On 5/7/25 5:34 PM, Ming Lei wrote:
> elevator_set_none() is called when deleting disk, in which queue has been
> un-registered, and elevator switch can't happen any more.
>
> So if q->elevator is NULL, it is not necessary to freeze queue and drain
> IO any more.
>
> Signed-off-by: Ming Lei <ming.lei@xxxxxxxxxx>
> ---
> block/elevator.c | 10 ++++++++++
> 1 file changed, 10 insertions(+)
>
> diff --git a/block/elevator.c b/block/elevator.c
> index e1386b84a415..35f4de749dcd 100644
> --- a/block/elevator.c
> +++ b/block/elevator.c
> @@ -754,8 +754,18 @@ void elevator_set_none(struct request_queue *q)
> .name = "none",
> .quiesce_queue = true,
> };
> + bool need_change;
> int err;
>
> + WARN_ON_ONCE(blk_queue_registered(q));
> +
> + /* queue has been unregisted, elevator can't be switched anymore */
> + mutex_lock(&q->elevator_lock);
> + need_change = !!q->elevator;
> + mutex_unlock(&q->elevator_lock);
> + if (!need_change)
> + return;
> +
> err = elevator_change(q, &ctx);
> if (err < 0)
> pr_warn("%s: set none elevator failed %d\n", __func__, err);
What in case blk_unregister_queue/elevator_set_none is called from
__add_disk ? The __add_disk and elv_iosched_store may run concurrently.
So I think we should remove sysfs entries (i.e. delete disk->queue_kobj)
before calling elevator_set_none() from blk_unregister_queue().
Thanks,
--Nilay
