From 88e72717c2de4181d8a6de1b04315953ad2bebdf Mon Sep 17 00:00:00 2001 From: Thierry Reding Date: Thu, 24 Sep 2015 18:35:31 +0200 Subject: drm/irq: Use unsigned int pipe in public API MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit This continues the pattern started in commit cc1ef118fc09 ("drm/irq: Make pipe unsigned and name consistent"). This is applied to the public APIs and driver callbacks, so pretty much all drivers need to be updated to match the new prototypes. Cc: Christian König Cc: Alex Deucher Cc: Russell King Cc: Inki Dae Cc: Jianwei Wang Cc: Alison Wang Cc: Patrik Jakobsson Cc: Daniel Vetter Cc: Jani Nikula Cc: Philipp Zabel Cc: David Airlie Cc: Rob Clark Cc: Ben Skeggs Cc: Tomi Valkeinen Cc: Laurent Pinchart Cc: Mark Yao Cc: Benjamin Gaignard Cc: Vincent Abriou Cc: Thomas Hellstrom Signed-off-by: Thierry Reding Reviewed-by: Ville Syrjälä Signed-off-by: Daniel Vetter --- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu.h') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h index 6647fb26ef25..5e43178d07d5 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h @@ -2349,10 +2349,10 @@ 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); -- cgit v1.2.3 From c113ea1c4f4a7592f17e53d658873b83af0dbb4b Mon Sep 17 00:00:00 2001 From: Alex Deucher Date: Thu, 8 Oct 2015 16:30:37 -0400 Subject: drm/amdgpu: rework sdma structures MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Rework the sdma structures in the driver to consolidate all of the sdma info into a single structure and allow for asics that may have different numbers of sdma instances. Reviewed-by: Christian König Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 22 +-- drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c | 4 +- drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c | 4 +- drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 7 +- drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c | 10 +- drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c | 4 +- drivers/gpu/drm/amd/amdgpu/cik_sdma.c | 130 ++++++++--------- drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c | 156 ++++++++++---------- drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c | 166 +++++++++++----------- 9 files changed, 245 insertions(+), 258 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu.h') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h index 6647fb26ef25..afc984806c4c 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h @@ -1708,7 +1708,7 @@ struct amdgpu_vce { /* * SDMA */ -struct amdgpu_sdma { +struct amdgpu_sdma_instance { /* SDMA firmware */ const struct firmware *fw; uint32_t fw_version; @@ -1718,6 +1718,13 @@ 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 */ @@ -2064,9 +2071,7 @@ 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; @@ -2203,17 +2208,18 @@ static inline void amdgpu_ring_write(struct amdgpu_ring *ring, uint32_t v) 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; } 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..0e1376317683 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c @@ -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..79fa5c7de856 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c @@ -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_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c index 749420f1ea6f..29fc45ce64dd 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c @@ -104,10 +104,11 @@ 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; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c index 78233225dd36..3f5f2d58ad94 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; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c index 30dce235ddeb..b13a74b273a6 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c @@ -540,8 +540,8 @@ 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/cik_sdma.c b/drivers/gpu/drm/amd/amdgpu/cik_sdma.c index 9ea9de457da3..814598e76c98 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++) @@ -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; @@ -327,8 +327,8 @@ 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 +336,7 @@ 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 +376,7 @@ 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 +402,8 @@ 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); @@ -502,26 +502,25 @@ 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; @@ -830,7 +829,7 @@ static void cik_sdma_vm_set_pte_pde(struct amdgpu_ib *ib, */ static void cik_sdma_vm_pad_ib(struct amdgpu_ib *ib) { - struct amdgpu_sdma *sdma = amdgpu_get_sdma_instance(ib->ring); + struct amdgpu_sdma_instance *sdma = amdgpu_get_sdma_instance(ib->ring); u32 pad_count; int i; @@ -934,6 +933,8 @@ 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 +947,7 @@ 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 +956,33 @@ 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 +990,10 @@ 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 +1070,7 @@ 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", @@ -1223,7 +1215,7 @@ 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 +1228,7 @@ 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 */ @@ -1334,8 +1326,10 @@ static const struct amdgpu_ring_funcs cik_sdma_ring_funcs = { 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 +1343,9 @@ 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 +1410,7 @@ 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; } } @@ -1431,7 +1425,7 @@ static void cik_sdma_set_vm_pte_funcs(struct amdgpu_device *adev) { 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 = &adev->sdma.instance[0].ring; adev->vm_manager.vm_pte_funcs_ring->is_pte_ring = true; } } diff --git a/drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c b/drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c index 14e87234171a..f8b868c7c496 100644 --- a/drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c +++ b/drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c @@ -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++) @@ -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); @@ -368,8 +368,8 @@ 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 +377,7 @@ 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 +419,7 @@ 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 +445,8 @@ 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); @@ -545,29 +545,23 @@ 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; @@ -894,7 +888,7 @@ static void sdma_v2_4_vm_set_pte_pde(struct amdgpu_ib *ib, */ static void sdma_v2_4_vm_pad_ib(struct amdgpu_ib *ib) { - struct amdgpu_sdma *sdma = amdgpu_get_sdma_instance(ib->ring); + struct amdgpu_sdma_instance *sdma = amdgpu_get_sdma_instance(ib->ring); u32 pad_count; int i; @@ -952,6 +946,8 @@ 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 +959,21 @@ 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 +983,20 @@ 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 +1004,10 @@ 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 +1086,7 @@ 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", @@ -1243,7 +1229,7 @@ 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 +1242,7 @@ 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 */ @@ -1345,8 +1331,10 @@ static const struct amdgpu_ring_funcs sdma_v2_4_ring_funcs = { 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 +1348,9 @@ 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 +1416,7 @@ 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; } } @@ -1443,7 +1431,7 @@ static void sdma_v2_4_set_vm_pte_funcs(struct amdgpu_device *adev) { 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 = &adev->sdma.instance[0].ring; adev->vm_manager.vm_pte_funcs_ring->is_pte_ring = true; } } diff --git a/drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c b/drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c index 9bfe92df15f7..670555a45da9 100644 --- a/drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c +++ b/drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c @@ -184,7 +184,7 @@ 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; @@ -204,27 +204,27 @@ static int sdma_v3_0_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); @@ -235,9 +235,9 @@ 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 +276,7 @@ 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 +300,7 @@ 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 +308,7 @@ 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++) @@ -369,7 +369,7 @@ 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); @@ -454,8 +454,8 @@ 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 +463,7 @@ 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 +500,7 @@ 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 +530,7 @@ 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 +557,8 @@ 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); @@ -669,23 +669,22 @@ 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 +700,21 @@ 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; + 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 */ @@ -1013,7 +1012,7 @@ static void sdma_v3_0_vm_set_pte_pde(struct amdgpu_ib *ib, */ static void sdma_v3_0_vm_pad_ib(struct amdgpu_ib *ib) { - struct amdgpu_sdma *sdma = amdgpu_get_sdma_instance(ib->ring); + struct amdgpu_sdma_instance *sdma = amdgpu_get_sdma_instance(ib->ring); u32 pad_count; int i; @@ -1071,6 +1070,12 @@ static int sdma_v3_0_early_init(void *handle) { struct amdgpu_device *adev = (struct amdgpu_device *)handle; + switch (adev->asic_type) { + 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 +1087,21 @@ 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 +1111,23 @@ 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 +1135,10 @@ 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 +1218,7 @@ 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", @@ -1367,7 +1363,7 @@ 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 +1376,7 @@ 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 */ @@ -1468,8 +1464,10 @@ static const struct amdgpu_ring_funcs sdma_v3_0_ring_funcs = { 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 +1481,9 @@ 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 +1549,7 @@ 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; } } @@ -1566,7 +1564,7 @@ static void sdma_v3_0_set_vm_pte_funcs(struct amdgpu_device *adev) { 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 = &adev->sdma.instance[0].ring; adev->vm_manager.vm_pte_funcs_ring->is_pte_ring = true; } } -- cgit v1.2.3 From d9c13156a628cc9f8f062f2c10e2bff55b92aaab Mon Sep 17 00:00:00 2001 From: Christian König Date: Mon, 28 Sep 2015 12:31:26 +0200 Subject: drm/amdgpu: add option to stop on VM fault MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Signed-off-by: Christian König Reviewed-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 6 +++++ drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c | 4 ++++ drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c | 40 ++++++++++++++++++++++++++++----- drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c | 36 +++++++++++++++++++++++++++++ 4 files changed, 80 insertions(+), 6 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu.h') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h index afc984806c4c..a3dbbd1fc060 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h @@ -79,6 +79,7 @@ extern int amdgpu_bapm; extern int