]> asedeno.scripts.mit.edu Git - linux.git/commitdiff
drm/amdgpu: improve AMDGPU_GEM_CREATE_VRAM_CLEARED handling (v2)
authorChristian König <christian.koenig@amd.com>
Thu, 17 Nov 2016 11:24:48 +0000 (12:24 +0100)
committerAlex Deucher <alexander.deucher@amd.com>
Wed, 23 Nov 2016 20:08:45 +0000 (15:08 -0500)
Drop this whole reserve/unreserve dance, just lock the
reservation object manually when creating the BO.

v2: rebase on dma_fence renaming

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
drivers/gpu/drm/amd/amdgpu/amdgpu_object.c

index 611da3bd29819096b3323000b726268a85d0c683..bf79b73e1538e3785317720018ffc8376e870610 100644 (file)
@@ -371,36 +371,36 @@ int amdgpu_bo_create_restricted(struct amdgpu_device *adev,
 
        amdgpu_fill_placement_to_bo(bo, placement);
        /* Kernel allocation are uninterruptible */
+
+       if (!resv) {
+               bool locked;
+
+               reservation_object_init(&bo->tbo.ttm_resv);
+               locked = ww_mutex_trylock(&bo->tbo.ttm_resv.lock);
+               WARN_ON(!locked);
+       }
        r = ttm_bo_init(&adev->mman.bdev, &bo->tbo, size, type,
                        &bo->placement, page_align, !kernel, NULL,
-                       acc_size, sg, resv, &amdgpu_ttm_bo_destroy);
-       if (unlikely(r != 0)) {
+                       acc_size, sg, resv ? resv : &bo->tbo.ttm_resv,
+                       &amdgpu_ttm_bo_destroy);
+       if (unlikely(r != 0))
                return r;
-       }
 
        if (flags & AMDGPU_GEM_CREATE_VRAM_CLEARED &&
            bo->tbo.mem.placement & TTM_PL_FLAG_VRAM) {
                struct dma_fence *fence;
 
-               r = amdgpu_bo_reserve(bo, false);
-               if (unlikely(r != 0))
-                       goto fail_free;
-
-               amdgpu_ttm_placement_from_domain(bo, AMDGPU_GEM_DOMAIN_VRAM);
-               r = ttm_bo_validate(&bo->tbo, &bo->placement, false, false);
-               if (unlikely(r != 0))
-                       goto fail_unreserve;
-
                r = amdgpu_fill_buffer(bo, 0, bo->tbo.resv, &fence);
                if (unlikely(r))
                        goto fail_unreserve;
 
                amdgpu_bo_fence(bo, fence, false);
-               amdgpu_bo_unreserve(bo);
                dma_fence_put(bo->tbo.moving);
                bo->tbo.moving = dma_fence_get(fence);
                dma_fence_put(fence);
        }
+       if (!resv)
+               ww_mutex_unlock(&bo->tbo.resv->lock);
        *bo_ptr = bo;
 
        trace_amdgpu_bo_create(bo);
@@ -408,8 +408,7 @@ int amdgpu_bo_create_restricted(struct amdgpu_device *adev,
        return 0;
 
 fail_unreserve:
-       amdgpu_bo_unreserve(bo);
-fail_free:
+       ww_mutex_unlock(&bo->tbo.resv->lock);
        amdgpu_bo_unref(&bo);
        return r;
 }