Re: [PATCH v5 4/7] scsi: don't wait for quiesce in scsi_stop_queue()

[Bart Van Assche <bvanassche@acm.org>]

On 6/12/23 09:50, mwilck@suse.com wrote:
> @@ -2800,9 +2792,17 @@ static void scsi_device_block(struct scsi_device *sdev, void *data)
>   
>   	mutex_lock(&sdev->state_mutex);
>   	err = __scsi_internal_device_block_nowait(sdev);
> -	if (err == 0)
> -		scsi_stop_queue(sdev, false);
> -	mutex_unlock(&sdev->state_mutex);
> +	if (err == 0) {
> +		/*
> +		 * scsi_stop_queue() must be called with the state_mutex
> +		 * held. Otherwise a simultaneous scsi_start_queue() call
> +		 * might unquiesce the queue before we quiesce it.
> +		 */
> +		scsi_stop_queue(sdev);
> +		mutex_unlock(&sdev->state_mutex);
> +		blk_mq_wait_quiesce_done(sdev->request_queue->tag_set);
> +	} else
> +		mutex_unlock(&sdev->state_mutex);
>   
>   	WARN_ONCE(err, "__scsi_internal_device_block_nowait(%s) failed: err = %d\n",
>   		  dev_name(&sdev->sdev_gendev), err);

Has it been considered to modify the above code such that there is a
single mutex_unlock() call instead of two? I wouldn't mind if
blk_mq_wait_quiesce_done() would be called if err != 0 since performance
is not that important if this function fails.

Thanks,

Bart.