drm/vmwgfx: Introduce a pin count to allow for recursive pinning v2

v2: Fix dma buffer validation on resource pinning.

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Sinclair Yeh <syeh@vmware.com>

9 files changed, 179 insertions, 196 deletions

diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_dmabuf.c b/drivers/gpu/drm/vmwgfx/vmwgfx_dmabuf.c
index 914b375763dc..4b9344dd6c27 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_dmabuf.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_dmabuf.c
@@ -32,25 +32,20 @@
 
 
 /**
- * vmw_dmabuf_to_placement - Validate a buffer to placement.
+ * vmw_dmabuf_pin_in_placement - Validate a buffer to placement.
  *
  * @dev_priv:  Driver private.
  * @buf:  DMA buffer to move.
- * @pin:  Pin buffer if true.
+ * @placement:  The placement to pin it.
  * @interruptible:  Use interruptible wait.
  *
- * May only be called by the current master since it assumes that the
- * master lock is the current master's lock.
- * This function takes the master's lock in write mode.
- * Flushes and unpins the query bo to avoid failures.
- *
  * Returns
  *  -ERESTARTSYS if interrupted by a signal.
  */
-int vmw_dmabuf_to_placement(struct vmw_private *dev_priv,
-			    struct vmw_dma_buffer *buf,
-			    struct ttm_placement *placement,
-			    bool interruptible)
+int vmw_dmabuf_pin_in_placement(struct vmw_private *dev_priv,
+				struct vmw_dma_buffer *buf,
+				struct ttm_placement *placement,
+				bool interruptible)
 {
 	struct ttm_buffer_object *bo = &buf->base;
 	int ret;
@@ -66,6 +61,8 @@ int vmw_dmabuf_to_placement(struct vmw_private *dev_priv,
 		goto err;
 
 	ret = ttm_bo_validate(bo, placement, interruptible, false);
+	if (!ret)
+		vmw_bo_pin_reserved(buf, true);
 
 	ttm_bo_unreserve(bo);
 
@@ -75,12 +72,10 @@ err:
 }
 
 /**
- * vmw_dmabuf_to_vram_or_gmr - Move a buffer to vram or gmr.
+ * vmw_dmabuf_pin_in_vram_or_gmr - Move a buffer to vram or gmr.
  *
- * May only be called by the current master since it assumes that the
- * master lock is the current master's lock.
- * This function takes the master's lock in write mode.
- * Flushes and unpins the query bo if @pin == true to avoid failures.
+ * This function takes the reservation_sem in write mode.
+ * Flushes and unpins the query bo to avoid failures.
  *
  * @dev_priv:  Driver private.
  * @buf:  DMA buffer to move.
@@ -90,55 +85,34 @@ err:
  * Returns
  * -ERESTARTSYS if interrupted by a signal.
  */
-int vmw_dmabuf_to_vram_or_gmr(struct vmw_private *dev_priv,
-			      struct vmw_dma_buffer *buf,
-			      bool pin, bool interruptible)
+int vmw_dmabuf_pin_in_vram_or_gmr(struct vmw_private *dev_priv,
+				  struct vmw_dma_buffer *buf,
+				  bool interruptible)
 {
 	struct ttm_buffer_object *bo = &buf->base;
-	struct ttm_placement *placement;
 	int ret;
 
 	ret = ttm_write_lock(&dev_priv->reservation_sem, interruptible);
 	if (unlikely(ret != 0))
 		return ret;
 
-	if (pin)
-		vmw_execbuf_release_pinned_bo(dev_priv);
+	vmw_execbuf_release_pinned_bo(dev_priv);
 
 	ret = ttm_bo_reserve(bo, interruptible, false, false, NULL);
 	if (unlikely(ret != 0))
 		goto err;
 
-	/**
-	 * Put BO in VRAM if there is space, otherwise as a GMR.
-	 * If there is no space in VRAM and GMR ids are all used up,
-	 * start evicting GMRs to make room. If the DMA buffer can't be
-	 * used as a GMR, this will return -ENOMEM.
-	 */
-
-	if (pin)
-		placement = &vmw_vram_gmr_ne_placement;
-	else
-		placement = &vmw_vram_gmr_placement;
-
-	ret = ttm_bo_validate(bo, placement, interruptible, false);
+	ret = ttm_bo_validate(bo, &vmw_vram_gmr_placement, interruptible,
+			      false);
 	if (likely(ret == 0) || ret == -ERESTARTSYS)
-		goto err_unreserve;
-
+		goto out_unreserve;
 
-	/**
-	 * If that failed, try VRAM again, this time evicting
-	 * previous contents.
-	 */
-
-	if (pin)
-		placement = &vmw_vram_ne_placement;
-	else
-		placement = &vmw_vram_placement;
+	ret = ttm_bo_validate(bo, &vmw_vram_placement, interruptible, false);
 
