drm/ttm: Fix a deadlock if the target BO is not idle during swap
authorxinhui pan <xinhui.pan@amd.com>
Tue, 7 Sep 2021 04:08:32 +0000 (12:08 +0800)
committerDave Airlie <airlied@redhat.com>
Fri, 10 Sep 2021 06:18:53 +0000 (16:18 +1000)
The ret value might be -EBUSY, caller will think lru lock is still
locked but actually NOT. So return -ENOSPC instead. Otherwise we hit
list corruption.

ttm_bo_cleanup_refs might fail too if BO is not idle. If we return 0,
caller(ttm_tt_populate -> ttm_global_swapout ->ttm_device_swapout) will
be stuck as we actually did not free any BO memory. This usually happens
when the fence is not signaled for a long time.

Signed-off-by: xinhui pan <xinhui.pan@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Fixes: ebd59851c796 ("drm/ttm: move swapout logic around v3")
Link: https://patchwork.freedesktop.org/patch/msgid/20210907040832.1107747-1-xinhui.pan@amd.com
Signed-off-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
drivers/gpu/drm/ttm/ttm_bo.c

index ea4add2b97179c47cf6ef93fd039ce5fb47d5c32..bb9e02c31946e906bd5c2533082fe6c76055689b 100644 (file)
@@ -1160,9 +1160,9 @@ int ttm_bo_swapout(struct ttm_buffer_object *bo, struct ttm_operation_ctx *ctx,
        }
 
        if (bo->deleted) {
-               ttm_bo_cleanup_refs(bo, false, false, locked);
+               ret = ttm_bo_cleanup_refs(bo, false, false, locked);
                ttm_bo_put(bo);
-               return 0;
+               return ret == -EBUSY ? -ENOSPC : ret;
        }
 
        ttm_bo_del_from_lru(bo);
@@ -1216,7 +1216,7 @@ out:
        if (locked)
                dma_resv_unlock(bo->base.resv);
        ttm_bo_put(bo);
-       return ret;
+       return ret == -EBUSY ? -ENOSPC : ret;
 }
 
 void ttm_bo_tt_destroy(struct ttm_buffer_object *bo)