Re: [PATCH] SH/Dreamcast - add support for GD-Rom device

[Jens Axboe <jens.axboe@oracle.com>]

On Thu, Dec 27 2007, Adrian McMenamin wrote:
> 
> On Thu, 2007-12-27 at 17:18 +0900, Paul Mundt wrote:
> > On Thu, Dec 27, 2007 at 01:26:47AM +0000, Adrian McMenamin wrote:
> 
> > 
> > > +		/* now seek to take the request spinlock
> > > + 		 * before handling ending the request */
> > > +		spin_lock(&gdrom_lock);
> > > +		list_del_init(&req->queuelist);
> > > +		blk_requeue_request(gd.gdrom_rq, req);
> > > +		if (err)
> > > +			end_request(req, 0);
> > > +		else
> > > +			end_request(req, 1);
> > > +	}
> > > +	spin_unlock(&gdrom_lock);
> > > +	kfree(read_command);
> > > +}	
> > > +
> > This locking is all over the place. What is this lock supposed to be
> > accomplishing?
> > -
> 
> I have to hold the lock to access the request queue. As the linked list
> of deferred requests is under the control of code also protected by the
> lock, it is also held to ensure manipulation of that list is serialised.
> 
> The first step of the loop manipulates that linked list - so it is held
> as we re-iterate over the loop.
> 
> This is pretty much the way Jens recommended I do it.

I didn't recommend the last requeue bit, it looks like a work-around due
to the way that end_request() works. The kerneldoc comment for that
function also tells you NOT to use this in new code. Use
end_dequeued_request() and get rid of the requeue, and streamline 'err'
so you can just pass it directly in.

The locking otherwise looks fine to me.

-- 
Jens Axboe