Re: [PATCH RFC v2 2/6] blk-throttle: delay initialization until configuration

[Yu Kuai]

Hi,

在 2024/04/13 1:59, Tejun Heo 写道:
> Hello,
>
> On Sat, Apr 06, 2024 at 04:00:55PM +0800, Yu Kuai wrote:
> > @@ -1480,6 +1547,9 @@ void blk_throtl_cancel_bios(struct gendisk *disk)
> >   	struct cgroup_subsys_state *pos_css;
> >   	struct blkcg_gq *blkg;
> >   
> > +	if (!q->td)
> > +		return;
>
> So, this naked test is safe because the interface functions are shut down by
> the time this function is called.
>
> >   static inline bool blk_should_throtl(struct bio *bio)
> >   {
> > -	struct throtl_grp *tg = blkg_to_tg(bio->bi_blkg);
> > +	struct throtl_grp *tg;
> >   	int rw = bio_data_dir(bio);
> >   
> > +	if (!bio->bi_bdev->bd_queue->td)
> > +		return false;
>
> and this one because ->td is set while the queue is frozen and this path
> shouldn't be running while it gets set, right?

Yes, this is called under bio_queue_enter()
> Can you please add comments explaining why those are safe? Otherwise, the
> patch looks generally sane to me on the first glance. Can you please also
> add how you tested the change?

And I realized that there are no tests for bkl-throttle from blktests,
and I'm using some other tests from our testers to cover basic
functionality. Perhaps will it make sense to add some tests to blktests?

Thanks,
Kuai

> Thanks.