Re: [PATCH V4 11/12] block: add poll_capable method to support bio-based IO polling

[Ming Lei <ming.lei@redhat.com>]

On Tue, Mar 30, 2021 at 02:50:51PM +0800, JeffleXu wrote:
> 
> 
> On 3/30/21 2:26 PM, Hannes Reinecke wrote:
> > On 3/29/21 5:26 PM, Ming Lei wrote:
> >> From: Jeffle Xu <jefflexu@linux.alibaba.com>
> >>
> >> This method can be used to check if bio-based device supports IO polling
> >> or not. For mq devices, checking for hw queue in polling mode is
> >> adequate, while the sanity check shall be implementation specific for
> >> bio-based devices. For example, dm device needs to check if all
> >> underlying devices are capable of IO polling.
> >>
> >> Though bio-based device may have done the sanity check during the
> >> device initialization phase, cacheing the result of this sanity check
> >> (such as by cacheing in the queue_flags) may not work. Because for dm
> >> devices, users could change the state of the underlying devices through
> >> '/sys/block/<dev>/io_poll', bypassing the dm device above. In this case,
> >> the cached result of the very beginning sanity check could be
> >> out-of-date. Thus the sanity check needs to be done every time 'io_poll'
> >> is to be modified.
> >>
> >> Signed-off-by: Jeffle Xu <jefflexu@linux.alibaba.com>
> >> ---
> >>   block/blk-sysfs.c      | 14 +++++++++++---
> >>   include/linux/blkdev.h |  1 +
> >>   2 files changed, 12 insertions(+), 3 deletions(-)
> >>
> >> diff --git a/block/blk-sysfs.c b/block/blk-sysfs.c
> >> index db3268d41274..c8e7e4af66cb 100644
> >> --- a/block/blk-sysfs.c
> >> +++ b/block/blk-sysfs.c
> >> @@ -426,9 +426,17 @@ static ssize_t queue_poll_store(struct
> >> request_queue *q, const char *page,
> >>       unsigned long poll_on;
> >>       ssize_t ret;
> >>   -    if (!q->tag_set || q->tag_set->nr_maps <= HCTX_TYPE_POLL ||
> >> -        !q->tag_set->map[HCTX_TYPE_POLL].nr_queues)
> >> -        return -EINVAL;
> >> +    if (queue_is_mq(q)) {
> >> +        if (!q->tag_set || q->tag_set->nr_maps <= HCTX_TYPE_POLL ||
> >> +            !q->tag_set->map[HCTX_TYPE_POLL].nr_queues)
> >> +            return -EINVAL;
> >> +    } else {
> >> +        struct gendisk *disk = queue_to_disk(q);
> >> +
> >> +        if (!disk->fops->poll_capable ||
> >> +            !disk->fops->poll_capable(disk))
> >> +            return -EINVAL;
> >> +    }
> >>         ret = queue_var_store(&poll_on, page, count);
> >>       if (ret < 0)
> >> diff --git a/include/linux/blkdev.h b/include/linux/blkdev.h
> >> index bfab74b45f15..a46f975f2a2f 100644
> >> --- a/include/linux/blkdev.h
> >> +++ b/include/linux/blkdev.h
> >> @@ -1881,6 +1881,7 @@ struct block_device_operations {
> >>       int (*report_zones)(struct gendisk *, sector_t sector,
> >>               unsigned int nr_zones, report_zones_cb cb, void *data);
> >>       char *(*devnode)(struct gendisk *disk, umode_t *mode);
> >> +    bool (*poll_capable)(struct gendisk *disk);
> >>       struct module *owner;
> >>       const struct pr_ops *pr_ops;
> >>   };
> >>
> > I really wonder how this would work for nvme multipath; but I guess it
> > doesn't change the current situation.

It should work for nvme multipath since the approach covers this case,
and bio submitted to underlying NVMe is marked with REQ_HIPRI and
REQ_POLL_CTX too.

> 
> I wonder, at least, md/dm, which is built upon other devices, or
> 'virtual device' in other words, should be distinguished from other
> 'original' bio-based device (e.g., nvme multipath) then. Maybe one extra
> flag or something.

There is REQ_NVME_MPATH, but not sure we need to deal with that.


Thanks,
Ming