On 7/27/23 04:34, Bart Van Assche wrote:
> Measurements have shown that limiting the queue depth to one for zoned
to one -> to one per zone
> writes has a significant negative performance impact on zoned UFS devices.
> Hence this patch that disables zone locking by the mq-deadline scheduler
> if zoned writes are submitted in order and if the storage controller
The "submitted in order" is actually never checked. So let's not use this in
this explanation.
> preserves the command order. This patch is based on the following
> assumptions:
> - It happens infrequently that zoned write requests are reordered by the
> block layer.
> - The I/O priority of all write requests is the same per zone.
> - Either no I/O scheduler is used or an I/O scheduler is used that
> submits write requests per zone in LBA order.
>
> Cc: Christoph Hellwig <hch@xxxxxx>
> Cc: Damien Le Moal <dlemoal@xxxxxxxxxx>
> Cc: Ming Lei <ming.lei@xxxxxxxxxx>
> Signed-off-by: Bart Van Assche <bvanassche@xxxxxxx>
> ---
> block/mq-deadline.c | 25 +++++++++++++++++++------
> 1 file changed, 19 insertions(+), 6 deletions(-)
>
> diff --git a/block/mq-deadline.c b/block/mq-deadline.c
> index 02a916ba62ee..9a64577fe942 100644
> --- a/block/mq-deadline.c
> +++ b/block/mq-deadline.c
> @@ -338,6 +338,17 @@ static struct request *deadline_skip_seq_writes(struct deadline_data *dd,
> return rq;
> }
>
> +/*
> + * Use write locking if either QUEUE_FLAG_NO_ZONE_WRITE_LOCK or
> + * REQ_NO_ZONE_WRITE_LOCK has not been set. Not using zone write locking is
> + * only safe if the submitter allocates and submit requests in LBA order per
> + * zone and if the block driver preserves the request order.
> + */
> +static bool dd_use_write_locking(struct request *rq)
Nit: maybe rename this to dd_use_zone_write_locking() to be clear this is for
zoned devices only ?
> +{
> + return blk_queue_is_zoned(rq->q) && !blk_no_zone_write_lock(rq);
> +}
> +
> /*
> * For the specified data direction, return the next request to
> * dispatch using arrival ordered lists.
> @@ -353,7 +364,7 @@ deadline_fifo_request(struct deadline_data *dd, struct dd_per_prio *per_prio,
> return NULL;
>
> rq = rq_entry_fifo(per_prio->fifo_list[data_dir].next);
> - if (data_dir == DD_READ || !blk_queue_is_zoned(rq->q))
> + if (data_dir == DD_READ || !dd_use_write_locking(rq))
> return rq;
>
> /*
> @@ -398,7 +409,7 @@ deadline_next_request(struct deadline_data *dd, struct dd_per_prio *per_prio,
> if (!rq)
> return NULL;
>
> - if (data_dir == DD_READ || !blk_queue_is_zoned(rq->q))
> + if (data_dir == DD_READ || !dd_use_write_locking(rq))
> return rq;
>
> /*
> @@ -526,8 +537,9 @@ static struct request *__dd_dispatch_request(struct deadline_data *dd,
> }
>
> /*
> - * For a zoned block device, if we only have writes queued and none of
> - * them can be dispatched, rq will be NULL.
> + * For a zoned block device that requires write serialization, if we
> + * only have writes queued and none of them can be dispatched, rq will
> + * be NULL.
> */
> if (!rq)
> return NULL;
> @@ -552,7 +564,8 @@ static struct request *__dd_dispatch_request(struct deadline_data *dd,
> /*
> * If the request needs its target zone locked, do it.
> */
> - blk_req_zone_write_lock(rq);
> + if (dd_use_write_locking(rq))
> + blk_req_zone_write_lock(rq);
> rq->rq_flags |= RQF_STARTED;
> return rq;
> }
> @@ -933,7 +946,7 @@ static void dd_finish_request(struct request *rq)
>
> atomic_inc(&per_prio->stats.completed);
>
> - if (blk_queue_is_zoned(q)) {
> + if (dd_use_write_locking(rq)) {
> unsigned long flags;
>
> spin_lock_irqsave(&dd->zone_lock, flags);
--
Damien Le Moal
Western Digital Research
