Re: [PATCH v2] mm, shrinker: make shrinker_list lockless

From: Shakeel Butt <shakeelb@google.com>
To: Tetsuo Handa <penguin-kernel@i-love.sakura.ne.jp>
Cc: Minchan Kim <minchan@kernel.org>,
	Huang Ying <ying.huang@intel.com>,
	Mel Gorman <mgorman@techsingularity.net>,
	Vladimir Davydov <vdavydov.dev@gmail.com>,
	Michal Hocko <mhocko@kernel.org>,
	Greg Thelen <gthelen@google.com>,
	Johannes Weiner <hannes@cmpxchg.org>,
	Andrew Morton <akpm@linux-foundation.org>,
	Linux MM <linux-mm@kvack.org>,
	LKML <linux-kernel@vger.kernel.org>
Subject: Re: [PATCH v2] mm, shrinker: make shrinker_list lockless
Date: Fri, 10 Nov 2017 10:16:14 -0800	[thread overview]
Message-ID: <CALvZod4HAVpmPyE7k_Brme=mzic8SDO8gyAFtDhW9r0oBurN4w@mail.gmail.com> (raw)
In-Reply-To: <201711100646.IJH39597.HOtMLJVSFOQFOF@I-love.SAKURA.ne.jp>

On Thu, Nov 9, 2017 at 1:46 PM, Tetsuo Handa
<penguin-kernel@i-love.sakura.ne.jp> wrote:
> Shakeel Butt wrote:
>> > If you can accept serialized register_shrinker()/unregister_shrinker(),
>> > I think that something like shown below can do it.
>>
>> If we assume that we will never do register_shrinker and
>> unregister_shrinker on the same object in parallel then do we still
>> need to do msleep & synchronize_rcu() within mutex?
>
> Doing register_shrinker() and unregister_shrinker() on the same object
> in parallel is wrong. This mutex is to ensure that we do not need to
> worry about ->list.next field. synchronize_rcu() should not be slow.
> If you want to avoid msleep() with mutex held, you can also apply
>
>> > If you want parallel register_shrinker()/unregister_shrinker(), something like
>> > shown below on top of shown above will do it.
>
> change.

Thanks for the explanation. Can you post the patch for others to
review without parallel register/unregister and SHRINKER_PERMANENT (we
can add when we need them)? You can use the motivation for the patch I
mentioned in my patch instead.