PPSMC_MSG_GPIO17 ((uint16_t)0x98) -#define PPSMC_MSG_API_SetSvi2Volt_Vddc ((uint16_t)0x99) -#define PPSMC_MSG_API_SetSvi2Volt_Vddci ((uint16_t)0x9A) -#define PPSMC_MSG_API_SetSvi2Volt_Mvdd ((uint16_t)0x9B) -#define PPSMC_MSG_API_GetSvi2Volt_Vddc ((uint16_t)0x9C) -#define PPSMC_MSG_API_GetSvi2Volt_Vddci ((uint16_t)0x9D) -#define PPSMC_MSG_API_GetSvi2Volt_Mvdd ((uint16_t)0x9E) - -#define PPSMC_MSG_BREAK ((uint16_t)0xF8) - -#define PPSMC_MSG_Test ((uint16_t)0x100) -#define PPSMC_MSG_DRV_DRAM_ADDR_HI ((uint16_t)0x250) -#define PPSMC_MSG_DRV_DRAM_ADDR_LO ((uint16_t)0x251) -#define PPSMC_MSG_SMU_DRAM_ADDR_HI ((uint16_t)0x252) -#define PPSMC_MSG_SMU_DRAM_ADDR_LO ((uint16_t)0x253) -#define PPSMC_MSG_LoadUcodes ((uint16_t)0x254) - -typedef uint16_t PPSMC_Msg; - -#define PPSMC_EVENT_STATUS_THERMAL 0x00000001 -#define PPSMC_EVENT_STATUS_REGULATORHOT 0x00000002 -#define PPSMC_EVENT_STATUS_DC 0x00000004 -#define PPSMC_EVENT_STATUS_GPIO17 0x00000008 - -#pragma pack(pop) - -#endif diff --git a/drivers/gpu/drm/amd/amdgpu/tonga_smc.c b/drivers/gpu/drm/amd/amdgpu/tonga_smc.c index 5421309c1862..083893dd68c0 100644 --- a/drivers/gpu/drm/amd/amdgpu/tonga_smc.c +++ b/drivers/gpu/drm/amd/amdgpu/tonga_smc.c @@ -25,7 +25,7 @@ #include "drmP.h" #include "amdgpu.h" #include "tonga_ppsmc.h" -#include "tonga_smumgr.h" +#include "tonga_smum.h" #include "smu_ucode_xfer_vi.h" #include "amdgpu_ucode.h" @@ -272,6 +272,12 @@ static int tonga_smu_upload_firmware_image(struct amdgpu_device *adev) if (!adev->pm.fw) return -EINVAL; + /* Skip SMC ucode loading on SR-IOV capable boards. + * vbios does this for us in asic_init in that case. + */ + if (adev->virtualization.supports_sr_iov) + return 0; + hdr = (const struct smc_firmware_header_v1_0 *)adev->pm.fw->data; amdgpu_ucode_print_smc_hdr(&hdr->header); diff --git a/drivers/gpu/drm/amd/amdgpu/tonga_smumgr.h b/drivers/gpu/drm/amd/amdgpu/tonga_smum.h index c031ff99fe3e..c031ff99fe3e 100644 --- a/drivers/gpu/drm/amd/amdgpu/tonga_smumgr.h +++ b/drivers/gpu/drm/amd/amdgpu/tonga_smum.h diff --git a/drivers/gpu/drm/amd/amdgpu/uvd_v4_2.c b/drivers/gpu/drm/amd/amdgpu/uvd_v4_2.c index ed50dd725788..c606ccb38d8b 100644 --- a/drivers/gpu/drm/amd/amdgpu/uvd_v4_2.c +++ b/drivers/gpu/drm/amd/amdgpu/uvd_v4_2.c @@ -164,7 +164,7 @@ static int uvd_v4_2_hw_init(void *handle) goto done; } - r = amdgpu_ring_lock(ring, 10); + r = amdgpu_ring_alloc(ring, 10); if (r) { DRM_ERROR("amdgpu: ring failed to lock UVD ring (%d).\n", r); goto done; @@ -189,7 +189,7 @@ static int uvd_v4_2_hw_init(void *handle) amdgpu_ring_write(ring, PACKET0(mmUVD_SEMA_CNTL, 0)); amdgpu_ring_write(ring, 3); - amdgpu_ring_unlock_commit(ring); + amdgpu_ring_commit(ring); done: /* lower clocks again */ @@ -439,33 +439,6 @@ static void uvd_v4_2_ring_emit_fence(struct amdgpu_ring *ring, u64 addr, u64 seq } /** - * uvd_v4_2_ring_emit_semaphore - emit semaphore command - * - * @ring: amdgpu_ring pointer - * @semaphore: semaphore to emit commands for - * @emit_wait: true if we should emit a wait command - * - * Emit a semaphore command (either wait or signal) to the UVD ring. - */ -static bool uvd_v4_2_ring_emit_semaphore(struct amdgpu_ring *ring, - struct amdgpu_semaphore *semaphore, - bool emit_wait) -{ - uint64_t addr = semaphore->gpu_addr; - - amdgpu_ring_write(ring, PACKET0(mmUVD_SEMA_ADDR_LOW, 0)); - amdgpu_ring_write(ring, (addr >> 3) & 0x000FFFFF); - - amdgpu_ring_write(ring, PACKET0(mmUVD_SEMA_ADDR_HIGH, 0)); - amdgpu_ring_write(ring, (addr >> 23) & 0x000FFFFF); - - amdgpu_ring_write(ring, PACKET0(mmUVD_SEMA_CMD, 0)); - amdgpu_ring_write(ring, 0x80 | (emit_wait ? 1 : 0)); - - return true; -} - -/** * uvd_v4_2_ring_test_ring - register write test * * @ring: amdgpu_ring pointer @@ -480,7 +453,7 @@ static int uvd_v4_2_ring_test_ring(struct amdgpu_ring *ring) int r; WREG32(mmUVD_CONTEXT_ID, 0xCAFEDEAD); - r = amdgpu_ring_lock(ring, 3); + r = amdgpu_ring_alloc(ring, 3); if (r) { DRM_ERROR("amdgpu: cp failed to lock ring %d (%d).\n", ring->idx, r); @@ -488,7 +461,7 @@ static int uvd_v4_2_ring_test_ring(struct amdgpu_ring *ring) } amdgpu_ring_write(ring, PACKET0(mmUVD_CONTEXT_ID, 0)); amdgpu_ring_write(ring, 0xDEADBEEF); - amdgpu_ring_unlock_commit(ring); + amdgpu_ring_commit(ring); for (i = 0; i < adev->usec_timeout; i++) { tmp = RREG32(mmUVD_CONTEXT_ID); if (tmp == 0xDEADBEEF) @@ -549,7 +522,7 @@ static int uvd_v4_2_ring_test_ib(struct amdgpu_ring *ring) goto error; } - r = amdgpu_uvd_get_destroy_msg(ring, 1, &fence); + r = amdgpu_uvd_get_destroy_msg(ring, 1, true, &fence); if (r) { DRM_ERROR("amdgpu: failed to get destroy ib (%d).\n", r); goto error; @@ -603,6 +576,10 @@ static void uvd_v4_2_mc_resume(struct amdgpu_device *adev) addr = (adev->uvd.gpu_addr >> 32) & 0xFF; WREG32(mmUVD_LMI_EXT40_ADDR, addr | (0x9 << 16) | (0x1 << 31)); + WREG32(mmUVD_UDEC_ADDR_CONFIG, adev->gfx.config.gb_addr_config); + WREG32(mmUVD_UDEC_DB_ADDR_CONFIG, adev->gfx.config.gb_addr_config); + WREG32(mmUVD_UDEC_DBW_ADDR_CONFIG, adev->gfx.config.gb_addr_config); + uvd_v4_2_init_cg(adev); } @@ -611,7 +588,7 @@ static void uvd_v4_2_enable_mgcg(struct amdgpu_device *adev, { u32 orig, data; - if (enable && (adev->cg_flags & AMDGPU_CG_SUPPORT_UVD_MGCG)) { + if (enable && (adev->cg_flags & AMD_CG_SUPPORT_UVD_MGCG)) { data = RREG32_UVD_CTX(ixUVD_CGC_MEM_CTRL); data = 0xfff; WREG32_UVD_CTX(ixUVD_CGC_MEM_CTRL, data); @@ -804,6 +781,13 @@ static void uvd_v4_2_print_status(void *handle) RREG32(mmUVD_SEMA_SIGNAL_INCOMPLETE_TIMEOUT_CNTL)); dev_info(adev->dev, " UVD_CONTEXT_ID=0x%08X\n", RREG32(mmUVD_CONTEXT_ID)); + dev_info(adev->dev, " UVD_UDEC_ADDR_CONFIG=0x%08X\n", + RREG32(mmUVD_UDEC_ADDR_CONFIG)); + dev_info(adev->dev, " UVD_UDEC_DB_ADDR_CONFIG=0x%08X\n", + RREG32(mmUVD_UDEC_DB_ADDR_CONFIG)); + dev_info(adev->dev, " UVD_UDEC_DBW_ADDR_CONFIG=0x%08X\n", + RREG32(mmUVD_UDEC_DBW_ADDR_CONFIG)); + } static int uvd_v4_2_set_interrupt_state(struct amdgpu_device *adev, @@ -830,6 +814,9 @@ static int uvd_v4_2_set_clockgating_state(void *handle, bool gate = false; struct amdgpu_device *adev = (struct amdgpu_device *)handle; + if (!(adev->cg_flags & AMD_CG_SUPPORT_UVD_MGCG)) + return 0; + if (state == AMD_CG_STATE_GATE) gate = true; @@ -848,7 +835,10 @@ static int uvd_v4_2_set_powergating_state(void *handle, * revisit this when there is a cleaner line between * the smc and the hw blocks */ - struct amdgpu_device *adev = (struct amdgpu_device *)handle; + struct amdgpu_device *adev = (struct amdgpu_device *)handle; + + if (!(adev->pg_flags & AMD_PG_SUPPORT_UVD)) + return 0; if (state == AMD_PG_STATE_GATE) { uvd_v4_2_stop(adev); @@ -882,11 +872,10 @@ static const struct amdgpu_ring_funcs uvd_v4_2_ring_funcs = { .parse_cs = amdgpu_uvd_ring_parse_cs, .emit_ib = uvd_v4_2_ring_emit_ib, .emit_fence = uvd_v4_2_ring_emit_fence, - .emit_semaphore = uvd_v4_2_ring_emit_semaphore, .test_ring = uvd_v4_2_ring_test_ring, .test_ib = uvd_v4_2_ring_test_ib, - .is_lockup = amdgpu_ring_test_lockup, .insert_nop = amdgpu_ring_insert_nop, + .pad_ib = amdgpu_ring_generic_pad_ib, }; static void uvd_v4_2_set_ring_funcs(struct amdgpu_device *adev) diff --git a/drivers/gpu/drm/amd/amdgpu/uvd_v5_0.c b/drivers/gpu/drm/amd/amdgpu/uvd_v5_0.c index 9ad8b9906c0b..e3c852d9d79a 100644 --- a/drivers/gpu/drm/amd/amdgpu/uvd_v5_0.c +++ b/drivers/gpu/drm/amd/amdgpu/uvd_v5_0.c @@ -160,7 +160,7 @@ static int uvd_v5_0_hw_init(void *handle) goto done; } - r = amdgpu_ring_lock(ring, 10); + r = amdgpu_ring_alloc(ring, 10); if (r) { DRM_ERROR("amdgpu: ring failed to lock UVD ring (%d).