[PATCH v2 1/2] blk-mq: add tagset quiesce interface

[Sagi Grimberg]

>> The only question in my mind is regarding patch 2/2 with the subtle
>> ordering of nvme_set_queue_dying...
> 
> I think we need to sort that out as a pre-patch.  There is quite some
> history there was a fair amount of dead lock potential earlier, but
> I think I sorted quite a lot out there, including a new lock for setting
> the gendisk size, which is what the comment refers to.
> 
> Something like this:
> 
> diff --git a/drivers/nvme/host/core.c b/drivers/nvme/host/core.c
> index 059737c1a2c19..cb7623b5d2c8b 100644
> --- a/drivers/nvme/host/core.c
> +++ b/drivers/nvme/host/core.c
> @@ -5105,21 +5105,21 @@ static void nvme_stop_ns_queue(struct nvme_ns *ns)
>   
>   /*
>    * Prepare a queue for teardown.
> - *
> - * This must forcibly unquiesce queues to avoid blocking dispatch, and only set
> - * the capacity to 0 after that to avoid blocking dispatchers that may be
> - * holding bd_butex.  This will end buffered writers dirtying pages that can't
> - * be synced.
>    */
>   static void nvme_set_queue_dying(struct nvme_ns *ns)
>   {
>   	if (test_and_set_bit(NVME_NS_DEAD, &ns->flags))
>   		return;
>   
> +	/*
> +	 * Mark the disk dead to prevent new opens, and set the capacity to 0
> +	 * to end buffered writers dirtying pages that can't be synced.
> +	 */
>   	blk_mark_disk_dead(ns->disk);
> -	nvme_start_ns_queue(ns);
> -
>   	set_capacity_and_notify(ns->disk, 0);
> +
> +	/* forcibly unquiesce queues to avoid blocking dispatch */
> +	nvme_start_ns_queue(ns);
>   }

If we no longer have this ordering requirement, then I don't see why
the unquiesce cannot move before or after nvme_set_queue_dying and apply
a tagset-wide quiesce/unquiesce...