drm/amdgpu: move GART recovery into GTT manager v2

The GTT manager handles the GART address space anyway, so it is
completely pointless to keep the same information around twice.

v2: rebased

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

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

@@ -1630,9 +1630,6 @@ struct amdgpu_device {
 	/* link all shadow bo */
 	struct list_head                shadow_list;
 	struct mutex                    shadow_list_lock;
-	/* link all gtt */
-	spinlock_t			gtt_list_lock;
-	struct list_head                gtt_list;
 	/* keep an lru list of rings by HW IP */
 	struct list_head		ring_lru_list;
 	spinlock_t			ring_lru_list_lock;

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

@@ -2180,9 +2180,6 @@ int amdgpu_device_init(struct amdgpu_device *adev,
 	INIT_LIST_HEAD(&adev->shadow_list);
 	mutex_init(&adev->shadow_list_lock);
 
-	INIT_LIST_HEAD(&adev->gtt_list);
-	spin_lock_init(&adev->gtt_list_lock);
-
 	INIT_LIST_HEAD(&adev->ring_lru_list);
 	spin_lock_init(&adev->ring_lru_list_lock);
 
@@ -2877,7 +2874,8 @@ retry:
 				atomic_inc(&adev->vram_lost_counter);
 			}
 
-			r = amdgpu_ttm_recover_gart(adev);
+			r = amdgpu_gtt_mgr_recover(
+				&adev->mman.bdev.man[TTM_PL_TT]);
 			if (r)
 				goto out;
 
@@ -2939,7 +2937,7 @@ static int amdgpu_reset_sriov(struct amdgpu_device *adev, uint64_t *reset_flags,
 		goto error;
 
 	/* we need recover gart prior to run SMC/CP/SDMA resume */
-	amdgpu_ttm_recover_gart(adev);
+	amdgpu_gtt_mgr_recover(&adev->mman.bdev.man[TTM_PL_TT]);
 
 	/* now we are okay to resume SMC/CP/SDMA */
 	r = amdgpu_sriov_reinit_late(adev);

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

@@ -31,6 +31,11 @@ struct amdgpu_gtt_mgr {
 	atomic64_t available;
 };
 
+struct amdgpu_gtt_node {
+	struct drm_mm_node node;
+	struct ttm_buffer_object *tbo;
+};
+
 /**
  * amdgpu_gtt_mgr_init - init GTT manager and DRM MM
  *
@@ -87,9 +92,9 @@ static int amdgpu_gtt_mgr_fini(struct ttm_mem_type_manager *man)
  */
 bool amdgpu_gtt_mgr_has_gart_addr(struct ttm_mem_reg *mem)
 {
-	struct drm_mm_node *node = mem->mm_node;
+	struct amdgpu_gtt_node *node = mem->mm_node;
 
-	return (node->start != AMDGPU_BO_INVALID_OFFSET);
+	return (node->node.start != AMDGPU_BO_INVALID_OFFSET);
 }
 
 /**
@@ -109,7 +114,7 @@ static int amdgpu_gtt_mgr_alloc(struct ttm_mem_type_manager *man,
 {
 	struct amdgpu_device *adev = amdgpu_ttm_adev(man->bdev);
 	struct amdgpu_gtt_mgr *mgr = man->priv;
-	struct drm_mm_node *node = mem->mm_node;
+	struct amdgpu_gtt_node *node = mem->mm_node;
 	enum drm_mm_insert_mode mode;
 	unsigned long fpfn, lpfn;
 	int r;
@@ -132,13 +137,13 @@ static int amdgpu_gtt_mgr_alloc(struct ttm_mem_type_manager *man,
 		mode = DRM_MM_INSERT_HIGH;
 
 	spin_lock(&mgr->lock);
-	r = drm_mm_insert_node_in_range(&mgr->mm, node,
-					mem->num_pages, mem->page_alignment, 0,
-					fpfn, lpfn, mode);
+	r = drm_mm_insert_node_in_range(&mgr->mm, &node->node, mem->num_pages,
+					mem->page_alignment, 0, fpfn, lpfn,
+					mode);
 	spin_unlock(&mgr->lock);
 
 	if (!r)
-		mem->start = node->start;
+		mem->start = node->node.start;
 
 	return r;
 }
@@ -159,7 +164,7 @@ static int amdgpu_gtt_mgr_new(struct ttm_mem_type_manager *man,
 			      struct ttm_mem_reg *mem)
 {
 	struct amdgpu_gtt_mgr *mgr = man->priv;
-	struct drm_mm_node *node;
+	struct amdgpu_gtt_node *node;
 	int r;
 
 	spin_lock(&mgr->lock);
@@ -177,8 +182,9 @@ static int amdgpu_gtt_mgr_new(struct ttm_mem_type_manager *man,
 		goto err_out;
 	}
 