\n", r); goto done; @@ -185,7 +185,7 @@ static int uvd_v5_0_hw_init(void *handle) amdgpu_ring_write(ring, PACKET0(mmUVD_SEMA_CNTL, 0)); amdgpu_ring_write(ring, 3); - amdgpu_ring_unlock_commit(ring); + amdgpu_ring_commit(ring); done: /* lower clocks again */ @@ -279,6 +279,10 @@ static void uvd_v5_0_mc_resume(struct amdgpu_device *adev) size = AMDGPU_UVD_HEAP_SIZE; WREG32(mmUVD_VCPU_CACHE_OFFSET2, offset >> 3); WREG32(mmUVD_VCPU_CACHE_SIZE2, size); + + WREG32(mmUVD_UDEC_ADDR_CONFIG, adev->gfx.config.gb_addr_config); + WREG32(mmUVD_UDEC_DB_ADDR_CONFIG, adev->gfx.config.gb_addr_config); + WREG32(mmUVD_UDEC_DBW_ADDR_CONFIG, adev->gfx.config.gb_addr_config); } /** @@ -483,33 +487,6 @@ static void uvd_v5_0_ring_emit_fence(struct amdgpu_ring *ring, u64 addr, u64 seq } /** - * uvd_v5_0_ring_emit_semaphore - emit semaphore command - * - * @ring: amdgpu_ring pointer - * @semaphore: semaphore to emit commands for - * @emit_wait: true if we should emit a wait command - * - * Emit a semaphore command (either wait or signal) to the UVD ring. - */ -static bool uvd_v5_0_ring_emit_semaphore(struct amdgpu_ring *ring, - struct amdgpu_semaphore *semaphore, - bool emit_wait) -{ - uint64_t addr = semaphore->gpu_addr; - - amdgpu_ring_write(ring, PACKET0(mmUVD_SEMA_ADDR_LOW, 0)); - amdgpu_ring_write(ring, (addr >> 3) & 0x000FFFFF); - - amdgpu_ring_write(ring, PACKET0(mmUVD_SEMA_ADDR_HIGH, 0)); - amdgpu_ring_write(ring, (addr >> 23) & 0x000FFFFF); - - amdgpu_ring_write(ring, PACKET0(mmUVD_SEMA_CMD, 0)); - amdgpu_ring_write(ring, 0x80 | (emit_wait ? 1 : 0)); - - return true; -} - -/** * uvd_v5_0_ring_test_ring - register write test * * @ring: amdgpu_ring pointer @@ -524,7 +501,7 @@ static int uvd_v5_0_ring_test_ring(struct amdgpu_ring *ring) int r; WREG32(mmUVD_CONTEXT_ID, 0xCAFEDEAD); - r = amdgpu_ring_lock(ring, 3); + r = amdgpu_ring_alloc(ring, 3); if (r) { DRM_ERROR("amdgpu: cp failed to lock ring %d (%d).\n", ring->idx, r); @@ -532,7 +509,7 @@ static int uvd_v5_0_ring_test_ring(struct amdgpu_ring *ring) } amdgpu_ring_write(ring, PACKET0(mmUVD_CONTEXT_ID, 0)); amdgpu_ring_write(ring, 0xDEADBEEF); - amdgpu_ring_unlock_commit(ring); + amdgpu_ring_commit(ring); for (i = 0; i < adev->usec_timeout; i++) { tmp = RREG32(mmUVD_CONTEXT_ID); if (tmp == 0xDEADBEEF) @@ -595,7 +572,7 @@ static int uvd_v5_0_ring_test_ib(struct amdgpu_ring *ring) goto error; } - r = amdgpu_uvd_get_destroy_msg(ring, 1, &fence); + r = amdgpu_uvd_get_destroy_msg(ring, 1, true, &fence); if (r) { DRM_ERROR("amdgpu: failed to get destroy ib (%d).\n", r); goto error; @@ -751,6 +728,12 @@ static void uvd_v5_0_print_status(void *handle) RREG32(mmUVD_SEMA_SIGNAL_INCOMPLETE_TIMEOUT_CNTL)); dev_info(adev->dev, " UVD_CONTEXT_ID=0x%08X\n", RREG32(mmUVD_CONTEXT_ID)); + dev_info(adev->dev, " UVD_UDEC_ADDR_CONFIG=0x%08X\n", + RREG32(mmUVD_UDEC_ADDR_CONFIG)); + dev_info(adev->dev, " UVD_UDEC_DB_ADDR_CONFIG=0x%08X\n", + RREG32(mmUVD_UDEC_DB_ADDR_CONFIG)); + dev_info(adev->dev, " UVD_UDEC_DBW_ADDR_CONFIG=0x%08X\n", + RREG32(mmUVD_UDEC_DBW_ADDR_CONFIG)); } static int uvd_v5_0_set_interrupt_state(struct amdgpu_device *adev, @@ -774,6 +757,11 @@ static int uvd_v5_0_process_interrupt(struct amdgpu_device *adev, static int uvd_v5_0_set_clockgating_state(void *handle, enum amd_clockgating_state state) { + struct amdgpu_device *adev = (struct amdgpu_device *)handle; + + if (!(adev->cg_flags & AMD_CG_SUPPORT_UVD_MGCG)) + return 0; + return 0; } @@ -789,6 +777,9 @@ static int uvd_v5_0_set_powergating_state(void *handle, */ struct amdgpu_device *adev = (struct amdgpu_device *)handle; + if (!(adev->pg_flags & AMD_PG_SUPPORT_UVD)) + return 0; + if (state == AMD_PG_STATE_GATE) { uvd_v5_0_stop(adev); return 0; @@ -821,11 +812,10 @@ static const struct amdgpu_ring_funcs uvd_v5_0_ring_funcs = { .parse_cs = amdgpu_uvd_ring_parse_cs, .emit_ib = uvd_v5_0_ring_emit_ib, .emit_fence = uvd_v5_0_ring_emit_fence, - .emit_semaphore = uvd_v5_0_ring_emit_semaphore, .test_ring = uvd_v5_0_ring_test_ring, .test_ib = uvd_v5_0_ring_test_ib, - .is_lockup = amdgpu_ring_test_lockup, .insert_nop = amdgpu_ring_insert_nop, + .pad_ib = amdgpu_ring_generic_pad_ib, }; static void uvd_v5_0_set_ring_funcs(struct amdgpu_device *adev) diff --git a/drivers/gpu/drm/amd/amdgpu/uvd_v6_0.c b/drivers/gpu/drm/amd/amdgpu/uvd_v6_0.c index 7e9934fa4193..3375e614ac67 100644 --- a/drivers/gpu/drm/amd/amdgpu/uvd_v6_0.c +++ b/drivers/gpu/drm/amd/amdgpu/uvd_v6_0.c @@ -157,7 +157,7 @@ static int uvd_v6_0_hw_init(void *handle) goto done; } - r = amdgpu_ring_lock(ring, 10); + r = amdgpu_ring_alloc(ring, 10); if (r) { DRM_ERROR("amdgpu: ring failed to lock UVD ring (%d).\n", r); goto done; @@ -182,7 +182,7 @@ static int uvd_v6_0_hw_init(void *handle) amdgpu_ring_write(ring, PACKET0(mmUVD_SEMA_CNTL, 0)); amdgpu_ring_write(ring, 3); - amdgpu_ring_unlock_commit(ring); + amdgpu_ring_commit(ring); done: if (!r) @@ -277,6 +277,238 @@ static void uvd_v6_0_mc_resume(struct amdgpu_device *adev) size = AMDGPU_UVD_HEAP_SIZE; WREG32(mmUVD_VCPU_CACHE_OFFSET2, offset >> 3); WREG32(mmUVD_VCPU_CACHE_SIZE2, size); + + WREG32(mmUVD_UDEC_ADDR_CONFIG, adev->gfx.config.gb_addr_config); + WREG32(mmUVD_UDEC_DB_ADDR_CONFIG, adev->gfx.config.gb_addr_config); + WREG32(mmUVD_UDEC_DBW_ADDR_CONFIG, adev->gfx.config.gb_addr_config); +} + +static void cz_set_uvd_clock_gating_branches(struct amdgpu_device *adev, + bool enable) +{ + u32 data, data1; + + data = RREG32(mmUVD_CGC_GATE); + data1 = RREG32(mmUVD_SUVD_CGC_GATE); + if (enable) { + data |= UVD_CGC_GATE__SYS_MASK | + UVD_CGC_GATE__UDEC_MASK | + UVD_CGC_GATE__MPEG2_MASK | + UVD_CGC_GATE__RBC_MASK | + UVD_CGC_GATE__LMI_MC_MASK | + UVD_CGC_GATE__IDCT_MASK | + UVD_CGC_GATE__MPRD_MASK | + UVD_CGC_GATE__MPC_MASK | + UVD_CGC_GATE__LBSI_MASK | + UVD_CGC_GATE__LRBBM_MASK | + UVD_CGC_GATE__UDEC_RE_MASK | + UVD_CGC_GATE__UDEC_CM_MASK | + UVD_CGC_GATE__UDEC_IT_MASK | + UVD_CGC_GATE__UDEC_DB_MASK | + UVD_CGC_GATE__UDEC_MP_MASK | + UVD_CGC_GATE__WCB_MASK | + UVD_CGC_GATE__VCPU_MASK | + UVD_CGC_GATE__SCPU_MASK; + data1 |= UVD_SUVD_CGC_GATE__SRE_MASK | + UVD_SUVD_CGC_GATE__SIT_MASK | + UVD_SUVD_CGC_GATE__SMP_MASK | + UVD_SUVD_CGC_GATE__SCM_MASK | + UVD_SUVD_CGC_GATE__SDB_MASK | + UVD_SUVD_CGC_GATE__SRE_H264_MASK | + UVD_SUVD_CGC_GATE__SRE_HEVC_MASK | + UVD_SUVD_CGC_GATE__SIT_H264_MASK | + UVD_SUVD_CGC_GATE__SIT_HEVC_MASK | + UVD_SUVD_CGC_GATE__SCM_H264_MASK | + UVD_SUVD_CGC_GATE__SCM_HEVC_MASK | + UVD_SUVD_CGC_GATE__SDB_H264_MASK | + UVD_SUVD_CGC_GATE__SDB_HEVC_MASK; + } else { + data &= ~(UVD_CGC_GATE__SYS_MASK | + UVD_CGC_GATE__UDEC_MASK | + UVD_CGC_GATE__MPEG2_MASK | + UVD_CGC_GATE__RBC_MASK | + UVD_CGC_GATE__LMI_MC_MASK | + UVD_CGC_GATE__LMI_UMC_MASK | + UVD_CGC_GATE__IDCT_MASK | + UVD_CGC_GATE__MPRD_MASK | + UVD_CGC_GATE__MPC_MASK | + UVD_CGC_GATE__LBSI_MASK | + UVD_CGC_GATE__LRBBM_MASK | + UVD_CGC_GATE__UDEC_RE_MASK | + UVD_CGC_GATE__UDEC_CM_MASK | + UVD_CGC_GATE__UDEC_IT_MASK | + UVD_CGC_GATE__UDEC_DB_MASK | + UVD_CGC_GATE__UDEC_MP_MASK | + UVD_CGC_GATE__WCB_MASK | + UVD_CGC_GATE__VCPU_MASK | + UVD_CGC_GATE__SCPU_MASK); + data1 &= ~(UVD_SUVD_CGC_GATE__SRE_MASK | + UVD_SUVD_CGC_GATE__SIT_MASK | + UVD_SUVD_CGC_GATE__SMP_MASK | + UVD_SUVD_CGC_GATE__SCM_MASK | + UVD_SUVD_CGC_GATE__SDB_MASK | + UVD_SUVD_CGC_GATE__SRE_H264_MASK | + UVD_SUVD_CGC_GATE__SRE_HEVC_MASK | + UVD_SUVD_CGC_GATE__SIT_H264_MASK | + UVD_SUVD_CGC_GATE__SIT_HEVC_MASK | + UVD_SUVD_CGC_GATE__SCM_H264_MASK | + UVD_SUVD_CGC_GATE__SCM_HEVC_MASK | + UVD_SUVD_CGC_GATE__SDB_H264_MASK | + UVD_SUVD_CGC_GATE__SDB_HEVC_MASK); + } + WREG32(mmUVD_CGC_GATE, data); + WREG32(mmUVD_SUVD_CGC_GATE, data1); +} + +static