-	ret = ttm_bo_validate(bo, placement, interruptible, false);
+out_unreserve:
+	if (!ret)
+		vmw_bo_pin_reserved(buf, true);
 
-err_unreserve:
 	ttm_bo_unreserve(bo);
 err:
 	ttm_write_unlock(&dev_priv->reservation_sem);
@@ -146,67 +120,50 @@ err:
 }
 
 /**
- * vmw_dmabuf_to_vram - Move a buffer to vram.
+ * vmw_dmabuf_pin_in_vram - Move a buffer to vram.
  *
- * May only be called by the current master since it assumes that the
- * master lock is the current master's lock.
- * This function takes the master's lock in write mode.
+ * This function takes the reservation_sem in write mode.
+ * Flushes and unpins the query bo to avoid failures.
  *
  * @dev_priv:  Driver private.
  * @buf:  DMA buffer to move.
- * @pin:  Pin buffer in vram if true.
  * @interruptible:  Use interruptible wait.
  *
  * Returns
  * -ERESTARTSYS if interrupted by a signal.
  */
-int vmw_dmabuf_to_vram(struct vmw_private *dev_priv,
-		       struct vmw_dma_buffer *buf,
-		       bool pin, bool interruptible)
+int vmw_dmabuf_pin_in_vram(struct vmw_private *dev_priv,
+			   struct vmw_dma_buffer *buf,
+			   bool interruptible)
 {
-	struct ttm_placement *placement;
-
-	if (pin)
-		placement = &vmw_vram_ne_placement;
-	else
-		placement = &vmw_vram_placement;
-
-	return vmw_dmabuf_to_placement(dev_priv, buf,
-				       placement,
-				       interruptible);
+	return vmw_dmabuf_pin_in_placement(dev_priv, buf, &vmw_vram_placement,
+					   interruptible);
 }
 
 /**
- * vmw_dmabuf_to_start_of_vram - Move a buffer to start of vram.
+ * vmw_dmabuf_pin_in_start_of_vram - Move a buffer to start of vram.
  *
- * May only be called by the current master since it assumes that the
- * master lock is the current master's lock.
- * This function takes the master's lock in write mode.
- * Flushes and unpins the query bo if @pin == true to avoid failures.
+ * This function takes the reservation_sem in write mode.
+ * Flushes and unpins the query bo to avoid failures.
  *
  * @dev_priv:  Driver private.
- * @buf:  DMA buffer to move.
- * @pin:  Pin buffer in vram if true.
+ * @buf:  DMA buffer to pin.
  * @interruptible:  Use interruptible wait.
  *
  * Returns
  * -ERESTARTSYS if interrupted by a signal.
  */
-int vmw_dmabuf_to_start_of_vram(struct vmw_private *dev_priv,
-				struct vmw_dma_buffer *buf,
-				bool pin, bool interruptible)
+int vmw_dmabuf_pin_in_start_of_vram(struct vmw_private *dev_priv,
+				    struct vmw_dma_buffer *buf,
+				    bool interruptible)
 {
 	struct ttm_buffer_object *bo = &buf->base;
 	struct ttm_placement placement;
 	struct ttm_place place;
 	int ret = 0;
 
-	if (pin)
-		place = vmw_vram_ne_placement.placement[0];
-	else
-		place = vmw_vram_placement.placement[0];
+	place = vmw_vram_placement.placement[0];
 	place.lpfn = bo->num_pages;
-
 	placement.num_placement = 1;
 	placement.placement = &place;
 	placement.num_busy_placement = 1;
@@ -216,13 +173,16 @@ int vmw_dmabuf_to_start_of_vram(struct vmw_private *dev_priv,
 	if (unlikely(ret != 0))
 		return ret;
 
-	if (pin)
-		vmw_execbuf_release_pinned_bo(dev_priv);
+	vmw_execbuf_release_pinned_bo(dev_priv);
 	ret = ttm_bo_reserve(bo, interruptible, false, false, NULL);
 	if (unlikely(ret != 0))
 		goto err_unlock;
 
-	/* Is this buffer already in vram but not at the start of it? */
+	/*
+	 * Is this buffer already in vram but not at the start of it?
+	 * In that case, evict it first because TTM isn't good at handling
+	 * that situation.
+	 */
 	if (bo->mem.mem_type == TTM_PL_VRAM &&
 	    bo->mem.start < bo->num_pages &&
 	    bo->mem.start > 0)
@@ -230,8 +190,10 @@ int vmw_dmabuf_to_start_of_vram(struct vmw_private *dev_priv,
 
 	ret = ttm_bo_validate(bo, &placement, interruptible, false);
 
