[Devel] [PATCH v14 16/18] vmpressure: in-kernel notifications

[Glauber Costa]

On Fri, Dec 20, 2013 at 8:53 PM, Luiz Capitulino <lcapitulino at redhat.com> wrote:
> On Fri, 20 Dec 2013 20:46:05 +0400
> Glauber Costa <glommer at gmail.com> wrote:
>
>> On Fri, Dec 20, 2013 at 8:44 PM, Luiz Capitulino <lcapitulino at redhat.com> wrote:
>> > On Fri, 20 Dec 2013 10:03:32 -0500
>> > Luiz Capitulino <lcapitulino at redhat.com> wrote:
>> >
>> >> > The answer for all of your questions above can be summarized by noting
>> >> > that for the lack of other users (at the time), this patch does the bare minimum
>> >> > for memcg needs. I agree, for instance, that it would be good to pass the level
>> >> > but since memcg won't do anything with thta, I didn't pass it.
>> >> >
>> >> > That should be extended if you need to.
>> >>
>> >> That works for me. That is, including this minimal version first and
>> >> extending it when we get in-tree users.
>> >
>> > Btw, there's something I was thinking just right now. If/when we
>> > convert shrink functions to use this API, they will come to depend
>> > on CONFIG_MEMCG=y. IOW, they won't work if CONFIG_MEMCG=n.
>> >
>> > Is this acceptable (this is an honest question)? Because today, they
>> > do work when CONFIG_MEMCG=n. Should those shrink functions use the
>> > shrinker API as a fallback?
>>
>> If you have a non-memcg user, that should obviously be available for
>> CONFIG_MEMCG=n
>
> OK, which means we'll have to change it, right? Because, if I'm not
> missing something, today vmpressure does depend on CONFIG_MEMCG=y.

You mean the main vmpressure mechanism?
Sorry, this was out of my mental cachelines. Yes, vmpressure depends
on MEMCG, because
the pressure interface is memcg-specific (global == root memcg)

You might want to change that so you can reuse the mechanism and let
only the user interface
depend on memcg.


-- 
E Mare, Libertas