From aac1e3caac343c05ab550fa611489bd206aa8a7b Mon Sep 17 00:00:00 2001 From: Alex Deucher Date: Wed, 24 Feb 2016 10:06:06 -0500 Subject: drm/amdgpu/gfx: fix off by one in rb rework (v2) When I reworked this code, I messed up num rb count. v2: use hweight32 Reviewed-by: Ken Wang Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c | 7 ++----- drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c | 7 ++----- 2 files changed, 4 insertions(+), 10 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu') diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c index 250bcbce7fdc..bc5bdaf3d2bb 100644 --- a/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c @@ -1635,7 +1635,7 @@ static u32 gfx_v7_0_get_rb_active_bitmap(struct amdgpu_device *adev) static void gfx_v7_0_setup_rb(struct amdgpu_device *adev) { int i, j; - u32 data, tmp, num_rbs = 0; + u32 data; u32 active_rbs = 0; mutex_lock(&adev->grbm_idx_mutex); @@ -1655,10 +1655,7 @@ static void gfx_v7_0_setup_rb(struct amdgpu_device *adev) mutex_unlock(&adev->grbm_idx_mutex); adev->gfx.config.backend_enable_mask = active_rbs; - tmp = active_rbs; - while (tmp >>= 1) - num_rbs++; - adev->gfx.config.num_rbs = num_rbs; + adev->gfx.config.num_rbs = hweight32(active_rbs); } /** diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c index 10c865087d0a..71d536e595a2 100644 --- a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c @@ -2613,7 +2613,7 @@ static u32 gfx_v8_0_get_rb_active_bitmap(struct amdgpu_device *adev) static void gfx_v8_0_setup_rb(struct amdgpu_device *adev) { int i, j; - u32 data, tmp, num_rbs = 0; + u32 data; u32 active_rbs = 0; mutex_lock(&adev->grbm_idx_mutex); @@ -2629,10 +2629,7 @@ static void gfx_v8_0_setup_rb(struct amdgpu_device *adev) mutex_unlock(&adev->grbm_idx_mutex); adev->gfx.config.backend_enable_mask = active_rbs; - tmp = active_rbs; - while (tmp >>= 1) - num_rbs++; - adev->gfx.config.num_rbs = num_rbs; + adev->gfx.config.num_rbs = hweight32(active_rbs); } /** -- cgit v1.2.3 From 70eacc72dbbc229013fee54550c03179f9a513d8 Mon Sep 17 00:00:00 2001 From: Christian König Date: Mon, 22 Feb 2016 15:11:56 +0100 Subject: drm/amdgpu: fix error handling in amdgpu_bo_list_set MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Don't leak BOs in case of some error. Signed-off-by: Christian König Reviewed-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c | 3 +++ 1 file changed, 3 insertions(+) (limited to 'drivers/gpu/drm/amd/amdgpu') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c index 90d6fc1618aa..4792f9d0b7d4 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.c @@ -118,6 +118,7 @@ static int amdgpu_bo_list_set(struct amdgpu_device *adev, usermm = amdgpu_ttm_tt_get_usermm(entry->robj->tbo.ttm); if (usermm) { if (usermm != current->mm) { + amdgpu_bo_unref(&entry->robj); r = -EPERM; goto error_free; } @@ -151,6 +152,8 @@ static int amdgpu_bo_list_set(struct amdgpu_device *adev, return 0; error_free: + while (i--) + amdgpu_bo_unref(&array[i].robj); drm_free_large(array); return r; } -- cgit v1.2.3 From a65ecc4070c75ed53553def79e5da862b47daf1f Mon Sep 17 00:00:00 2001 From: Dan Carpenter Date: Thu, 25 Feb 2016 10:47:10 +0300 Subject: drm/amd: cleanup get_mfd_cell_dev() It's simpler to just use snprintf() to print this to one buffer instead of using strcpy() and strcat(). Also using snprintf() is slightly safer than using sprintf(). Signed-off-by: Dan Carpenter Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_acp.c | 6 ++---- 1 file changed, 2 insertions(+), 4 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_acp.