[PATCH] fix: infiniband/hfi1: __sdma_process_event: fix refcount underflow on s00_hw_down + e40_sw_cleaned event path

[WenTao Liang]

When the SDMA engine is in the s00_hw_down state and receives the
  e40_sw_cleaned event, the code directly calls sdma_sw_tear_down which
  unconditionally performs a kref_put via sdma_put. However, no matching
  sdma_get was ever executed on this path (sdma_get is only called in the
  e10/e30 event paths). This causes a refcount underflow.

Skip the sdma_sw_tear_down call in the s00_hw_down + e40_sw_cleaned path
  since no reference was acquired.

Cc: stable@xxxxxxxxxxxxxxx
Fixes: 7724105686e7 ("IB/hfi1: add driver files")
Signed-off-by: WenTao Liang <vulab@xxxxxxxxxxx>
---
 drivers/infiniband/hw/hfi1/sdma.c | 1 -
 1 file changed, 1 deletion(-)

diff --git a/drivers/infiniband/hw/hfi1/sdma.c b/drivers/infiniband/hw/hfi1/sdma.c
index cfd9dd0f7e81..bf35deb8aa6f 100644
--- a/drivers/infiniband/hw/hfi1/sdma.c
+++ b/drivers/infiniband/hw/hfi1/sdma.c
@@ -2539,7 +2539,6 @@ static void __sdma_process_event(struct sdma_engine *sde,
 		case sdma_event_e25_hw_clean_up_done:
 			break;
 		case sdma_event_e40_sw_cleaned:
-			sdma_sw_tear_down(sde);
 			break;
 		case sdma_event_e50_hw_cleaned:
 			break;
-- 
2.39.5 (Apple Git-154)