-	/* For some reason we didn't up at the start of vram */
+	/* For some reason we didn't end up at the start of vram */
 	WARN_ON(ret == 0 && bo->offset != 0);
+	if (!ret)
+		vmw_bo_pin_reserved(buf, true);
 
 	ttm_bo_unreserve(bo);
 err_unlock:
@@ -240,13 +202,10 @@ err_unlock:
 	return ret;
 }
 
-
 /**
- * vmw_dmabuf_upin - Unpin the buffer given buffer, does not move the buffer.
+ * vmw_dmabuf_unpin - Unpin the buffer given buffer, does not move the buffer.
  *
- * May only be called by the current master since it assumes that the
- * master lock is the current master's lock.
- * This function takes the master's lock in write mode.
+ * This function takes the reservation_sem in write mode.
  *
  * @dev_priv:  Driver private.
  * @buf:  DMA buffer to unpin.
@@ -259,16 +218,25 @@ int vmw_dmabuf_unpin(struct vmw_private *dev_priv,
 		     struct vmw_dma_buffer *buf,
 		     bool interruptible)
 {
-	/*
-	 * We could in theory early out if the buffer is
-	 * unpinned but we need to lock and reserve the buffer
-	 * anyways so we don't gain much by that.
-	 */
-	return vmw_dmabuf_to_placement(dev_priv, buf,
-				       &vmw_evictable_placement,
-				       interruptible);
-}
+	struct ttm_buffer_object *bo = &buf->base;
+	int ret;
 
+	ret = ttm_read_lock(&dev_priv->reservation_sem, interruptible);
+	if (unlikely(ret != 0))
+		return ret;
+
+	ret = ttm_bo_reserve(bo, interruptible, false, false, 0);
+	if (unlikely(ret != 0))
+		goto err;
+
+	vmw_bo_pin_reserved(buf, false);
+
+	ttm_bo_unreserve(bo);
+
+err:
+	ttm_read_unlock(&dev_priv->reservation_sem);
+	return ret;
+}
 
 /**
  * vmw_bo_get_guest_ptr - Get the guest ptr representing the current placement
@@ -291,21 +259,31 @@ void vmw_bo_get_guest_ptr(const struct ttm_buffer_object *bo,
 
 
 /**
- * vmw_bo_pin - Pin or unpin a buffer object without moving it.
+ * vmw_bo_pin_reserved - Pin or unpin a buffer object without moving it.
  *
- * @bo: The buffer object. Must be reserved.
+ * @vbo: The buffer object. Must be reserved.
  * @pin: Whether to pin or unpin.
  *
  */
-void vmw_bo_pin(struct ttm_buffer_object *bo, bool pin)
+void vmw_bo_pin_reserved(struct vmw_dma_buffer *vbo, bool pin)
 {
 	struct ttm_place pl;
 	struct ttm_placement placement;
+	struct ttm_buffer_object *bo = &vbo->base;
 	uint32_t old_mem_type = bo->mem.mem_type;
 	int ret;
 
 	lockdep_assert_held(&bo->resv->lock.base);
 
+	if (pin) {
+		if (vbo->pin_count++ > 0)
+			return;
+	} else {
+		WARN_ON(vbo->pin_count <= 0);
+		if (--vbo->pin_count > 0)
+			return;
+	}
+
 	pl.fpfn = 0;
 	pl.lpfn = 0;
 	pl.flags = TTM_PL_FLAG_VRAM | VMW_PL_FLAG_GMR | VMW_PL_FLAG_MOB
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c
index ab1b70ce19c1..e55db3fdf601 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c
@@ -298,30 +298,31 @@ static void vmw_print_capabilities(uint32_t capabilities)
 static int vmw_dummy_query_bo_create(struct vmw_private *dev_priv)
 {
 	int ret;
-	struct ttm_buffer_object *bo;
+	struct vmw_dma_buffer *vbo;
 	struct ttm_bo_kmap_obj map;
 	volatile SVGA3dQueryResult *result;
 	bool dummy;
 
 	/*
-	 * Create the bo as pinned, so that a tryreserve will
+	 * Create the vbo as pinned, so that a tryreserve will
 	 * immediately succeed. This is because we're the only
 	 * user of the bo currently.
 	 */
-	ret = ttm_bo_create(&dev_priv->bdev,
-			    PAGE_SIZE,
-			    ttm_bo_type_device,
-			    &vmw_sys_ne_placement,
-			    0, false, NULL,
-			    &bo);
+	vbo = kzalloc(sizeof(*vbo), GFP_KERNEL);
+	if (!vbo)
+		return -ENOMEM;
 
+	ret = vmw_dmabuf_init(dev_priv, vbo, PAGE_SIZE,
+			      &vmw_sys_ne_placement, false,
+			      &vmw_dmabuf_bo_free);
 	if (unlikely(ret != 0))
 		return ret;
 