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_acp.c index 9f8cfaab3004..d6b0bff510aa 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_acp.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_acp.c @@ -240,12 +240,10 @@ static int acp_poweron(struct generic_pm_domain *genpd) static struct device *get_mfd_cell_dev(const char *device_name, int r) { char auto_dev_name[25]; - char buf[8]; struct device *dev; - sprintf(buf, ".%d.auto", r); - strcpy(auto_dev_name, device_name); - strcat(auto_dev_name, buf); + snprintf(auto_dev_name, sizeof(auto_dev_name), + "%s.%d.auto", device_name, r); dev = bus_find_device_by_name(&platform_bus_type, NULL, auto_dev_name); dev_info(dev, "device %s added to pm domain\n", auto_dev_name); -- cgit v1.2.3 From 4ff37a83f19dab4e67299325ee22e98346eee857 Mon Sep 17 00:00:00 2001 From: Christian König Date: Fri, 26 Feb 2016 16:18:26 +0100 Subject: drm/amdgpu: fix VM faults caused by vm_grab_id() v4 MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit The owner must be per ring as long as we don't support sharing VMIDs per process. Also move the assigned VMID and page directory address into the IB structure. v3: assign the VMID to all IBs, not just the first one. v4: use correct pointer for owner Signed-off-by: Christian König Reviewed-by: Chunming Zhou Acked-by: Alex Deucher Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 16 +++-- drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c | 7 +- drivers/gpu/drm/amd/amdgpu/amdgpu_job.c | 15 +++-- drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 115 +++++++++++++++++--------------- drivers/gpu/drm/amd/amdgpu/cik_sdma.c | 2 +- drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c | 6 +- drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c | 6 +- drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c | 2 +- drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c | 2 +- 9 files changed, 91 insertions(+), 80 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h index f5bac97a438b..0c42a85ca5a5 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h @@ -769,8 +769,9 @@ struct amdgpu_ib { uint32_t *ptr; struct amdgpu_fence *fence; struct amdgpu_user_fence *user; - bool grabbed_vmid; struct amdgpu_vm *vm; + unsigned vm_id; + uint64_t vm_pd_addr; struct amdgpu_ctx *ctx; uint32_t gds_base, gds_size; uint32_t gws_base, gws_size; @@ -877,10 +878,10 @@ struct amdgpu_vm_pt { }; struct amdgpu_vm_id { - unsigned id; - uint64_t pd_gpu_addr; + struct amdgpu_vm_manager_id *mgr_id; + uint64_t pd_gpu_addr; /* last flushed PD/PT update */ - struct fence *flushed_updates; + struct fence *flushed_updates; }; struct amdgpu_vm { @@ -954,10 +955,11 @@ void amdgpu_vm_get_pt_bos(struct amdgpu_vm *vm, struct list_head *duplicates); void amdgpu_vm_move_pt_bos_in_lru(struct amdgpu_device *adev, struct amdgpu_vm *vm); int amdgpu_vm_grab_id(struct amdgpu_vm *vm, struct amdgpu_ring *ring, - struct amdgpu_sync *sync, struct fence *fence); + struct amdgpu_sync *sync, struct fence *fence, + unsigned *vm_id, uint64_t *vm_pd_addr); void amdgpu_vm_flush(struct amdgpu_ring *ring, - struct amdgpu_vm *vm, - struct fence *updates); + unsigned vmid, + uint64_t pd_addr); uint64_t amdgpu_vm_map_gart(const dma_addr_t *pages_addr, uint64_t addr); int amdgpu_vm_update_page_directory(struct amdgpu_device *adev, struct amdgpu_vm *vm); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c index b5bdd5d59b58..db14a7bbb8f4 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c @@ -75,6 +75,7 @@ int