On Mon, Nov 27, 2017 at 05:14:46PM +0000, Bart Van Assche wrote:

On Mon, 2017-11-27 at 13:07 +0800, Ming Lei wrote:

quoted

If .queue_rq() returns BLK_STS_RESOURCE, blk-mq will rerun the queue in
the three situations:

1) if BLK_MQ_S_SCHED_RESTART is set
- queue is rerun after one rq is completed, see blk_mq_sched_restart()
which is run from blk_mq_free_request()

2) run out of driver tag
- queue is rerun after one tag is freed

3) otherwise
- queue is run immediately in blk_mq_dispatch_rq_list()

This random dealy of running hw queue is introduced by commit 6077c2d706097c0
(dm rq: Avoid that request processing stalls sporadically), which claimed
one request processing stalling is fixed, but never explained the behind
idea, and it is a workaound at most. Even the question isn't explained by
anyone in recent discussion.

Also calling blk_mq_delay_run_hw_queue() inside .queue_rq() is a horrible
hack because it makes BLK_MQ_S_SCHED_RESTART not working, and degrades I/O
peformance a lot.

Finally this patch makes sure that dm-rq returns BLK_STS_RESOURCE to blk-mq
only when underlying queue is out of resource, so we switch to return
DM_MAPIO_DELAY_REQUEU if either MPATHF_QUEUE_IO or MPATHF_PG_INIT_REQUIRED
is set in multipath_clone_and_map().

Sorry but in my opinion the above description shows that you don't understand
the dm-mpath driver completely.

I have to treat your above comment as a noop since you never provide a explanation.

Also I don't think it is wrong to deal with MPATHF_QUEUE_IO/MPATHF_PG_INIT_REQUIRED
via DM_MAPIO_DELAY_REQUEUE, since both can seldom happen, and the delay
won't cause performance issue.

The idea behind this change is that this patchset switches to return BLK_STS_RESOURCE
to blk-mq only when we run out of resource, but the above two(MPATHF_QUEUE_IO and
MPATHF_PG_INIT_REQUIRED) don't belong to 'run out of resource'.

quoted

diff --git a/drivers/md/dm-mpath.c b/drivers/md/dm-mpath.c
index c8faa2b85842..8fe3f45407ce 100644
--- a/drivers/md/dm-mpath.c
+++ b/drivers/md/dm-mpath.c

@@ -484,9 +484,7 @@ static int multipath_clone_and_map(struct dm_target *ti, struct request *rq,
 		return DM_MAPIO_KILL;
 	} else if (test_bit(MPATHF_QUEUE_IO, &m->flags) ||
 		   test_bit(MPATHF_PG_INIT_REQUIRED, &m->flags)) {
-		if (pg_init_all_paths(m))
-			return DM_MAPIO_DELAY_REQUEUE;
-		return DM_MAPIO_REQUEUE;
+		return DM_MAPIO_DELAY_REQUEUE;
 	}

This patch removes a pg_init_all_paths() call but you don't explain why you
think it is allowed to remove that call. Did you perhaps remove that call by
mistake?

OK, that is a problem, will fix it in V2.

-- 
Ming