drm/ttm: enable swapout for reserved BOs during allocation

if the bo shares same reservation object then not lock it again
at swapout time to make it possible to swap out.

v2: refine the commmit message
Reviewed-by: Thomas Hellström <thellstrom@vmware.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Chuming Zhou <david1.zhou@amd.com>
Signed-off-by: Roger He <Hongbo.He@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drivers/gpu/drm/ttm/ttm_bo.c

@@ -1699,19 +1699,21 @@ EXPORT_SYMBOL(ttm_bo_synccpu_write_release);
  * A buffer object shrink method that tries to swap out the first
  * buffer object on the bo_global::swap_lru list.
  */
-int ttm_bo_swapout(struct ttm_bo_global *glob)
+int ttm_bo_swapout(struct ttm_bo_global *glob, struct ttm_operation_ctx *ctx)
 {
 	struct ttm_buffer_object *bo;
 	int ret = -EBUSY;
+	bool locked;
 	unsigned i;
 
 	spin_lock(&glob->lru_lock);
 	for (i = 0; i < TTM_MAX_BO_PRIORITY; ++i) {
 		list_for_each_entry(bo, &glob->swap_lru[i], swap) {
-			ret = reservation_object_trylock(bo->resv) ? 0 : -EBUSY;
-			if (!ret)
+			if (ttm_bo_evict_swapout_allowable(bo, ctx, &locked)) {
+				ret = 0;
 				break;
 			}
+		}
 		if (!ret)
 			break;
 	}
@@ -1786,7 +1788,12 @@ EXPORT_SYMBOL(ttm_bo_swapout);
 
 void ttm_bo_swapout_all(struct ttm_bo_device *bdev)
 {
-	while (ttm_bo_swapout(bdev->glob) == 0)
+	struct ttm_operation_ctx ctx = {
+		.interruptible = false,
+		.no_wait_gpu = false
+	};
+
+	while (ttm_bo_swapout(bdev->glob, &ctx) == 0)
 		;
 }
 EXPORT_SYMBOL(ttm_bo_swapout_all);

drivers/gpu/drm/ttm/ttm_memory.c

@@ -211,7 +211,7 @@ static bool ttm_zones_above_swap_target(struct ttm_mem_global *glob,
  */
 
 static void ttm_shrink(struct ttm_mem_global *glob, bool from_wq,
-		       uint64_t extra)
+			uint64_t extra, struct ttm_operation_ctx *ctx)
 {
 	int ret;
 
@@ -219,7 +219,7 @@ static void ttm_shrink(struct ttm_mem_global *glob, bool from_wq,
 
 	while (ttm_zones_above_swap_target(glob, from_wq, extra)) {
 		spin_unlock(&glob->lock);
-		ret = ttm_bo_swapout(glob->bo_glob);
+		ret = ttm_bo_swapout(glob->bo_glob, ctx);
 		spin_lock(&glob->lock);
 		if (unlikely(ret != 0))
 			break;
@@ -230,10 +230,14 @@ static void ttm_shrink(struct ttm_mem_global *glob, bool from_wq,
 
 static void ttm_shrink_work(struct work_struct *work)
 {
+	struct ttm_operation_ctx ctx = {
+		.interruptible = false,
+		.no_wait_gpu = false
+	};
 	struct ttm_mem_global *glob =
 	    container_of(work, struct ttm_mem_global, work);
 
-	ttm_shrink(glob, true, 0ULL);
+	ttm_shrink(glob, true, 0ULL, &ctx);
 }
 
 static int ttm_mem_init_kernel_zone(struct ttm_mem_global *glob,
@@ -520,7 +524,7 @@ static int ttm_mem_global_alloc_zone(struct ttm_mem_global *glob,
 			return -ENOMEM;
 		if (unlikely(count-- == 0))
 			return -ENOMEM;
-		ttm_shrink(glob, false, memory + (memory >> 2) + 16);
+		ttm_shrink(glob, false, memory + (memory >> 2) + 16, ctx);
 	}
 
 	return 0;

include/drm/ttm/ttm_bo_api.h

@@ -752,7 +752,8 @@ ssize_t ttm_bo_io(struct ttm_bo_device *bdev, struct file *filp,
 		  const char __user *wbuf, char __user *rbuf,
 		  size_t count, loff_t *f_pos, bool write);
 
-int ttm_bo_swapout(struct ttm_bo_global *glob);
+int ttm_bo_swapout(struct ttm_bo_global *glob,
+			struct ttm_operation_ctx *ctx);
 void ttm_bo_swapout_all(struct ttm_bo_device *bdev);
 int ttm_bo_wait_unreserved(struct ttm_buffer_object *bo);
 #endif