void tonga_set_uvd_clock_gating_branches(struct amdgpu_device *adev, + bool enable) +{ + u32 data, data1; + + data = RREG32(mmUVD_CGC_GATE); + data1 = RREG32(mmUVD_SUVD_CGC_GATE); + if (enable) { + data |= UVD_CGC_GATE__SYS_MASK | + UVD_CGC_GATE__UDEC_MASK | + UVD_CGC_GATE__MPEG2_MASK | + UVD_CGC_GATE__RBC_MASK | + UVD_CGC_GATE__LMI_MC_MASK | + UVD_CGC_GATE__IDCT_MASK | + UVD_CGC_GATE__MPRD_MASK | + UVD_CGC_GATE__MPC_MASK | + UVD_CGC_GATE__LBSI_MASK | + UVD_CGC_GATE__LRBBM_MASK | + UVD_CGC_GATE__UDEC_RE_MASK | + UVD_CGC_GATE__UDEC_CM_MASK | + UVD_CGC_GATE__UDEC_IT_MASK | + UVD_CGC_GATE__UDEC_DB_MASK | + UVD_CGC_GATE__UDEC_MP_MASK | + UVD_CGC_GATE__WCB_MASK | + UVD_CGC_GATE__VCPU_MASK | + UVD_CGC_GATE__SCPU_MASK; + data1 |= UVD_SUVD_CGC_GATE__SRE_MASK | + UVD_SUVD_CGC_GATE__SIT_MASK | + UVD_SUVD_CGC_GATE__SMP_MASK | + UVD_SUVD_CGC_GATE__SCM_MASK | + UVD_SUVD_CGC_GATE__SDB_MASK; + } else { + data &= ~(UVD_CGC_GATE__SYS_MASK | + UVD_CGC_GATE__UDEC_MASK | + UVD_CGC_GATE__MPEG2_MASK | + UVD_CGC_GATE__RBC_MASK | + UVD_CGC_GATE__LMI_MC_MASK | + UVD_CGC_GATE__LMI_UMC_MASK | + UVD_CGC_GATE__IDCT_MASK | + UVD_CGC_GATE__MPRD_MASK | + UVD_CGC_GATE__MPC_MASK | + UVD_CGC_GATE__LBSI_MASK | + UVD_CGC_GATE__LRBBM_MASK | + UVD_CGC_GATE__UDEC_RE_MASK | + UVD_CGC_GATE__UDEC_CM_MASK | + UVD_CGC_GATE__UDEC_IT_MASK | + UVD_CGC_GATE__UDEC_DB_MASK | + UVD_CGC_GATE__UDEC_MP_MASK | + UVD_CGC_GATE__WCB_MASK | + UVD_CGC_GATE__VCPU_MASK | + UVD_CGC_GATE__SCPU_MASK); + data1 &= ~(UVD_SUVD_CGC_GATE__SRE_MASK | + UVD_SUVD_CGC_GATE__SIT_MASK | + UVD_SUVD_CGC_GATE__SMP_MASK | + UVD_SUVD_CGC_GATE__SCM_MASK | + UVD_SUVD_CGC_GATE__SDB_MASK); + } + WREG32(mmUVD_CGC_GATE, data); + WREG32(mmUVD_SUVD_CGC_GATE, data1); +} + +static void uvd_v6_0_set_uvd_dynamic_clock_mode(struct amdgpu_device *adev, + bool swmode) +{ + u32 data, data1 = 0, data2; + + /* Always un-gate UVD REGS bit */ + data = RREG32(mmUVD_CGC_GATE); + data &= ~(UVD_CGC_GATE__REGS_MASK); + WREG32(mmUVD_CGC_GATE, data); + + data = RREG32(mmUVD_CGC_CTRL); + data &= ~(UVD_CGC_CTRL__CLK_OFF_DELAY_MASK | + UVD_CGC_CTRL__CLK_GATE_DLY_TIMER_MASK); + data |= UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK | + 1 << REG_FIELD_SHIFT(UVD_CGC_CTRL, CLK_GATE_DLY_TIMER) | + 4 << REG_FIELD_SHIFT(UVD_CGC_CTRL, CLK_OFF_DELAY); + + data2 = RREG32(mmUVD_SUVD_CGC_CTRL); + if (swmode) { + data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK | + UVD_CGC_CTRL__UDEC_CM_MODE_MASK | + UVD_CGC_CTRL__UDEC_IT_MODE_MASK | + UVD_CGC_CTRL__UDEC_DB_MODE_MASK | + UVD_CGC_CTRL__UDEC_MP_MODE_MASK | + UVD_CGC_CTRL__SYS_MODE_MASK | + UVD_CGC_CTRL__UDEC_MODE_MASK | + UVD_CGC_CTRL__MPEG2_MODE_MASK | + UVD_CGC_CTRL__REGS_MODE_MASK | + UVD_CGC_CTRL__RBC_MODE_MASK | + UVD_CGC_CTRL__LMI_MC_MODE_MASK | + UVD_CGC_CTRL__LMI_UMC_MODE_MASK | + UVD_CGC_CTRL__IDCT_MODE_MASK | + UVD_CGC_CTRL__MPRD_MODE_MASK | + UVD_CGC_CTRL__MPC_MODE_MASK | + UVD_CGC_CTRL__LBSI_MODE_MASK | + UVD_CGC_CTRL__LRBBM_MODE_MASK | + UVD_CGC_CTRL__WCB_MODE_MASK | + UVD_CGC_CTRL__VCPU_MODE_MASK | + UVD_CGC_CTRL__JPEG_MODE_MASK | + UVD_CGC_CTRL__SCPU_MODE_MASK); + data1 |= UVD_CGC_CTRL2__DYN_OCLK_RAMP_EN_MASK | + UVD_CGC_CTRL2__DYN_RCLK_RAMP_EN_MASK; + data1 &= ~UVD_CGC_CTRL2__GATER_DIV_ID_MASK; + data1 |= 7 << REG_FIELD_SHIFT(UVD_CGC_CTRL2, GATER_DIV_ID); + data2 &= ~(UVD_SUVD_CGC_CTRL__SRE_MODE_MASK | + UVD_SUVD_CGC_CTRL__SIT_MODE_MASK | + UVD_SUVD_CGC_CTRL__SMP_MODE_MASK | + UVD_SUVD_CGC_CTRL__SCM_MODE_MASK | + UVD_SUVD_CGC_CTRL__SDB_MODE_MASK); + } else { + data |= UVD_CGC_CTRL__UDEC_RE_MODE_MASK | + UVD_CGC_CTRL__UDEC_CM_MODE_MASK | + UVD_CGC_CTRL__UDEC_IT_MODE_MASK | + UVD_CGC_CTRL__UDEC_DB_MODE_MASK | + UVD_CGC_CTRL__UDEC_MP_MODE_MASK | + UVD_CGC_CTRL__SYS_MODE_MASK | + UVD_CGC_CTRL__UDEC_MODE_MASK | + UVD_CGC_CTRL__MPEG2_MODE_MASK | + UVD_CGC_CTRL__REGS_MODE_MASK | + UVD_CGC_CTRL__RBC_MODE_MASK | + UVD_CGC_CTRL__LMI_MC_MODE_MASK | + UVD_CGC_CTRL__LMI_UMC_MODE_MASK | + UVD_CGC_CTRL__IDCT_MODE_MASK | + UVD_CGC_CTRL__MPRD_MODE_MASK | + UVD_CGC_CTRL__MPC_MODE_MASK | + UVD_CGC_CTRL__LBSI_MODE_MASK | + UVD_CGC_CTRL__LRBBM_MODE_MASK | + UVD_CGC_CTRL__WCB_MODE_MASK | + UVD_CGC_CTRL__VCPU_MODE_MASK | + UVD_CGC_CTRL__SCPU_MODE_MASK; + data2 |= UVD_SUVD_CGC_CTRL__SRE_MODE_MASK | + UVD_SUVD_CGC_CTRL__SIT_MODE_MASK | + UVD_SUVD_CGC_CTRL__SMP_MODE_MASK | + UVD_SUVD_CGC_CTRL__SCM_MODE_MASK | + UVD_SUVD_CGC_CTRL__SDB_MODE_MASK; + } + WREG32(mmUVD_CGC_CTRL, data); + WREG32(mmUVD_SUVD_CGC_CTRL, data2); + + data = RREG32_UVD_CTX(ixUVD_CGC_CTRL2); + data &= ~(REG_FIELD_MASK(UVD_CGC_CTRL2, DYN_OCLK_RAMP_EN) | + REG_FIELD_MASK(UVD_CGC_CTRL2, DYN_RCLK_RAMP_EN) | + REG_FIELD_MASK(UVD_CGC_CTRL2, GATER_DIV_ID)); + data1 &= (REG_FIELD_MASK(UVD_CGC_CTRL2, DYN_OCLK_RAMP_EN) | + REG_FIELD_MASK(UVD_CGC_CTRL2, DYN_RCLK_RAMP_EN) | + REG_FIELD_MASK(UVD_CGC_CTRL2, GATER_DIV_ID)); + data |= data1; + WREG32_UVD_CTX(ixUVD_CGC_CTRL2, data); } /** @@ -303,8 +535,19 @@ static int uvd_v6_0_start(struct amdgpu_device *adev) uvd_v6_0_mc_resume(adev); - /* disable clock gating */ - WREG32(mmUVD_CGC_GATE, 0); + /* Set dynamic clock gating in S/W control mode */ + if (adev->cg_flags & AMD_CG_SUPPORT_UVD_MGCG) { + if (adev->flags & AMD_IS_APU) + cz_set_uvd_clock_gating_branches(adev, false); + else + tonga_set_uvd_clock_gating_branches(adev, false); + uvd_v6_0_set_uvd_dynamic_clock_mode(adev, true); + } else { + /* disable clock gating */ + uint32_t data = RREG32(mmUVD_CGC_CTRL); + data &= ~UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK; + WREG32(mmUVD_CGC_CTRL, data); + } /* disable interupt */ WREG32_P(mmUVD_MASTINT_EN, 0, ~(1 << 1)); @@ -483,33 +726,6 @@ static void uvd_v6_0_ring_emit_fence(struct amdgpu_ring *ring, u64 addr, u64 seq } /** - * uvd_v6_0_ring_emit_semaphore - emit semaphore command - * - * @ring: amdgpu_ring pointer - * @semaphore: semaphore to emit commands for - * @emit_wait: true if we should emit a wait command - * - * Emit a semaphore command (either wait or signal) to the UVD ring. - */ -static bool uvd_v6_0_ring_emit_semaphore(struct amdgpu_ring *ring, - struct amdgpu_semaphore *semaphore, - bool emit_wait) -{ - uint64_t addr = semaphore->gpu_addr; - - amdgpu_ring_write(ring, PACKET0(mmUVD_SEMA_ADDR_LOW, 0)); - amdgpu_ring_write(ring, (addr >> 3) & 0x000FFFFF); - - amdgpu_ring_write(ring, PACKET0(mmUVD_SEMA_ADDR_HIGH, 0)); - amdgpu_ring_write(ring, (addr >> 23) & 0x000FFFFF); - - amdgpu_ring_write(ring, PACKET0(mmUVD_SEMA_CMD, 0)); - amdgpu_ring_write(ring, 0x80 | (emit_wait ? 1 : 0)); - - return true; -} - -/** * uvd_v6_0_ring_test_ring - register write test * * @ring: amdgpu_ring pointer @@ -524,7 +740,7 @@ static int uvd_v6_0_ring_test_ring(struct amdgpu_ring *ring) int r; WREG32(mmUVD_CONTEXT_ID, 0xCAFEDEAD); - r = amdgpu_ring_lock(ring, 3); + r = amdgpu_ring_alloc(ring, 3); if (r) { DRM_ERROR("amdgpu: cp failed to lock ring %d (%d).\n", ring->idx, r); @@ -532,7 +748,7 @@ static int uvd_v6_0_ring_test_ring(struct amdgpu_ring *ring) } amdgpu_ring_write(ring, PACKET0(mmUVD_CONTEXT_ID, 0)); amdgpu_ring_write(ring, 0xDEADBEEF); - amdgpu_ring_unlock_commit(ring); + amdgpu_ring_commit(ring); for (i = 0; i < adev->usec_timeout; i++) { tmp = RREG32(mmUVD_CONTEXT_ID); if (tmp == 0xDEADBEEF) @@ -588,7 +804,7 @@ static int uvd_v6_0_ring_test_ib(struct amdgpu_ring *ring) goto error; } - r = amdgpu_uvd_get_destroy_msg(ring, 1, &fence); + r = amdgpu_uvd_get_destroy_msg(ring, 1, true, &fence); if (r) { DRM_ERROR("amdgpu: failed to get destroy ib (%d).