Re: [PATCH 1/2] blk-mq: introduce blk_mq_tagset_wait_request_completed()

[yi sun]

In my case, I want all hw queues owned by this device to be clean.
Because in the virtio device, each hw queue corresponds to a virtqueue,
and all virtqueues will be deleted when vdev suspends.

The blk_mq_tagset_wait_request_completed() function can ensure that
the device has processed all in_flight requests , and these requests have
become the complete state. I don’t understand the blk_mq_freeze_queue()
function very well. Can the function ensure that all requests have become
complete status? How should I use the function to achieve the same effect?

ps: requests become in_flight status in virtio_queue_rq() and become complete
status in virtblk_done().

Yi

On Wed, Jan 24, 2024 at 3:14 AM Keith Busch <kbusch@xxxxxxxxxx> wrote:
>
> On Mon, Jan 22, 2024 at 07:07:21PM +0800, Yi Sun wrote:
> > In some cases, it is necessary to wait for all requests to become complete
> > status before performing other operations. Otherwise, these requests will never
> > be processed successfully.
> >
> > For example, when the virtio device is in hibernation, the virtqueues
> > will be deleted. It must be ensured that virtqueue is not in use before being deleted.
> > Otherwise the requests in the virtqueue will be lost. This function can ensure
> > that all requests have been taken out of the virtqueues.
> >
> > Prepare for fixing this kind of issue by introducing
> > blk_mq_tagset_wait_request_completed().
>
> Does blk_mq_freeze_queue() not work for your use case? I think that
> should work unless you have some driver specific requests entered that
> don't ever get released.
>
> > +static bool blk_mq_tagset_count_inflight_rqs(struct request *rq, void *data)
> > +{
> > +     unsigned int *count = data;
> > +
> > +     if (blk_mq_request_started(rq) && !blk_mq_request_completed(rq))
> > +             (*count)++;
> > +     return true;
> > +}
> > +
> > +/**
> > + * blk_mq_tagset_wait_request_completed - Wait for all inflight requests
> > + * to become completed.
> > + *
> > + * Note: This function has to be run after all IO queues are shutdown.
> > + */
> > +void blk_mq_tagset_wait_request_completed(struct blk_mq_tag_set *tagset)
> > +{
> > +     while (true) {
> > +             unsigned int count = 0;
> > +
> > +             blk_mq_tagset_busy_iter(tagset,
> > +                             blk_mq_tagset_count_inflight_rqs, &count);
>
> If the tagset is shared, then one active user can prevent this from ever
> completing. It sounds like your use case cares about just one specific
> request_queue, not all of them.
>
> > +             if (!count)
> > +                     break;
> > +             msleep(20);
> > +     }
> > +}
> > +EXPORT_SYMBOL(blk_mq_tagset_wait_request_completed);