On 2019-08-28 06:40, Damien Le Moal wrote:
> Commit 7211aef86f79 ("block: mq-deadline: Fix write completion
> handling") added a call to blk_mq_sched_mark_restart_hctx() in
> dd_dispatch_request() to make sure that write request dispatching does
> not stall when all target zones are locked. This fix left a subtle race
> when a write completion happens during a dispatch execution on another
> CPU:
>
> CPU 0: Dispatch CPU1: write completion
>
> dd_dispatch_request()
> lock(&dd->lock);
> ...
> lock(&dd->zone_lock); dd_finish_request()
> rq = find request lock(&dd->zone_lock);
> unlock(&dd->zone_lock);
> zone write unlock
> unlock(&dd->zone_lock);
> ...
> __blk_mq_free_request
> check restart flag (not set)
> -> queue not run
> ...
> if (!rq && have writes)
> blk_mq_sched_mark_restart_hctx()
> unlock(&dd->lock)
>
> Since the dispatch context finishes after the write request completion
> handling, marking the queue as needing a restart is not seen from
> __blk_mq_free_request() and blk_mq_sched_restart() not executed leading
> to the dispatch stall under 100% write workloads.
>
> Fix this by moving the call to blk_mq_sched_mark_restart_hctx() from
> dd_dispatch_request() into dd_finish_request() under the zone lock to
> ensure full mutual exclusion between write request dispatch selection
> and zone unlock on write request completion.
>
> Fixes: 7211aef86f79 ("block: mq-deadline: Fix write completion handling")
> Cc: stable@xxxxxxxxxxxxxxx
> Reported-by: Hans Holmberg <Hans.Holmberg@xxxxxxx>
> Signed-off-by: Damien Le Moal <damien.lemoal@xxxxxxx>
> ---
> block/mq-deadline.c | 19 +++++++++----------
> 1 file changed, 9 insertions(+), 10 deletions(-)
>
> diff --git a/block/mq-deadline.c b/block/mq-deadline.c
> index a17466f310f4..b490f47fd553 100644
> --- a/block/mq-deadline.c
> +++ b/block/mq-deadline.c
> @@ -377,13 +377,6 @@ static struct request *__dd_dispatch_request(struct deadline_data *dd)
> * hardware queue, but we may return a request that is for a
> * different hardware queue. This is because mq-deadline has shared
> * state for all hardware queues, in terms of sorting, FIFOs, etc.
> - *
> - * For a zoned block device, __dd_dispatch_request() may return NULL
> - * if all the queued write requests are directed at zones that are already
> - * locked due to on-going write requests. In this case, make sure to mark
> - * the queue as needing a restart to ensure that the queue is run again
> - * and the pending writes dispatched once the target zones for the ongoing
> - * write requests are unlocked in dd_finish_request().
> */
> static struct request *dd_dispatch_request(struct blk_mq_hw_ctx *hctx)
> {
> @@ -392,9 +385,6 @@ static struct request *dd_dispatch_request(struct blk_mq_hw_ctx *hctx)
>
> spin_lock(&dd->lock);
> rq = __dd_dispatch_request(dd);
> - if (!rq && blk_queue_is_zoned(hctx->queue) &&
> - !list_empty(&dd->fifo_list[WRITE]))
> - blk_mq_sched_mark_restart_hctx(hctx);
> spin_unlock(&dd->lock);
>
> return rq;
> @@ -561,6 +551,13 @@ static void dd_prepare_request(struct request *rq, struct bio *bio)
> * spinlock so that the zone is never unlocked while deadline_fifo_request()
> * or deadline_next_request() are executing. This function is called for
> * all requests, whether or not these requests complete successfully.
> + *
> + * For a zoned block device, __dd_dispatch_request() may have stopped
> + * dispatching requests if all the queued requests are write requests directed
> + * at zones that are already locked due to on-going write requests. To ensure
> + * write request dispatch progress in this case, mark the queue as needing a
> + * restart to ensure that the queue is run again after completion of the
> + * request and zones being unlocked.
> */
> static void dd_finish_request(struct request *rq)
> {
> @@ -572,6 +569,8 @@ static void dd_finish_request(struct request *rq)
>
> spin_lock_irqsave(&dd->zone_lock, flags);
> blk_req_zone_write_unlock(rq);
> + if (!list_empty(&dd->fifo_list[WRITE]))
> + blk_mq_sched_mark_restart_hctx(rq->mq_hctx);
> spin_unlock_irqrestore(&dd->zone_lock, flags);
> }
> }
>
Looks good to me.
Reviewed-by: Hans Holmberg <hans.holmberg@xxxxxxx>
