Re: [PATCH 12/12] blkcg: kill the mind-bending blkg->dev

[Vivek Goyal <vgoyal@redhat.com>]

On Thu, Jan 19, 2012 at 08:30:56AM -0800, Tejun Heo wrote:
> Hello,
> 
> On Thu, Jan 19, 2012 at 10:49:19AM -0500, Vivek Goyal wrote:
> > > Let stat and conf read functions get device name from
> > > blkg->q->backing_dev_info.dev and pass it down to printing functions
> > > and remove blkg->dev.
> > 
> > I think one of the reasons for not using blkg->q->backing_dev_info.dev was
> > that there were still drivers where multiple gendisks were sharing the
> > request queue.
> 
> Hmmm... is that relevant?  The configurations / stats are per
> cgroup-request_queue pair.  How does multiple genhd's sharing a queue
> make any difference?  Also, the fill_dev_details functions were using
> the same q->backing_dev_info to fill it in, so it's not like it was
> using anything different.

You are right. Looks like it should not make any difference. Also looks
like bdi->dev points to first disk on the queue and ignores later ones.

int bdi_register(struct backing_dev_info *bdi, struct device *parent,
                const char *fmt, ...)
{       
        if (bdi->dev)   /* The driver needs to use separate queues per
device */
                return 0;
}

So if a queue is being shared by multiple disks, then any throttling rule
on the device will practically trated a rule on the queue and will be
effective on all the devices cumulatively. So sounds fine.

Thanks
Vivek