On Wed, Jun 01, 2022 at 09:14:29AM +0200, Christoph Hellwig wrote:
> On Wed, Jun 01, 2022 at 03:09:26PM +0800, Ming Lei wrote:
> > > Yes, we probably need a blk_mq_quiesce_queue call like in the incremental
> > > patch below. Do you have any good reproducer, though?
> >
> > blktests block/027 should cover this.
>
> That did not trigger the problem for me.
This kind of issue is often not 100% duplicated.
>
> > > if (q->elevator) {
> > > + blk_mq_quiesce_queue(q);
> > > +
> > > mutex_lock(&q->sysfs_lock);
> > > elevator_exit(q);
> > > mutex_unlock(&q->sysfs_lock);
> > > +
> > > + blk_mq_unquiesce_queue(q);
> > > }
> > >
> >
> > I am afraid the above way may slow down disk shutdown a lot, see
> > the following commit, that is also the reason why I moved it into disk
> > release handler, when any sync io submission are done.
>
> SCSI devices that are just probed and never had a disk attached will
> not have q->elevator set and not hit this quiesce at all.
Yes, but host with hundreds of real LUNs may be shutdown slowly too
since sd_remove() won't be called in async way.
Thanks,
Ming
