On Tue, Jan 09, 2018 at 05:29:27PM -0700, Jens Axboe wrote:
> Move completion related items (like the call single data) near the
> end of the struct, instead of mixing them in with the initial
> queueing related fields.
>
> Move queuelist below the bio structures. Then we have all
> queueing related bits in the first cache line.
>
> This yields a 1.5-2% increase in IOPS for a null_blk test, both for
> sync and for high thread count access. Sync test goes form 975K to
> 992K, 32-thread case from 20.8M to 21.2M IOPS.
One nit below, otherwise
Reviewed-by: Omar Sandoval <osandov@xxxxxx>
> Signed-off-by: Jens Axboe <axboe@xxxxxxxxx>
> ---
> block/blk-mq.c | 19 ++++++++++---------
> include/linux/blkdev.h | 28 +++++++++++++++-------------
> 2 files changed, 25 insertions(+), 22 deletions(-)
>
> diff --git a/block/blk-mq.c b/block/blk-mq.c
> index 7248ee043651..ec128001ea8b 100644
> --- a/block/blk-mq.c
> +++ b/block/blk-mq.c
> @@ -270,8 +270,6 @@ static struct request *blk_mq_rq_ctx_init(struct blk_mq_alloc_data *data,
> struct blk_mq_tags *tags = blk_mq_tags_from_data(data);
> struct request *rq = tags->static_rqs[tag];
>
> - rq->rq_flags = 0;
> -
> if (data->flags & BLK_MQ_REQ_INTERNAL) {
> rq->tag = -1;
> rq->internal_tag = tag;
> @@ -285,26 +283,23 @@ static struct request *blk_mq_rq_ctx_init(struct blk_mq_alloc_data *data,
> data->hctx->tags->rqs[rq->tag] = rq;
> }
>
> - INIT_LIST_HEAD(&rq->queuelist);
> /* csd/requeue_work/fifo_time is initialized before use */
> rq->q = data->q;
> rq->mq_ctx = data->ctx;
> + rq->rq_flags = 0;
> + rq->cpu = -1;
> rq->cmd_flags = op;
> if (data->flags & BLK_MQ_REQ_PREEMPT)
> rq->rq_flags |= RQF_PREEMPT;
> if (blk_queue_io_stat(data->q))
> rq->rq_flags |= RQF_IO_STAT;
> - rq->cpu = -1;
> + /* do not touch atomic flags, it needs atomic ops against the timer */
This comment was just removed in a previous patch but it snuck back in.