amdgpu_ib_get(struct amdgpu_device *adev, struct amdgpu_vm *vm, } ib->vm = vm; + ib->vm_id = 0; return 0; } @@ -139,7 +140,7 @@ int amdgpu_ib_schedule(struct amdgpu_ring *ring, unsigned num_ibs, return -EINVAL; } - if (vm && !ibs->grabbed_vmid) { + if (vm && !ibs->vm_id) { dev_err(adev->dev, "VM IB without ID\n"); return -EINVAL; } @@ -152,10 +153,10 @@ int amdgpu_ib_schedule(struct amdgpu_ring *ring, unsigned num_ibs, if (vm) { /* do context switch */ - amdgpu_vm_flush(ring, vm, last_vm_update); + amdgpu_vm_flush(ring, ib->vm_id, ib->vm_pd_addr); if (ring->funcs->emit_gds_switch) - amdgpu_ring_emit_gds_switch(ring, ib->vm->ids[ring->idx].id, + amdgpu_ring_emit_gds_switch(ring, ib->vm_id, ib->gds_base, ib->gds_size, ib->gws_base, ib->gws_size, ib->oa_base, ib->oa_size); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c index f29bbb96a881..90e52f7e17a0 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c @@ -105,16 +105,23 @@ static struct fence *amdgpu_job_dependency(struct amd_sched_job *sched_job) struct fence *fence = amdgpu_sync_get_fence(&job->sync); - if (fence == NULL && vm && !job->ibs->grabbed_vmid) { + if (fence == NULL && vm && !job->ibs->vm_id) { struct amdgpu_ring *ring = job->ring; + unsigned i, vm_id; + uint64_t vm_pd_addr; int r; r = amdgpu_vm_grab_id(vm, ring, &job->sync, - &job->base.s_fence->base); + &job->base.s_fence->base, + &vm_id, &vm_pd_addr); if (r) DRM_ERROR("Error getting VM ID (%d)\n", r); - else - job->ibs->grabbed_vmid = true; + else { + for (i = 0; i < job->num_ibs; ++i) { + job->ibs[i].vm_id = vm_id; + job->ibs[i].vm_pd_addr = vm_pd_addr; + } + } fence = amdgpu_sync_get_fence(&job->sync); } diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c index 264c5968a1d3..ba909245fef5 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c @@ -50,6 +50,9 @@ * SI supports 16. */ +/* Special value that no flush is necessary */ +#define AMDGPU_VM_NO_FLUSH (~0ll) + /** * amdgpu_vm_num_pde - return the number of page directory entries * @@ -157,50 +160,69 @@ void amdgpu_vm_move_pt_bos_in_lru(struct amdgpu_device *adev, * Allocate an id for the vm, adding fences to the sync obj as necessary. */ int amdgpu_vm_grab_id(struct amdgpu_vm *vm, struct amdgpu_ring *ring, - struct amdgpu_sync *sync, struct fence *fence) + struct amdgpu_sync *sync, struct fence *fence, + unsigned *vm_id, uint64_t *vm_pd_addr) { - struct amdgpu_vm_id *vm_id = &vm->ids[ring->idx]; + uint64_t pd_addr = amdgpu_bo_gpu_offset(vm->page_directory); struct amdgpu_device *adev = ring->adev; - struct amdgpu_vm_manager_id *id; + struct amdgpu_vm_id *id = &vm->ids[ring->idx]; + struct fence *updates = sync->last_vm_update; int r; mutex_lock(&adev->vm_manager.lock); /* check if the id is still valid */ - if (vm_id->id) { + if (id->mgr_id) { + struct fence *flushed = id->flushed_updates; + bool is_later; long owner; - id = &adev->vm_manager.ids[vm_id->id]; - owner = atomic_long_read(&id->owner); - if (owner == (long)vm) { - list_move_tail(&id->list, &adev->vm_manager.ids_lru); - trace_amdgpu_vm_grab_id(vm, vm_id->id, ring->idx); + if (!flushed) + is_later = true; + else if (!updates) + is_later = false; + else + is_later = fence_is_later(updates, flushed); + + owner = atomic_long_read(&id->mgr_id->owner); + if (!is_later && owner == (long)id && + pd_addr == id->pd_gpu_addr) { + + fence_put(id->mgr_id->active); + id->mgr_id->active = fence_get(fence); + + list_move_tail(&id->mgr_id->list, + &adev->vm_manager.ids_lru); - fence_put(id->active); - id->active = fence_get(fence); + *vm_id = id->mgr_id - adev->vm_manager.ids; + *vm_pd_addr = AMDGPU_VM_NO_FLUSH; + trace_amdgpu_vm_grab_id(vm, *vm_id, ring->idx); mutex_unlock(&adev->vm_manager.lock); return 0; } } - /* we definately need to flush */ - vm_id->pd_gpu_addr = ~0ll; + id->mgr_id = list_first_entry(&adev->vm_manager.ids_lru, + struct amdgpu_vm_manager_id, + list); - id = list_first_entry(&adev->vm_manager.ids_lru, - struct amdgpu_vm_manager_id, - list); - list_move_tail(&id->list, &adev->vm_manager.ids_lru); - atomic_long_set(&id->owner, (long)vm); + r = amdgpu_sync_fence(ring->adev, sync, id->mgr_id->active); + if (!r) { + fence_put(id->mgr_id->active); + id->mgr_id->active = fence_get(fence); - vm_id->id = id - adev->vm_manager.ids; - trace_amdgpu_vm_grab_id(vm, vm_id->id, ring->idx); + fence_put(id->flushed_updates); + id->flushed_updates = fence_get(updates); - r = amdgpu_sync_fence(ring->adev, sync, id->active); + id->pd_gpu_addr = pd_addr; - if (!r) { - fence_put(id->active); - id->active = fence_get(fence); + list_move_tail(&id->mgr_id->list, &adev->vm_manager.ids_lru); + atomic_long_set(&id->mgr_id->owner, (long)id); + + *vm_id = id->mgr_id - adev->vm_manager.ids; + *vm_pd_addr = pd_addr; + trace_amdgpu_vm_grab_id(vm, *vm_id, ring->idx); } mutex_unlock(&adev->vm_manager.lock); @@ -211,35 +233,18 @@ int amdgpu_vm_grab_id(struct amdgpu_vm *vm, struct amdgpu_ring *ring, * amdgpu_vm_flush - hardware flush the vm * * @ring: ring to use for flush - * @vm: vm we want to flush - * @updates: last vm update that we waited for + * @vmid: vmid number to use + * @pd_addr: address of the page directory * - * Flush the vm. + * Emit a VM flush when it is necessary. */ void amdgpu_vm_flush(struct amdgpu_ring *ring, - struct amdgpu_vm *vm, - struct fence *updates) + unsigned vmid, + uint64_t pd_addr) { - uint64_t pd_addr = amdgpu_bo_gpu_offset(vm->page_directory); - struct amdgpu_vm_id *vm_id = &vm->ids[ring->idx]; - struct fence *flushed_updates = vm_id->flushed_updates; - bool is_later; - - if (!flushed_updates) - is_later = true; - else if (!updates) - is_later = false; - else - is_later = fence_is_later(updates, flushed_updates); - - if (pd_addr != vm_id->pd_gpu_addr || is_later) { - trace_amdgpu_vm_flush(pd_addr, ring->idx, vm_id->id); - if (is_later) { - vm_id->flushed_updates = fence_get(updates); - fence_put(flushed_updates); - } - vm_id->pd_gpu_addr = pd_addr; - amdgpu_ring_emit_vm_flush(ring, vm_id->id, vm_id->pd_gpu_addr); + if (pd_addr != AMDGPU_VM_NO_FLUSH) { + trace_amdgpu_vm_flush(pd_addr, ring->idx, vmid); + amdgpu_ring_emit_vm_flush(ring, vmid, pd_addr); } } @@ -1284,7 +1289,7 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm) int i, r; for (i = 0; i < AMDGPU_MAX_RINGS; ++i) { - vm->ids[i].id = 0; + vm->ids[i].mgr_id = NULL; vm->ids[i].flushed_updates = NULL; } vm->va = RB_ROOT; @@ -1381,13 +1386,13 @@ 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; + struct amdgpu_vm_id *id = &vm->ids[i]; - atomic_long_cmpxchg(&adev->vm_manager.ids[id].owner, - (long)vm, 0); - fence_put(vm->ids[i].flushed_updates); + if (id->mgr_id) + atomic_long_cmpxchg(&id->mgr_id->owner, + (long)id, 0); + fence_put(id->flushed_updates); } - } /** diff --git a/drivers/gpu/drm/amd/amdgpu/cik_sdma.c b/drivers/gpu/drm/amd/amdgpu/cik_sdma.c index 675f34916aab..e4e4b2ac77b7 100644 --- a/drivers/gpu/drm/amd/amdgpu/cik_sdma.c +++ b/drivers/gpu/drm/amd/amdgpu/cik_sdma.c @@ -212,7 +212,7 @@ static void cik_sdma_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count) static void cik_sdma_ring_emit_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib) { - u32 