-	node->start = AMDGPU_BO_INVALID_OFFSET;
-	node->size = mem->num_pages;
+	node->node.start = AMDGPU_BO_INVALID_OFFSET;
+	node->node.size = mem->num_pages;
+	node->tbo = tbo;
 	mem->mm_node = node;
 
 	if (place->fpfn || place->lpfn || place->flags & TTM_PL_FLAG_TOPDOWN) {
@@ -190,7 +196,7 @@ static int amdgpu_gtt_mgr_new(struct ttm_mem_type_manager *man,
 			goto err_out;
 		}
 	} else {
-		mem->start = node->start;
+		mem->start = node->node.start;
 	}
 
 	return 0;
@@ -214,14 +220,14 @@ static void amdgpu_gtt_mgr_del(struct ttm_mem_type_manager *man,
 			       struct ttm_mem_reg *mem)
 {
 	struct amdgpu_gtt_mgr *mgr = man->priv;
-	struct drm_mm_node *node = mem->mm_node;
+	struct amdgpu_gtt_node *node = mem->mm_node;
 
 	if (!node)
 		return;
 
 	spin_lock(&mgr->lock);
-	if (node->start != AMDGPU_BO_INVALID_OFFSET)
-		drm_mm_remove_node(node);
+	if (node->node.start != AMDGPU_BO_INVALID_OFFSET)
+		drm_mm_remove_node(&node->node);
 	spin_unlock(&mgr->lock);
 	atomic64_add(mem->num_pages, &mgr->available);
 
@@ -244,6 +250,25 @@ uint64_t amdgpu_gtt_mgr_usage(struct ttm_mem_type_manager *man)
 	return (result > 0 ? result : 0) * PAGE_SIZE;
 }
 
+int amdgpu_gtt_mgr_recover(struct ttm_mem_type_manager *man)
+{
+	struct amdgpu_gtt_mgr *mgr = man->priv;
+	struct amdgpu_gtt_node *node;
+	struct drm_mm_node *mm_node;
+	int r = 0;
+
+	spin_lock(&mgr->lock);
+	drm_mm_for_each_node(mm_node, &mgr->mm) {
+		node = container_of(mm_node, struct amdgpu_gtt_node, node);
+		r = amdgpu_ttm_recover_gart(node->tbo);
+		if (r)
+			break;
+	}
+	spin_unlock(&mgr->lock);
+
+	return r;
+}
+
 /**
  * amdgpu_gtt_mgr_debug - dump VRAM table
  *

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

@@ -689,7 +689,6 @@ struct amdgpu_ttm_tt {
 	struct list_head        guptasks;
 	atomic_t		mmu_invalidations;
 	uint32_t		last_set_pages;
-	struct list_head        list;
 };
 
 int amdgpu_ttm_tt_get_user_pages(struct ttm_tt *ttm, struct page **pages)
@@ -865,21 +864,14 @@ static int amdgpu_ttm_backend_bind(struct ttm_tt *ttm,
 		return 0;
 	}
 
-	spin_lock(&gtt->adev->gtt_list_lock);
 	flags = amdgpu_ttm_tt_pte_flags(gtt->adev, ttm, bo_mem);
 	gtt->offset = (u64)bo_mem->start << PAGE_SHIFT;
 	r = amdgpu_gart_bind(gtt->adev, gtt->offset, ttm->num_pages,
 		ttm->pages, gtt->ttm.dma_address, flags);
 
-	if (r) {
+	if (r)
 		DRM_ERROR("failed to bind %lu pages at 0x%08llX\n",
 			  ttm->num_pages, gtt->offset);
-		goto error_gart_bind;
-	}
-
-	list_add_tail(&gtt->list, &gtt->adev->gtt_list);
-error_gart_bind:
-	spin_unlock(&gtt->adev->gtt_list_lock);
 	return r;
 }
 
