Re: [PATCH] blk-mq: fix blk_mq_tagset_busy_iter

From: Ming Lei
Date: Thu Aug 02 2018 - 13:06:23 EST


On Thu, Aug 02, 2018 at 09:54:06AM -0700, James Bottomley wrote:
> On Fri, 2018-08-03 at 00:43 +0800, Ming Lei wrote:
> > Commit d250bf4e776ff09d5("blk-mq: only iterate over inflight requests
> > in blk_mq_tagset_busy_iter") uses 'blk_mq_rq_state(rq) ==
> > MQ_RQ_IN_FLIGHT'
> > to replace 'blk_mq_request_started(req)', this way is wrong, and
> > causes
> > lots of test system hang during booting.
> >
> > Fix the issue by using blk_mq_request_started(req) inside
> > bt_tags_iter().
> >
> > Fixes: d250bf4e776ff09d5 ("blk-mq: only iterate over inflight
> > requests in blk_mq_tagset_busy_iter")
> > Cc: Josef Bacik <josef@xxxxxxxxxxxxxx>
> > Cc: Christoph Hellwig <hch@xxxxxx>
> > Cc: Guenter Roeck <linux@xxxxxxxxxxxx>
> > Cc: Mark Brown <broonie@xxxxxxxxxx>
> > Cc: Matt Hart <matthew.hart@xxxxxxxxxx>
> > Cc: Johannes Thumshirn <jthumshirn@xxxxxxx>
> > Cc: John Garry <john.garry@xxxxxxxxxx>
> > Cc: Hannes Reinecke <hare@xxxxxxxx>,
> > Cc: "Martin K. Petersen" <martin.petersen@xxxxxxxxxx>,
> > Cc: James Bottomley <James.Bottomley@xxxxxxxxxxxxxxxxxxxxx>
> > Cc: linux-scsi@xxxxxxxxxxxxxxx
> > Cc: linux-kernel@xxxxxxxxxxxxxxx
> > Signed-off-by: Ming Lei <ming.lei@xxxxxxxxxx>
> > ---
> >  block/blk-mq-tag.c | 2 +-
> >  1 file changed, 1 insertion(+), 1 deletion(-)
> >
> > diff --git a/block/blk-mq-tag.c b/block/blk-mq-tag.c
> > index 09b2ee6694fb..3de0836163c2 100644
> > --- a/block/blk-mq-tag.c
> > +++ b/block/blk-mq-tag.c
> > @@ -271,7 +271,7 @@ static bool bt_tags_iter(struct sbitmap *bitmap,
> > unsigned int bitnr, void *data)
> >    * test and set the bit before assining ->rqs[].
> >    */
> >   rq = tags->rqs[bitnr];
> > - if (rq && blk_mq_rq_state(rq) == MQ_RQ_IN_FLIGHT)
> > + if (rq && blk_mq_request_started(rq))
>
> So now we have dueling versions of this patch:
>
> https://marc.info/?l=linux-scsi&m=153322802207688
>
> Can we at least make sure we've root caused the problem and confirmed
> we've got it fixed before we start the formal patch process? When we

EH uses scsi_host_busy to check if the error handler needs to be waken
up. And blk_mq_tagset_busy_iter() is used for implementing scsi_host_busy(),
so causes EH not waken up, then this timed-out request can't be handled.

> do start the formal patch process, please give appropriate credit to
> the reporter(s) since this has been a royal pain for them to help us
> track down.

Sure.

Jens, could you add reported-by if you are fine with this version? Or please
just let me know if new version is needed, then I can add it.


Thanks,
Ming