\n", r); goto error; @@ -735,6 +951,12 @@ static void uvd_v6_0_print_status(void *handle) RREG32(mmUVD_SEMA_SIGNAL_INCOMPLETE_TIMEOUT_CNTL)); dev_info(adev->dev, " UVD_CONTEXT_ID=0x%08X\n", RREG32(mmUVD_CONTEXT_ID)); + dev_info(adev->dev, " UVD_UDEC_ADDR_CONFIG=0x%08X\n", + RREG32(mmUVD_UDEC_ADDR_CONFIG)); + dev_info(adev->dev, " UVD_UDEC_DB_ADDR_CONFIG=0x%08X\n", + RREG32(mmUVD_UDEC_DB_ADDR_CONFIG)); + dev_info(adev->dev, " UVD_UDEC_DBW_ADDR_CONFIG=0x%08X\n", + RREG32(mmUVD_UDEC_DBW_ADDR_CONFIG)); } static int uvd_v6_0_set_interrupt_state(struct amdgpu_device *adev, @@ -758,6 +980,24 @@ static int uvd_v6_0_process_interrupt(struct amdgpu_device *adev, static int uvd_v6_0_set_clockgating_state(void *handle, enum amd_clockgating_state state) { + struct amdgpu_device *adev = (struct amdgpu_device *)handle; + bool enable = (state == AMD_CG_STATE_GATE) ? true : false; + + if (!(adev->cg_flags & AMD_CG_SUPPORT_UVD_MGCG)) + return 0; + + if (enable) { + if (adev->flags & AMD_IS_APU) + cz_set_uvd_clock_gating_branches(adev, enable); + else + tonga_set_uvd_clock_gating_branches(adev, enable); + uvd_v6_0_set_uvd_dynamic_clock_mode(adev, true); + } else { + uint32_t data = RREG32(mmUVD_CGC_CTRL); + data &= ~UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK; + WREG32(mmUVD_CGC_CTRL, data); + } + return 0; } @@ -773,6 +1013,9 @@ static int uvd_v6_0_set_powergating_state(void *handle, */ struct amdgpu_device *adev = (struct amdgpu_device *)handle; + if (!(adev->pg_flags & AMD_PG_SUPPORT_UVD)) + return 0; + if (state == AMD_PG_STATE_GATE) { uvd_v6_0_stop(adev); return 0; @@ -805,11 +1048,10 @@ static const struct amdgpu_ring_funcs uvd_v6_0_ring_funcs = { .parse_cs = amdgpu_uvd_ring_parse_cs, .emit_ib = uvd_v6_0_ring_emit_ib, .emit_fence = uvd_v6_0_ring_emit_fence, - .emit_semaphore = uvd_v6_0_ring_emit_semaphore, .test_ring = uvd_v6_0_ring_test_ring, .test_ib = uvd_v6_0_ring_test_ib, - .is_lockup = amdgpu_ring_test_lockup, .insert_nop = amdgpu_ring_insert_nop, + .pad_ib = amdgpu_ring_generic_pad_ib, }; static void uvd_v6_0_set_ring_funcs(struct amdgpu_device *adev) diff --git a/drivers/gpu/drm/amd/amdgpu/vce_v2_0.c b/drivers/gpu/drm/amd/amdgpu/vce_v2_0.c index cd16df543f64..c7e885bcfd41 100644 --- a/drivers/gpu/drm/amd/amdgpu/vce_v2_0.c +++ b/drivers/gpu/drm/amd/amdgpu/vce_v2_0.c @@ -373,7 +373,7 @@ static void vce_v2_0_enable_mgcg(struct amdgpu_device *adev, bool enable) { bool sw_cg = false; - if (enable && (adev->cg_flags & AMDGPU_CG_SUPPORT_VCE_MGCG)) { + if (enable && (adev->cg_flags & AMD_CG_SUPPORT_VCE_MGCG)) { if (sw_cg) vce_v2_0_set_sw_cg(adev, true); else @@ -608,6 +608,9 @@ static int vce_v2_0_set_powergating_state(void *handle, */ struct amdgpu_device *adev = (struct amdgpu_device *)handle; + if (!(adev->pg_flags & AMD_PG_SUPPORT_VCE)) + return 0; + if (state == AMD_PG_STATE_GATE) /* XXX do we need a vce_v2_0_stop()? */ return 0; @@ -639,11 +642,10 @@ static const struct amdgpu_ring_funcs vce_v2_0_ring_funcs = { .parse_cs = amdgpu_vce_ring_parse_cs, .emit_ib = amdgpu_vce_ring_emit_ib, .emit_fence = amdgpu_vce_ring_emit_fence, - .emit_semaphore = amdgpu_vce_ring_emit_semaphore, .test_ring = amdgpu_vce_ring_test_ring, .test_ib = amdgpu_vce_ring_test_ib, - .is_lockup = amdgpu_ring_test_lockup, .insert_nop = amdgpu_ring_insert_nop, + .pad_ib = amdgpu_ring_generic_pad_ib, }; static void vce_v2_0_set_ring_funcs(struct amdgpu_device *adev) diff --git a/drivers/gpu/drm/amd/amdgpu/vce_v3_0.c b/drivers/gpu/drm/amd/amdgpu/vce_v3_0.c index f0656dfb53f3..ce468ee5da2a 100644 --- a/drivers/gpu/drm/amd/amdgpu/vce_v3_0.c +++ b/drivers/gpu/drm/amd/amdgpu/vce_v3_0.c @@ -40,6 +40,9 @@ #define GRBM_GFX_INDEX__VCE_INSTANCE__SHIFT 0x04 #define GRBM_GFX_INDEX__VCE_INSTANCE_MASK 0x10 +#define mmVCE_LMI_VCPU_CACHE_40BIT_BAR0 0x8616 +#define mmVCE_LMI_VCPU_CACHE_40BIT_BAR1 0x8617 +#define mmVCE_LMI_VCPU_CACHE_40BIT_BAR2 0x8618 #define VCE_V3_0_FW_SIZE (384 * 1024) #define VCE_V3_0_STACK_SIZE (64 * 1024) @@ -100,6 +103,108 @@ static void vce_v3_0_ring_set_wptr(struct amdgpu_ring *ring) WREG32(mmVCE_RB_WPTR2, ring->wptr); } +static void vce_v3_0_override_vce_clock_gating(struct amdgpu_device *adev, bool override) +{ + u32 tmp, data; + + tmp = data = RREG32(mmVCE_RB_ARB_CTRL); + if (override) + data |= VCE_RB_ARB_CTRL__VCE_CGTT_OVERRIDE_MASK; + else + data &= ~VCE_RB_ARB_CTRL__VCE_CGTT_OVERRIDE_MASK; + + if (tmp != data) + WREG32(mmVCE_RB_ARB_CTRL, data); +} + +static void vce_v3_0_set_vce_sw_clock_gating(struct amdgpu_device *adev, + bool gated) +{ + u32 tmp, data; + /* Set Override to disable Clock Gating */ + vce_v3_0_override_vce_clock_gating(adev, true); + + if (!gated) { + /* Force CLOCK ON for VCE_CLOCK_GATING_B, + * {*_FORCE_ON, *_FORCE_OFF} = {1, 0} + * VREG can be FORCE ON or set to Dynamic, but can't be OFF + */ + tmp = data = RREG32(mmVCE_CLOCK_GATING_B); + data |= 0x1ff; + data &= ~0xef0000; + if (tmp != data) + WREG32(mmVCE_CLOCK_GATING_B, data); + + /* Force CLOCK ON for VCE_UENC_CLOCK_GATING, + * {*_FORCE_ON, *_FORCE_OFF} = {1, 0} + */ + tmp = data = RREG32(mmVCE_UENC_CLOCK_GATING); + data |= 0x3ff000; + data &= ~0xffc00000; + if (tmp != data) + WREG32(mmVCE_UENC_CLOCK_GATING, data); + + /* set VCE_UENC_CLOCK_GATING_2 */ + tmp = data = RREG32(mmVCE_UENC_CLOCK_GATING_2); + data |= 0x2; + data &= ~0x2; + if (tmp != data) + WREG32(mmVCE_UENC_CLOCK_GATING_2, data); + + /* Force CLOCK ON for VCE_UENC_REG_CLOCK_GATING */ + tmp = data = RREG32(mmVCE_UENC_REG_CLOCK_GATING); + data |= 0x37f; + if (tmp != data) + WREG32(mmVCE_UENC_REG_CLOCK_GATING, data); + + /* Force VCE_UENC_DMA_DCLK_CTRL Clock ON */ + tmp = data = RREG32(mmVCE_UENC_DMA_DCLK_CTRL); + data |= VCE_UENC_DMA_DCLK_CTRL__WRDMCLK_FORCEON_MASK | + VCE_UENC_DMA_DCLK_CTRL__RDDMCLK_FORCEON_MASK | + VCE_UENC_DMA_DCLK_CTRL__REGCLK_FORCEON_MASK | + 0x8; + if (tmp != data) + WREG32(mmVCE_UENC_DMA_DCLK_CTRL, data); + } else { + /* Force CLOCK OFF for VCE_CLOCK_GATING_B, + * {*, *_FORCE_OFF} = {*, 1} + * set VREG to Dynamic, as it can't be OFF + */ + tmp = data = RREG32(mmVCE_CLOCK_GATING_B); + data &= ~0x80010; + data |= 0xe70008; + if (tmp != data) + WREG32(mmVCE_CLOCK_GATING_B, data); + /* Force CLOCK OFF for VCE_UENC_CLOCK_GATING, + * Force ClOCK OFF takes precedent over Force CLOCK ON setting. + * {*_FORCE_ON, *_FORCE_OFF} = {*, 1} + */ + tmp = data = RREG32(mmVCE_UENC_CLOCK_GATING); + data |= 0xffc00000; + if (tmp != data) + WREG32(mmVCE_UENC_CLOCK_GATING, data); + /* Set VCE_UENC_CLOCK_GATING_2 */ + tmp = data = RREG32(mmVCE_UENC_CLOCK_GATING_2); + data |= 0x10000; + if (tmp != data) + WREG32(mmVCE_UENC_CLOCK_GATING_2, data); + /* Set VCE_UENC_REG_CLOCK_GATING to dynamic */ + tmp = data = RREG32(mmVCE_UENC_REG_CLOCK_GATING); + data &= ~0xffc00000; + if (tmp != data) + WREG32(mmVCE_UENC_REG_CLOCK_GATING, data); + /* Set VCE_UENC_DMA_DCLK_CTRL CG always in dynamic mode */ + tmp = data = RREG32(mmVCE_UENC_DMA_DCLK_CTRL); + data &= ~(VCE_UENC_DMA_DCLK_CTRL__WRDMCLK_FORCEON_MASK | + VCE_UENC_DMA_DCLK_CTRL__RDDMCLK_FORCEON_MASK | + VCE_UENC_DMA_DCLK_CTRL__REGCLK_FORCEON_MASK | + 0x8); + if (tmp != data) + WREG32(mmVCE_UENC_DMA_DCLK_CTRL, data); + } + vce_v3_0_override_vce_clock_gating(adev, false); +} + /** * vce_v3_0_start - start VCE block * @@ -118,7 +223,7 @@ static int vce_v3_0_start(struct amdgpu_device *adev) if (adev->vce.harvest_config & (1 << idx)) continue; - if(idx == 0) + if (idx == 0) WREG32_P(mmGRBM_GFX_INDEX, 0, ~GRBM_GFX_INDEX__VCE_INSTANCE_MASK); else @@ -130,9 +235,11 @@ static int vce_v3_0_start(struct amdgpu_device *adev) /* set BUSY flag */ WREG32_P(mmVCE_STATUS, 1, ~1); - - WREG32_P(mmVCE_VCPU_CNTL, VCE_VCPU_CNTL__CLK_EN_MASK, - ~VCE_VCPU_CNTL__CLK_EN_MASK); + if (adev->asic_type >= CHIP_STONEY) + WREG32_P(mmVCE_VCPU_CNTL, 1, ~0x200001); + else + WREG32_P(mmVCE_VCPU_CNTL, VCE_VCPU_CNTL__CLK_EN_MASK, + ~VCE_VCPU_CNTL__CLK_EN_MASK); WREG32_P(mmVCE_SOFT_RESET, VCE_SOFT_RESET__ECPU_SOFT_RESET_MASK, @@ -169,6 +276,10 @@ static int vce_v3_0_start(struct amdgpu_device *adev) /* clear BUSY flag */ WREG32_P(mmVCE_STATUS, 0, ~1); + /* Set Clock-Gating off */ + if (adev->cg_flags & AMD_CG_SUPPORT_VCE_MGCG) + vce_v3_0_set_vce_sw_clock_gating(adev, false); + if (r) { DRM_ERROR("VCE not responding, giving up!!!