@@ -920,29 +912,23 @@ int amdgpu_ttm_bind(struct ttm_buffer_object *bo)
 	return r;
 }
 
-int amdgpu_ttm_recover_gart(struct amdgpu_device *adev)
+int amdgpu_ttm_recover_gart(struct ttm_buffer_object *tbo)
 {
-	struct amdgpu_ttm_tt *gtt, *tmp;
-	struct ttm_mem_reg bo_mem;
+	struct amdgpu_device *adev = amdgpu_ttm_adev(tbo->bdev);
+	struct amdgpu_ttm_tt *gtt = (void *)tbo->ttm;
 	uint64_t flags;
 	int r;
 
-	bo_mem.mem_type = TTM_PL_TT;
-	spin_lock(&adev->gtt_list_lock);
-	list_for_each_entry_safe(gtt, tmp, &adev->gtt_list, list) {
-		flags = amdgpu_ttm_tt_pte_flags(gtt->adev, &gtt->ttm.ttm, &bo_mem);
-		r = amdgpu_gart_bind(adev, gtt->offset, gtt->ttm.ttm.num_pages,
-				     gtt->ttm.ttm.pages, gtt->ttm.dma_address,
-				     flags);
-		if (r) {
-			spin_unlock(&adev->gtt_list_lock);
-			DRM_ERROR("failed to bind %lu pages at 0x%08llX\n",
-				  gtt->ttm.ttm.num_pages, gtt->offset);
-			return r;
-		}
-	}
-	spin_unlock(&adev->gtt_list_lock);
-	return 0;
+	if (!gtt)
+		return 0;
+
+	flags = amdgpu_ttm_tt_pte_flags(adev, &gtt->ttm.ttm, &tbo->mem);
+	r = amdgpu_gart_bind(adev, gtt->offset, gtt->ttm.ttm.num_pages,
+			     gtt->ttm.ttm.pages, gtt->ttm.dma_address, flags);
+	if (r)
+		DRM_ERROR("failed to bind %lu pages at 0x%08llX\n",
+			  gtt->ttm.ttm.num_pages, gtt->offset);
+	return r;
 }
 
 static int amdgpu_ttm_backend_unbind(struct ttm_tt *ttm)
@@ -957,16 +943,10 @@ static int amdgpu_ttm_backend_unbind(struct ttm_tt *ttm)
 		return 0;
 
 	/* unbind shouldn't be done for GDS/GWS/OA in ttm_bo_clean_mm */
-	spin_lock(&gtt->adev->gtt_list_lock);
 	r = amdgpu_gart_unbind(gtt->adev, gtt->offset, ttm->num_pages);
-	if (r) {
+	if (r)
 		DRM_ERROR("failed to unbind %lu pages at 0x%08llX\n",
 			  gtt->ttm.ttm.num_pages, gtt->offset);
-		goto error_unbind;
-	}
-	list_del_init(&gtt->list);
-error_unbind:
-	spin_unlock(&gtt->adev->gtt_list_lock);
 	return r;
 }
 
@@ -1003,7 +983,6 @@ static struct ttm_tt *amdgpu_ttm_tt_create(struct ttm_bo_device *bdev,
 		kfree(gtt);
 		return NULL;
 	}
-	INIT_LIST_HEAD(&gtt->list);
 	return &gtt->ttm.ttm;
 }
 

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

@@ -69,6 +69,7 @@ extern const struct ttm_mem_type_manager_func amdgpu_vram_mgr_func;
 
 bool amdgpu_gtt_mgr_has_gart_addr(struct ttm_mem_reg *mem);
 uint64_t amdgpu_gtt_mgr_usage(struct ttm_mem_type_manager *man);
+int amdgpu_gtt_mgr_recover(struct ttm_mem_type_manager *man);
 
 uint64_t amdgpu_vram_mgr_usage(struct ttm_mem_type_manager *man);
 uint64_t amdgpu_vram_mgr_vis_usage(struct ttm_mem_type_manager *man);
@@ -91,7 +92,7 @@ int amdgpu_fill_buffer(struct amdgpu_bo *bo,
 
 int amdgpu_mmap(struct file *filp, struct vm_area_struct *vma);
 int amdgpu_ttm_bind(struct ttm_buffer_object *bo);
-int amdgpu_ttm_recover_gart(struct amdgpu_device *adev);
+int amdgpu_ttm_recover_gart(struct ttm_buffer_object *tbo);
 
 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);