diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c
index f1a857ec1021..8d5a646ebe6a 100644
--- a/drivers/gpu/drm/ttm/ttm_bo.c
+++ b/drivers/gpu/drm/ttm/ttm_bo.c
@@ -429,8 +429,20 @@ static void ttm_bo_cleanup_refs_or_queue(struct ttm_buffer_object *bo)
 		sync_obj = driver->sync_obj_ref(bo->sync_obj);
 	spin_unlock(&bdev->fence_lock);
 
-	if (!ret)
+	if (!ret) {
+
+		/*
+		 * Make NO_EVICT bos immediately available to
+		 * shrinkers, now that they are queued for
+		 * destruction.
+		 */
+		if (bo->mem.placement & TTM_PL_FLAG_NO_EVICT) {
+			bo->mem.placement &= ~TTM_PL_FLAG_NO_EVICT;
+			ttm_bo_add_to_lru(bo);
+		}
+
 		ww_mutex_unlock(&bo->resv->lock);
+	}
 
 	kref_get(&bo->list_kref);
 	list_add_tail(&bo->ddestroy, &bdev->ddestroy);
@@ -986,24 +998,32 @@ int ttm_bo_move_buffer(struct ttm_buffer_object *bo,
 	return ret;
 }
 
-static int ttm_bo_mem_compat(struct ttm_placement *placement,
-			     struct ttm_mem_reg *mem)
+static bool ttm_bo_mem_compat(struct ttm_placement *placement,
+			      struct ttm_mem_reg *mem,
+			      uint32_t *new_flags)
 {
 	int i;
 
 	if (mem->mm_node && placement->lpfn != 0 &&
 	    (mem->start < placement->fpfn ||
 	     mem->start + mem->num_pages > placement->lpfn))
-		return -1;
+		return false;
 
 	for (i = 0; i < placement->num_placement; i++) {
-		if ((placement->placement[i] & mem->placement &
-			TTM_PL_MASK_CACHING) &&
-			(placement->placement[i] & mem->placement &
-			TTM_PL_MASK_MEM))
-			return i;
+		*new_flags = placement->placement[i];
+		if ((*new_flags & mem->placement & TTM_PL_MASK_CACHING) &&
+		    (*new_flags & mem->placement & TTM_PL_MASK_MEM))
+			return true;
 	}
-	return -1;
+
+	for (i = 0; i < placement->num_busy_placement; i++) {
+		*new_flags = placement->busy_placement[i];
+		if ((*new_flags & mem->placement & TTM_PL_MASK_CACHING) &&
+		    (*new_flags & mem->placement & TTM_PL_MASK_MEM))
+			return true;
+	}
+
+	return false;
 }
 
 int ttm_bo_validate(struct ttm_buffer_object *bo,
@@ -1012,6 +1032,7 @@ int ttm_bo_validate(struct ttm_buffer_object *bo,
 			bool no_wait_gpu)
 {
 	int ret;
+	uint32_t new_flags;
 
 	lockdep_assert_held(&bo->resv->lock.base);
 	/* Check that range is valid */
@@ -1022,8 +1043,7 @@ int ttm_bo_validate(struct ttm_buffer_object *bo,
 	/*
 	 * Check whether we need to move buffer.
 	 */
-	ret = ttm_bo_mem_compat(placement, &bo->mem);
-	if (ret < 0) {
+	if (!ttm_bo_mem_compat(placement, &bo->mem, &new_flags)) {
 		ret = ttm_bo_move_buffer(bo, placement, interruptible,
 					 no_wait_gpu);
 		if (ret)
@@ -1033,7 +1053,7 @@ int ttm_bo_validate(struct ttm_buffer_object *bo,
 		 * Use the access and other non-mapping-related flag bits from
 		 * the compatible memory placement flags to the active flags
 		 */
-		ttm_flag_masked(&bo->mem.placement, placement->placement[ret],
+		ttm_flag_masked(&bo->mem.placement, new_flags,
 				~TTM_PL_MASK_MEMTYPE);
 	}
 	/*
diff --git a/drivers/gpu/drm/ttm/ttm_bo_util.c b/drivers/gpu/drm/ttm/ttm_bo_util.c
index 7cc904d3a4d1..4834c463c38b 100644
--- a/drivers/gpu/drm/ttm/ttm_bo_util.c
+++ b/drivers/gpu/drm/ttm/ttm_bo_util.c
@@ -343,19 +343,25 @@ int ttm_bo_move_memcpy(struct ttm_buffer_object *bo,
 	if (ret)
 		goto out;
 
+	/*
+	 * Single TTM move. NOP.
+	 */
 	if (old_iomap == NULL && new_iomap == NULL)
 		goto out2;
+
+	/*
+	 * Move nonexistent data. NOP.
+	 */
 	if (old_iomap == NULL && ttm == NULL)
 		goto out2;
 
