On Fri, Aug 15, 2025 at 05:34:23PM +0800, Yu Kuai wrote:
> Hi,
>
> 在 2025/08/15 17:15, Yu Kuai 写道:
> > Will it be simpler if we move blk_mq_freeze_queue_nomemsave() into
> > blk_mq_elv_switch_none(), after elevator is succeed switching to none
> > then freeze the queue.
> >
> > Later in blk_mq_elv_switch_back we'll know if xa_load() return valid
> > elevator_type, related queue is already freezed.
>
> Like following:
>
> diff --git a/block/blk-mq.c b/block/blk-mq.c
> index e9f037a25fe3..3640fae5707b 100644
> --- a/block/blk-mq.c
> +++ b/block/blk-mq.c
> @@ -5010,7 +5010,13 @@ static int blk_mq_elv_switch_none(struct
> request_queue *q,
> __elevator_get(q->elevator->type);
>
> elevator_set_none(q);
> + } else {
> + ret = xa_insert(elv_tbl, q->id, xa_mk_value(1), GFP_KERNEL);
> + if (WARN_ON_ONCE(ret))
> + return ret;
> }
> +
> + blk_mq_freeze_queue_nomemsave(q);
> return ret;
> }
>
> @@ -5045,9 +5051,6 @@ static void __blk_mq_update_nr_hw_queues(struct
> blk_mq_tag_set *set,
> blk_mq_sysfs_unregister_hctxs(q);
> }
>
> - list_for_each_entry(q, &set->tag_list, tag_set_list)
> - blk_mq_freeze_queue_nomemsave(q);
> -
> /*
> * Switch IO scheduler to 'none', cleaning up the data associated
> * with the previous scheduler. We will switch back once we are done
> diff --git a/block/elevator.c b/block/elevator.c
> index e2ebfbf107b3..9400ea9ec024 100644
> --- a/block/elevator.c
> +++ b/block/elevator.c
> @@ -715,16 +715,21 @@ void elv_update_nr_hw_queues(struct request_queue *q,
> struct elevator_type *e,
>
> WARN_ON_ONCE(q->mq_freeze_depth == 0);
>
> - if (e && !blk_queue_dying(q) && blk_queue_registered(q)) {
> - ctx.name = e->elevator_name;
> - ctx.et = t;
> -
> - mutex_lock(&q->elevator_lock);
> - /* force to reattach elevator after nr_hw_queue is updated
> */
> - ret = elevator_switch(q, &ctx);
> - mutex_unlock(&q->elevator_lock);
> + if (e) {
> + if (!xa_is_value(e) && !blk_queue_dying(q) &&
> + blk_queue_registered(q)) {
> + ctx.name = e->elevator_name;
> + ctx.et = t;
> +
> + mutex_lock(&q->elevator_lock);
> + /* force to reattach elevator after nr_hw_queue is
> updated */
> + ret = elevator_switch(q, &ctx);
> + mutex_unlock(&q->elevator_lock);
> + }
> +
> + blk_mq_unfreeze_queue_nomemrestore(q);
> }
> - blk_mq_unfreeze_queue_nomemrestore(q);
> +
I feel it doesn't become simpler, :-(
However we still can avoid the change in elv_update_nr_hw_queues() by moving
freeze/unfree queue to blk_mq_elv_switch_back(), which looks more readable.
Thanks,
Ming