amdgpu_deep_color; extern int amdgpu_vm_size; extern int amdgpu_vm_block_size; +extern int amdgpu_vm_fault_stop; extern int amdgpu_enable_scheduler; extern int amdgpu_sched_jobs; extern int amdgpu_sched_hw_submission; @@ -960,6 +961,11 @@ 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; uint64_t addr; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c index c183772de7c1..bec091632194 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c @@ -75,6 +75,7 @@ int amdgpu_bapm = -1; int amdgpu_deep_color = 0; int amdgpu_vm_size = 8; int amdgpu_vm_block_size = -1; +int amdgpu_vm_fault_stop = 0; int amdgpu_exp_hw_support = 0; int amdgpu_enable_scheduler = 1; int amdgpu_sched_jobs = 16; @@ -141,6 +142,9 @@ 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(exp_hw_support, "experimental hw support (1 = enable, 0 = disable (default))"); module_param_named(exp_hw_support, amdgpu_exp_hw_support, int, 0444); diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c b/drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c index fab5471d25d7..488348272c4d 100644 --- a/drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c @@ -435,6 +435,33 @@ static int gmc_v7_0_gart_set_pte_pde(struct amdgpu_device *adev, return 0; } +/** + * 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 * @@ -523,15 +550,13 @@ 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); @@ -1268,6 +1293,9 @@ 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..42b5ff827055 100644 --- a/drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c @@ -549,6 +549,35 @@ static int gmc_v8_0_gart_set_pte_pde(struct amdgpu_device *adev, return 0; } +/** + * 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 * @@ -663,6 +692,10 @@ 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", @@ -1268,6 +1301,9 @@ 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", -- cgit v1.2.3 From b495bd3a54e732e846bfde49ba2ea93b62b1de91 Mon Sep 17 00:00:00 2001 From: Christian König Date: Thu, 10 Sep 2015 14:00:35 +0200 Subject: drm/amdgpu: add option to clear VM page tables after every submit MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit This makes it much easier to find when userspace misses to send some buffers. Signed-off-by: Christian König Reviewed-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 1 + drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 17 ++++++++++++++++- drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c | 4 ++++ 3 files changed, 21 insertions(+), 1 deletion(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu.h') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h index a3dbbd1fc060..736eb4560fd5 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h @@ -80,6 +80,7 @@ extern int amdgpu_deep_color; extern int amdgpu_vm_size; extern int amdgpu_vm_block_size; extern int amdgpu_vm_fault_stop; +extern int amdgpu_vm_debug; extern int amdgpu_enable_scheduler; extern int amdgpu_sched_jobs; extern int amdgpu_sched_hw_submission; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c index 29fc45ce64dd..25012c790f8f 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c @@ -567,9 +567,24 @@ static int amdgpu_bo_vm_update_pte(struct amdgpu_cs_parser *p, if (r) return r; } + + } + + r = amdgpu_vm_clear_invalids(adev, vm, &p->ibs[0].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, diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c index bec091632194..6134b1ece17f 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c @@ -76,6 +76,7 @@ int amdgpu_deep_color = 0; int amdgpu_vm_size = 8; 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 = 1; int amdgpu_sched_jobs = 16; @@ -145,6 +146,9 @@ 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); -- cgit v1.2.3 From b7e4dad3e1fc5d3909737fb72e57aedeb0072c7f Mon Sep 17 00:00:00 2001 From: Christian König Date: Tue, 1 Sep 2015 10:50:26 +0200 Subject: drm/amdgpu: remove old lockup detection infrastructure MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit It didn't worked to well anyway. Signed-off-by: Christian König Reviewed-by: Chunming Zhou Reviewed-by: Junwei Zhang --- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 7 ----- drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 6 ----- drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c | 20 +------------- drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c | 1 - drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c | 43 ------------------------------ drivers/gpu/drm/amd/amdgpu/cik_sdma.c | 19 ------------- drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c | 20 -------------- drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c | 11 -------- drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c | 19 ------------- drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c | 19 ------------- drivers/gpu/drm/amd/amdgpu/uvd_v4_2.c | 1 - drivers/gpu/drm/amd/amdgpu/uvd_v5_0.c | 1 - drivers/gpu/drm/amd/amdgpu/uvd_v6_0.c | 1 - drivers/gpu/drm/amd/amdgpu/vce_v2_0.c | 1 - drivers/gpu/drm/amd/amdgpu/vce_v3_0.c | 1 - 15 files changed, 1 insertion(+), 169 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu.h') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h index 736eb4560fd5..911d67032d57 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h @@ -345,7 +345,6 @@ 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); }; @@ -907,8 +906,6 @@ struct amdgpu_ring { unsigned ring_size; unsigned ring_free_dw; int count_dw; - atomic_t last_rptr; - atomic64_t last_activity; uint64_t gpu_addr; uint32_t align_mask; uint32_t ptr_mask; @@ -1230,8 +1227,6 @@ 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, @@ -1974,7 +1969,6 @@ 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]; @@ -2253,7 +2247,6 @@ amdgpu_get_sdma_instance(struct amdgpu_ring *ring) #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)) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c index 901a460b2c55..28c9ee334dd7 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c @@ -1816,12 +1816,6 @@ int amdgpu_gpu_reset(struct amdgpu_device *adev) 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 */ diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c index fcad7e060938..4010aa6b4e53 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c @@ -269,17 +269,6 @@ static void amdgpu_fence_check_lockup(struct work_struct *work) 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); } @@ -380,7 +369,6 @@ static bool amdgpu_fence_enable_signaling(struct fence *f) */ static int amdgpu_fence_ring_wait_seq(struct amdgpu_ring *ring, uint64_t seq) { - struct amdgpu_device *adev = ring->adev; bool signaled = false; BUG_ON(!ring); @@ -391,8 +379,7 @@ static int amdgpu_fence_ring_wait_seq(struct amdgpu_ring *ring, uint64_t seq) return 0; wait_event(ring->fence_drv.fence_queue, ( - (signaled = amdgpu_fence_seq_signaled(ring, seq)) - || adev->needs_reset)); + (signaled = amdgpu_fence_seq_signaled(ring, seq)))); if (signaled) return 0; @@ -939,11 +926,6 @@ signed long amdgpu_fence_wait_any(struct amdgpu_device *adev, if (amdgpu_test_signaled_any(array, count)) break; - if (adev->needs_reset) { - t = -EDEADLK; - break; - } - t = schedule_timeout(t); if (t > 0 && intr && signal_pending(current)) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c index c439735ee670..aad4c1c69448 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c @@ -298,7 +298,6 @@ 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_ring.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c index b13a74b273a6..b2df348aa223 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c @@ -67,8 +67,6 @@ void amdgpu_ring_free_size(struct amdgpu_ring *ring) 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); } } @@ -208,46 +206,6 @@ void amdgpu_ring_unlock_undo(struct amdgpu_ring *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 * @@ -436,7 +394,6 @@ int amdgpu_ring_init(struct amdgpu_device *adev, struct amdgpu_ring *ring, if (amdgpu_debugfs_ring_init(adev, ring)) { DRM_ERROR("Failed to register debugfs file for rings !\n"); } - amdgpu_ring_lockup_update(ring); return 0; } diff --git a/drivers/gpu/drm/amd/amdgpu/cik_sdma.c b/drivers/gpu/drm/amd/amdgpu/cik_sdma.c index 814598e76c98..5f712ceddf08 100644 --- a/drivers/gpu/drm/amd/amdgpu/cik_sdma.c +++ b/drivers/gpu/drm/amd/amdgpu/cik_sdma.c @@ -1290,24 +1290,6 @@ 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, @@ -1320,7 +1302,6 @@ static const struct amdgpu_ring_funcs cik_sdma_ring_funcs = { .emit_hdp_flush = cik_sdma_ring_emit_hdp_flush, .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, }; diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c index e992bf2ff66c..72793f93e2fc 100644 --- a/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c @@ -5542,24 +5542,6 @@ 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, @@ -5573,7 +5555,6 @@ static const struct amdgpu_ring_funcs gfx_v7_0_ring_funcs_gfx = { .emit_hdp_flush = gfx_v7_0_ring_emit_hdp_flush, .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, }; @@ -5590,7 +5571,6 @@ static const struct amdgpu_ring_funcs gfx_v7_0_ring_funcs_compute = { .emit_hdp_flush = gfx_v7_0_ring_emit_hdp_flush, .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, }; diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c index 7dd7651b0ba7..e0355eaddabe 100644 --- a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c @@ -4075,15 +4075,6 @@ 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]; @@ -4365,7 +4356,6 @@ static const struct amdgpu_ring_funcs gfx_v8_0_ring_funcs_gfx = { .emit_hdp_flush = gfx_v8_0_ring_emit_hdp_flush, .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, }; @@ -4382,7 +4372,6 @@ static const struct amdgpu_ring_funcs gfx_v8_0_ring_funcs_compute = { .emit_hdp_flush = gfx_v8_0_ring_emit_hdp_flush, .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, }; diff --git a/drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c b/drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c index f8b868c7c496..2cf50180cc51 100644 --- a/drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c +++ b/drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c @@ -1295,24 +1295,6 @@ 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, @@ -1325,7 +1307,6 @@ static const struct amdgpu_ring_funcs sdma_v2_4_ring_funcs = { .emit_hdp_flush = sdma_v2_4_ring_emit_hdp_flush, .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, }; diff --git a/drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c b/drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c index 670555a45da9..0271bf9c0609 100644 --- a/drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c +++ b/drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c @@ -1428,24 +1428,6 @@ 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, @@ -1458,7 +1440,6 @@ static const struct amdgpu_ring_funcs sdma_v3_0_ring_funcs = { .emit_hdp_flush = sdma_v3_0_ring_emit_hdp_flush, .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, }; diff --git a/drivers/gpu/drm/amd/amdgpu/uvd_v4_2.c b/drivers/gpu/drm/amd/amdgpu/uvd_v4_2.c index ed50dd725788..5e9f73af83a8 100644 --- a/drivers/gpu/drm/amd/amdgpu/uvd_v4_2.c +++ b/drivers/gpu/drm/amd/amdgpu/uvd_v4_2.c @@ -885,7 +885,6 @@ static const struct amdgpu_ring_funcs uvd_v4_2_ring_funcs = { .