-	ret = ttm_bo_reserve(bo, false, true, false, NULL);
+	ret = ttm_bo_reserve(&vbo->base, false, true, false, NULL);
 	BUG_ON(ret != 0);
+	vmw_bo_pin_reserved(vbo, true);
 
-	ret = ttm_bo_kmap(bo, 0, 1, &map);
+	ret = ttm_bo_kmap(&vbo->base, 0, 1, &map);
 	if (likely(ret == 0)) {
 		result = ttm_kmap_obj_virtual(&map, &dummy);
 		result->totalSize = sizeof(*result);
@@ -329,14 +330,14 @@ static int vmw_dummy_query_bo_create(struct vmw_private *dev_priv)
 		result->result32 = 0xff;
 		ttm_bo_kunmap(&map);
 	}
-	vmw_bo_pin(bo, false);
-	ttm_bo_unreserve(bo);
+	vmw_bo_pin_reserved(vbo, false);
+	ttm_bo_unreserve(&vbo->base);
 
 	if (unlikely(ret != 0)) {
 		DRM_ERROR("Dummy query buffer map failed.\n");
-		ttm_bo_unref(&bo);
+		vmw_dmabuf_unreference(&vbo);
 	} else
-		dev_priv->dummy_query_bo = bo;
+		dev_priv->dummy_query_bo = vbo;
 
 	return ret;
 }
@@ -434,7 +435,7 @@ static void vmw_release_device_early(struct vmw_private *dev_priv)
 
 	BUG_ON(dev_priv->pinned_bo != NULL);
 
-	ttm_bo_unref(&dev_priv->dummy_query_bo);
+	vmw_dmabuf_unreference(&dev_priv->dummy_query_bo);
 	if (dev_priv->cman)
 		vmw_cmdbuf_remove_pool(dev_priv->cman);
 
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
index 5d04859a472d..12eaa6c805d8 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
@@ -85,6 +85,7 @@ struct vmw_fpriv {
 struct vmw_dma_buffer {
 	struct ttm_buffer_object base;
 	struct list_head res_list;
+	s32 pin_count;
 };
 
 /**
@@ -358,7 +359,7 @@ struct vmw_sw_context{
 	uint32_t *cmd_bounce;
 	uint32_t cmd_bounce_size;
 	struct list_head resource_list;
-	struct ttm_buffer_object *cur_query_bo;
+	struct vmw_dma_buffer *cur_query_bo;
 	struct list_head res_relocations;
 	uint32_t *buf_start;
 	struct vmw_res_cache_entry res_cache[vmw_res_max];
@@ -533,8 +534,8 @@ struct vmw_private {
 	 * are protected by the cmdbuf mutex.
 	 */
 
-	struct ttm_buffer_object *dummy_query_bo;
-	struct ttm_buffer_object *pinned_bo;
+	struct vmw_dma_buffer *dummy_query_bo;
+	struct vmw_dma_buffer *pinned_bo;
 	uint32_t query_cid;
 	uint32_t query_cid_valid;
 	bool dummy_query_bo_pinned;
@@ -700,25 +701,25 @@ extern void vmw_resource_evict_all(struct vmw_private *dev_priv);
 /**
  * DMA buffer helper routines - vmwgfx_dmabuf.c
  */
-extern int vmw_dmabuf_to_placement(struct vmw_private *vmw_priv,
-				   struct vmw_dma_buffer *bo,
-				   struct ttm_placement *placement,
-				   bool interruptible);
-extern int vmw_dmabuf_to_vram(struct vmw_private *dev_priv,
-			      struct vmw_dma_buffer *buf,
-			      bool pin, bool interruptible);
-extern int vmw_dmabuf_to_vram_or_gmr(struct vmw_private *dev_priv,
-				     struct vmw_dma_buffer *buf,
-				     bool pin, bool interruptible);
-extern int vmw_dmabuf_to_start_of_vram(struct vmw_private *vmw_priv,
+extern int vmw_dmabuf_pin_in_placement(struct vmw_private *vmw_priv,
 				       struct vmw_dma_buffer *bo,
-				       bool pin, bool interruptible);
+				       struct ttm_placement *placement,
+				       bool interruptible);
+extern int vmw_dmabuf_pin_in_vram(struct vmw_private *dev_priv,
+				  struct vmw_dma_buffer *buf,
+				  bool interruptible);
+extern int vmw_dmabuf_pin_in_vram_or_gmr(struct vmw_private *dev_priv,
+					 struct vmw_dma_buffer *buf,
+					 bool interruptible);
+extern int vmw_dmabuf_pin_in_start_of_vram(struct vmw_private *vmw_priv,
+					   struct vmw_dma_buffer *bo,
+					   bool interruptible);
 extern int vmw_dmabuf_unpin(struct vmw_private *vmw_priv,
 			    struct vmw_dma_buffer *bo,
 			    bool interruptible);
 extern void vmw_bo_get_guest_ptr(const struct ttm_buffer_object *buf,
 				 SVGAGuestPtr *ptr);
-extern void vmw_bo_pin(struct ttm_buffer_object *bo, bool pin);
+extern void vmw_bo_pin_reserved(struct vmw_dma_buffer *bo, bool pin);
 
