Re: [PATCHv2] scsi: use 'scsi_device_from_queue()' for scsi_dh

[Bart Van Assche <Bart.VanAssche@sandisk.com>]

On Thu, 2017-02-16 at 16:12 +0100, Hannes Reinecke wrote:
> +struct scsi_device *scsi_device_from_queue(struct request_queue *q)
> +{
> +	struct scsi_device *sdev = NULL;
> +	unsigned long flags;
> +
> +	spin_lock_irqsave(q->queue_lock, flags);
> +	if (q->mq_ops) {
> +		if (q->mq_ops == &scsi_mq_ops)
> +			sdev = q->queuedata;
> +	} else if (q->request_fn == scsi_request_fn)
> +		sdev = q->queuedata;
> +	if (!sdev || !get_device(&sdev->sdev_gendev))
> +		sdev = NULL;
> +	spin_unlock_irqrestore(q->queue_lock, flags);
> +
> +	return sdev;
> +}

Hello Hannes,

Do we need to take the queue lock? Neither q->mq_ops nor q->request_fn are
modified after a block device has been created. q->queuedata is not modified
by any SCSI driver after it has been set. And since the caller of
scsi_device_from_queue() has to guarantee that the queue does not disappear
while this function is in progress, the queue lock does not have to be held
around the get_device() call either. Otherwise this patch looks fine to me.

Bart.