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, }; diff --git a/drivers/gpu/drm/amd/amdgpu/uvd_v5_0.c b/drivers/gpu/drm/amd/amdgpu/uvd_v5_0.c index 9ad8b9906c0b..38864f562981 100644 --- a/drivers/gpu/drm/amd/amdgpu/uvd_v5_0.c +++ b/drivers/gpu/drm/amd/amdgpu/uvd_v5_0.c @@ -824,7 +824,6 @@ static const struct amdgpu_ring_funcs uvd_v5_0_ring_funcs = { .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, }; diff --git a/drivers/gpu/drm/amd/amdgpu/uvd_v6_0.c b/drivers/gpu/drm/amd/amdgpu/uvd_v6_0.c index 7e9934fa4193..121915bbc3b6 100644 --- a/drivers/gpu/drm/amd/amdgpu/uvd_v6_0.c +++ b/drivers/gpu/drm/amd/amdgpu/uvd_v6_0.c @@ -808,7 +808,6 @@ static const struct amdgpu_ring_funcs uvd_v6_0_ring_funcs = { .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, }; diff --git a/drivers/gpu/drm/amd/amdgpu/vce_v2_0.c b/drivers/gpu/drm/amd/amdgpu/vce_v2_0.c index cd16df543f64..52ac7a8f1e58 100644 --- a/drivers/gpu/drm/amd/amdgpu/vce_v2_0.c +++ b/drivers/gpu/drm/amd/amdgpu/vce_v2_0.c @@ -642,7 +642,6 @@ static const struct amdgpu_ring_funcs vce_v2_0_ring_funcs = { .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, }; diff --git a/drivers/gpu/drm/amd/amdgpu/vce_v3_0.c b/drivers/gpu/drm/amd/amdgpu/vce_v3_0.c index f0656dfb53f3..b1bf27bb8f9e 100644 --- a/drivers/gpu/drm/amd/amdgpu/vce_v3_0.c +++ b/drivers/gpu/drm/amd/amdgpu/vce_v3_0.c @@ -643,7 +643,6 @@ static const struct amdgpu_ring_funcs vce_v3_0_ring_funcs = { .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, }; -- cgit v1.2.3 From 0c418f10104d4aa1d6b83698790898dc9ef1c12d Mon Sep 17 00:00:00 2001 From: Christian König Date: Tue, 1 Sep 2015 15:13:53 +0200 Subject: drm/amdgpu: remove the exclusive lock MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Finally getting rid of it. Signed-off-by: Christian König --- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 1 - drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 10 ++-------- drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 4 ---- drivers/gpu/drm/amd/amdgpu/amdgpu_display.c | 7 +------ drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c | 21 +++++---------------- drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c | 7 ------- 6 files changed, 8 insertions(+), 42 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu.h') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h index 911d67032d57..39d96088f7b4 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h @@ -1955,7 +1955,6 @@ struct amdgpu_device { struct device *dev; struct drm_device *ddev; struct pci_dev *pdev; - struct rw_semaphore exclusive_lock; /* ASIC */ enum amd_asic_type asic_type; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c index 25012c790f8f..5fdc0394561e 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c @@ -831,11 +831,8 @@ int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp) 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) @@ -843,8 +840,7 @@ int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *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; } @@ -915,14 +911,12 @@ int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp) mutex_unlock(&job->job_lock); amdgpu_cs_parser_fini_late(parser); - up_read(&adev->exclusive_lock); return 0; } cs->out.handle = parser->ibs[parser->num_ibs - 1].sequence; out: amdgpu_cs_parser_fini(parser, r, reserved_buffers); - up_read(&adev->exclusive_lock); r = amdgpu_cs_handle_lockup(adev, r); return r; } diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c index 28c9ee334dd7..19e185ec3707 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c @@ -1418,7 +1418,6 @@ int amdgpu_device_init(struct amdgpu_device *adev, 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); @@ -1814,8 +1813,6 @@ int amdgpu_gpu_reset(struct amdgpu_device *adev) int i, r; int resched; - down_write(&adev->exclusive_lock); - atomic_inc(&adev->gpu_reset_counter); /* block TTM */ @@ -1879,7 +1876,6 @@ retry: dev_info(adev->dev, "GPU reset failed\n"); } - up_write(&adev->exclusive_lock); return r; } diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_display.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_display.c index 9b34a3410c32..fdf7525cf120 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_display.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_display.c @@ -47,11 +47,8 @@ static void amdgpu_flip_wait_fence(struct amdgpu_device *adev, fence = to_amdgpu_fence(*f); if (fence) { r = fence_wait(&fence->base, false); - if (r == -EDEADLK) { - up_read(&adev->exclusive_lock); + if (r == -EDEADLK) r = amdgpu_gpu_reset(adev); - down_read(&adev->exclusive_lock); - } } else r = fence_wait(*f, false); @@ -77,7 +74,6 @@ static void amdgpu_flip_work_func(struct work_struct *__work) unsigned long flags; unsigned i; - 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]); @@ -93,7 +89,6 @@ static void amdgpu_flip_work_func(struct work_struct *__work) amdgpuCrtc->pflip_status = AMDGPU_FLIP_SUBMITTED; spin_unlock_irqrestore(&crtc->dev->event_lock, flags); - up_read(&adev->exclusive_lock); } /* diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c index 4010aa6b4e53..1fadc15e64ae 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c @@ -260,16 +260,8 @@ static void amdgpu_fence_check_lockup(struct work_struct *work) 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)) { + if (amdgpu_fence_activity(ring)) wake_up_all(&ring->fence_drv.fence_queue); - } - up_read(&ring->adev->exclusive_lock); } /** @@ -317,18 +309,15 @@ 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); + amdgpu_fence_process(ring); + + if (atomic64_read(&ring->fence_drv.last_seq) >= fence->seq) + return true; - if (atomic64_read(&ring->fence_drv.last_seq) >= fence->seq) - return true; - } return false; } diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c index 7297ca3a0ba7..d81ab785368a 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c @@ -181,7 +181,6 @@ 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 +213,9 @@ 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; } @@ -250,8 +247,6 @@ int amdgpu_gem_userptr_ioctl(struct drm_device *dev, void *data, 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, @@ -293,14 +288,12 @@ int amdgpu_gem_userptr_ioctl(struct drm_device *dev, void *data, goto handle_lockup; args->handle = handle; - up_read(&adev->exclusive_lock); return 0; release_object: drm_gem_object_unreference_unlocked(gobj); handle_lockup: - up_read(&adev->exclusive_lock); r = amdgpu_gem_handle_lockup(adev, r); return r; -- cgit v1.2.3 From ce16b0e5a32a157abd6446214e8b91c55064204e Mon Sep 17 00:00:00 2001 From: Junwei Zhang Date: Sun, 6 Sep 2015 13:55:03 +0800 Subject: drm/amdgpu: remove the unnecessary parameter adev for amdgpu_fence_wait_any() MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Signed-off-by: Junwei Zhang Reviewed-by: Christian König --- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 3 +-- drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c | 9 ++------- drivers/gpu/drm/amd/amdgpu/amdgpu_sa.c | 2 +- 3 files changed, 4 insertions(+), 10 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu.h') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h index 39d96088f7b4..a9c0def6ce31 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h @@ -447,8 +447,7 @@ 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, +signed long amdgpu_fence_wait_any(struct fence **array, uint32_t count, bool intr, signed long t); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c index 1fadc15e64ae..003a219943f1 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c @@ -857,16 +857,12 @@ static void amdgpu_fence_wait_cb(struct fence *fence, struct fence_cb *cb) static signed long amdgpu_fence_default_wait(struct fence *f, bool intr, signed long t) { - 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); + return amdgpu_fence_wait_any(&f, 1, intr, t); } /** * Wait the fence array with timeout * - * @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 @@ -874,8 +870,7 @@ static signed long amdgpu_fence_default_wait(struct fence *f, bool intr, * * It will return when any fence is signaled or timeout. */ -signed long amdgpu_fence_wait_any(struct amdgpu_device *adev, - struct fence **array, uint32_t count, +signed long amdgpu_fence_wait_any(struct fence **array, uint32_t count, bool intr, signed long t) { struct amdgpu_wait_cb *cb; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_sa.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_sa.c index e90712443fe9..5cb27d525e43 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_sa.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_sa.c @@ -372,7 +372,7 @@ int amdgpu_sa_bo_new(struct amdgpu_device *adev, } 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, + t = amdgpu_fence_wait_any(fences, AMDGPU_MAX_RINGS, false, MAX_SCHEDULE_TIMEOUT); r = (t > 0) ? 0 : t; spin_lock(&sa_manager->wq.lock); -- cgit v1.2.3 From ee327caf1a2cdba9313167c36db2d7ff02d534bc Mon Sep 17 00:00:00 2001 From: Christian König Date: Tue, 20 Oct 2015 17:38:07 +0200 Subject: drm/amdgpu: switch to common fence_wait_any_timeout v2 MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit No need to duplicate the functionality any more. v2: fix handling if no fence is available. Signed-off-by: Christian König Reviewed-by: Alex Deucher (v1) --- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 4 -- drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c | 98 ------------------------------- drivers/gpu/drm/amd/amdgpu/amdgpu_sa.c | 20 ++++--- 3 files changed, 13 insertions(+), 109 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu.h') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h index a9c0def6ce31..dd7d2ce2355c 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h @@ -447,10 +447,6 @@ 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 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); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c index 663caa9d1175..c4bb28292b9f 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c @@ -822,104 +822,6 @@ static const char *amdgpu_fence_get_timeline_name(struct fence *f) return (const char *)fence->ring->name; } -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; -} - -struct amdgpu_wait_cb { - struct fence_cb base; - struct task_struct *task; -}; - -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); -} - -/** - * Wait the fence array with timeout - * - * @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 - * - * It will return when any fence is signaled or timeout. - */ -signed long amdgpu_fence_wait_any(struct fence **array, uint32_t count, - bool intr, signed long t) -{ - 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; - } - } - } - - while (t > 0) { - if (intr) - set_current_state(TASK_INTERRUPTIBLE); - else - set_current_state(TASK_UNINTERRUPTIBLE); - - /* - * 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; - - t = schedule_timeout(t); - - if (t > 0 && intr && signal_pending(current)) - t = -ERESTARTSYS; - } - - __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); - } - -err_free_cb: - kfree(cb); - - return t; -} - const struct fence_ops amdgpu_fence_ops = { .get_driver_name = amdgpu_fence_get_driver_name, .get_timeline_name = amdgpu_fence_get_timeline_name, diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_sa.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_sa.c index 5cb27d525e43..3f48759793de 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_sa.