-	if (ttm->state == tt_unpopulated) {
+	/*
+	 * TTM might be null for moves within the same region.
+	 */
+	if (ttm && ttm->state == tt_unpopulated) {
 		ret = ttm->bdev->driver->ttm_tt_populate(ttm);
-		if (ret) {
-			/* if we fail here don't nuke the mm node
-			 * as the bo still owns it */
-			old_copy.mm_node = NULL;
+		if (ret)
 			goto out1;
-		}
 	}
 
 	add = 0;
@@ -381,11 +387,8 @@ int ttm_bo_move_memcpy(struct ttm_buffer_object *bo,
 						   prot);
 		} else
 			ret = ttm_copy_io_page(new_iomap, old_iomap, page);
-		if (ret) {
-			/* failing here, means keep old copy as-is */
-			old_copy.mm_node = NULL;
+		if (ret)
 			goto out1;
-		}
 	}
 	mb();
 out2:
@@ -403,7 +406,12 @@ int ttm_bo_move_memcpy(struct ttm_buffer_object *bo,
 	ttm_mem_reg_iounmap(bdev, old_mem, new_iomap);
 out:
 	ttm_mem_reg_iounmap(bdev, &old_copy, old_iomap);
-	ttm_bo_mem_put(bo, &old_copy);
+
+	/*
+	 * On error, keep the mm node!
+	 */
+	if (!ret)
+		ttm_bo_mem_put(bo, &old_copy);
 	return ret;
 }
 EXPORT_SYMBOL(ttm_bo_move_memcpy);
diff --git a/drivers/gpu/drm/ttm/ttm_bo_vm.c b/drivers/gpu/drm/ttm/ttm_bo_vm.c
index 1006c15445e9..c03514b93f9c 100644
--- a/drivers/gpu/drm/ttm/ttm_bo_vm.c
+++ b/drivers/gpu/drm/ttm/ttm_bo_vm.c
@@ -41,6 +41,51 @@
 
 #define TTM_BO_VM_NUM_PREFAULT 16
 
+static int ttm_bo_vm_fault_idle(struct ttm_buffer_object *bo,
+				struct vm_area_struct *vma,
+				struct vm_fault *vmf)
+{
+	struct ttm_bo_device *bdev = bo->bdev;
+	int ret = 0;
+
+	spin_lock(&bdev->fence_lock);
+	if (likely(!test_bit(TTM_BO_PRIV_FLAG_MOVING, &bo->priv_flags)))
+		goto out_unlock;
+
+	/*
+	 * Quick non-stalling check for idle.
+	 */
+	ret = ttm_bo_wait(bo, false, false, true);
+	if (likely(ret == 0))
+		goto out_unlock;
+
+	/*
+	 * If possible, avoid waiting for GPU with mmap_sem
+	 * held.
+	 */
+	if (vmf->flags & FAULT_FLAG_ALLOW_RETRY) {
+		ret = VM_FAULT_RETRY;
+		if (vmf->flags & FAULT_FLAG_RETRY_NOWAIT)
+			goto out_unlock;
+
+		up_read(&vma->vm_mm->mmap_sem);
+		(void) ttm_bo_wait(bo, false, true, false);
+		goto out_unlock;
+	}
+
+	/*
+	 * Ordinary wait.
+	 */
+	ret = ttm_bo_wait(bo, false, true, false);
+	if (unlikely(ret != 0))
+		ret = (ret != -ERESTARTSYS) ? VM_FAULT_SIGBUS :
+			VM_FAULT_NOPAGE;
+
+out_unlock:
+	spin_unlock(&bdev->fence_lock);
+	return ret;
+}
+
 static int ttm_bo_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 {
 	struct ttm_buffer_object *bo = (struct ttm_buffer_object *)
@@ -91,18 +136,11 @@ static int ttm_bo_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 	 * Wait for buffer data in transit, due to a pipelined
 	 * move.
 	 */
-
-	spin_lock(&bdev->fence_lock);
-	if (test_bit(TTM_BO_PRIV_FLAG_MOVING, &bo->priv_flags)) {
-		ret = ttm_bo_wait(bo, false, true, false);
-		spin_unlock(&bdev->fence_lock);
-		if (unlikely(ret != 0)) {
-			retval = (ret != -ERESTARTSYS) ?
-			    VM_FAULT_SIGBUS : VM_FAULT_NOPAGE;
-			goto out_unlock;
-		}
-	} else
-		spin_unlock(&bdev->fence_lock);
+	ret = ttm_bo_vm_fault_idle(bo, vma, vmf);
+	if (unlikely(ret != 0)) {
+		retval = ret;
+		goto out_unlock;
+	}
 
 	ret = ttm_mem_io_lock(man, true);
 	if (unlikely(ret != 0)) {