\n"); mutex_unlock(&adev->grbm_idx_mutex); @@ -203,13 +314,11 @@ static int vce_v3_0_start(struct amdgpu_device *adev) static unsigned vce_v3_0_get_harvest_config(struct amdgpu_device *adev) { u32 tmp; - unsigned ret; - /* Fiji is single pipe */ - if (adev->asic_type == CHIP_FIJI) { - ret = AMDGPU_VCE_HARVEST_VCE1; - return ret; - } + /* Fiji, Stoney are single pipe */ + if ((adev->asic_type == CHIP_FIJI) || + (adev->asic_type == CHIP_STONEY)) + return AMDGPU_VCE_HARVEST_VCE1; /* Tonga and CZ are dual or single pipe */ if (adev->flags & AMD_IS_APU) @@ -223,19 +332,14 @@ static unsigned vce_v3_0_get_harvest_config(struct amdgpu_device *adev) switch (tmp) { case 1: - ret = AMDGPU_VCE_HARVEST_VCE0; - break; + return AMDGPU_VCE_HARVEST_VCE0; case 2: - ret = AMDGPU_VCE_HARVEST_VCE1; - break; + return AMDGPU_VCE_HARVEST_VCE1; case 3: - ret = AMDGPU_VCE_HARVEST_VCE0 | AMDGPU_VCE_HARVEST_VCE1; - break; + return AMDGPU_VCE_HARVEST_VCE0 | AMDGPU_VCE_HARVEST_VCE1; default: - ret = 0; + return 0; } - - return ret; } static int vce_v3_0_early_init(void *handle) @@ -310,28 +414,22 @@ static int vce_v3_0_sw_fini(void *handle) static int vce_v3_0_hw_init(void *handle) { - struct amdgpu_ring *ring; - int r; + int r, i; struct amdgpu_device *adev = (struct amdgpu_device *)handle; r = vce_v3_0_start(adev); if (r) return r; - ring = &adev->vce.ring[0]; - ring->ready = true; - r = amdgpu_ring_test_ring(ring); - if (r) { - ring->ready = false; - return r; - } + adev->vce.ring[0].ready = false; + adev->vce.ring[1].ready = false; - ring = &adev->vce.ring[1]; - ring->ready = true; - r = amdgpu_ring_test_ring(ring); - if (r) { - ring->ready = false; - return r; + for (i = 0; i < 2; i++) { + r = amdgpu_ring_test_ring(&adev->vce.ring[i]); + if (r) + return r; + else + adev->vce.ring[i].ready = true; } DRM_INFO("VCE initialized successfully.\n"); @@ -390,8 +488,12 @@ static void vce_v3_0_mc_resume(struct amdgpu_device *adev, int idx) WREG32(mmVCE_LMI_SWAP_CNTL, 0); WREG32(mmVCE_LMI_SWAP_CNTL1, 0); WREG32(mmVCE_LMI_VM_CTRL, 0); - - WREG32(mmVCE_LMI_VCPU_CACHE_40BIT_BAR, (adev->vce.gpu_addr >> 8)); + if (adev->asic_type >= CHIP_STONEY) { + WREG32(mmVCE_LMI_VCPU_CACHE_40BIT_BAR0, (adev->vce.gpu_addr >> 8)); + WREG32(mmVCE_LMI_VCPU_CACHE_40BIT_BAR1, (adev->vce.gpu_addr >> 8)); + WREG32(mmVCE_LMI_VCPU_CACHE_40BIT_BAR2, (adev->vce.gpu_addr >> 8)); + } else + WREG32(mmVCE_LMI_VCPU_CACHE_40BIT_BAR, (adev->vce.gpu_addr >> 8)); offset = AMDGPU_VCE_FIRMWARE_OFFSET; size = VCE_V3_0_FW_SIZE; WREG32(mmVCE_VCPU_CACHE_OFFSET0, offset & 0x7fffffff); @@ -427,17 +529,9 @@ static bool vce_v3_0_is_idle(void *handle) { struct amdgpu_device *adev = (struct amdgpu_device *)handle; u32 mask = 0; - int idx; - for (idx = 0; idx < 2; ++idx) { - if (adev->vce.harvest_config & (1 << idx)) - continue; - - if (idx == 0) - mask |= SRBM_STATUS2__VCE0_BUSY_MASK; - else - mask |= SRBM_STATUS2__VCE1_BUSY_MASK; - } + mask |= (adev->vce.harvest_config & AMDGPU_VCE_HARVEST_VCE0) ? 0 : SRBM_STATUS2__VCE0_BUSY_MASK; + mask |= (adev->vce.harvest_config & AMDGPU_VCE_HARVEST_VCE1) ? 0 : SRBM_STATUS2__VCE1_BUSY_MASK; return !(RREG32(mmSRBM_STATUS2) & mask); } @@ -446,23 +540,11 @@ static int vce_v3_0_wait_for_idle(void *handle) { unsigned i; struct amdgpu_device *adev = (struct amdgpu_device *)handle; - u32 mask = 0; - int idx; - for (idx = 0; idx < 2; ++idx) { - if (adev->vce.harvest_config & (1 << idx)) - continue; - - if (idx == 0) - mask |= SRBM_STATUS2__VCE0_BUSY_MASK; - else - mask |= SRBM_STATUS2__VCE1_BUSY_MASK; - } - - for (i = 0; i < adev->usec_timeout; i++) { - if (!(RREG32(mmSRBM_STATUS2) & mask)) + for (i = 0; i < adev->usec_timeout; i++) + if (vce_v3_0_is_idle(handle)) return 0; - } + return -ETIMEDOUT; } @@ -470,17 +552,10 @@ static int vce_v3_0_soft_reset(void *handle) { struct amdgpu_device *adev = (struct amdgpu_device *)handle; u32 mask = 0; - int idx; - for (idx = 0; idx < 2; ++idx) { - if (adev->vce.harvest_config & (1 << idx)) - continue; + mask |= (adev->vce.harvest_config & AMDGPU_VCE_HARVEST_VCE0) ? 0 : SRBM_SOFT_RESET__SOFT_RESET_VCE0_MASK; + mask |= (adev->vce.harvest_config & AMDGPU_VCE_HARVEST_VCE1) ? 0 : SRBM_SOFT_RESET__SOFT_RESET_VCE1_MASK; - if (idx == 0) - mask |= SRBM_SOFT_RESET__SOFT_RESET_VCE0_MASK; - else - mask |= SRBM_SOFT_RESET__SOFT_RESET_VCE1_MASK; - } WREG32_P(mmSRBM_SOFT_RESET, mask, ~(SRBM_SOFT_RESET__SOFT_RESET_VCE0_MASK | SRBM_SOFT_RESET__SOFT_RESET_VCE1_MASK)); @@ -575,12 +650,15 @@ static int vce_v3_0_process_interrupt(struct amdgpu_device *adev, struct amdgpu_iv_entry *entry) { DRM_DEBUG("IH: VCE\n"); + + WREG32_P(mmVCE_SYS_INT_STATUS, + VCE_SYS_INT_STATUS__VCE_SYS_INT_TRAP_INTERRUPT_INT_MASK, + ~VCE_SYS_INT_STATUS__VCE_SYS_INT_TRAP_INTERRUPT_INT_MASK); + switch (entry->src_data) { case 0: - amdgpu_fence_process(&adev->vce.ring[0]); - break; case 1: - amdgpu_fence_process(&adev->vce.ring[1]); + amdgpu_fence_process(&adev->vce.ring[entry->src_data]); break; default: DRM_ERROR("Unhandled interrupt: %d %d\n", @@ -594,6 +672,47 @@ static int vce_v3_0_process_interrupt(struct amdgpu_device *adev, static int vce_v3_0_set_clockgating_state(void *handle, enum amd_clockgating_state state) { + struct amdgpu_device *adev = (struct amdgpu_device *)handle; + bool enable = (state == AMD_CG_STATE_GATE) ? true : false; + int i; + + if (!(adev->cg_flags & AMD_CG_SUPPORT_VCE_MGCG)) + return 0; + + mutex_lock(&adev->grbm_idx_mutex); + for (i = 0; i < 2; i++) { + /* Program VCE Instance 0 or 1 if not harvested */ + if (adev->vce.harvest_config & (1 << i)) + continue; + + if (i == 0) + WREG32_P(mmGRBM_GFX_INDEX, 0, + ~GRBM_GFX_INDEX__VCE_INSTANCE_MASK); + else + WREG32_P(mmGRBM_GFX_INDEX, + GRBM_GFX_INDEX__VCE_INSTANCE_MASK, + ~GRBM_GFX_INDEX__VCE_INSTANCE_MASK); + + if (enable) { + /* initialize VCE_CLOCK_GATING_A: Clock ON/OFF delay */ + uint32_t data = RREG32(mmVCE_CLOCK_GATING_A); + data &= ~(0xf | 0xff0); + data |= ((0x0 << 0) | (0x04 << 4)); + WREG32(mmVCE_CLOCK_GATING_A, data); + + /* initialize VCE_UENC_CLOCK_GATING: Clock ON/OFF delay */ + data = RREG32(mmVCE_UENC_CLOCK_GATING); + data &= ~(0xf | 0xff0); + data |= ((0x0 << 0) | (0x04 << 4)); + WREG32(mmVCE_UENC_CLOCK_GATING, data); + } + + vce_v3_0_set_vce_sw_clock_gating(adev, enable); + } + + WREG32_P(mmGRBM_GFX_INDEX, 0, ~GRBM_GFX_INDEX__VCE_INSTANCE_MASK); + mutex_unlock(&adev->grbm_idx_mutex); + return 0; } @@ -609,6 +728,9 @@ static int vce_v3_0_set_powergating_state(void *handle, */ struct amdgpu_device *adev = (struct amdgpu_device *)handle; + if (!