On Thu, Apr 10, 2025 at 05:34:18PM +0200, Christoph Hellwig wrote:
> On Thu, Apr 10, 2025 at 09:30:17PM +0800, Ming Lei wrote:
> > @@ -5071,9 +4984,15 @@ static void __blk_mq_update_nr_hw_queues(struct blk_mq_tag_set *set,
> > blk_mq_debugfs_register_hctxs(q);
> > }
> >
> > -switch_back:
> > - list_for_each_entry(q, &set->tag_list, tag_set_list)
> > - blk_mq_elv_switch_back(&head, q);
> > + list_for_each_entry(q, &set->tag_list, tag_set_list) {
> > + const char *name = "none";
> > +
> > + mutex_lock(&q->elevator_lock);
> > + if (q->elevator && !blk_queue_dying(q))
> > + name = q->elevator->type->elevator_name;
> > + __elevator_change(q, name, true);
> > + mutex_unlock(&q->elevator_lock);
> > + }
>
> Coming back to this after looking through the next patches.
>
> Why do we even need the __elevator_change call here? We've not
> actually disabled the elevator, and we prevent other callers
> from changing it.
>
> As you pass in the force argument this now always calls
> elevator_switch and thus blk_mq_init_sched. But why?
sched tags is built over hctx and depends on ->nr_hw_queues,
when nr_hw_queues is changed, sched tags has to be rebuilt.
Thanks,
Ming
