blk-mq: don't call io sched's .requeue_request when requeueing rq to ->dispatch
authorMing Lei <ming.lei@redhat.com>
Fri, 23 Feb 2018 15:36:56 +0000 (23:36 +0800)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Fri, 9 Mar 2018 06:41:20 +0000 (22:41 -0800)
commit 105976f517791aed3b11f8f53b308a2069d42055 upstream.

__blk_mq_requeue_request() covers two cases:

- one is that the requeued request is added to hctx->dispatch, such as
blk_mq_dispatch_rq_list()

- another case is that the request is requeued to io scheduler, such as
blk_mq_requeue_request().

We should call io sched's .requeue_request callback only for the 2nd
case.

Cc: Paolo Valente <paolo.valente@linaro.org>
Cc: Omar Sandoval <osandov@fb.com>
Fixes: bd166ef183c2 ("blk-mq-sched: add framework for MQ capable IO schedulers")
Cc: stable@vger.kernel.org
Reviewed-by: Bart Van Assche <bart.vanassche@wdc.com>
Acked-by: Paolo Valente <paolo.valente@linaro.org>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
block/blk-mq.c

index b60798a..f1fb126 100644 (file)
@@ -638,7 +638,6 @@ static void __blk_mq_requeue_request(struct request *rq)
 
        trace_block_rq_requeue(q, rq);
        wbt_requeue(q->rq_wb, &rq->issue_stat);
-       blk_mq_sched_requeue_request(rq);
 
        if (test_and_clear_bit(REQ_ATOM_STARTED, &rq->atomic_flags)) {
                if (q->dma_drain_size && blk_rq_bytes(rq))
@@ -650,6 +649,9 @@ void blk_mq_requeue_request(struct request *rq, bool kick_requeue_list)
 {
        __blk_mq_requeue_request(rq);
 
+       /* this request will be re-inserted to io scheduler queue */
+       blk_mq_sched_requeue_request(rq);
+
        BUG_ON(blk_queued_rq(rq));
        blk_mq_add_to_requeue_list(rq, true, kick_requeue_list);
 }