(adev->pg_flags & AMD_PG_SUPPORT_VCE)) + return 0; + if (state == AMD_PG_STATE_GATE) /* XXX do we need a vce_v3_0_stop()? */ return 0; @@ -640,11 +762,10 @@ static const struct amdgpu_ring_funcs vce_v3_0_ring_funcs = { .parse_cs = amdgpu_vce_ring_parse_cs, .emit_ib = amdgpu_vce_ring_emit_ib, .emit_fence = amdgpu_vce_ring_emit_fence, - .emit_semaphore = amdgpu_vce_ring_emit_semaphore, .test_ring = amdgpu_vce_ring_test_ring, .test_ib = amdgpu_vce_ring_test_ib, - .is_lockup = amdgpu_ring_test_lockup, .insert_nop = amdgpu_ring_insert_nop, + .pad_ib = amdgpu_ring_generic_pad_ib, }; static void vce_v3_0_set_ring_funcs(struct amdgpu_device *adev) diff --git a/drivers/gpu/drm/amd/amdgpu/vi.c b/drivers/gpu/drm/amd/amdgpu/vi.c index 0bac8702e934..1c120efa292c 100644 --- a/drivers/gpu/drm/amd/amdgpu/vi.c +++ b/drivers/gpu/drm/amd/amdgpu/vi.c @@ -31,6 +31,7 @@ #include "amdgpu_vce.h" #include "amdgpu_ucode.h" #include "atom.h" +#include "amd_pcie.h" #include "gmc/gmc_8_1_d.h" #include "gmc/gmc_8_1_sh_mask.h" @@ -60,6 +61,7 @@ #include "vi.h" #include "vi_dpm.h" #include "gmc_v8_0.h" +#include "gmc_v7_0.h" #include "gfx_v8_0.h" #include "sdma_v2_4.h" #include "sdma_v3_0.h" @@ -71,6 +73,10 @@ #include "uvd_v5_0.h" #include "uvd_v6_0.h" #include "vce_v3_0.h" +#include "amdgpu_powerplay.h" +#if defined(CONFIG_DRM_AMD_ACP) +#include "amdgpu_acp.h" +#endif /* * Indirect registers accessor @@ -232,6 +238,13 @@ static const u32 cz_mgcg_cgcg_init[] = mmHDP_XDP_CGTT_BLK_CTRL, 0xc0000fff, 0x00000104, }; +static const u32 stoney_mgcg_cgcg_init[] = +{ + mmCGTT_DRM_CLK_CTRL0, 0xffffffff, 0x00000100, + mmHDP_XDP_CGTT_BLK_CTRL, 0xffffffff, 0x00000104, + mmHDP_HOST_PATH_CNTL, 0xffffffff, 0x0f000027, +}; + static void vi_init_golden_registers(struct amdgpu_device *adev) { /* Some of the registers might be dependent on GRBM_GFX_INDEX */ @@ -258,6 +271,11 @@ static void vi_init_golden_registers(struct amdgpu_device *adev) cz_mgcg_cgcg_init, (const u32)ARRAY_SIZE(cz_mgcg_cgcg_init)); break; + case CHIP_STONEY: + amdgpu_program_register_sequence(adev, + stoney_mgcg_cgcg_init, + (const u32)ARRAY_SIZE(stoney_mgcg_cgcg_init)); + break; default: break; } @@ -364,6 +382,38 @@ static bool vi_read_disabled_bios(struct amdgpu_device *adev) WREG32_SMC(ixROM_CNTL, rom_cntl); return r; } + +static bool vi_read_bios_from_rom(struct amdgpu_device *adev, + u8 *bios, u32 length_bytes) +{ + u32 *dw_ptr; + unsigned long flags; + u32 i, length_dw; + + if (bios == NULL) + return false; + if (length_bytes == 0) + return false; + /* APU vbios image is part of sbios image */ + if (adev->flags & AMD_IS_APU) + return false; + + dw_ptr = (u32 *)bios; + length_dw = ALIGN(length_bytes, 4) / 4; + /* take the smc lock since we are using the smc index */ + spin_lock_irqsave(&adev->smc_idx_lock, flags); + /* set rom index to 0 */ + WREG32(mmSMC_IND_INDEX_0, ixROM_INDEX); + WREG32(mmSMC_IND_DATA_0, 0); + /* set index to data for continous read */ + WREG32(mmSMC_IND_INDEX_0, ixROM_DATA); + for (i = 0; i < length_dw; i++) + dw_ptr[i] = RREG32(mmSMC_IND_DATA_0); + spin_unlock_irqrestore(&adev->smc_idx_lock, flags); + + return true; +} + static struct amdgpu_allowed_register_entry tonga_allowed_read_registers[] = { {mmGB_MACROTILE_MODE7, true}, }; @@ -488,6 +538,7 @@ static int vi_read_register(struct amdgpu_device *adev, u32 se_num, case CHIP_FIJI: case CHIP_TONGA: case CHIP_CARRIZO: + case CHIP_STONEY: asic_register_table = cz_allowed_read_registers; size = ARRAY_SIZE(cz_allowed_read_registers); break; @@ -523,370 +574,12 @@ static int vi_read_register(struct amdgpu_device *adev, u32 se_num, return -EINVAL; } -static void vi_print_gpu_status_regs(struct amdgpu_device *adev) -{ - dev_info(adev->dev, " GRBM_STATUS=0x%08X\n", - RREG32(mmGRBM_STATUS)); - dev_info(adev->dev, " GRBM_STATUS2=0x%08X\n", - RREG32(mmGRBM_STATUS2)); - dev_info(adev->dev, " GRBM_STATUS_SE0=0x%08X\n", - RREG32(mmGRBM_STATUS_SE0)); - dev_info(adev->dev, " GRBM_STATUS_SE1=0x%08X\n", - RREG32(mmGRBM_STATUS_SE1)); - dev_info(adev->dev, " GRBM_STATUS_SE2=0x%08X\n", - RREG32(mmGRBM_STATUS_SE2)); - dev_info(adev->dev, " GRBM_STATUS_SE3=0x%08X\n", - RREG32(mmGRBM_STATUS_SE3)); - dev_info(adev->dev, " SRBM_STATUS=0x%08X\n", - RREG32(mmSRBM_STATUS)); - dev_info(adev->dev, " SRBM_STATUS2=0x%08X\n", - RREG32(mmSRBM_STATUS2)); - dev_info(adev->dev, " SDMA0_STATUS_REG = 0x%08X\n", - RREG32(mmSDMA0_STATUS_REG + SDMA0_REGISTER_OFFSET)); - dev_info(adev->dev, " SDMA1_STATUS_REG = 0x%08X\n", - RREG32(mmSDMA0_STATUS_REG + SDMA1_REGISTER_OFFSET)); - dev_info(adev->dev, " CP_STAT = 0x%08x\n", RREG32(mmCP_STAT)); - dev_info(adev->dev, " CP_STALLED_STAT1 = 0x%08x\n", - RREG32(mmCP_STALLED_STAT1)); - dev_info(adev->dev, " CP_STALLED_STAT2 = 0x%08x\n", - RREG32(mmCP_STALLED_STAT2)); - dev_info(adev->dev, " CP_STALLED_STAT3 = 0x%08x\n", - RREG32(mmCP_STALLED_STAT3)); - dev_info(adev->dev, " CP_CPF_BUSY_STAT = 0x%08x\n", - RREG32(mmCP_CPF_BUSY_STAT)); - dev_info(adev->dev, " CP_CPF_STALLED_STAT1 = 0x%08x\n", - RREG32(mmCP_CPF_STALLED_STAT1)); - dev_info(adev->dev, " CP_CPF_STATUS = 0x%08x\n", RREG32(mmCP_CPF_STATUS)); - dev_info(adev->dev, " CP_CPC_BUSY_STAT = 0x%08x\n", RREG32(mmCP_CPC_BUSY_STAT)); - dev_info(adev->dev, " CP_CPC_STALLED_STAT1 = 0x%08x\n", - RREG32(mmCP_CPC_STALLED_STAT1)); - dev_info(adev->dev, " CP_CPC_STATUS = 0x%08x\n", RREG32(mmCP_CPC_STATUS)); -} - -/** - * vi_gpu_check_soft_reset - check which blocks are busy - * - * @adev: amdgpu_device pointer - * - * Check which blocks are busy and return the relevant reset - * mask to be used by vi_gpu_soft_reset(). - * Returns a mask of the blocks to be reset. - */ -u32 vi_gpu_check_soft_reset(struct amdgpu_device *adev) -{ - u32 reset_mask = 0; - u32 tmp; - - /* GRBM_STATUS */ - tmp = RREG32(mmGRBM_STATUS); - if (tmp & (GRBM_STATUS__PA_BUSY_MASK | GRBM_STATUS__SC_BUSY_MASK | - GRBM_STATUS__BCI_BUSY_MASK | GRBM_STATUS__SX_BUSY_MASK | - GRBM_STATUS__TA_BUSY_MASK | GRBM_STATUS__VGT_BUSY_MASK | - GRBM_STATUS__DB_BUSY_MASK | GRBM_STATUS__CB_BUSY_MASK | - GRBM_STATUS__GDS_BUSY_MASK | GRBM_STATUS__SPI_BUSY_MASK | - GRBM_STATUS__IA_BUSY_MASK | GRBM_STATUS__IA_BUSY_NO_DMA_MASK)) - reset_mask |= AMDGPU_RESET_GFX; - - if (tmp & (GRBM_STATUS__CP_BUSY_MASK | GRBM_STATUS__CP_COHERENCY_BUSY_MASK)) - reset_mask |= AMDGPU_RESET_CP; - - /* GRBM_STATUS2 */ - tmp = RREG32(mmGRBM_STATUS2); - if (tmp & GRBM_STATUS2__RLC_BUSY_MASK) - reset_mask |= AMDGPU_RESET_RLC; - - if (tmp & (GRBM_STATUS2__CPF_BUSY_MASK | - GRBM_STATUS2__CPC_BUSY_MASK | - GRBM_STATUS2__CPG_BUSY_MASK)) - reset_mask |= AMDGPU_RESET_CP; - - /* SRBM_STATUS2 */ - tmp = RREG32(mmSRBM_STATUS2); - if (tmp & SRBM_STATUS2__SDMA_BUSY_MASK) - reset_mask |= AMDGPU_RESET_DMA; - - if (tmp & SRBM_STATUS2__SDMA1_BUSY_MASK) - reset_mask |= AMDGPU_RESET_DMA1; - - /* SRBM_STATUS */ - tmp = RREG32(mmSRBM_STATUS); - - if (tmp & SRBM_STATUS__IH_BUSY_MASK) - reset_mask |= AMDGPU_RESET_IH; - - if (tmp & SRBM_STATUS__SEM_BUSY_MASK) - reset_mask |= AMDGPU_RESET_SEM; - - if (tmp & SRBM_STATUS__GRBM_RQ_PENDING_MASK) - reset_mask |= AMDGPU_RESET_GRBM; - - if (adev->asic_type != CHIP_TOPAZ) { - if (tmp & (SRBM_STATUS__UVD_RQ_PENDING_MASK | - SRBM_STATUS__UVD_BUSY_MASK)) - reset_mask |= AMDGPU_RESET_UVD; - } - - if (tmp & SRBM_STATUS__VMC_BUSY_MASK) - reset_mask |= AMDGPU_RESET_VMC; - - if (tmp & (SRBM_STATUS__MCB_BUSY_MASK | SRBM_STATUS__MCB_NON_DISPLAY_BUSY_MASK | - SRBM_STATUS__MCC_BUSY_MASK | SRBM_STATUS__MCD_BUSY_MASK)) - reset_mask |= AMDGPU_RESET_MC; - - /* SDMA0_STATUS_REG */ - tmp = RREG32(mmSDMA0_STATUS_REG + SDMA0_REGISTER_OFFSET); - if (!(tmp & SDMA0_STATUS_REG__IDLE_MASK)) - reset_mask |= AMDGPU_RESET_DMA; - - /* SDMA1_STATUS_REG */ - tmp = RREG32(mmSDMA0_STATUS_REG + SDMA1_REGISTER_OFFSET); - if (!