drm/amdgpu: only move VM BOs in the LRU during validation v2

This should save us a bunch of command submission overhead.

v2: move the LRU move to the right place to avoid the move for the root BO
    and handle the shadow BOs as well. This turned out to be a bug fix because
    the move needs to happen before the kmap.
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
Acked-by: Chunming Zhou <david1.zhou@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

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

@@ -673,10 +673,8 @@ static int amdgpu_cs_parser_bos(struct amdgpu_cs_parser *p,
 	}
 
 error_validate:
-	if (r) {
-		amdgpu_vm_move_pt_bos_in_lru(p->adev, &fpriv->vm);
+	if (r)
 		ttm_eu_backoff_reservation(&p->ticket, &p->validated);
-	}
 
 error_free_pages:
 
@@ -724,21 +722,18 @@ static int amdgpu_cs_sync_rings(struct amdgpu_cs_parser *p)
  * If error is set than unvalidate buffer, otherwise just free memory
  * used by parsing context.
  **/
-static void amdgpu_cs_parser_fini(struct amdgpu_cs_parser *parser, int error, bool backoff)
+static void amdgpu_cs_parser_fini(struct amdgpu_cs_parser *parser, int error,
+				  bool backoff)
 {
-	struct amdgpu_fpriv *fpriv = parser->filp->driver_priv;
 	unsigned i;
 
-	if (!error) {
-		amdgpu_vm_move_pt_bos_in_lru(parser->adev, &fpriv->vm);
-
+	if (!error)
 		ttm_eu_fence_buffer_objects(&parser->ticket,
 					    &parser->validated,
 					    parser->fence);
-	} else if (backoff) {
+	else if (backoff)
 		ttm_eu_backoff_reservation(&parser->ticket,
 					   &parser->validated);
-	}
 
 	for (i = 0; i < parser->num_post_dep_syncobjs; i++)
 		drm_syncobj_put(parser->post_dep_syncobjs[i]);

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

@@ -159,7 +159,8 @@ void amdgpu_vm_get_pd_bo(struct amdgpu_vm *vm,
  */
 static int amdgpu_vm_validate_level(struct amdgpu_vm_pt *parent,
 				    int (*validate)(void *, struct amdgpu_bo *),
-				    void *param, bool use_cpu_for_update)
+				    void *param, bool use_cpu_for_update,
+				    struct ttm_bo_global *glob)
 {
 	unsigned i;
 	int r;
@@ -183,12 +184,18 @@ static int amdgpu_vm_validate_level(struct amdgpu_vm_pt *parent,
 		if (r)
 			return r;
 
+		spin_lock(&glob->lru_lock);
+		ttm_bo_move_to_lru_tail(&entry->bo->tbo);
+		if (entry->bo->shadow)
+			ttm_bo_move_to_lru_tail(&entry->bo->shadow->tbo);
+		spin_unlock(&glob->lru_lock);
+
 		/*
 		 * Recurse into the sub directory. This is harmless because we
 		 * have only a maximum of 5 layers.
 		 */
 		r = amdgpu_vm_validate_level(entry, validate, param,
-					     use_cpu_for_update);
+					     use_cpu_for_update, glob);
 		if (r)
 			return r;
 	}
@@ -220,54 +227,11 @@ int amdgpu_vm_validate_pt_bos(struct amdgpu_device *adev, struct amdgpu_vm *vm,
 		return 0;
 
 	return amdgpu_vm_validate_level(&vm->root, validate, param,
-					vm->use_cpu_for_update);
+					vm->use_cpu_for_update,
+					adev->mman.bdev.glob);
 }
 
 /**
- * amdgpu_vm_move_level_in_lru - move one level of PT BOs to the LRU tail
- *
- * @adev: amdgpu device instance
- * @vm: vm providing the BOs
- *
- * Move the PT BOs to the tail of the LRU.
- */
-static void amdgpu_vm_move_level_in_lru(struct amdgpu_vm_pt *parent)
-{
-	unsigned i;
-
-	if (!parent->entries)
-		return;
-
-	for (i = 0; i <= parent->last_entry_used; ++i) {
-		struct amdgpu_vm_pt *entry = &parent->entries[i];
-
-		if (!entry->bo)
-			continue;
-
-		ttm_bo_move_to_lru_tail(&entry->bo->tbo);
-		amdgpu_vm_move_level_in_lru(entry);
-	}
-}
-
-/**
- * amdgpu_vm_move_pt_bos_in_lru - move the PT BOs to the LRU tail
- *
- * @adev: amdgpu device instance
- * @vm: vm providing the BOs
- *
- * Move the PT BOs to the tail of the LRU.
- */
-void amdgpu_vm_move_pt_bos_in_lru(struct amdgpu_device *adev,
-				  struct amdgpu_vm *vm)
-{
-	struct ttm_bo_global *glob = adev->mman.bdev.glob;
-
-	spin_lock(&glob->lru_lock);
-	amdgpu_vm_move_level_in_lru(&vm->root);
-	spin_unlock(&glob->lru_lock);
-}
-
- /**
  * amdgpu_vm_alloc_levels - allocate the PD/PT levels
  *
  * @adev: amdgpu_device pointer

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

@@ -223,8 +223,6 @@ void amdgpu_vm_get_pd_bo(struct amdgpu_vm *vm,
 int amdgpu_vm_validate_pt_bos(struct amdgpu_device *adev, struct amdgpu_vm *vm,
 			      int (*callback)(void *p, struct amdgpu_bo *bo),
 			      void *param);
-void amdgpu_vm_move_pt_bos_in_lru(struct amdgpu_device *adev,
-				  struct amdgpu_vm *vm);
 int amdgpu_vm_alloc_pts(struct amdgpu_device *adev,
 			struct amdgpu_vm *vm,
 			uint64_t saddr, uint64_t size);