diff options
Diffstat (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c')
| -rw-r--r-- | drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 46 |
1 files changed, 30 insertions, 16 deletions
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c index 06d07502a1f6..d0d0ea565e3d 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c @@ -32,6 +32,7 @@ #include <drm/amdgpu_drm.h> #include <drm/drm_syncobj.h> +#include "amdgpu_cs.h" #include "amdgpu.h" #include "amdgpu_trace.h" #include "amdgpu_gmc.h" @@ -314,7 +315,7 @@ static void amdgpu_cs_get_threshold_for_moves(struct amdgpu_device *adev, } total_vram = adev->gmc.real_vram_size - atomic64_read(&adev->vram_pin_size); - used_vram = amdgpu_vram_mgr_usage(&adev->mman.vram_mgr); + used_vram = ttm_resource_manager_usage(&adev->mman.vram_mgr.manager); free_vram = used_vram >= total_vram ? 0 : total_vram - used_vram; spin_lock(&adev->mm_stats.lock); @@ -341,7 +342,7 @@ static void amdgpu_cs_get_threshold_for_moves(struct amdgpu_device *adev, if (free_vram >= 128 * 1024 * 1024 || free_vram >= total_vram / 8) { s64 min_us; - /* Be more aggresive on dGPUs. Try to fill a portion of free + /* Be more aggressive on dGPUs. Try to fill a portion of free * VRAM now. */ if (!(adev->flags & AMD_IS_APU)) @@ -585,6 +586,16 @@ static int amdgpu_cs_parser_bos(struct amdgpu_cs_parser *p, } } + /* Move fence waiting after getting reservation lock of + * PD root. Then there is no need on a ctx mutex lock. + */ + r = amdgpu_ctx_wait_prev_fence(p->ctx, p->entity); + if (unlikely(r != 0)) { + if (r != -ERESTARTSYS) + DRM_ERROR("amdgpu_ctx_wait_prev_fence failed.\n"); + goto error_validate; + } + amdgpu_cs_get_threshold_for_moves(p->adev, &p->bytes_moved_threshold, &p->bytes_moved_vis_threshold); p->bytes_moved = 0; @@ -775,12 +786,12 @@ static int amdgpu_cs_vm_handling(struct amdgpu_cs_parser *p) memcpy(ib->ptr, kptr, chunk_ib->ib_bytes); amdgpu_bo_kunmap(aobj); - r = amdgpu_ring_parse_cs(ring, p, j); + r = amdgpu_ring_parse_cs(ring, p, p->job, ib); if (r) return r; } else { ib->ptr = (uint32_t *)kptr; - r = amdgpu_ring_patch_cs_in_place(ring, p, j); + r = amdgpu_ring_patch_cs_in_place(ring, p, p->job, ib); amdgpu_bo_kunmap(aobj); if (r) return r; @@ -944,7 +955,7 @@ static int amdgpu_cs_ib_fill(struct amdgpu_device *adev, if (parser->job->uf_addr && ring->funcs->no_user_fence) return -EINVAL; - return amdgpu_ctx_wait_prev_fence(parser->ctx, parser->entity); + return 0; } static int amdgpu_cs_process_fence_dep(struct amdgpu_cs_parser *p, @@ -1149,6 +1160,9 @@ static int amdgpu_cs_dependencies(struct amdgpu_device *adev, { int i, r; + /* TODO: Investigate why we still need the context lock */ + mutex_unlock(&p->ctx->lock); + for (i = 0; i < p->nchunks; ++i) { struct amdgpu_cs_chunk *chunk; @@ -1159,32 +1173,34 @@ static int amdgpu_cs_dependencies(struct amdgpu_device *adev, case AMDGPU_CHUNK_ID_SCHEDULED_DEPENDENCIES: r = amdgpu_cs_process_fence_dep(p, chunk); if (r) - return r; + goto out; break; case AMDGPU_CHUNK_ID_SYNCOBJ_IN: r = amdgpu_cs_process_syncobj_in_dep(p, chunk); if (r) - return r; + goto out; break; case AMDGPU_CHUNK_ID_SYNCOBJ_OUT: r = amdgpu_cs_process_syncobj_out_dep(p, chunk); if (r) - return r; + goto out; break; case AMDGPU_CHUNK_ID_SYNCOBJ_TIMELINE_WAIT: r = amdgpu_cs_process_syncobj_timeline_in_dep(p, chunk); if (r) - return r; + goto out; break; case AMDGPU_CHUNK_ID_SYNCOBJ_TIMELINE_SIGNAL: r = amdgpu_cs_process_syncobj_timeline_out_dep(p, chunk); if (r) - return r; + goto out; break; } } - return 0; +out: + mutex_lock(&p->ctx->lock); + return r; } static void amdgpu_cs_post_dependencies(struct amdgpu_cs_parser *p) @@ -1272,16 +1288,13 @@ static int amdgpu_cs_submit(struct amdgpu_cs_parser *p, continue; /* - * Work around dma_resv shortcommings by wrapping up the + * Work around dma_resv shortcomings by wrapping up the * submission in a dma_fence_chain and add it as exclusive - * fence, but first add the submission as shared fence to make - * sure that shared fences never signal before the exclusive - * one. + * fence. */ dma_fence_chain_init(chain, dma_resv_excl_fence(resv), dma_fence_get(p->fence), 1); - dma_resv_add_shared_fence(resv, p->fence); rcu_assign_pointer(resv->fence_excl, &chain->base); e->chain = NULL; } @@ -1509,6 +1522,7 @@ int amdgpu_cs_fence_to_handle_ioctl(struct drm_device *dev, void *data, return 0; default: + dma_fence_put(fence); return -EINVAL; } } |