(tmp & SDMA0_STATUS_REG__IDLE_MASK)) - reset_mask |= AMDGPU_RESET_DMA1; -#if 0 - /* VCE_STATUS */ - if (adev->asic_type != CHIP_TOPAZ) { - tmp = RREG32(mmVCE_STATUS); - if (tmp & VCE_STATUS__VCPU_REPORT_RB0_BUSY_MASK) - reset_mask |= AMDGPU_RESET_VCE; - if (tmp & VCE_STATUS__VCPU_REPORT_RB1_BUSY_MASK) - reset_mask |= AMDGPU_RESET_VCE1; - - } - - if (adev->asic_type != CHIP_TOPAZ) { - if (amdgpu_display_is_display_hung(adev)) - reset_mask |= AMDGPU_RESET_DISPLAY; - } -#endif - - /* Skip MC reset as it's mostly likely not hung, just busy */ - if (reset_mask & AMDGPU_RESET_MC) { - DRM_DEBUG("MC busy: 0x%08X, clearing.\n", reset_mask); - reset_mask &= ~AMDGPU_RESET_MC; - } - - return reset_mask; -} - -/** - * vi_gpu_soft_reset - soft reset GPU - * - * @adev: amdgpu_device pointer - * @reset_mask: mask of which blocks to reset - * - * Soft reset the blocks specified in @reset_mask. - */ -static void vi_gpu_soft_reset(struct amdgpu_device *adev, u32 reset_mask) -{ - struct amdgpu_mode_mc_save save; - u32 grbm_soft_reset = 0, srbm_soft_reset = 0; - u32 tmp; - - if (reset_mask == 0) - return; - - dev_info(adev->dev, "GPU softreset: 0x%08X\n", reset_mask); - - vi_print_gpu_status_regs(adev); - dev_info(adev->dev, " VM_CONTEXT1_PROTECTION_FAULT_ADDR 0x%08X\n", - RREG32(mmVM_CONTEXT1_PROTECTION_FAULT_ADDR)); - dev_info(adev->dev, " VM_CONTEXT1_PROTECTION_FAULT_STATUS 0x%08X\n", - RREG32(mmVM_CONTEXT1_PROTECTION_FAULT_STATUS)); - - /* disable CG/PG */ - - /* stop the rlc */ - //XXX - //gfx_v8_0_rlc_stop(adev); - - /* Disable GFX parsing/prefetching */ - tmp = RREG32(mmCP_ME_CNTL); - tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, ME_HALT, 1); - tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, PFP_HALT, 1); - tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, CE_HALT, 1); - WREG32(mmCP_ME_CNTL, tmp); - - /* Disable MEC parsing/prefetching */ - tmp = RREG32(mmCP_MEC_CNTL); - tmp = REG_SET_FIELD(tmp, CP_MEC_CNTL, MEC_ME1_HALT, 1); - tmp = REG_SET_FIELD(tmp, CP_MEC_CNTL, MEC_ME2_HALT, 1); - WREG32(mmCP_MEC_CNTL, tmp); - - if (reset_mask & AMDGPU_RESET_DMA) { - /* sdma0 */ - tmp = RREG32(mmSDMA0_F32_CNTL + SDMA0_REGISTER_OFFSET); - tmp = REG_SET_FIELD(tmp, SDMA0_F32_CNTL, HALT, 1); - WREG32(mmSDMA0_F32_CNTL + SDMA0_REGISTER_OFFSET, tmp); - } - if (reset_mask & AMDGPU_RESET_DMA1) { - /* sdma1 */ - tmp = RREG32(mmSDMA0_F32_CNTL + SDMA1_REGISTER_OFFSET); - tmp = REG_SET_FIELD(tmp, SDMA0_F32_CNTL, HALT, 1); - WREG32(mmSDMA0_F32_CNTL + SDMA1_REGISTER_OFFSET, tmp); - } - - gmc_v8_0_mc_stop(adev, &save); - if (amdgpu_asic_wait_for_mc_idle(adev)) { - dev_warn(adev->dev, "Wait for MC idle timedout !\n"); - } - - if (reset_mask & (AMDGPU_RESET_GFX | AMDGPU_RESET_COMPUTE | AMDGPU_RESET_CP)) { - grbm_soft_reset = - REG_SET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CP, 1); - grbm_soft_reset = - REG_SET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_GFX, 1); - } - - if (reset_mask & AMDGPU_RESET_CP) { - grbm_soft_reset = - REG_SET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CP, 1); - srbm_soft_reset = - REG_SET_FIELD(srbm_soft_reset, SRBM_SOFT_RESET, SOFT_RESET_GRBM, 1); - } - - if (reset_mask & AMDGPU_RESET_DMA) - srbm_soft_reset = - REG_SET_FIELD(srbm_soft_reset, SRBM_SOFT_RESET, SOFT_RESET_SDMA, 1); - - if (reset_mask & AMDGPU_RESET_DMA1) - srbm_soft_reset = - REG_SET_FIELD(srbm_soft_reset, SRBM_SOFT_RESET, SOFT_RESET_SDMA1, 1); - - if (reset_mask & AMDGPU_RESET_DISPLAY) - srbm_soft_reset = - REG_SET_FIELD(srbm_soft_reset, SRBM_SOFT_RESET, SOFT_RESET_DC, 1); - - if (reset_mask & AMDGPU_RESET_RLC) - grbm_soft_reset = - REG_SET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_RLC, 1); - - if (reset_mask & AMDGPU_RESET_SEM) - srbm_soft_reset = - REG_SET_FIELD(srbm_soft_reset, SRBM_SOFT_RESET, SOFT_RESET_SEM, 1); - - if (reset_mask & AMDGPU_RESET_IH) - srbm_soft_reset = - REG_SET_FIELD(srbm_soft_reset, SRBM_SOFT_RESET, SOFT_RESET_IH, 1); - - if (reset_mask & AMDGPU_RESET_GRBM) - srbm_soft_reset = - REG_SET_FIELD(srbm_soft_reset, SRBM_SOFT_RESET, SOFT_RESET_GRBM, 1); - - if (reset_mask & AMDGPU_RESET_VMC) - srbm_soft_reset = - REG_SET_FIELD(srbm_soft_reset, SRBM_SOFT_RESET, SOFT_RESET_VMC, 1); - - if (reset_mask & AMDGPU_RESET_UVD) - srbm_soft_reset = - REG_SET_FIELD(srbm_soft_reset, SRBM_SOFT_RESET, SOFT_RESET_UVD, 1); - - if (reset_mask & AMDGPU_RESET_VCE) - srbm_soft_reset = - REG_SET_FIELD(srbm_soft_reset, SRBM_SOFT_RESET, SOFT_RESET_VCE0, 1); - - if (reset_mask & AMDGPU_RESET_VCE) - srbm_soft_reset = - REG_SET_FIELD(srbm_soft_reset, SRBM_SOFT_RESET, SOFT_RESET_VCE1, 1); - - if (!(adev->flags & AMD_IS_APU)) { - if (reset_mask & AMDGPU_RESET_MC) - srbm_soft_reset = - REG_SET_FIELD(srbm_soft_reset, SRBM_SOFT_RESET, SOFT_RESET_MC, 1); - } - - if (grbm_soft_reset) { - tmp = RREG32(mmGRBM_SOFT_RESET); - tmp |= grbm_soft_reset; - dev_info(adev->dev, "GRBM_SOFT_RESET=0x%08X\n", tmp); - WREG32(mmGRBM_SOFT_RESET, tmp); - tmp = RREG32(mmGRBM_SOFT_RESET); - - udelay(50); - - tmp &= ~grbm_soft_reset; - WREG32(mmGRBM_SOFT_RESET, tmp); - tmp = RREG32(mmGRBM_SOFT_RESET); - } - - if (srbm_soft_reset) { - tmp = RREG32(mmSRBM_SOFT_RESET); - tmp |= srbm_soft_reset; - dev_info(adev->dev, "SRBM_SOFT_RESET=0x%08X\n", tmp); - WREG32(mmSRBM_SOFT_RESET, tmp); - tmp = RREG32(mmSRBM_SOFT_RESET); - - udelay(50); - - tmp &= ~srbm_soft_reset; - WREG32(mmSRBM_SOFT_RESET, tmp); - tmp = RREG32(mmSRBM_SOFT_RESET); - } - - /* Wait a little for things to settle down */ - udelay(50); - - gmc_v8_0_mc_resume(adev, &save); - udelay(50); - - vi_print_gpu_status_regs(adev); -} - static void vi_gpu_pci_config_reset(struct amdgpu_device *adev) { - struct amdgpu_mode_mc_save save; - u32 tmp, i; + u32 i; dev_info(adev->dev, "GPU pci config reset\n"); - /* disable dpm? */ - - /* disable cg/pg */ - - /* Disable GFX parsing/prefetching */ - tmp = RREG32(mmCP_ME_CNTL); - tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, ME_HALT, 1); - tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, PFP_HALT, 1); - tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, CE_HALT, 1); - WREG32(mmCP_ME_CNTL, tmp); - - /* Disable MEC parsing/prefetching */ - tmp = RREG32(mmCP_MEC_CNTL); - tmp = REG_SET_FIELD(tmp, CP_MEC_CNTL, MEC_ME1_HALT, 1); - tmp = REG_SET_FIELD(tmp, CP_MEC_CNTL, MEC_ME2_HALT, 1); - WREG32(mmCP_MEC_CNTL, tmp); - - /* Disable GFX parsing/prefetching */ - WREG32(mmCP_ME_CNTL, CP_ME_CNTL__ME_HALT_MASK | - CP_ME_CNTL__PFP_HALT_MASK | CP_ME_CNTL__CE_HALT_MASK); - - /* Disable MEC parsing/prefetching */ - WREG32(mmCP_MEC_CNTL, - CP_MEC_CNTL__MEC_ME1_HALT_MASK | CP_MEC_CNTL__MEC_ME2_HALT_MASK); - - /* sdma0 */ - tmp = RREG32(mmSDMA0_F32_CNTL + SDMA0_REGISTER_OFFSET); - tmp = REG_SET_FIELD(tmp, SDMA0_F32_CNTL, HALT, 1); - WREG32(mmSDMA0_F32_CNTL + SDMA0_REGISTER_OFFSET, tmp); - - /* sdma1 */ - tmp = RREG32(mmSDMA0_F32_CNTL + SDMA1_REGISTER_OFFSET); - tmp = REG_SET_FIELD(tmp, SDMA0_F32_CNTL, HALT, 1); - WREG32(mmSDMA0_F32_CNTL + SDMA1_REGISTER_OFFSET, tmp); - - /* XXX other engines? */ - - /* halt the rlc, disable cp internal ints */ - //XXX - //gfx_v8_0_rlc_stop(adev); - - udelay(50); - - /* disable mem access */ - gmc_v8_0_mc_stop(adev, &save); - if (amdgpu_asic_wait_for_mc_idle(adev)) { - dev_warn(adev->dev, "Wait for MC idle timed out !