One request may get driver tag several times, and active request is only
de-accounted in blk_mq_put_driver_tag() in case of io scheduler, and it
won't be done for none, so repeated accounting of active request may
happen.
Fix this issue by only accounting active request in case that RQF_MQ_INFLIGHT
isn't set.
Cc: Qian Cai <cai@xxxxxx>
Tested-by: Sachin Sant <sachinp@xxxxxxxxxxxxxxxxxx>
Fixes: 37f4a24c2469 ("blk-mq: centralise related handling into blk_mq_get_driver_tag")
Signed-off-by: Ming Lei <ming.lei@xxxxxxxxxx>
---
block/blk-mq.c | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/block/blk-mq.c b/block/blk-mq.c
index 6b36969220c1..656d92de9814 100644
--- a/block/blk-mq.c
+++ b/block/blk-mq.c
@@ -1149,7 +1149,8 @@ static bool blk_mq_get_driver_tag(struct request *rq)
if (rq->tag == BLK_MQ_NO_TAG && !__blk_mq_get_driver_tag(rq))
return false;
- if (hctx->flags & BLK_MQ_F_TAG_SHARED) {
+ if ((hctx->flags & BLK_MQ_F_TAG_SHARED) &&
+ !(rq->rq_flags & RQF_MQ_INFLIGHT)) {
rq->rq_flags |= RQF_MQ_INFLIGHT;
atomic_inc(&hctx->nr_active);
}
--
2.25.2