 /**
  * Misc Ioctl functionality - vmwgfx_ioctl.c
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c b/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c
index 0ec5fd6c71f4..92e89987b0d7 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c
@@ -308,7 +308,7 @@ static int vmw_cmd_ok(struct vmw_private *dev_priv,
  * submission is reached.
  */
 static int vmw_bo_to_validate_list(struct vmw_sw_context *sw_context,
-				   struct ttm_buffer_object *bo,
+				   struct vmw_dma_buffer *vbo,
 				   bool validate_as_mob,
 				   uint32_t *p_val_node)
 {
@@ -318,7 +318,7 @@ static int vmw_bo_to_validate_list(struct vmw_sw_context *sw_context,
 	struct drm_hash_item *hash;
 	int ret;
 
-	if (likely(drm_ht_find_item(&sw_context->res_ht, (unsigned long) bo,
+	if (likely(drm_ht_find_item(&sw_context->res_ht, (unsigned long) vbo,
 				    &hash) == 0)) {
 		vval_buf = container_of(hash, struct vmw_validate_buffer,
 					hash);
@@ -336,7 +336,7 @@ static int vmw_bo_to_validate_list(struct vmw_sw_context *sw_context,
 			return -EINVAL;
 		}
 		vval_buf = &sw_context->val_bufs[val_node];
-		vval_buf->hash.key = (unsigned long) bo;
+		vval_buf->hash.key = (unsigned long) vbo;
 		ret = drm_ht_insert_item(&sw_context->res_ht, &vval_buf->hash);
 		if (unlikely(ret != 0)) {
 			DRM_ERROR("Failed to initialize a buffer validation "
@@ -345,7 +345,7 @@ static int vmw_bo_to_validate_list(struct vmw_sw_context *sw_context,
 		}
 		++sw_context->cur_val_buf;
 		val_buf = &vval_buf->base;
-		val_buf->bo = ttm_bo_reference(bo);
+		val_buf->bo = ttm_bo_reference(&vbo->base);
 		val_buf->shared = false;
 		list_add_tail(&val_buf->head, &sw_context->validate_nodes);
 		vval_buf->validate_as_mob = validate_as_mob;
@@ -380,10 +380,10 @@ static int vmw_resources_reserve(struct vmw_sw_context *sw_context)
 			return ret;
 
 		if (res->backup) {
-			struct ttm_buffer_object *bo = &res->backup->base;
+			struct vmw_dma_buffer *vbo = res->backup;
 
 			ret = vmw_bo_to_validate_list
-				(sw_context, bo,
+				(sw_context, vbo,
 				 vmw_resource_needs_backup(res), NULL);
 
 			if (unlikely(ret != 0))
@@ -759,7 +759,7 @@ static int vmw_cmd_present_check(struct vmw_private *dev_priv,
  * command batch.
  */
 static int vmw_query_bo_switch_prepare(struct vmw_private *dev_priv,
-				       struct ttm_buffer_object *new_query_bo,
+				       struct vmw_dma_buffer *new_query_bo,
 				       struct vmw_sw_context *sw_context)
 {
 	struct vmw_res_cache_entry *ctx_entry =
@@ -771,7 +771,7 @@ static int vmw_query_bo_switch_prepare(struct vmw_private *dev_priv,
 
 	if (unlikely(new_query_bo != sw_context->cur_query_bo)) {
 
-		if (unlikely(new_query_bo->num_pages > 4)) {
+		if (unlikely(new_query_bo->base.num_pages > 4)) {
 			DRM_ERROR("Query buffer too large.\n");
 			return -EINVAL;
 		}
@@ -840,12 +840,12 @@ static void vmw_query_bo_switch_commit(struct vmw_private *dev_priv,
 
 	if (dev_priv->pinned_bo != sw_context->cur_query_bo) {
 		if (dev_priv->pinned_bo) {
-			vmw_bo_pin(dev_priv->pinned_bo, false);
-			ttm_bo_unref(&dev_priv->pinned_bo);
+			vmw_bo_pin_reserved(dev_priv->pinned_bo, false);
+			vmw_dmabuf_unreference(&dev_priv->pinned_bo);
 		}
 
 		if (!sw_context->needs_post_query_barrier) {
-			vmw_bo_pin(sw_context->cur_query_bo, true);
+			vmw_bo_pin_reserved(sw_context->cur_query_bo, true);
 