\n"); - } - /* disable BM */ pci_clear_master(adev->pdev); /* reset */ @@ -926,26 +619,11 @@ static void vi_set_bios_scratch_engine_hung(struct amdgpu_device *adev, bool hun */ static int vi_asic_reset(struct amdgpu_device *adev) { - u32 reset_mask; - - reset_mask = vi_gpu_check_soft_reset(adev); + vi_set_bios_scratch_engine_hung(adev, true); - if (reset_mask) - vi_set_bios_scratch_engine_hung(adev, true); + vi_gpu_pci_config_reset(adev); - /* try soft reset */ - vi_gpu_soft_reset(adev, reset_mask); - - reset_mask = vi_gpu_check_soft_reset(adev); - - /* try pci config reset */ - if (reset_mask && amdgpu_hard_reset) - vi_gpu_pci_config_reset(adev); - - reset_mask = vi_gpu_check_soft_reset(adev); - - if (!reset_mask) - vi_set_bios_scratch_engine_hung(adev, false); + vi_set_bios_scratch_engine_hung(adev, false); return 0; } @@ -1002,9 +680,6 @@ static int vi_set_vce_clocks(struct amdgpu_device *adev, u32 evclk, u32 ecclk) static void vi_pcie_gen3_enable(struct amdgpu_device *adev) { - u32 mask; - int ret; - if (pci_is_root_bus(adev->pdev->bus)) return; @@ -1014,11 +689,8 @@ static void vi_pcie_gen3_enable(struct amdgpu_device *adev) if (adev->flags & AMD_IS_APU) return; - ret = drm_pcie_get_speed_cap_mask(adev->ddev, &mask); - if (ret != 0) - return; - - if (!(mask & (DRM_PCIE_SPEED_50 | DRM_PCIE_SPEED_80))) + if (!(adev->pm.pcie_gen_mask & (CAIL_PCIE_LINK_SPEED_SUPPORT_GEN2 | + CAIL_PCIE_LINK_SPEED_SUPPORT_GEN3))) return; /* todo */ @@ -1064,10 +736,10 @@ static const struct amdgpu_ip_block_version topaz_ip_blocks[] = }, { .type = AMD_IP_BLOCK_TYPE_GMC, - .major = 8, - .minor = 0, + .major = 7, + .minor = 4, .rev = 0, - .funcs = &gmc_v8_0_ip_funcs, + .funcs = &gmc_v7_0_ip_funcs, }, { .type = AMD_IP_BLOCK_TYPE_IH, @@ -1081,7 +753,7 @@ static const struct amdgpu_ip_block_version topaz_ip_blocks[] = .major = 7, .minor = 1, .rev = 0, - .funcs = &iceland_dpm_ip_funcs, + .funcs = &amdgpu_pp_ip_funcs, }, { .type = AMD_IP_BLOCK_TYPE_GFX, @@ -1128,7 +800,7 @@ static const struct amdgpu_ip_block_version tonga_ip_blocks[] = .major = 7, .minor = 1, .rev = 0, - .funcs = &tonga_dpm_ip_funcs, + .funcs = &amdgpu_pp_ip_funcs, }, { .type = AMD_IP_BLOCK_TYPE_DCE, @@ -1196,7 +868,7 @@ static const struct amdgpu_ip_block_version fiji_ip_blocks[] = .major = 7, .minor = 1, .rev = 0, - .funcs = &fiji_dpm_ip_funcs, + .funcs = &amdgpu_pp_ip_funcs, }, { .type = AMD_IP_BLOCK_TYPE_DCE, @@ -1264,7 +936,7 @@ static const struct amdgpu_ip_block_version cz_ip_blocks[] = .major = 8, .minor = 0, .rev = 0, - .funcs = &cz_dpm_ip_funcs, + .funcs = &amdgpu_pp_ip_funcs }, { .type = AMD_IP_BLOCK_TYPE_DCE, @@ -1301,6 +973,15 @@ static const struct amdgpu_ip_block_version cz_ip_blocks[] = .rev = 0, .funcs = &vce_v3_0_ip_funcs, }, +#if defined(CONFIG_DRM_AMD_ACP) + { + .type = AMD_IP_BLOCK_TYPE_ACP, + .major = 2, + .minor = 2, + .rev = 0, + .funcs = &acp_ip_funcs, + }, +#endif }; int vi_set_ip_blocks(struct amdgpu_device *adev) @@ -1319,6 +1000,7 @@ int vi_set_ip_blocks(struct amdgpu_device *adev) adev->num_ip_blocks = ARRAY_SIZE(tonga_ip_blocks); break; case CHIP_CARRIZO: + case CHIP_STONEY: adev->ip_blocks = cz_ip_blocks; adev->num_ip_blocks = ARRAY_SIZE(cz_ip_blocks); break; @@ -1330,19 +1012,24 @@ int vi_set_ip_blocks(struct amdgpu_device *adev) return 0; } +#define ATI_REV_ID_FUSE_MACRO__ADDRESS 0xC0014044 +#define ATI_REV_ID_FUSE_MACRO__SHIFT 9 +#define ATI_REV_ID_FUSE_MACRO__MASK 0x00001E00 + static uint32_t vi_get_rev_id(struct amdgpu_device *adev) { - if (adev->asic_type == CHIP_TOPAZ) + if (adev->flags & AMD_IS_APU) + return (RREG32_SMC(ATI_REV_ID_FUSE_MACRO__ADDRESS) & ATI_REV_ID_FUSE_MACRO__MASK) + >> ATI_REV_ID_FUSE_MACRO__SHIFT; + else return (RREG32(mmPCIE_EFUSE4) & PCIE_EFUSE4__STRAP_BIF_ATI_REV_ID_MASK) >> PCIE_EFUSE4__STRAP_BIF_ATI_REV_ID__SHIFT; - else - return (RREG32(mmCC_DRM_ID_STRAPS) & CC_DRM_ID_STRAPS__ATI_REV_ID_MASK) - >> CC_DRM_ID_STRAPS__ATI_REV_ID__SHIFT; } static const struct amdgpu_asic_funcs vi_asic_funcs = { .read_disabled_bios = &vi_read_disabled_bios, + .read_bios_from_rom = &vi_read_bios_from_rom, .read_register = &vi_read_register, .reset = &vi_asic_reset, .set_vga_state = &vi_vga_set_state, @@ -1384,36 +1071,36 @@ static int vi_common_early_init(void *handle) adev->external_rev_id = 0xFF; switch (adev->asic_type) { case CHIP_TOPAZ: - adev->has_uvd = false; adev->cg_flags = 0; adev->pg_flags = 0; adev->external_rev_id = 0x1; - if (amdgpu_smc_load_fw && smc_enabled) - adev->firmware.smu_load = true; break; case CHIP_FIJI: + adev->cg_flags = 0; + adev->pg_flags = 0; + adev->external_rev_id = adev->rev_id + 0x3c; + break; case CHIP_TONGA: - adev->has_uvd = true; adev->cg_flags = 0; adev->pg_flags = 0; adev->external_rev_id = adev->rev_id + 0x14; - if (amdgpu_smc_load_fw && smc_enabled) - adev->firmware.smu_load = true; break; case CHIP_CARRIZO: - adev->has_uvd = true; + case CHIP_STONEY: adev->cg_flags = 0; - /* Disable UVD pg */ - adev->pg_flags = /* AMDGPU_PG_SUPPORT_UVD | */AMDGPU_PG_SUPPORT_VCE; + adev->pg_flags = 0; adev->external_rev_id = adev->rev_id + 0x1; - if (amdgpu_smc_load_fw && smc_enabled) - adev->firmware.smu_load = true; break; default: /* FIXME: not supported yet */ return -EINVAL; } + if (amdgpu_smc_load_fw && smc_enabled) + adev->firmware.smu_load = true; + + amdgpu_get_pcie_info(adev); + return 0; } @@ -1487,9 +1174,95 @@ static int vi_common_soft_reset(void *handle) return 0; } +static void fiji_update_bif_medium_grain_light_sleep(struct amdgpu_device *adev, + bool enable) +{ + uint32_t temp, data; + + temp = data = RREG32_PCIE(ixPCIE_CNTL2); + + if (enable) + data |= PCIE_CNTL2__SLV_MEM_LS_EN_MASK | + PCIE_CNTL2__MST_MEM_LS_EN_MASK | + PCIE_CNTL2__REPLAY_MEM_LS_EN_MASK; + else + data &= ~(PCIE_CNTL2__SLV_MEM_LS_EN_MASK | + PCIE_CNTL2__MST_MEM_LS_EN_MASK | + PCIE_CNTL2__REPLAY_MEM_LS_EN_MASK); + + if (temp != data) + WREG32_PCIE(ixPCIE_CNTL2, data); +} + +static void fiji_update_hdp_medium_grain_clock_gating(struct amdgpu_device *adev, + bool enable) +{ + uint32_t temp, data; + + temp = data = RREG32(mmHDP_HOST_PATH_CNTL); + + if (enable) + data &= ~HDP_HOST_PATH_CNTL__CLOCK_GATING_DIS_MASK; + else + data |= HDP_HOST_PATH_CNTL__CLOCK_GATING_DIS_MASK; + + if (temp != data) + WREG32(mmHDP_HOST_PATH_CNTL, data); +} + +static void fiji_update_hdp_light_sleep(struct amdgpu_device *adev, + bool enable) +{ + uint32_t temp, data; + + temp = data = RREG32(mmHDP_MEM_POWER_LS); + + if (enable) + data |= HDP_MEM_POWER_LS__LS_ENABLE_MASK; + else + data &= ~HDP_MEM_POWER_LS__LS_ENABLE_MASK; + + if (temp != data) + WREG32(mmHDP_MEM_POWER_LS, data); +} + +static void fiji_update_rom_medium_grain_clock_gating(struct amdgpu_device *adev, + bool enable) +{ + uint32_t temp, data; + + temp = data = RREG32_SMC(ixCGTT_ROM_CLK_CTRL0); + + if (enable) + data &= ~(CGTT_ROM_CLK_CTRL0__SOFT_OVERRIDE0_MASK | + CGTT_ROM_CLK_CTRL0__SOFT_OVERRIDE1_MASK); + else + data |= CGTT_ROM_CLK_CTRL0__SOFT_OVERRIDE0_MASK | + CGTT_ROM_CLK_CTRL0__SOFT_OVERRIDE1_MASK; + + if (temp != data) + WREG32_SMC(ixCGTT_ROM_CLK_CTRL0, data); +} + static int vi_common_set_clockgating_state(void *handle, enum amd_clockgating_state state) { + struct amdgpu_device *adev = (struct amdgpu_device *)handle; + + switch (adev->asic_type) { + case CHIP_FIJI: + fiji_update_bif_medium_grain_light_sleep(adev, + state == AMD_CG_STATE_GATE ? true : false); + fiji_update_hdp_medium_grain_clock_gating(adev, + state == AMD_CG_STATE_GATE ? true : false); + fiji_update_hdp_light_sleep(adev, + state == AMD_CG_STATE_GATE ? true : false); + fiji_update_rom_medium_grain_clock_gating(adev, + state == AMD_CG_STATE_GATE ? true : false); + break; + default: + break; + } return 0; } diff --git a/drivers/gpu/drm/amd/amdgpu/vid.h b/drivers/gpu/drm/amd/amdgpu/vid.h index d98aa9d82fa1..ace49976f7be 100644 --- a/drivers/gpu/drm/amd/amdgpu/vid.h +++ b/drivers/gpu/drm/amd/amdgpu/vid.h @@ -71,8 +71,6 @@ #define VMID(x) ((x) << 4) #define QUEUEID(x) ((x) << 8) -#define RB_BITMAP_WIDTH_PER_SH 2 - #define MC_SEQ_MISC0__MT__MASK 0xf0000000 #define MC_SEQ_MISC0__MT__GDDR1 0x10000000 #define MC_SEQ_MISC0__MT__DDR2 0x20000000 |