extra_bits = (ib->vm ? ib->vm->ids[ring->idx].id : 0) & 0xf; + u32 extra_bits = ib->vm_id & 0xf; u32 next_rptr = ring->wptr + 5; while ((next_rptr & 7) != 4) diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c index bc5bdaf3d2bb..9cdf59518533 100644 --- a/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c @@ -2043,8 +2043,7 @@ static void gfx_v7_0_ring_emit_ib_gfx(struct amdgpu_ring *ring, else header = PACKET3(PACKET3_INDIRECT_BUFFER, 2); - control |= ib->length_dw | - (ib->vm ? (ib->vm->ids[ring->idx].id << 24) : 0); + control |= ib->length_dw | (ib->vm_id << 24); amdgpu_ring_write(ring, header); amdgpu_ring_write(ring, @@ -2072,8 +2071,7 @@ static void gfx_v7_0_ring_emit_ib_compute(struct amdgpu_ring *ring, header = PACKET3(PACKET3_INDIRECT_BUFFER, 2); - control |= ib->length_dw | - (ib->vm ? (ib->vm->ids[ring->idx].id << 24) : 0); + control |= ib->length_dw | (ib->vm_id << 24); amdgpu_ring_write(ring, header); amdgpu_ring_write(ring, diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c index 71d536e595a2..5f67a189bce9 100644 --- a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c @@ -4619,8 +4619,7 @@ static void gfx_v8_0_ring_emit_ib_gfx(struct amdgpu_ring *ring, else header = PACKET3(PACKET3_INDIRECT_BUFFER, 2); - control |= ib->length_dw | - (ib->vm ? (ib->vm->ids[ring->idx].id << 24) : 0); + control |= ib->length_dw | (ib->vm_id << 24); amdgpu_ring_write(ring, header); amdgpu_ring_write(ring, @@ -4649,8 +4648,7 @@ static void gfx_v8_0_ring_emit_ib_compute(struct amdgpu_ring *ring, header = PACKET3(PACKET3_INDIRECT_BUFFER, 2); - control |= ib->length_dw | - (ib->vm ? (ib->vm->ids[ring->idx].id << 24) : 0); + control |= ib->length_dw | (ib->vm_id << 24); amdgpu_ring_write(ring, header); amdgpu_ring_write(ring, diff --git a/drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c b/drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c index 29ec986dd6fc..dddb8d6a81f3 100644 --- a/drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c +++ b/drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c @@ -244,7 +244,7 @@ static void sdma_v2_4_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count) static void sdma_v2_4_ring_emit_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib) { - u32 vmid = (ib->vm ? ib->vm->ids[ring->idx].id : 0) & 0xf; + u32 vmid = ib->vm_id & 0xf; u32 next_rptr = ring->wptr + 5; while ((next_rptr & 7) != 2) diff --git a/drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c b/drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c index 6f064d7076e6..19e02f7a06f3 100644 --- a/drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c +++ b/drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c @@ -355,7 +355,7 @@ static void sdma_v3_0_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count) static void sdma_v3_0_ring_emit_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib) { - u32 vmid = (ib->vm ? ib->vm->ids[ring->idx].id : 0) & 0xf; + u32 vmid = ib->vm_id & 0xf; u32 next_rptr = ring->wptr + 5; while ((next_rptr & 7) != 2) -- cgit v1.2.3 From 22073fe764c9ff2742c27a8c06f28ef6cd9a56e3 Mon Sep 17 00:00:00 2001 From: Christian König Date: Fri, 26 Feb 2016 16:18:36 +0100 Subject: drm/amdgpu: trace the pd_addr in vm_grab_id as well MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Makes matching it to the flushes much easier. Signed-off-by: Christian König Reviewed-by: Chunming Zhou Acked-by: Alex Deucher Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_trace.h | 19 +++++++++++-------- drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 5 +++-- 2 files changed, 14 insertions(+), 10 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_trace.