 			/*
 			 * We pin also the dummy_query_bo buffer so that we
@@ -853,14 +853,17 @@ static void vmw_query_bo_switch_commit(struct vmw_private *dev_priv,
 			 * dummy queries in context destroy paths.
 			 */
 
-			vmw_bo_pin(dev_priv->dummy_query_bo, true);
-			dev_priv->dummy_query_bo_pinned = true;
+			if (!dev_priv->dummy_query_bo_pinned) {
+				vmw_bo_pin_reserved(dev_priv->dummy_query_bo,
+						    true);
+				dev_priv->dummy_query_bo_pinned = true;
+			}
 
 			BUG_ON(sw_context->last_query_ctx == NULL);
 			dev_priv->query_cid = sw_context->last_query_ctx->id;
 			dev_priv->query_cid_valid = true;
 			dev_priv->pinned_bo =
-				ttm_bo_reference(sw_context->cur_query_bo);
+				vmw_dmabuf_reference(sw_context->cur_query_bo);
 		}
 	}
 }
@@ -889,7 +892,6 @@ static int vmw_translate_mob_ptr(struct vmw_private *dev_priv,
 				 struct vmw_dma_buffer **vmw_bo_p)
 {
 	struct vmw_dma_buffer *vmw_bo = NULL;
-	struct ttm_buffer_object *bo;
 	uint32_t handle = *id;
 	struct vmw_relocation *reloc;
 	int ret;
@@ -900,7 +902,6 @@ static int vmw_translate_mob_ptr(struct vmw_private *dev_priv,
 		ret = -EINVAL;
 		goto out_no_reloc;
 	}
-	bo = &vmw_bo->base;
 
 	if (unlikely(sw_context->cur_reloc >= VMWGFX_MAX_RELOCATIONS)) {
 		DRM_ERROR("Max number relocations per submission"
@@ -913,7 +914,7 @@ static int vmw_translate_mob_ptr(struct vmw_private *dev_priv,
 	reloc->mob_loc = id;
 	reloc->location = NULL;
 
-	ret = vmw_bo_to_validate_list(sw_context, bo, true, &reloc->index);
+	ret = vmw_bo_to_validate_list(sw_context, vmw_bo, true, &reloc->index);
 	if (unlikely(ret != 0))
 		goto out_no_reloc;
 
@@ -951,7 +952,6 @@ static int vmw_translate_guest_ptr(struct vmw_private *dev_priv,
 				   struct vmw_dma_buffer **vmw_bo_p)
 {
 	struct vmw_dma_buffer *vmw_bo = NULL;
-	struct ttm_buffer_object *bo;
 	uint32_t handle = ptr->gmrId;
 	struct vmw_relocation *reloc;
 	int ret;
@@ -962,7 +962,6 @@ static int vmw_translate_guest_ptr(struct vmw_private *dev_priv,
 		ret = -EINVAL;
 		goto out_no_reloc;
 	}
-	bo = &vmw_bo->base;
 
 	if (unlikely(sw_context->cur_reloc >= VMWGFX_MAX_RELOCATIONS)) {
 		DRM_ERROR("Max number relocations per submission"
@@ -974,7 +973,7 @@ static int vmw_translate_guest_ptr(struct vmw_private *dev_priv,
 	reloc = &sw_context->relocs[sw_context->cur_reloc++];
 	reloc->location = ptr;
 
-	ret = vmw_bo_to_validate_list(sw_context, bo, false, &reloc->index);
+	ret = vmw_bo_to_validate_list(sw_context, vmw_bo, false, &reloc->index);
 	if (unlikely(ret != 0))
 		goto out_no_reloc;
 
@@ -1081,7 +1080,7 @@ static int vmw_cmd_end_gb_query(struct vmw_private *dev_priv,
 	if (unlikely(ret != 0))
 		return ret;
 
-	ret = vmw_query_bo_switch_prepare(dev_priv, &vmw_bo->base, sw_context);
+	ret = vmw_query_bo_switch_prepare(dev_priv, vmw_bo, sw_context);
 
 	vmw_dmabuf_unreference(&vmw_bo);
 	return ret;
@@ -1135,7 +1134,7 @@ static int vmw_cmd_end_query(struct vmw_private *dev_priv,
 	if (unlikely(ret != 0))
 		return ret;
 
-	ret = vmw_query_bo_switch_prepare(dev_priv, &vmw_bo->base, sw_context);
+	ret = vmw_query_bo_switch_prepare(dev_priv, vmw_bo, sw_context);
 
 	vmw_dmabuf_unreference(&vmw_bo);
 	return ret;
@@ -2239,16 +2238,11 @@ static int vmw_validate_single_buffer(struct vmw_private *dev_priv,
 				      struct ttm_buffer_object *bo,
 				      bool validate_as_mob)
 {
+	struct vmw_dma_buffer *vbo = container_of(bo, struct vmw_dma_buffer,
+						  base);
 	int ret;
 
