drm/amdgpu: move userptr BOs to CPU domain during CS v2

Instead of moving them in the MMU notifier move them during CS.

v2: still mark pages as accessed/dirty
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com> (v1)
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drivers/gpu/drm/amd/amdgpu/amdgpu.h

@@ -1794,6 +1794,7 @@ void amdgpu_ttm_placement_from_domain(struct amdgpu_bo *abo, u32 domain);
 bool amdgpu_ttm_bo_is_amdgpu_bo(struct ttm_buffer_object *bo);
 int amdgpu_ttm_tt_get_user_pages(struct ttm_tt *ttm, struct page **pages);
 void amdgpu_ttm_tt_set_user_pages(struct ttm_tt *ttm, struct page **pages);
+void amdgpu_ttm_tt_mark_user_pages(struct ttm_tt *ttm);
 int amdgpu_ttm_tt_set_userptr(struct ttm_tt *ttm, uint64_t addr,
 				     uint32_t flags);
 bool amdgpu_ttm_tt_has_userptr(struct ttm_tt *ttm);

drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c

@@ -475,6 +475,12 @@ static int amdgpu_cs_list_validate(struct amdgpu_cs_parser *p,
 		/* Check if we have user pages and nobody bound the BO already */
 		if (amdgpu_ttm_tt_userptr_needs_pages(bo->tbo.ttm) &&
 		    lobj->user_pages) {
+			amdgpu_ttm_placement_from_domain(bo,
+							 AMDGPU_GEM_DOMAIN_CPU);
+			r = ttm_bo_validate(&bo->tbo, &bo->placement, true,
+					    false);
+			if (r)
+				return r;
 			amdgpu_ttm_tt_set_user_pages(bo->tbo.ttm,
 						     lobj->user_pages);
 			binding_userptr = true;

drivers/gpu/drm/amd/amdgpu/amdgpu_mn.c

@@ -137,10 +137,7 @@ static void amdgpu_mn_invalidate_node(struct amdgpu_mn_node *node,
 		if (r <= 0)
 			DRM_ERROR("(%ld) failed to wait for user bo\n", r);
 
-		amdgpu_ttm_placement_from_domain(bo, AMDGPU_GEM_DOMAIN_CPU);
-		r = ttm_bo_validate(&bo->tbo, &bo->placement, false, false);
-		if (r)
-			DRM_ERROR("(%ld) failed to validate user bo\n", r);
+		amdgpu_ttm_tt_mark_user_pages(bo->tbo.ttm);
 
 		amdgpu_bo_unreserve(bo);
 	}

drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c

@@ -685,6 +685,24 @@ void amdgpu_ttm_tt_set_user_pages(struct ttm_tt *ttm, struct page **pages)
 	}
 }
 
+void amdgpu_ttm_tt_mark_user_pages(struct ttm_tt *ttm)
+{
+	struct amdgpu_ttm_tt *gtt = (void *)ttm;
+	unsigned i;
+
+	for (i = 0; i < ttm->num_pages; ++i) {
+		struct page *page = ttm->pages[i];
+
+		if (!page)
+			continue;
+
+		if (!(gtt->userflags & AMDGPU_GEM_USERPTR_READONLY))
+			set_page_dirty(page);
+
+		mark_page_accessed(page);
+	}
+}
+
 static void amdgpu_trace_dma_map(struct ttm_tt *ttm)
 {
 	struct amdgpu_device *adev = amdgpu_ttm_adev(ttm->bdev);
@@ -740,7 +758,6 @@ static void amdgpu_ttm_tt_unpin_userptr(struct ttm_tt *ttm)
 {
 	struct amdgpu_device *adev = amdgpu_ttm_adev(ttm->bdev);
 	struct amdgpu_ttm_tt *gtt = (void *)ttm;
-	struct sg_page_iter sg_iter;
 
 	int write = !(gtt->userflags & AMDGPU_GEM_USERPTR_READONLY);
 	enum dma_data_direction direction = write ?
@@ -753,13 +770,7 @@ static void amdgpu_ttm_tt_unpin_userptr(struct ttm_tt *ttm)
 	/* free the sg table and pages again */
 	dma_unmap_sg(adev->dev, ttm->sg->sgl, ttm->sg->nents, direction);
 
-	for_each_sg_page(ttm->sg->sgl, &sg_iter, ttm->sg->nents, 0) {
-		struct page *page = sg_page_iter_page(&sg_iter);
-		if (!(gtt->userflags & AMDGPU_GEM_USERPTR_READONLY))
-			set_page_dirty(page);
-
-		mark_page_accessed(page);
-	}
+	amdgpu_ttm_tt_mark_user_pages(ttm);
 
 	amdgpu_trace_dma_unmap(ttm);