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_trace.h index 9ca3735c563c..26a5f4acf584 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_trace.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_trace.h @@ -100,21 +100,24 @@ TRACE_EVENT(amdgpu_sched_run_job, TRACE_EVENT(amdgpu_vm_grab_id, - TP_PROTO(struct amdgpu_vm *vm, unsigned vmid, int ring), - TP_ARGS(vm, vmid, ring), + TP_PROTO(struct amdgpu_vm *vm, int ring, unsigned vmid, + uint64_t pd_addr), + TP_ARGS(vm, ring, vmid, pd_addr), TP_STRUCT__entry( __field(struct amdgpu_vm *, vm) - __field(u32, vmid) __field(u32, ring) + __field(u32, vmid) + __field(u64, pd_addr) ), TP_fast_assign( __entry->vm = vm; - __entry->vmid = vmid; __entry->ring = ring; + __entry->vmid = vmid; + __entry->pd_addr = pd_addr; ), - TP_printk("vm=%p, id=%u, ring=%u", __entry->vm, __entry->vmid, - __entry->ring) + TP_printk("vm=%p, ring=%u, id=%u, pd_addr=%010Lx", __entry->vm, + __entry->ring, __entry->vmid, __entry->pd_addr) ); TRACE_EVENT(amdgpu_vm_bo_map, @@ -231,8 +234,8 @@ TRACE_EVENT(amdgpu_vm_flush, __entry->ring = ring; __entry->id = id; ), - TP_printk("pd_addr=%010Lx, ring=%u, id=%u", - __entry->pd_addr, __entry->ring, __entry->id) + TP_printk("ring=%u, id=%u, pd_addr=%010Lx", + __entry->ring, __entry->id, __entry->pd_addr) ); TRACE_EVENT(amdgpu_bo_list_set, diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c index ba909245fef5..d9dc8bea5e98 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c @@ -196,7 +196,8 @@ int amdgpu_vm_grab_id(struct amdgpu_vm *vm, struct amdgpu_ring *ring, *vm_id = id->mgr_id - adev->vm_manager.ids; *vm_pd_addr = AMDGPU_VM_NO_FLUSH; - trace_amdgpu_vm_grab_id(vm, *vm_id, ring->idx); + trace_amdgpu_vm_grab_id(vm, ring->idx, *vm_id, + *vm_pd_addr); mutex_unlock(&adev->vm_manager.lock); return 0; @@ -222,7 +223,7 @@ int amdgpu_vm_grab_id(struct amdgpu_vm *vm, struct amdgpu_ring *ring, *vm_id = id->mgr_id - adev->vm_manager.ids; *vm_pd_addr = pd_addr; - trace_amdgpu_vm_grab_id(vm, *vm_id, ring->idx); + trace_amdgpu_vm_grab_id(vm, ring->idx, *vm_id, *vm_pd_addr); } mutex_unlock(&adev->vm_manager.lock); -- cgit v1.2.3 From 6157bd7a1009c2a6944fb3eee8ed2b3dea091fd8 Mon Sep 17 00:00:00 2001 From: Flora Cui Date: Thu, 3 Mar 2016 12:59:49 +0800 Subject: drm/amdgpu: fix rb bitmap & cu bitmap calculation Fix some copy paste typos. Signed-off-by: Flora Cui Reviewed-by: Alex Deucher Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/cikd.h | 3 --- drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c | 15 +++++++-------- drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c | 9 ++++++--- drivers/gpu/drm/amd/amdgpu/vid.h | 2 -- 4 files changed, 13 insertions(+), 16 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu') diff --git a/drivers/gpu/drm/amd/amdgpu/cikd.h b/drivers/gpu/drm/amd/amdgpu/cikd.h index 7f6d457f250a..60d4493206dd 100644 --- a/drivers/gpu/drm/amd/amdgpu/cikd.h +++ b/drivers/gpu/drm/amd/amdgpu/cikd.h @@ -46,9 +46,6 @@ #define BONAIRE_GB_ADDR_CONFIG_GOLDEN 0x12010001 #define HAWAII_GB_ADDR_CONFIG_GOLDEN 0x12011003 -#define CIK_RB_BITMAP_WIDTH_PER_SH 2 -#define HAWAII_RB_BITMAP_WIDTH_PER_SH 4 - #define AMDGPU_NUM_OF_VMIDS 8 #define PIPEID(x) ((x) << 0) diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c index 9cdf59518533..8fb7ebf3be3e 100644 --- a/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c @@ -1637,18 +1637,16 @@ static void gfx_v7_0_setup_rb(struct amdgpu_device *adev) int i, j; u32 data; u32 active_rbs = 0; + u32 rb_bitmap_width_per_sh = adev->gfx.config.max_backends_per_se / + adev->gfx.config.max_sh_per_se; mutex_lock(&adev->grbm_idx_mutex); for (i = 0; i < adev->gfx.config.max_shader_engines; i++) { for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) { gfx_v7_0_select_se_sh(adev, i, j); data = gfx_v7_0_get_rb_active_bitmap(adev); - if (adev->asic_type == CHIP_HAWAII) - active_rbs |= data << ((i * adev->gfx.config.max_sh_per_se + j) * - HAWAII_RB_BITMAP_WIDTH_PER_SH); - else - active_rbs |= data << ((i * adev->gfx.config.max_sh_per_se + j) * - CIK_RB_BITMAP_WIDTH_PER_SH); + active_rbs |= data << ((i * adev->gfx.config.max_sh_per_se + j) * + rb_bitmap_width_per_sh); } } gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff); @@ -3820,8 +3818,7 @@ static u32 gfx_v7_0_get_cu_active_bitmap(struct amdgpu_device *adev) data &= CC_GC_SHADER_ARRAY_CONFIG__INACTIVE_CUS_MASK; data >>= CC_GC_SHADER_ARRAY_CONFIG__INACTIVE_CUS__SHIFT; - mask = gfx_v7_0_create_bitmask(adev->gfx.config.max_backends_per_se / - adev->gfx.config.max_sh_per_se); + mask = gfx_v7_0_create_bitmask(adev->gfx.config.max_cu_per_sh); return (~data) & mask; } @@ -5232,6 +5229,8 @@ int gfx_v7_0_get_cu_info(struct amdgpu_device *adev, if (!adev || !cu_info) return -EINVAL; + memset(cu_info, 0, sizeof(*cu_info)); + mutex_lock(&adev->grbm_idx_mutex); for (i = 0; i < adev->gfx.config.max_shader_engines; i++) { for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) { diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c index 5f67a189bce9..e37378fe1edc 100644 --- a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c @@ -2615,6 +2615,8 @@ static void gfx_v8_0_setup_rb(struct amdgpu_device *adev) int i, j; u32 data; u32 active_rbs = 0; + u32 rb_bitmap_width_per_sh = adev->gfx.config.max_backends_per_se / + adev->gfx.config.max_sh_per_se; mutex_lock(&adev->grbm_idx_mutex); for (i = 0; i < adev->gfx.config.max_shader_engines; i++) { @@ -2622,7 +2624,7 @@ static void gfx_v8_0_setup_rb(struct amdgpu_device *adev) gfx_v8_0_select_se_sh(adev, i, j); data = gfx_v8_0_get_rb_active_bitmap(adev); active_rbs |= data << ((i * adev->gfx.config.max_sh_per_se + j) * - RB_BITMAP_WIDTH_PER_SH); + rb_bitmap_width_per_sh); } } gfx_v8_0_select_se_sh(adev, 0xffffffff, 0xffffffff); @@ -5126,8 +5128,7 @@ static u32 gfx_v8_0_get_cu_active_bitmap(struct amdgpu_device *adev) data &= CC_GC_SHADER_ARRAY_CONFIG__INACTIVE_CUS_MASK; data >>= CC_GC_SHADER_ARRAY_CONFIG__INACTIVE_CUS__SHIFT; - mask = gfx_v8_0_create_bitmask(adev->gfx.config.max_backends_per_se / - adev->gfx.config.max_sh_per_se); + mask = gfx_v8_0_create_bitmask(adev->gfx.config.max_cu_per_sh); return (~data) & mask; } @@ -5141,6 +5142,8 @@ int gfx_v8_0_get_cu_info(struct amdgpu_device *adev, if (!adev || !cu_info) return -EINVAL; + memset(cu_info, 0, sizeof(*cu_info)); + mutex_lock(&adev->grbm_idx_mutex); for (i = 0; i < adev->gfx.config.max_shader_engines; i++) { for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) { diff --git a/drivers/gpu/drm/amd/amdgpu/vid.h b/drivers/gpu/drm/amd/amdgpu/vid.h index d98aa9d82fa1..ace49976f7be 100644 --- a/drivers/gpu/drm/amd/amdgpu/vid.h +++ b/drivers/gpu/drm/amd/amdgpu/vid.h @@ -71,8 +71,6 @@ #define VMID(x) ((x) << 4) #define QUEUEID(x) ((x) << 8) -#define RB_BITMAP_WIDTH_PER_SH 2 - #define MC_SEQ_MISC0__MT__MASK 0xf0000000 #define MC_SEQ_MISC0__MT__GDDR1 0x10000000 #define MC_SEQ_MISC0__MT__DDR2 0x20000000 -- cgit v1.2.3