-
-	/*
-	 * Don't validate pinned buffers.
-	 */
-
-	if (bo == dev_priv->pinned_bo ||
-	    (bo == dev_priv->dummy_query_bo &&
-	     dev_priv->dummy_query_bo_pinned))
+	if (vbo->pin_count > 0)
 		return 0;
 
 	if (validate_as_mob)
@@ -2767,9 +2761,11 @@ static void vmw_execbuf_unpin_panic(struct vmw_private *dev_priv)
 	DRM_ERROR("Can't unpin query buffer. Trying to recover.\n");
 
 	(void) vmw_fallback_wait(dev_priv, false, true, 0, false, 10*HZ);
-	vmw_bo_pin(dev_priv->pinned_bo, false);
-	vmw_bo_pin(dev_priv->dummy_query_bo, false);
-	dev_priv->dummy_query_bo_pinned = false;
+	vmw_bo_pin_reserved(dev_priv->pinned_bo, false);
+	if (dev_priv->dummy_query_bo_pinned) {
+		vmw_bo_pin_reserved(dev_priv->dummy_query_bo, false);
+		dev_priv->dummy_query_bo_pinned = false;
+	}
 }
 
 
@@ -2811,11 +2807,11 @@ void __vmw_execbuf_release_pinned_bo(struct vmw_private *dev_priv,
 
 	INIT_LIST_HEAD(&validate_list);
 
-	pinned_val.bo = ttm_bo_reference(dev_priv->pinned_bo);
+	pinned_val.bo = ttm_bo_reference(&dev_priv->pinned_bo->base);
 	pinned_val.shared = false;
 	list_add_tail(&pinned_val.head, &validate_list);
 
-	query_val.bo = ttm_bo_reference(dev_priv->dummy_query_bo);
+	query_val.bo = ttm_bo_reference(&dev_priv->dummy_query_bo->base);
 	query_val.shared = false;
 	list_add_tail(&query_val.head, &validate_list);
 
@@ -2836,10 +2832,11 @@ void __vmw_execbuf_release_pinned_bo(struct vmw_private *dev_priv,
 		dev_priv->query_cid_valid = false;
 	}
 
-	vmw_bo_pin(dev_priv->pinned_bo, false);
-	vmw_bo_pin(dev_priv->dummy_query_bo, false);
-	dev_priv->dummy_query_bo_pinned = false;
-
+	vmw_bo_pin_reserved(dev_priv->pinned_bo, false);
+	if (dev_priv->dummy_query_bo_pinned) {
+		vmw_bo_pin_reserved(dev_priv->dummy_query_bo, false);
+		dev_priv->dummy_query_bo_pinned = false;
+	}
 	if (fence == NULL) {
 		(void) vmw_execbuf_fence_commands(NULL, dev_priv, &lfence,
 						  NULL);
@@ -2851,7 +2848,9 @@ void __vmw_execbuf_release_pinned_bo(struct vmw_private *dev_priv,
 
 	ttm_bo_unref(&query_val.bo);
 	ttm_bo_unref(&pinned_val.bo);
-	ttm_bo_unref(&dev_priv->pinned_bo);
+	vmw_dmabuf_unreference(&dev_priv->pinned_bo);
+	DRM_INFO("Dummy query bo pin count: %d\n",
+		 dev_priv->dummy_query_bo->pin_count);
 
 out_unlock:
 	return;
@@ -2861,7 +2860,7 @@ out_no_emit:
 out_no_reserve:
 	ttm_bo_unref(&query_val.bo);
 	ttm_bo_unref(&pinned_val.bo);
-	ttm_bo_unref(&dev_priv->pinned_bo);
+	vmw_dmabuf_unreference(&dev_priv->pinned_bo);
 }
 
 /**
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_fb.c b/drivers/gpu/drm/vmwgfx/vmwgfx_fb.c
index d0a3bcf5c0d2..b54d99bca9bf 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_fb.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_fb.c
@@ -636,7 +636,7 @@ int vmw_fb_on(struct vmw_private *vmw_priv)
 	/* Make sure that all overlays are stoped when we take over */
 	vmw_overlay_stop_all(vmw_priv);
 
-	ret = vmw_dmabuf_to_start_of_vram(vmw_priv, par->vmw_bo, true, false);
+	ret = vmw_dmabuf_pin_in_start_of_vram(vmw_priv, par->vmw_bo, false);
 	if (unlikely(ret != 0)) {
 		DRM_ERROR("could not move buffer to start of VRAM\n");
 		goto err_no_buffer;
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c b/drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c
index 239815c8b073..9b8b09f8135b 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_fifo.c
@@ -604,7 +604,7 @@ static int vmw_fifo_emit_dummy_legacy_query(struct vmw_private *dev_priv,
 	 * without writing to the query result structure.
 	 */
 