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_sa.c @@ -337,6 +337,7 @@ int amdgpu_sa_bo_new(struct amdgpu_device *adev, { struct fence *fences[AMDGPU_MAX_RINGS]; unsigned tries[AMDGPU_MAX_RINGS]; + unsigned count; int i, r; signed long t; @@ -371,13 +372,18 @@ 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(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_MAX_RINGS; ++i) + if (fences[i]) + fences[count++] = fences[i]; + + if (count) { + spin_unlock(&sa_manager->wq.lock); + t = fence_wait_any_timeout(fences, count, false, + MAX_SCHEDULE_TIMEOUT); + 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) -- cgit v1.2.3 From 8120b61fdf3d3aa31d6b7e4a4f633bec1d2edd38 Mon Sep 17 00:00:00 2001 From: Christian König Date: Thu, 22 Oct 2015 11:29:33 +0200 Subject: drm/amdgpu: move ring_from_fence to common code MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Going to need that elsewhere as well. Signed-off-by: Christian König Reviewed-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 1 + drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c | 24 ++++++++++++++++++++++++ drivers/gpu/drm/amd/amdgpu/amdgpu_sa.c | 23 ++--------------------- 3 files changed, 27 insertions(+), 21 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu.h') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h index dd7d2ce2355c..6d9c929e6d51 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h @@ -1231,6 +1231,7 @@ int amdgpu_ring_init(struct amdgpu_device *adev, struct amdgpu_ring *ring, struct amdgpu_irq_src *irq_src, unsigned irq_type, enum amdgpu_ring_type ring_type); void amdgpu_ring_fini(struct amdgpu_ring *ring); +struct amdgpu_ring *amdgpu_ring_from_fence(struct fence *f); /* * CS. diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c index b2df348aa223..78e9b0f14661 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c @@ -436,6 +436,30 @@ void amdgpu_ring_fini(struct amdgpu_ring *ring) } } +/** + * amdgpu_ring_from_fence - get ring from fence + * + * @f: fence structure + * + * Extract the ring a fence belongs to. Handles both scheduler as + * well as hardware fences. + */ +struct amdgpu_ring *amdgpu_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) + return container_of(s_fence->sched, struct amdgpu_ring, sched); + + a_fence = to_amdgpu_fence(f); + if (a_fence) + return a_fence->ring; + + return NULL; +} + /* * Debugfs info */ diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_sa.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_sa.c index 3f48759793de..0212b31dc194 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_sa.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_sa.c @@ -139,25 +139,6 @@ 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; @@ -318,7 +299,7 @@ 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 = amdgpu_ring_from_fence(best_bo->fence)->idx; ++tries[idx]; sa_manager->hole = best_bo->olist.prev; @@ -412,7 +393,7 @@ void amdgpu_sa_bo_free(struct amdgpu_device *adev, struct amdgpu_sa_bo **sa_bo, if (fence && !fence_is_signaled(fence)) { uint32_t idx; (*sa_bo)->fence = fence_get(fence); - idx = amdgpu_sa_get_ring_from_fence(fence); + idx = amdgpu_ring_from_fence(fence)->idx; list_add_tail(&(*sa_bo)->flist, &sa_manager->flist[idx]); } else { amdgpu_sa_bo_remove_locked(*sa_bo); -- cgit v1.2.3 From d52832986aa52ea18940e817d5d04c081c2f42f3 Mon Sep 17 00:00:00 2001 From: Christian König Date: Thu, 22 Oct 2015 11:55:58 +0200 Subject: drm/amdgpu: use common fences for VMID management v2 MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit v2: add missing NULL check. Signed-off-by: Christian König Reviewed-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 4 ++-- drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 27 +++++++++++++++------------ drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c | 4 ++-- drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c | 4 ++-- 4 files changed, 21 insertions(+), 18 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu.h') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h index 6d9c929e6d51..371994c4645f 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h @@ -970,7 +970,7 @@ struct amdgpu_vm_id { /* last flushed PD/PT update */ struct fence *flushed_updates; /* last use of vmid */ - struct amdgpu_fence *last_id_use; + struct fence *last_id_use; }; struct amdgpu_vm { @@ -1003,7 +1003,7 @@ struct amdgpu_vm { }; struct amdgpu_vm_manager { - struct amdgpu_fence *active[AMDGPU_NUM_VM]; + struct fence *active[AMDGPU_NUM_VM]; uint32_t max_pfn; /* number of VMIDs */ unsigned nvm; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c index 06e207fd007b..a12c726f7781 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c @@ -135,7 +135,7 @@ struct amdgpu_bo_list_entry *amdgpu_vm_get_bos(struct amdgpu_device *adev, int amdgpu_vm_grab_id(struct amdgpu_vm *vm, struct amdgpu_ring *ring, struct amdgpu_sync *sync) { - struct amdgpu_fence *best[AMDGPU_MAX_RINGS] = {}; + struct fence *best[AMDGPU_MAX_RINGS] = {}; struct amdgpu_vm_id *vm_id = &vm->ids[ring->idx]; struct amdgpu_device *adev = ring->adev; @@ -154,7 +154,8 @@ int amdgpu_vm_grab_id(struct amdgpu_vm *vm, struct amdgpu_ring *ring, /* 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]; + struct fence *fence = adev->vm_manager.active[i]; + struct amdgpu_ring *fring; if (fence == NULL) { /* found a free one */ @@ -163,21 +164,23 @@ int amdgpu_vm_grab_id(struct amdgpu_vm *vm, struct amdgpu_ring *ring, return 0; } - if (amdgpu_fence_is_earlier(fence, best[fence->ring->idx])) { - best[fence->ring->idx] = fence; - choices[fence->ring == ring ? 0 : 1] = i; + fring = amdgpu_ring_from_fence(fence); + if (best[fring->idx] == NULL || + fence_is_later(best[fring->idx], fence)) { + best[fring->idx] = fence; + choices[fring == ring ? 0 : 1] = i; } } for (i = 0; i < 2; ++i) { if (choices[i]) { - struct amdgpu_fence *fence; + struct fence *fence; fence = adev->vm_manager.active[choices[i]]; vm_id->id = choices[i]; trace_amdgpu_vm_grab_id(choices[i], ring->idx); - return amdgpu_sync_fence(ring->adev, sync, &fence->base); + return amdgpu_sync_fence(ring->adev, sync, fence); } } @@ -246,11 +249,11 @@ void amdgpu_vm_fence(struct amdgpu_device *adev, 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); + fence_put(adev->vm_manager.active[vm_id]); + adev->vm_manager.active[vm_id] = fence_get(&fence->base); - amdgpu_fence_unref(&vm->ids[ridx].last_id_use); - vm->ids[ridx].last_id_use = amdgpu_fence_ref(fence); + fence_put(vm->ids[ridx].last_id_use); + vm->ids[ridx].last_id_use = fence_get(&fence->base); } /** @@ -1311,7 +1314,7 @@ void amdgpu_vm_fini(struct amdgpu_device *adev, struct amdgpu_vm *vm) for (i = 0; i < AMDGPU_MAX_RINGS; ++i) { fence_put(vm->ids[i].flushed_updates); - amdgpu_fence_unref(&vm->ids[i].last_id_use); + fence_put(vm->ids[i].last_id_use); } mutex_destroy(&vm->mutex); diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c b/drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c index 488348272c4d..85bbcdc73fff 100644 --- a/drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c @@ -965,7 +965,7 @@ static int gmc_v7_0_sw_fini(void *handle) if (adev->vm_manager.enabled) { for (i = 0; i < AMDGPU_NUM_VM; ++i) - amdgpu_fence_unref(&adev->vm_manager.active[i]); + fence_put(adev->vm_manager.active[i]); gmc_v7_0_vm_fini(adev); adev->vm_manager.enabled = false; } @@ -1015,7 +1015,7 @@ static int gmc_v7_0_suspend(void *handle) if (adev->vm_manager.enabled) { for (i = 0; i < AMDGPU_NUM_VM; ++i) - amdgpu_fence_unref(&adev->vm_manager.active[i]); + fence_put(adev->vm_manager.active[i]); gmc_v7_0_vm_fini(adev); adev->vm_manager.enabled = false; } diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c b/drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c index 72e977b1685d..1bcc4e74e3b4 100644 --- a/drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c @@ -984,7 +984,7 @@ static int gmc_v8_0_sw_fini(void *handle) if (adev->vm_manager.enabled) { for (i = 0; i < AMDGPU_NUM_VM; ++i) - amdgpu_fence_unref(&adev->vm_manager.active[i]); + fence_put(adev->vm_manager.active[i]); gmc_v8_0_vm_fini(adev); adev->vm_manager.enabled = false; } @@ -1036,7 +1036,7 @@ static int gmc_v8_0_suspend(void *handle) if (adev->vm_manager.enabled) { for (i = 0; i < AMDGPU_NUM_VM; ++i) - amdgpu_fence_unref(&adev->vm_manager.active[i]); + fence_put(adev->vm_manager.active[i]); gmc_v8_0_vm_fini(adev); adev->vm_manager.enabled = false; } -- cgit v1.2.3 From 16545c32a5242634a51c6906533750cf1db54879 Mon Sep 17 00:00:00 2001 From: Christian König Date: Thu, 22 Oct 2015 15:04:50 +0200 Subject: drm/amdgpu: use common fence for sync MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Stop using the driver internal functions. Signed-off-by: Christian König Reviewed-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 2 +- drivers/gpu/drm/amd/amdgpu/amdgpu_sync.c | 21 +++++++++++---------- 2 files changed, 12 insertions(+), 11 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu.h') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h index 371994c4645f..b986ea123994 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h @@ -704,7 +704,7 @@ void amdgpu_semaphore_free(struct amdgpu_device *adev, */ struct amdgpu_sync { struct amdgpu_semaphore *semaphores[AMDGPU_NUM_SYNCS]; - struct amdgpu_fence *sync_to[AMDGPU_MAX_RINGS]; + struct fence *sync_to[AMDGPU_MAX_RINGS]; DECLARE_HASHTABLE(fences, 4); struct fence *last_vm_update; }; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_sync.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_sync.c index 0499d2b06eb0..c6489b198a99 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_sync.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_sync.c @@ -108,7 +108,6 @@ int amdgpu_sync_fence(struct amdgpu_device *adev, struct amdgpu_sync *sync, { struct amdgpu_sync_entry *e; struct amdgpu_fence *fence; - struct amdgpu_fence *other; if (!f) return 0; @@ -136,10 +135,7 @@ int amdgpu_sync_fence(struct amdgpu_device *adev, struct amdgpu_sync *sync, 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); + amdgpu_sync_keep_later(&sync->sync_to[fence->ring->idx], f); return 0; } @@ -258,11 +254,11 @@ int amdgpu_sync_wait(struct amdgpu_sync *sync) return 0; for (i = 0; i < AMDGPU_MAX_RINGS; ++i) { - struct amdgpu_fence *fence = sync->sync_to[i]; + struct fence *fence = sync->sync_to[i]; if (!fence) continue; - r = fence_wait(&fence->base, false); + r = fence_wait(fence, false); if (r) return r; } @@ -287,9 +283,14 @@ int amdgpu_sync_rings(struct amdgpu_sync *sync, 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]; + struct amdgpu_semaphore *semaphore; + struct amdgpu_fence *fence; + + if (!sync->sync_to[i]) + continue; + + fence = to_amdgpu_fence(sync->sync_to[i]); /* check if we really need to sync */ if (!amdgpu_fence_need_sync(fence, ring)) @@ -374,7 +375,7 @@ void amdgpu_sync_free(struct amdgpu_device *adev, amdgpu_semaphore_free(adev, &sync->semaphores[i], fence); for (i = 0; i < AMDGPU_MAX_RINGS; ++i) - amdgpu_fence_unref(&sync->sync_to[i]); + fence_put(sync->sync_to[i]); fence_put(sync->last_vm_update); } -- cgit v1.2.3 From 6ef68c17d40e1e7e291ca513627a0d2a13ae095f Mon Sep 17 00:00:00 2001 From: Christian König Date: Thu, 22 Oct 2015 15:16:22 +0200 Subject: drm/amdgpu: remove amdgpu_fence_ref/unref MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Just move the remaining users to fence_put/get. Signed-off-by: Christian König Reviewed-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 3 --- drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c | 30 ------------------------------ drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c | 3 ++- drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c | 3 ++- 4 files changed, 4 insertions(+), 35 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu.h') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h index b986ea123994..8305a6ccf796 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h @@ -447,9 +447,6 @@ 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); -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, diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c index c4bb28292b9f..b912539420e4 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c @@ -416,36 +416,6 @@ int amdgpu_fence_wait_empty(struct amdgpu_ring *ring) 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; - - *fence = NULL; - if (tmp) - fence_put(&tmp->base); -} - /** * amdgpu_fence_count_emitted - get the count of emitted fences * diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c index aad4c1c69448..e65987743871 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c @@ -95,7 +95,8 @@ void amdgpu_ib_free(struct amdgpu_device *adev, struct amdgpu_ib *ib) { 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); + if (ib->fence) + fence_put(&ib->fence->base); } /** diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c index 2e946b2cad88..dcf4a8aca680 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c @@ -54,7 +54,8 @@ static struct fence *amdgpu_sched_run_job(struct amd_sched_job *sched_job) goto err; } - fence = amdgpu_fence_ref(job->ibs[job->num_ibs - 1].fence); + fence = job->ibs[job->num_ibs - 1].fence; + fence_get(&fence->base); err: if (job->free_job) -- cgit v1.2.3 From 1d1106b0f6b5cb4bc1b88d7bd4c41d0413331c5d Mon Sep 17 00:00:00 2001 From: Harry Wentland Date: Wed, 15 Jul 2015 07:10:41 -0400 Subject: drm/amdgpu: Make amdgpu_mn functions inline MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Unused amdgpu_mn functions threw warnings for every file that includes amdgpu.h. It makes sense to inline this amdgpu_mn stubs to avoid the warning. Signed-off-by: Harry Wentland Reviewed-by: Christian König Cc: stable@vger.kernel.org --- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu.h') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h index 8305a6ccf796..510cadeac7b1 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h @@ -1752,11 +1752,11 @@ 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 /* -- cgit v1.2.3 From a67bcd34f6257d5519971a84705e9101ed1160e8 Mon Sep 17 00:00:00 2001 From: Christian König Date: Mon, 26 Oct 2015 13:22:33 +0100 Subject: drm/amdgpu: remove now unused fence functions MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Signed-off-by: Christian König Reviewed-by: Chunming Zhou Reviewed-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 39 ------------------------------------- 1 file changed, 39 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu.h') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h index 510cadeac7b1..e615fffdb3f2 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h @@ -452,45 +452,6 @@ bool amdgpu_fence_need_sync(struct amdgpu_fence *fence, 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. */ -- cgit v1.2.3 From 7a91d6cb3c8f93288865f98ffa03485aff6dbab4 Mon Sep 17 00:00:00 2001 From: Christian König Date: Tue, 27 Oct 2015 17:28:24 +0100 Subject: drm/amdgpu: remove AMDGPU_FENCE_OWNER_MOVE MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Moves are exclusive operations anyway, just use the undefined owner for those. Signed-off-by: Christian König Reviewed-by: Chunming Zhou Reviewed-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 1 - drivers/gpu/drm/amd/amdgpu/amdgpu_sync.c | 4 ++-- drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 2 +- 3 files changed, 3 insertions(+), 4 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu.h') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h index e615fffdb3f2..5961d63f8913 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h @@ -405,7 +405,6 @@ struct amdgpu_fence_driver { /* 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) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_sync.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_sync.c index c6489b198a99..a6697fd05217 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_sync.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_sync.c @@ -191,8 +191,8 @@ 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; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c index a089e69e9927..81bb8e9fc26d 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c @@ -1041,7 +1041,7 @@ int amdgpu_copy_buffer(struct amdgpu_ring *ring, 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, + AMDGPU_FENCE_OWNER_UNDEFINED, fence); if (r) goto error_free; -- cgit v1.2.3 From c2776afe740db5598c4c457dcacb94d4427b13f9 Mon Sep 17 00:00:00 2001 From: Christian König Date: Tue, 3 Nov 2015 13:27:39 +0100 Subject: drm/amdgpu: use a timer for fence fallback MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Less overhead than a work item and also adds proper cleanup handling. Signed-off-by: Christian König Reviewed-by: Chunming Zhou Acked-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 3 +- drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c | 78 ++++++++++++++----------------- 2 files changed, 35 insertions(+), 46 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu.h') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h index 615ce6d464fb..3630cb545d0d 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h @@ -389,7 +389,6 @@ 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 */ @@ -398,7 +397,7 @@ struct amdgpu_fence_driver { bool initialized; struct amdgpu_irq_src *irq_src; unsigned irq_type; - struct delayed_work lockup_work; + struct timer_list fallback_timer; wait_queue_head_t fence_queue; }; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c index b92c1937543a..257fce356319 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c @@ -84,24 +84,6 @@ static u32 amdgpu_fence_read(struct amdgpu_ring *ring) 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 * @@ -135,6 +117,19 @@ int amdgpu_fence_emit(struct amdgpu_ring *ring, void *owner, return 0; } +/** + * amdgpu_fence_schedule_fallback - schedule fallback check + * + * @ring: pointer to struct amdgpu_ring + * + * Start a timer as fallback to our interrupts. + */ +static void amdgpu_fence_schedule_fallback(struct amdgpu_ring *ring) +{ + mod_timer(&ring->fence_drv.fallback_timer, + jiffies + AMDGPU_FENCE_JIFFIES_TIMEOUT); +} + /** * amdgpu_fence_activity - check for fence activity * @@ -201,45 +196,38 @@ static bool amdgpu_fence_activity(struct amdgpu_ring *ring) } while (atomic64_xchg(&ring->fence_drv.last_seq, seq) > seq); if (seq < last_emitted) - amdgpu_fence_schedule_check(ring); + amdgpu_fence_schedule_fallback(ring); return wake; } /** - * amdgpu_fence_check_lockup - check for hardware lockup + * amdgpu_fence_process - process a fence * - * @work: delayed work item + * @adev: amdgpu_device pointer + * @ring: ring index the fence is associated with * - * Checks for fence activity and if there is none probe - * the hardware if a lockup occured. + * Checks the current fence value and wakes the fence queue + * if the sequence number has increased (all asics). */ -static void amdgpu_fence_check_lockup(struct work_struct *work) +void amdgpu_fence_process(struct amdgpu_ring *ring) { - 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 (amdgpu_fence_activity(ring)) wake_up_all(&ring->fence_drv.fence_queue); } /** - * amdgpu_fence_process - process a fence + * amdgpu_fence_fallback - fallback for hardware interrupts * - * @adev: amdgpu_device pointer - * @ring: ring index the fence is associated with + * @work: delayed work item * - * Checks the current fence value and wakes the fence queue - * if the sequence number has increased (all asics). + * Checks for fence activity. */ -void amdgpu_fence_process(struct amdgpu_ring *ring) +static void amdgpu_fence_fallback(unsigned long arg) { - if (amdgpu_fence_activity(ring)) - wake_up_all(&ring->fence_drv.fence_queue); + struct amdgpu_ring *ring = (void *)arg; + + amdgpu_fence_process(ring); } /** @@ -289,7 +277,7 @@ static int amdgpu_fence_ring_wait_seq(struct amdgpu_ring *ring, uint64_t seq) if (atomic64_read(&ring->fence_drv.last_seq) >= seq) return 0; - amdgpu_fence_schedule_check(ring); + amdgpu_fence_schedule_fallback(ring); wait_event(ring->fence_drv.fence_queue, ( (signaled = amdgpu_fence_seq_signaled(ring, seq)))); @@ -490,9 +478,8 @@ int amdgpu_fence_driver_init_ring(struct amdgpu_ring *ring) atomic64_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); @@ -556,6 +543,7 @@ void amdgpu_fence_driver_fini(struct amdgpu_device *adev) 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); @@ -567,6 +555,7 @@ void amdgpu_fence_driver_fini(struct amdgpu_device *adev) 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); ring->fence_drv.initialized = false; } mutex_unlock(&adev->ring_lock); @@ -750,7 +739,8 @@ static bool amdgpu_fence_enable_signaling(struct fence *f) fence->fence_wake.func = amdgpu_fence_check_signaled; __add_wait_queue(&ring->fence_drv.fence_queue, &fence->fence_wake); fence_get(f); - amdgpu_fence_schedule_check(ring); + if (!timer_pending(&ring->fence_drv.fallback_timer)) + amdgpu_fence_schedule_fallback(ring); FENCE_TRACE(&fence->base, "armed on ring %i!\n", ring->idx); return true; } -- cgit v1.2.3 From 16ae42feb0ce6cc80473d35dd82a1cd07e2e7178 Mon Sep 17 00:00:00 2001 From: Christian König Date: Tue, 3 Nov 2015 14:53:28 +0100 Subject: drm/amdgpu: use common fence for amdgpu_vm_fence MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Just cleanup the function parameters. Signed-off-by: Christian König Reviewed-by: Chunming Zhou Reviewed-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 2 +- drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c | 2 +- drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 12 ++++++------ 3 files changed, 8 insertions(+), 8 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu.h') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h index 3630cb545d0d..b801b6702812 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h @@ -2332,7 +2332,7 @@ void amdgpu_vm_flush(struct amdgpu_ring *ring, struct fence *updates); void amdgpu_vm_fence(struct amdgpu_device *adev, struct amdgpu_vm *vm, - struct amdgpu_fence *fence); + struct 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); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c index e65987743871..3902e7a9f8cc 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c @@ -216,7 +216,7 @@ int amdgpu_ib_schedule(struct amdgpu_device *adev, unsigned num_ibs, } if (ib->vm) - amdgpu_vm_fence(adev, ib->vm, ib->fence); + amdgpu_vm_fence(adev, ib->vm, &ib->fence->base); amdgpu_ring_unlock_commit(ring); return 0; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c index 8c80381b1ea9..a9fcc995d27e 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c @@ -241,16 +241,16 @@ void amdgpu_vm_flush(struct amdgpu_ring *ring, */ void amdgpu_vm_fence(struct amdgpu_device *adev, struct amdgpu_vm *vm, - struct amdgpu_fence *fence) + struct fence *fence) { - unsigned ridx = fence->ring->idx; - unsigned vm_id = vm->ids[ridx].id; + struct amdgpu_ring *ring = amdgpu_ring_from_fence(fence); + unsigned vm_id = vm->ids[ring->idx].id; fence_put(adev->vm_manager.active[vm_id]); - adev->vm_manager.active[vm_id] = fence_get(&fence->base); + adev->vm_manager.active[vm_id] = fence_get(fence); - fence_put(vm->ids[ridx].last_id_use); - vm->ids[ridx].last_id_use = fence_get(&fence->base); + fence_put(vm->ids[ring->idx].last_id_use); + vm->ids[ring->idx].last_id_use = fence_get(fence); } /** -- cgit v1.2.3 From 7e52a81c2f0326a85d3ebc005829bcd604731c6d Mon Sep 17 00:00:00 2001 From: Christian König Date: Wed, 4 Nov 2015 15:44:39 +0100 Subject: drm/amdgpu: cleanup amdgpu_cs_parser handling MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit No need any more to allocate that structure dynamically, just put it on the stack. This is a start to cleanup some of the scheduler fallouts. Signed-off-by: Christian König Reviewed-by: Junwei Zhang Reviewed-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 5 -- drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 88 +++++++++++++--------------------- 2 files changed, 33 insertions(+), 60 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu.h') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h index b801b6702812..7b02e3455172 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h @@ -2256,11 +2256,6 @@ 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, diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c index ecc82dfe83f8..bf32096b8eb7 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c @@ -127,30 +127,6 @@ 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) -{ - struct amdgpu_cs_parser *parser; - int i; - - parser = kzalloc(sizeof(struct amdgpu_cs_parser), GFP_KERNEL); - if (!parser) - return NULL; - - 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; - - return parser; -} - int amdgpu_cs_parser_init(struct amdgpu_cs_parser *p, void *data) { union drm_amdgpu_cs *cs = data; @@ -490,6 +466,7 @@ static void amdgpu_cs_parser_fini_early(struct amdgpu_cs_parser *parser, int err 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) @@ -505,7 +482,6 @@ static void amdgpu_cs_parser_fini_late(struct amdgpu_cs_parser *parser) kfree(parser->ibs); if (parser->uf.bo) drm_gem_object_unreference_unlocked(&parser->uf.bo->gem_base); - kfree(parser); } /** @@ -824,36 +800,36 @@ int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp) union drm_amdgpu_cs *cs = data; struct amdgpu_fpriv *fpriv = filp->driver_priv; struct amdgpu_vm *vm = &fpriv->vm; - struct amdgpu_cs_parser *parser; + struct amdgpu_cs_parser parser = {}; bool reserved_buffers = false; int i, r; 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"); - amdgpu_cs_parser_fini(parser, r, false); + amdgpu_cs_parser_fini(&parser, r, false); r = amdgpu_cs_handle_lockup(adev, r); return r; } mutex_lock(&vm->mutex); - r = amdgpu_cs_parser_relocs(parser); + r = amdgpu_cs_parser_relocs(&parser); 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); } @@ -861,36 +837,38 @@ 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.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) { + if (amdgpu_enable_scheduler && parser.num_ibs) { struct amdgpu_job *job; - struct amdgpu_ring * ring = parser->ibs->ring; + struct amdgpu_ring * ring = parser.ibs->ring; + job = kzalloc(sizeof(struct amdgpu_job), GFP_KERNEL); if (!job) { r = -ENOMEM; goto out; } + 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; + 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) { - job->uf = parser->uf; + job->uf = parser.uf; job->ibs[job->num_ibs - 1].user = &job->uf; - parser->uf.bo = NULL; + parser.uf.bo = NULL; } - parser->ibs = NULL; - parser->num_ibs = 0; + parser.ibs = NULL; + parser.num_ibs = 0; job->free_job = amdgpu_cs_free_job; mutex_lock(&job->job_lock); @@ -902,24 +880,24 @@ int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp) goto out; } cs->out.handle = - amdgpu_ctx_add_fence(parser->ctx, ring, + amdgpu_ctx_add_fence(parser.ctx, ring, &job->base.s_fence->base); job->ibs[job->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, + 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); + amdgpu_cs_parser_fini_late(&parser); mutex_unlock(&vm->mutex); return 0; } - cs->out.handle = parser->ibs[parser->num_ibs - 1].sequence; + cs->out.handle = parser.ibs[parser.num_ibs - 1].sequence; out: - amdgpu_cs_parser_fini(parser, r, reserved_buffers); + amdgpu_cs_parser_fini(&parser, r, reserved_buffers); mutex_unlock(&vm->mutex); r = amdgpu_cs_handle_lockup(adev, r); return r; -- cgit v1.2.3 From e284022163716ecf11c37fd1057c35d689ef2c11 Mon Sep 17 00:00:00 2001 From: Christian König Date: Thu, 5 Nov 2015 19:49:48 +0100 Subject: drm/amdgpu: fix incorrect mutex usage v3 MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Before this patch the scheduler fence was created when we push the job into the queue, so we could only get the fence after pushing it. The mutex now was necessary to prevent the thread pushing the jobs to the hardware from running faster than the thread pushing the jobs into the queue. Otherwise the thread pushing jobs into the queue would have accessed possible freed up memory when it tries to get a reference to the fence. So what you get in the end is thread A: mutex_lock(&job->lock); ... Kick of thread B. ... mutex_unlock(&job->lock); And thread B: mutex_lock(&job->lock); .... mutex_unlock(&job->lock); kfree(job); I'm actually not sure if I'm still up to date on this, but this usage pattern used to be not allowed with mutexes. See here as well https://lwn.net/Articles/575460/. v2: remove unrelated changes, fix missing owner v3: rebased, add more commit message Signed-off-by: Christian König Reviewed-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 2 +- drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 43 +++++++++++++++------------ drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c | 27 +++++++---------- drivers/gpu/drm/amd/scheduler/gpu_scheduler.c | 10 +------ drivers/gpu/drm/amd/scheduler/gpu_scheduler.h | 3 +- 5 files changed, 37 insertions(+), 48 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu.h') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h index 7b02e3455172..0f187027c753 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h @@ -1225,7 +1225,7 @@ struct amdgpu_job { struct amdgpu_device *adev; struct amdgpu_ib *ibs; uint32_t num_ibs; - struct mutex job_lock; + void *owner; struct amdgpu_user_fence uf; int (*free_job)(struct amdgpu_job *job); }; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c index 2ae73d5232dd..44cf977ae4f6 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c @@ -845,8 +845,9 @@ int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp) goto out; if (amdgpu_enable_scheduler && parser.num_ibs) { - struct amdgpu_job *job; struct amdgpu_ring * ring = parser.ibs->ring; + struct amd_sched_fence *fence; + struct amdgpu_job *job; job = kzalloc(sizeof(struct amdgpu_job), GFP_KERNEL); if (!job) { @@ -859,37 +860,41 @@ int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp) job->adev = parser.adev; job->ibs = parser.ibs; job->num_ibs = parser.num_ibs; - job->base.owner = parser.filp; - mutex_init(&job->job_lock); + job->owner = parser.filp; + job->free_job = amdgpu_cs_free_job; + if (job->ibs[job->num_ibs - 1].user) { job->uf = parser.uf; job->ibs[job->num_ibs - 1].user = &job->uf; parser.uf.bo = NULL; } - parser.ibs = NULL; - parser.num_ibs = 0; - - 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); + fence = amd_sched_fence_create(job->base.s_entity, + parser.filp); + if (!fence) { + r = -ENOMEM; amdgpu_cs_free_job(job); kfree(job); goto out; } - cs->out.handle = - amdgpu_ctx_add_fence(parser.ctx, ring, - &job->base.s_fence->base); + job->base.s_fence = fence; + fence_get(&fence->base); + + cs->out.handle = amdgpu_ctx_add_fence(parser.ctx, ring, + &fence->base); job->ibs[job->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); + parser.ibs = NULL; + parser.num_ibs = 0; + trace_amdgpu_cs_ioctl(job); - mutex_unlock(&job->job_lock); + amd_sched_entity_push_job(&job->base); + + list_sort(NULL, &parser.validated, cmp_size_smaller_first); + ttm_eu_fence_buffer_objects(&parser.ticket, &parser.validated, + &fence->base); + fence_put(&fence->base); + amdgpu_cs_parser_fini_late(&parser); mutex_unlock(&vm->mutex); return 0; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c index 8ef9e4415fcc..438c05254695 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c @@ -45,12 +45,8 @@ static struct fence *amdgpu_sched_run_job(struct amd_sched_job *sched_job) return NULL; } job = to_amdgpu_job(sched_job); - mutex_lock(&job->job_lock); trace_amdgpu_sched_run_job(job); - r = amdgpu_ib_schedule(job->adev, - job->num_ibs, - job->ibs, - job->base.owner); + r = amdgpu_ib_schedule(job->adev, job->num_ibs, job->ibs, job->owner); if (r) { DRM_ERROR("Error scheduling IBs (%d)\n", r); goto err; @@ -63,7 +59,6 @@ err: if (job->free_job) job->free_job(job); - mutex_unlock(&job->job_lock); kfree(job); return fence ? &fence->base : NULL; } @@ -89,21 +84,19 @@ int amdgpu_sched_ib_submit_kernel_helper(struct amdgpu_device *adev, return -ENOMEM; job->base.sched = &ring->sched; job->base.s_entity = &adev->kernel_ctx.rings[ring->idx].entity; + job->base.s_fence = amd_sched_fence_create(job->base.s_entity, owner); + if (!job->base.s_fence) { + kfree(job); + return -ENOMEM; + } + *f = fence_get(&job->base.s_fence->base); + job->adev = adev; job->ibs = ibs; job->num_ibs = num_ibs; - job->base.owner = owner; - mutex_init(&job->job_lock); + job->owner = owner; 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); + amd_sched_entity_push_job(&job->base); } else { r = amdgpu_ib_schedule(adev, num_ibs, ibs, owner); if (r) diff --git a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c index ccb7c1554f5e..ea30d6ad4c13 100644 --- a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c +++ b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c @@ -276,21 +276,13 @@ static bool amd_sched_entity_in(struct amd_sched_job *sched_job) * * Returns 0 for success, negative error code otherwise. */ -int amd_sched_entity_push_job(struct amd_sched_job *sched_job) +void amd_sched_entity_push_job(struct amd_sched_job *sched_job) { struct amd_sched_entity *entity = sched_job->s_entity; - struct amd_sched_fence *fence = amd_sched_fence_create( - entity, sched_job->owner); - - if (!fence) - return -ENOMEM; - - sched_job->s_fence = fence; wait_event(entity->sched->job_scheduled, amd_sched_entity_in(sched_job)); trace_amd_sched_job(sched_job); - return 0; } /** diff --git a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h index 4d05ca6fb057..939692b14f4b 100644 --- a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h +++ b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h @@ -79,7 +79,6 @@ struct amd_sched_job { struct amd_gpu_scheduler *sched; struct amd_sched_entity *s_entity; struct amd_sched_fence *s_fence; - void *owner; }; extern const struct fence_ops amd_sched_fence_ops; @@ -131,7 +130,7 @@ int amd_sched_entity_init(struct amd_gpu_scheduler *sched, uint32_t jobs); void amd_sched_entity_fini(struct amd_gpu_scheduler *sched, struct amd_sched_entity *entity); -int amd_sched_entity_push_job(struct amd_sched_job *sched_job); +void amd_sched_entity_push_job(struct amd_sched_job *sched_job); struct amd_sched_fence *amd_sched_fence_create( struct amd_sched_entity *s_entity, void *owner); -- cgit v1.2.3 From 984810fc45389c545719fbb4219e8a12b27032a4 Mon Sep 17 00:00:00 2001 From: Christian König Date: Sat, 14 Nov 2015 21:05:35 +0100 Subject: drm/amdgpu: cleanup scheduler command submission MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Unify the two code path again, cause they do pretty much the same thing. Signed-off-by: Christian König Reviewed-by: Chunming Zhou Reviewed-by: Junwei Zhang --- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 1 + drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 63 +++++++++++++--------------------- 2 files changed, 25 insertions(+), 39 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu.h') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h index 0f187027c753..0ce5c1e45ec8 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h @@ -1210,6 +1210,7 @@ struct amdgpu_cs_parser { /* relocations */ struct amdgpu_bo_list_entry *vm_bos; struct list_head validated; + struct fence *fence; struct amdgpu_ib *ibs; uint32_t num_ibs; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c index 6096effd6a56..3afcf0237c25 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c @@ -439,8 +439,18 @@ 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) { + unsigned i; + if (!error) { /* Sort the buffer list from the smallest to largest buffer, * which affects the order of buffers in the LRU list. @@ -455,17 +465,13 @@ 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); } -} - -static void amdgpu_cs_parser_fini_late(struct amdgpu_cs_parser *parser) -{ - unsigned i; + fence_put(parser->fence); if (parser->ctx) amdgpu_ctx_put(parser->ctx); @@ -484,20 +490,6 @@ static void amdgpu_cs_parser_fini_late(struct amdgpu_cs_parser *parser) drm_gem_object_unreference_unlocked(&parser->uf.bo->gem_base); } -/** - * 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); -} - static int amdgpu_bo_vm_update_pte(struct amdgpu_cs_parser *p, struct amdgpu_vm *vm) { @@ -582,15 +574,9 @@ static int amdgpu_cs_ib_vm_chunk(struct amdgpu_device *adev, } 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); + if (!r) + amdgpu_cs_sync_rings(parser); -out: return r; } @@ -881,7 +867,7 @@ int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp) goto out; } job->base.s_fence = fence; - fence_get(&fence->base); + parser.fence = fence_get(&fence->base); cs->out.handle = amdgpu_ctx_add_fence(parser.ctx, ring, &fence->base); @@ -890,17 +876,16 @@ int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp) trace_amdgpu_cs_ioctl(job); amd_sched_entity_push_job(&job->base); - list_sort(NULL, &parser.validated, cmp_size_smaller_first); - ttm_eu_fence_buffer_objects(&parser.ticket, &parser.validated, - &fence->base); - fence_put(&fence->base); + } else { + struct amdgpu_fence *fence; - amdgpu_cs_parser_fini_late(&parser); - mutex_unlock(&vm->mutex); - return 0; + r = amdgpu_ib_schedule(adev, parser.num_ibs, parser.ibs, + parser.filp); + fence = parser.ibs[parser.num_ibs - 1].fence; + parser.fence = fence_get(&fence->base); + cs->out.handle = parser.ibs[parser.num_ibs - 1].sequence; } - cs->out.handle = parser.ibs[parser.num_ibs - 1].sequence; out: amdgpu_cs_parser_fini(&parser, r, reserved_buffers); mutex_unlock(&vm->mutex); -- cgit v1.2.3 From eeed25ab83bbd68b2c9dba3cc2ce5913537594f6 Mon Sep 17 00:00:00 2001 From: Christian König Date: Sun, 15 Nov 2015 15:50:07 +0100 Subject: drm/amdgpu: remove unused VM manager field MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Signed-off-by: Christian König Reviewed-by: Chunming Zhou Reviewed-by: Junwei Zhang --- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 2 -- 1 file changed, 2 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu.h') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h index 0ce5c1e45ec8..819d1473de30 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h @@ -967,8 +967,6 @@ struct amdgpu_vm_manager { 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]; /* vm pte handling */ const struct amdgpu_vm_pte_funcs *vm_pte_funcs; struct amdgpu_ring *vm_pte_funcs_ring; -- cgit v1.2.3 From 8b4fb00b5db271bb1e117a078d5d7a161d8d8af1 Mon Sep 17 00:00:00 2001 From: Christian König Date: Sun, 15 Nov 2015 16:04:16 +0100 Subject: drm/amdgpu: cleanup VM coding style MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Fix the indentation and move the VM functions to the structures. Signed-off-by: Christian König Reviewed-by: Chunming Zhou Reviewed-by: Junwei Zhang Reviewed-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 98 ++++++++++++++++++------------------- 1 file changed, 48 insertions(+), 50 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu.h') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h index 819d1473de30..d1e16726e225 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h @@ -916,8 +916,8 @@ struct amdgpu_ring { #define AMDGPU_VM_FAULT_STOP_ALWAYS 2 struct amdgpu_vm_pt { - struct amdgpu_bo *bo; - uint64_t addr; + struct amdgpu_bo *bo; + uint64_t addr; }; struct amdgpu_vm_id { @@ -959,19 +959,60 @@ struct amdgpu_vm { }; struct amdgpu_vm_manager { - struct fence *active[AMDGPU_NUM_VM]; - uint32_t max_pfn; + struct fence *active[AMDGPU_NUM_VM]; + uint32_t max_pfn; /* number of VMIDs */ - unsigned nvm; + unsigned nvm; /* vram base address for page table entry */ - u64 vram_base_offset; + u64 vram_base_offset; /* is vm enabled? */ - bool enabled; + bool enabled; /* vm pte handling */ const struct amdgpu_vm_pte_funcs *vm_pte_funcs; struct amdgpu_ring *vm_pte_funcs_ring; }; +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 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); + /* * context related structures */ @@ -2311,49 +2352,6 @@ int amdgpu_get_vblank_timestamp_kms(struct drm_device *dev, unsigned int pipe, 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 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 */ -- cgit v1.2.3 From ea89f8c9e8ba8a7b75446eef36917da50d2186d9 Mon Sep 17 00:00:00 2001 From: Christian König Date: Sun, 15 Nov 2015 20:52:06 +0100 Subject: drm/amdgpu: move VM manager clean into the VM code again MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit It's not a good idea to duplicate that code. Signed-off-by: Christian König Reviewed-by: Chunming Zhou Reviewed-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 1 + drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 15 +++++++++++++++ drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c | 8 ++------ drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c | 8 ++------ 4 files changed, 20 insertions(+), 12 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu.h') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h index d1e16726e225..beb74854a8a3 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h @@ -972,6 +972,7 @@ struct amdgpu_vm_manager { struct amdgpu_ring *vm_pte_funcs_ring; }; +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); struct amdgpu_bo_list_entry *amdgpu_vm_get_bos(struct amdgpu_device *adev, diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c index a9fcc995d27e..6bb209bc0d36 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c @@ -1318,3 +1318,18 @@ void amdgpu_vm_fini(struct amdgpu_device *adev, struct amdgpu_vm *vm) mutex_destroy(&vm->mutex); } + +/** + * 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.active[i]); +} diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c b/drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c index 98ee39001c45..7427d8cd4c43 100644 --- a/drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c @@ -961,12 +961,10 @@ 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) - fence_put(adev->vm_manager.active[i]); + amdgpu_vm_manager_fini(adev); gmc_v7_0_vm_fini(adev); adev->vm_manager.enabled = false; } @@ -1011,12 +1009,10 @@ 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) - fence_put(adev->vm_manager.active[i]); + amdgpu_vm_manager_fini(adev); gmc_v7_0_vm_fini(adev); adev->vm_manager.enabled = false; } diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c b/drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c index c9209b45d901..cb0e50ebb528 100644 --- a/drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c @@ -980,12 +980,10 @@ 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) - fence_put(adev->vm_manager.active[i]); + amdgpu_vm_manager_fini(adev); gmc_v8_0_vm_fini(adev); adev->vm_manager.enabled = false; } @@ -1032,12 +1030,10 @@ 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) - fence_put(adev->vm_manager.active[i]); + amdgpu_vm_manager_fini(adev); gmc_v8_0_vm_fini(adev); adev->vm_manager.enabled = false; } -- cgit v1.2.3 From 1c16c0a7b26c6c905dc79c4194135ca2f360f0f5 Mon Sep 17 00:00:00 2001 From: Christian König Date: Sat, 14 Nov 2015 21:31:40 +0100 Subject: drm/amdgpu: keep the owner for VMIDs MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit We don't need the last VM use any more, keep the owner directly. Signed-off-by: Christian König Reviewed-by: Chunming Zhou --- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 8 +++++--- drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 33 +++++++++++++++++++-------------- 2 files changed, 24 insertions(+), 17 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu.h') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h index beb74854a8a3..a5692624070a 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h @@ -925,8 +925,6 @@ struct amdgpu_vm_id { uint64_t pd_gpu_addr; /* last flushed PD/PT update */ struct fence *flushed_updates; - /* last use of vmid */ - struct fence *last_id_use; }; struct amdgpu_vm { @@ -959,7 +957,11 @@ struct amdgpu_vm { }; struct amdgpu_vm_manager { - struct fence *active[AMDGPU_NUM_VM]; + struct { + struct fence *active; + atomic_long_t owner; + } ids[AMDGPU_NUM_VM]; + uint32_t max_pfn; /* number of VMIDs */ unsigned nvm; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c index 6bb209bc0d36..0bdbb2480f9b 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c @@ -143,10 +143,15 @@ int amdgpu_vm_grab_id(struct amdgpu_vm *vm, struct amdgpu_ring *ring, unsigned i; /* 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]) { - trace_amdgpu_vm_grab_id(vm_id->id, ring->idx); - return 0; + if (vm_id->id) { + unsigned id = vm_id->id; + long owner; + + owner = atomic_long_read(&adev->vm_manager.ids[id].owner); + if (owner == (long)vm) { + trace_amdgpu_vm_grab_id(vm_id->id, ring->idx); + return 0; + } } /* we definately need to flush */ @@ -154,7 +159,7 @@ int amdgpu_vm_grab_id(struct amdgpu_vm *vm, struct amdgpu_ring *ring, /* skip over VMID 0, since it is the system VM */ for (i = 1; i < adev->vm_manager.nvm; ++i) { - struct fence *fence = adev->vm_manager.active[i]; + struct fence *fence = adev->vm_manager.ids[i].active; struct amdgpu_ring *fring; if (fence == NULL) { @@ -176,7 +181,7 @@ int amdgpu_vm_grab_id(struct amdgpu_vm *vm, struct amdgpu_ring *ring, if (choices[i]) { struct fence *fence; - fence = adev->vm_manager.active[choices[i]]; + fence = adev->vm_manager.ids[choices[i]].active; vm_id->id = choices[i]; trace_amdgpu_vm_grab_id(choices[i], ring->idx); @@ -246,11 +251,9 @@ void amdgpu_vm_fence(struct amdgpu_device *adev, struct amdgpu_ring *ring = amdgpu_ring_from_fence(fence); unsigned vm_id = vm->ids[ring->idx].id; - fence_put(adev->vm_manager.active[vm_id]); - adev->vm_manager.active[vm_id] = fence_get(fence); - - fence_put(vm->ids[ring->idx].last_id_use); - vm->ids[ring->idx].last_id_use = fence_get(fence); + fence_put(adev->vm_manager.ids[vm_id].active); + adev->vm_manager.ids[vm_id].active = fence_get(fence); + atomic_long_set(&adev->vm_manager.ids[vm_id].owner, (long)vm); } /** @@ -1238,7 +1241,6 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm) for (i = 0; i < AMDGPU_MAX_RINGS; ++i) { vm->ids[i].id = 0; vm->ids[i].flushed_updates = NULL; - vm->ids[i].last_id_use = NULL; } mutex_init(&vm->mutex); vm->va = RB_ROOT; @@ -1312,8 +1314,11 @@ void amdgpu_vm_fini(struct amdgpu_device *adev, struct amdgpu_vm *vm) fence_put(vm->page_directory_fence); for (i = 0; i < AMDGPU_MAX_RINGS; ++i) { + unsigned id = vm->ids[i].id; + + atomic_long_cmpxchg(&adev->vm_manager.ids[id].owner, + (long)vm, 0); fence_put(vm->ids[i].flushed_updates); - fence_put(vm->ids[i].last_id_use); } mutex_destroy(&vm->mutex); @@ -1331,5 +1336,5 @@ void amdgpu_vm_manager_fini(struct amdgpu_device *adev) unsigned i; for (i = 0; i < AMDGPU_NUM_VM; ++i) - fence_put(adev->vm_manager.active[i]); + fence_put(adev->vm_manager.ids[i].active); } -- cgit v1.2.3 From c25867dfabf045a8148fd179fa759bb17f670e42 Mon Sep 17 00:00:00 2001 From: Chunming Zhou Date: Fri, 13 Nov 2015 13:32:01 +0800 Subject: drm/amdgpu: add lock for interval tree in vm MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Change-Id: I62b892a22af37b32e6b4aefca80a25cf45426ed2 Signed-off-by: Chunming Zhou Reviewed-by: Christian König --- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 2 ++ drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 15 +++++++++++++-- 2 files changed, 15 insertions(+), 2 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu.h') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h index a5692624070a..306f75700bf8 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h @@ -954,6 +954,8 @@ struct amdgpu_vm { /* for id and flush management per ring */ struct amdgpu_vm_id ids[AMDGPU_MAX_RINGS]; + /* for interval tree */ + spinlock_t it_lock; }; struct amdgpu_vm_manager { diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c index 0bdbb2480f9b..0513f3fed2c2 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c @@ -1028,7 +1028,9 @@ int amdgpu_vm_bo_map(struct amdgpu_device *adev, saddr /= AMDGPU_GPU_PAGE_SIZE; eaddr /= AMDGPU_GPU_PAGE_SIZE; + spin_lock(&vm->it_lock); it = interval_tree_iter_first(&vm->va, saddr, eaddr - 1); + spin_unlock(&vm->it_lock); if (it) { struct amdgpu_bo_va_mapping *tmp; tmp = container_of(it, struct amdgpu_bo_va_mapping, it); @@ -1055,7 +1057,9 @@ int amdgpu_vm_bo_map(struct amdgpu_device *adev, mapping->flags = flags; list_add(&mapping->list, &bo_va->invalids); + spin_lock(&vm->it_lock); interval_tree_insert(&mapping->it, &vm->va); + spin_unlock(&vm->it_lock); trace_amdgpu_vm_bo_map(bo_va, mapping); /* Make sure the page tables are allocated */ @@ -1101,7 +1105,9 @@ int amdgpu_vm_bo_map(struct amdgpu_device *adev, error_free: list_del(&mapping->list); + spin_lock(&vm->it_lock); interval_tree_remove(&mapping->it, &vm->va); + spin_unlock(&vm->it_lock); trace_amdgpu_vm_bo_unmap(bo_va, mapping); kfree(mapping); @@ -1151,7 +1157,9 @@ int amdgpu_vm_bo_unmap(struct amdgpu_device *adev, } list_del(&mapping->list); + spin_lock(&vm->it_lock); interval_tree_remove(&mapping->it, &vm->va); + spin_unlock(&vm->it_lock); trace_amdgpu_vm_bo_unmap(bo_va, mapping); if (valid) @@ -1187,13 +1195,17 @@ void amdgpu_vm_bo_rmv(struct amdgpu_device *adev, list_for_each_entry_safe(mapping, next, &bo_va->valids, list) { list_del(&mapping->list); + spin_lock(&vm->it_lock); interval_tree_remove(&mapping->it, &vm->va); + spin_unlock(&vm->it_lock); trace_amdgpu_vm_bo_unmap(bo_va, mapping); list_add(&mapping->list, &vm->freed); } list_for_each_entry_safe(mapping, next, &bo_va->invalids, list) { list_del(&mapping->list); + spin_lock(&vm->it_lock); interval_tree_remove(&mapping->it, &vm->va); + spin_unlock(&vm->it_lock); kfree(mapping); } @@ -1248,7 +1260,7 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm) INIT_LIST_HEAD(&vm->invalidated); INIT_LIST_HEAD(&vm->cleared); INIT_LIST_HEAD(&vm->freed); - + spin_lock_init(&vm->it_lock); pd_size = amdgpu_vm_directory_size(adev); pd_entries = amdgpu_vm_num_pdes(adev); @@ -1312,7 +1324,6 @@ void amdgpu_vm_fini(struct amdgpu_device *adev, struct amdgpu_vm *vm) amdgpu_bo_unref(&vm->page_directory); fence_put(vm->page_directory_fence); - for (i = 0; i < AMDGPU_MAX_RINGS; ++i) { unsigned id = vm->ids[i].id; -- cgit v1.2.3 From 69b576a1bc8b466ae7bff0208f1c139dbaaf802c Mon Sep 17 00:00:00 2001 From: Chunming Zhou Date: Wed, 18 Nov 2015 11:17:39 +0800 Subject: drm/amdgpu: add mutex for ba_va->valids/invalids MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Signed-off-by: Chunming Zhou Reviewed-by: Christian König --- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 1 + drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 17 +++++++++++------ 2 files changed, 12 insertions(+), 6 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu.h') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h index 306f75700bf8..50672bb19798 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h @@ -496,6 +496,7 @@ struct amdgpu_bo_va_mapping { /* bo virtual addresses in a specific vm */ struct amdgpu_bo_va { + struct mutex mutex; /* protected by bo being reserved */ struct list_head bo_list; struct fence *last_pt_update; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c index 159ce54bbd8d..d6904ef742f0 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c @@ -922,8 +922,9 @@ int amdgpu_vm_clear_invalids(struct amdgpu_device *adev, bo_va = list_first_entry(&vm->invalidated, struct amdgpu_bo_va, vm_status); spin_unlock(&vm->status_lock); - + mutex_lock(&bo_va->mutex); r = amdgpu_vm_bo_update(adev, bo_va, NULL); + mutex_unlock(&bo_va->mutex); if (r) return r; @@ -967,7 +968,7 @@ struct amdgpu_bo_va *amdgpu_vm_bo_add(struct amdgpu_device *adev, INIT_LIST_HEAD(&bo_va->valids); INIT_LIST_HEAD(&bo_va->invalids); INIT_LIST_HEAD(&bo_va->vm_status); - + mutex_init(&bo_va->mutex); list_add_tail(&bo_va->bo_list, &bo->va); return bo_va; @@ -1045,7 +1046,9 @@ int amdgpu_vm_bo_map(struct amdgpu_device *adev, mapping->offset = offset; mapping->flags = flags; + mutex_lock(&bo_va->mutex); list_add(&mapping->list, &bo_va->invalids); + mutex_unlock(&bo_va->mutex); spin_lock(&vm->it_lock); interval_tree_insert(&mapping->it, &vm->va); spin_unlock(&vm->it_lock); @@ -1121,7 +1124,7 @@ int amdgpu_vm_bo_unmap(struct amdgpu_device *adev, bool valid = true; saddr /= AMDGPU_GPU_PAGE_SIZE; - + mutex_lock(&bo_va->mutex); list_for_each_entry(mapping, &bo_va->valids, list) { if (mapping->it.start == saddr) break; @@ -1135,10 +1138,12 @@ int amdgpu_vm_bo_unmap(struct amdgpu_device *adev, break; } - if (&mapping->list == &bo_va->invalids) + if (&mapping->list == &bo_va->invalids) { + mutex_unlock(&bo_va->mutex); return -ENOENT; + } } - + mutex_unlock(&bo_va->mutex); list_del(&mapping->list); spin_lock(&vm->it_lock); interval_tree_remove(&mapping->it, &vm->va); @@ -1190,8 +1195,8 @@ void amdgpu_vm_bo_rmv(struct amdgpu_device *adev, spin_unlock(&vm->it_lock); kfree(mapping); } - fence_put(bo_va->last_pt_update); + mutex_destroy(&bo_va->mutex); kfree(bo_va); } -- cgit v1.2.3 From e98c1b0de6fe73f488df62d83d83f377b1b6e2b8 Mon Sep 17 00:00:00 2001 From: Chunming Zhou Date: Fri, 13 Nov 2015 15:22:04 +0800 Subject: drm/amdgpu: remove vm->mutex MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Signed-off-by: Chunming Zhou Reviewed-by: Christian König --- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 2 -- drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 4 ---- drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c | 14 ++------------ drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 2 -- 4 files changed, 2 insertions(+), 20 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu.h') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h index 50672bb19798..251b14736de9 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h @@ -929,8 +929,6 @@ struct amdgpu_vm_id { }; struct amdgpu_vm { - struct mutex mutex; - struct rb_root va; /* protecting invalidated */ diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c index 3afcf0237c25..1d44d508d4d4 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c @@ -784,8 +784,6 @@ 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_fpriv *fpriv = filp->driver_priv; - struct amdgpu_vm *vm = &fpriv->vm; struct amdgpu_cs_parser parser = {}; bool reserved_buffers = false; int i, r; @@ -803,7 +801,6 @@ int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp) r = amdgpu_cs_handle_lockup(adev, r); return r; } - mutex_lock(&vm->mutex); r = amdgpu_cs_parser_relocs(&parser); if (r == -ENOMEM) DRM_ERROR("Not enough memory for command submission!\n"); @@ -888,7 +885,6 @@ int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp) out: amdgpu_cs_parser_fini(&parser, r, reserved_buffers); - mutex_unlock(&vm->mutex); r = amdgpu_cs_handle_lockup(adev, r); return r; } diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c index 00c5b580f56c..fc32fc01a64b 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c @@ -115,12 +115,9 @@ 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; - mutex_lock(&vm->mutex); r = amdgpu_bo_reserve(rbo, false); - if (r) { - mutex_unlock(&vm->mutex); + if (r) return r; - } bo_va = amdgpu_vm_bo_find(vm, rbo); if (!bo_va) { @@ -129,7 +126,6 @@ int amdgpu_gem_object_open(struct drm_gem_object *obj, struct drm_file *file_pri ++bo_va->ref_count; } amdgpu_bo_unreserve(rbo); - mutex_unlock(&vm->mutex); return 0; } @@ -142,10 +138,8 @@ void amdgpu_gem_object_close(struct drm_gem_object *obj, struct amdgpu_vm *vm = &fpriv->vm; struct amdgpu_bo_va *bo_va; int r; - mutex_lock(&vm->mutex); r = amdgpu_bo_reserve(rbo, true); if (r) { - mutex_unlock(&vm->mutex); dev_err(adev->dev, "leaking bo va because " "we fail to reserve bo (%d)\n", r); return; @@ -157,7 +151,6 @@ void amdgpu_gem_object_close(struct drm_gem_object *obj, } } amdgpu_bo_unreserve(rbo); - mutex_unlock(&vm->mutex); } static int amdgpu_gem_handle_lockup(struct amdgpu_device *adev, int r) @@ -553,7 +546,6 @@ 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; - mutex_lock(&fpriv->vm.mutex); rbo = gem_to_amdgpu_bo(gobj); INIT_LIST_HEAD(&list); INIT_LIST_HEAD(&duplicates); @@ -568,7 +560,6 @@ int amdgpu_gem_va_ioctl(struct drm_device *dev, void *data, } r = ttm_eu_reserve_buffers(&ticket, &list, true, &duplicates); if (r) { - mutex_unlock(&fpriv->vm.mutex); drm_gem_object_unreference_unlocked(gobj); return r; } @@ -577,7 +568,6 @@ int amdgpu_gem_va_ioctl(struct drm_device *dev, void *data, if (!bo_va) { ttm_eu_backoff_reservation(&ticket, &list); drm_gem_object_unreference_unlocked(gobj); - mutex_unlock(&fpriv->vm.mutex); return -ENOENT; } @@ -602,7 +592,7 @@ int amdgpu_gem_va_ioctl(struct drm_device *dev, void *data, ttm_eu_backoff_reservation(&ticket, &list); if (!r && !(args->flags & AMDGPU_VM_DELAY_UPDATE)) amdgpu_gem_va_update_vm(adev, bo_va, args->operation); - mutex_unlock(&fpriv->vm.mutex); + drm_gem_object_unreference_unlocked(gobj); return r; } diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c index d6904ef742f0..ae037e5b6ad0 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c @@ -1241,7 +1241,6 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm) vm->ids[i].id = 0; vm->ids[i].flushed_updates = NULL; } - mutex_init(&vm->mutex); vm->va = RB_ROOT; spin_lock_init(&vm->status_lock); INIT_LIST_HEAD(&vm->invalidated); @@ -1325,7 +1324,6 @@ void amdgpu_vm_fini(struct amdgpu_device *adev, struct amdgpu_vm *vm) fence_put(vm->ids[i].flushed_updates); } - mutex_destroy(&vm->mutex); } /** -- cgit v1.2.3