Re: [PATCH 5/5] block: don't verify IO lock for freeze/unfreeze in elevator_init_mq()

[Christoph Hellwig <hch@lst.de>]

On Wed, Oct 30, 2024 at 08:42:37PM +0800, Ming Lei wrote:
> --- a/block/elevator.c
> +++ b/block/elevator.c
> @@ -598,13 +598,17 @@ void elevator_init_mq(struct request_queue *q)
>  	 * drain any dispatch activities originated from passthrough
>  	 * requests, then no need to quiesce queue which may add long boot
>  	 * latency, especially when lots of disks are involved.
> +	 *
> +	 * Disk isn't added yet, so verifying queue lock only manually.
>  	 */
> -	blk_mq_freeze_queue(q);
> +	blk_mq_freeze_queue_non_owner(q);
> +	blk_freeze_acquire_lock(q, true, false);
>  	blk_mq_cancel_work_sync(q);
>  
>  	err = blk_mq_init_sched(q, e);
>  
> -	blk_mq_unfreeze_queue(q);
> +	blk_unfreeze_release_lock(q, true, false);
> +	blk_mq_unfreeze_queue_non_owner(q);

Why do we need to free at all from the add_disk case?  The passthrough
command should never hit the elevator, or am I missing something?