-	struct ttm_buffer_object *bo = dev_priv->dummy_query_bo;
+	struct ttm_buffer_object *bo = &dev_priv->dummy_query_bo->base;
 	struct {
 		SVGA3dCmdHeader header;
 		SVGA3dCmdWaitForQuery body;
@@ -653,7 +653,7 @@ static int vmw_fifo_emit_dummy_gb_query(struct vmw_private *dev_priv,
 	 * without writing to the query result structure.
 	 */
 
-	struct ttm_buffer_object *bo = dev_priv->dummy_query_bo;
+	struct ttm_buffer_object *bo = &dev_priv->dummy_query_bo->base;
 	struct {
 		SVGA3dCmdHeader header;
 		SVGA3dCmdWaitForGBQuery body;
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c b/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c
index 615ff6cfc4f9..99e2f5b9a023 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_kms.c
@@ -678,7 +678,7 @@ static int vmw_framebuffer_dmabuf_pin(struct vmw_framebuffer *vfb)
 
 	vmw_overlay_pause_all(dev_priv);
 
-	ret = vmw_dmabuf_to_start_of_vram(dev_priv, vfbd->buffer, true, false);
+	ret = vmw_dmabuf_pin_in_start_of_vram(dev_priv, vfbd->buffer, false);
 
 	vmw_overlay_resume_all(dev_priv);
 
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_overlay.c b/drivers/gpu/drm/vmwgfx/vmwgfx_overlay.c
index 7f4b2f072c6f..d839051cc1cb 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_overlay.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_overlay.c
@@ -232,9 +232,9 @@ static int vmw_overlay_move_buffer(struct vmw_private *dev_priv,
 		return vmw_dmabuf_unpin(dev_priv, buf, inter);
 
 	if (dev_priv->active_display_unit == vmw_du_legacy)
-		return vmw_dmabuf_to_vram(dev_priv, buf, true, inter);
+		return vmw_dmabuf_pin_in_vram(dev_priv, buf, inter);
 
-	return vmw_dmabuf_to_vram_or_gmr(dev_priv, buf, true, inter);
+	return vmw_dmabuf_pin_in_vram_or_gmr(dev_priv, buf, inter);
 }
 
 /**
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_resource.c b/drivers/gpu/drm/vmwgfx/vmwgfx_resource.c
index 9dcbe8ba08ea..271bc900d83a 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_resource.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_resource.c
@@ -1596,25 +1596,29 @@ int vmw_resource_pin(struct vmw_resource *res)
 		goto out_no_reserve;
 
 	if (res->pin_count == 0) {
-		struct ttm_buffer_object *bo = NULL;
+		struct vmw_dma_buffer *vbo = NULL;
 
 		if (res->backup) {
-			bo = &res->backup->base;
-
-			ttm_bo_reserve(bo, false, false, false, NULL);
-			ret = ttm_bo_validate(bo, res->func->backup_placement,
-					      false, false);
-			if (ret) {
-				ttm_bo_unreserve(bo);
-				goto out_no_validate;
+			vbo = res->backup;
+
+			ttm_bo_reserve(&vbo->base, false, false, false, NULL);
+			if (!vbo->pin_count) {
+				ret = ttm_bo_validate
+					(&vbo->base,
+					 res->func->backup_placement,
+					 false, false);
+				if (ret) {
+					ttm_bo_unreserve(&vbo->base);
+					goto out_no_validate;
+				}
 			}
 
 			/* Do we really need to pin the MOB as well? */
-			vmw_bo_pin(bo, true);
+			vmw_bo_pin_reserved(vbo, true);
 		}
 		ret = vmw_resource_validate(res);
-		if (bo)
-			ttm_bo_unreserve(bo);
+		if (vbo)
+			ttm_bo_unreserve(&vbo->base);
 		if (ret)
 			goto out_no_validate;
 	}
@@ -1650,11 +1654,11 @@ void vmw_resource_unpin(struct vmw_resource *res)
 
 	WARN_ON(res->pin_count == 0);
 	if (--res->pin_count == 0 && res->backup) {
-		struct ttm_buffer_object *bo = &res->backup->base;
+		struct vmw_dma_buffer *vbo = res->backup;
 
-		ttm_bo_reserve(bo, false, false, false, NULL);
-		vmw_bo_pin(bo, false);
-		ttm_bo_unreserve(bo);
+		ttm_bo_reserve(&vbo->base, false, false, false, NULL);
+		vmw_bo_pin_reserved(vbo, false);
+		ttm_bo_unreserve(&vbo->base);
 	}
 
 	vmw_resource_unreserve(res, NULL, 0UL);