Re: [PATCH v2 2/4] drm/ttm: Don't shadow the operation context

From: Christian König
Date: Mon Jun 26 2023 - 11:15:45 EST


Am 26.06.23 um 11:14 schrieb Thomas Hellström:
ttm_bo_swapout() shadows the ttm operation context which may cause
major confusion in driver callbacks when swapping out !TTM_PL_SYSTEM
memory. Fix this by reusing the operation context argument to
ttm_bo_swapout().

Cc: "Christian König" <christian.koenig@xxxxxxx>
Cc: Roger He <Hongbo.He@xxxxxxx>
Cc: <dri-devel@xxxxxxxxxxxxxxxxxxxxx>
Cc: <intel-gfx@xxxxxxxxxxxxxxxxxxxxx>
Cc: <stable@xxxxxxxxxxxxxxx> # v4.16+
Fixes: dc947770cf34 ("drm/ttm: enable swapout for reserved BOs during allocation")
Signed-off-by: Thomas Hellström <thomas.hellstrom@xxxxxxxxxxxxxxx>
Acked-by: Matthew Brost <matthew.brost@xxxxxxxxx>

We intentionally didn't used the parameter here, but I absolutely can't figure out why.

Feel free to add my rb, but let's give it some time upstream before you base anything on top of this. Just in case we missed something.

Regards,
Christian.

---
drivers/gpu/drm/ttm/ttm_bo.c | 3 +--
1 file changed, 1 insertion(+), 2 deletions(-)

diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c
index bd5dae4d1624..615d30c4262d 100644
--- a/drivers/gpu/drm/ttm/ttm_bo.c
+++ b/drivers/gpu/drm/ttm/ttm_bo.c
@@ -1154,7 +1154,6 @@ int ttm_bo_swapout(struct ttm_buffer_object *bo, struct ttm_operation_ctx *ctx,
* Move to system cached
*/
if (bo->resource->mem_type != TTM_PL_SYSTEM) {
- struct ttm_operation_ctx ctx = { false, false };
struct ttm_resource *evict_mem;
struct ttm_place hop;
@@ -1164,7 +1163,7 @@ int ttm_bo_swapout(struct ttm_buffer_object *bo, struct ttm_operation_ctx *ctx,
if (unlikely(ret))
goto out;
- ret = ttm_bo_handle_move_mem(bo, evict_mem, true, &ctx, &hop);
+ ret = ttm_bo_handle_move_mem(bo, evict_mem, true, ctx, &hop);
if (unlikely(ret != 0)) {
WARN(ret == -EMULTIHOP, "Unexpected multihop in swaput - likely driver bug.\n");
goto out;