Re: [PATCH] blk-mq: Fix blk_mq_tagset_busy_iter() for shared tags

[John Garry]

On 18/10/2021 10:07, Ming Lei wrote:
> On Mon, Oct 18, 2021 at 09:08:57AM +0100, John Garry wrote:
> > On 13/10/2021 16:13, John Garry wrote:
> > > > diff --git a/block/blk-mq-tag.c b/block/blk-mq-tag.c
> > > > index 72a2724a4eee..2a2ad6dfcc33 100644
> > > > --- a/block/blk-mq-tag.c
> > > > +++ b/block/blk-mq-tag.c
> > > > @@ -232,8 +232,9 @@ static bool bt_iter(struct sbitmap *bitmap,
> > > > unsigned int bitnr, void *data)
> > > >        if (!rq)
> > > >            return true;
> > > > -    if (rq->q == hctx->queue && rq->mq_hctx == hctx)
> > > > -        ret = iter_data->fn(hctx, rq, iter_data->data, reserved);
> > > > +    if (rq->q == hctx->queue && (rq->mq_hctx == hctx ||
> > > > +                blk_mq_is_shared_tags(hctx->flags)))
> > > > +        ret = iter_data->fn(rq->mq_hctx, rq, iter_data->data, reserved);
> > > >        blk_mq_put_rq_ref(rq);
> > > >        return ret;
> > > >    }
> > > > @@ -460,6 +461,9 @@ void blk_mq_queue_tag_busy_iter(struct
> > > > request_queue *q, busy_iter_fn *fn,
> > > >            if (tags->nr_reserved_tags)
> > > >                bt_for_each(hctx, &tags->breserved_tags, fn, priv, true);
> > > >            bt_for_each(hctx, &tags->bitmap_tags, fn, priv, false);
> > > > +
> > > > +        if (blk_mq_is_shared_tags(hctx->flags))
> > > > +            break;
> > > >        }
> > > >        blk_queue_exit(q);
> > > >    }
> > > I suppose that is ok, and means that we iter once.
> > >
> > > However, I have to ask, where is the big user of
> > > blk_mq_queue_tag_busy_iter() coming from? I saw this from Kashyap's
> > > mail:
> > >
> > >   > 1.31%     1.31%  kworker/57:1H-k  [kernel.vmlinux]
> > >   >       native_queued_spin_lock_slowpath
> > >   >       ret_from_fork
> > >   >       kthread
> > >   >       worker_thread
> > >   >       process_one_work
> > >   >       blk_mq_timeout_work
> > >   >       blk_mq_queue_tag_busy_iter
> > >   >       bt_iter
> > >   >       blk_mq_find_and_get_req
> > >   >       _raw_spin_lock_irqsave
> > >   >       native_queued_spin_lock_slowpath
> > >
> > > How or why blk_mq_timeout_work()?
> > Just some update: I tried hisi_sas with 10x SAS SSDs, megaraid sas with 1x
> > SATA HDD (that's all I have), and null blk with lots of devices, and I still
> > can't see high usage of blk_mq_queue_tag_busy_iter().
> It should be triggered easily in case of heavy io accounting:
>
> while true; do cat /proc/diskstats; done

Let me check that.

> > So how about we get this patch processed (to fix blk_mq_tagset_busy_iter()),
> > as it is independent of blk_mq_queue_tag_busy_iter()? And then wait for some
> > update or some more info from Kashyap regarding blk_mq_queue_tag_busy_iter()
> Looks fine:
>
> Reviewed-by: Ming Lei<ming.lei@xxxxxxxxxx>

Thanks, I'll just send a v2 with your tag for clarity, as there has been 
much discussion here.

John