blk-mq: insert request not through ->queue_rq into sw/scheduler queue
authorMing Lei <ming.lei@redhat.com>
Tue, 18 Aug 2020 09:07:28 +0000 (17:07 +0800)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Thu, 3 Sep 2020 09:29:27 +0000 (11:29 +0200)
commitb5ce818355a832da2c01add29208135f1b136e84
treee51f19288e19dd113fdef5d5f90c0a985236fc9c
parent3cea38246ad47a2cd95628d0c24936138313fec8
blk-mq: insert request not through ->queue_rq into sw/scheduler queue

[ Upstream commit db03f88fae8a2c8007caafa70287798817df2875 ]

c616cbee97ae ("blk-mq: punt failed direct issue to dispatch list") supposed
to add request which has been through ->queue_rq() to the hw queue dispatch
list, however it adds request running out of budget or driver tag to hw queue
too. This way basically bypasses request merge, and causes too many request
dispatched to LLD, and system% is unnecessary increased.

Fixes this issue by adding request not through ->queue_rq into sw/scheduler
queue, and this way is safe because no ->queue_rq is called on this request
yet.

High %system can be observed on Azure storvsc device, and even soft lock
is observed. This patch reduces %system during heavy sequential IO,
meantime decreases soft lockup risk.

Fixes: c616cbee97ae ("blk-mq: punt failed direct issue to dispatch list")
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Bart Van Assche <bvanassche@acm.org>
Cc: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Sasha Levin <sashal@kernel.org>
block/blk-mq.c