Re: [PATCH v5 3/6] mm/memcg: Protect per-CPU counter by disabling preemption on PREEMPT_RT where needed.

[Sebastian Andrzej Siewior <bigeasy@linutronix.de>]

On 2022-02-28 12:23:12 [+0100], Michal Hocko wrote:
> On Mon 28-02-22 12:08:40, Sebastian Andrzej Siewior wrote:
> > On 2022-02-28 09:05:45 [+0100], Michal Hocko wrote:
> > > Acked-by: Michal Hocko <mhocko@suse.com>
> > > 
> > > TBH I am not a fan of the counter special casing for the debugging
> > > enabled warnings but I do not feel strong enough to push you trhough an
> > > additional version round.
> > 
> > do you want to get rid of the warnings completely? Since we had the
> > check in memcg_stats_lock() it kinda felt useful to add something in
> > __memcg_stats_lock() case, too.
> 
> The thing that I dislike is that there is no other way for potential
> users of those counters to know these expectations. This is not
> documented anywhere so it mostly describes the _current_ state of the
> code which might change in the future. We can be more thorough and
> document counters wrt. to the context they can be used in which would
> make this less of a concern of course. But this can be done on top hence
> I do not want to push you for another versions. It is good to have your
> current work merged in the next merge window as long as there are no
> fallouts and for that it would be good to have it in the linux next and
> exposed for some more testing rather than dealing with this deatails.

Okay. Then I would suggest then I document the usage context of these
counters once things settled down. Now there is no validation at all, so
if someone uses the wrong context for a counter and you don't spot it
during the review then there will be no warning at runtime.

Sebastian