Re: [PATCH] sg block layer tcqing fix

[James Bottomley <James.Bottomley@steeleye.com>]

On Tue, 2004-01-06 at 09:22, Jens Axboe wrote:
> On Tue, Jan 06 2004, James Bottomley wrote:
> > +	/* unlikely because the tag was usually ended earlier
> > +	 *
> > +	 * NOTE: the lock should be held while checking the
> > +	 * flag.  However, any race here requiring the lock
> > +	 * would be a gross error. */
> 
> I don't agree with this comment - why would you need to have the queue
> lock to check req->flags?

Because the usual scsi method of doing this is:

	spin_lock_irqsave(q->queue_lock, flags);
	if (blk_rq_tagged(req))
		blk_queue_end_tag(q, req);
	spin_unlock_irqrestore(q->queue_lock, flags);

which was done because the REQ_QUEUE flag is altered by the
blk_queue_end_tag(), so I wanted to note that I'd done this one
differently deliberately.

On the other hand, I don't see the possibility of a legal race to do
this in any of our other block tag ending code either...perhaps I should
just convert them all to this form.

James