Re: [PATCH v4 1/5] mm/memcg: Move mod_objcg_state() to memcontrol.c

From: Johannes Weiner <hannes@cmpxchg.org>
To: Waiman Long <llong@redhat.com>
Cc: Michal Hocko <mhocko@kernel.org>,
	Vladimir Davydov <vdavydov.dev@gmail.com>,
	Andrew Morton <akpm@linux-foundation.org>,
	Tejun Heo <tj@kernel.org>, Christoph Lameter <cl@linux.com>,
	Pekka Enberg <penberg@kernel.org>,
	David Rientjes <rientjes@google.com>,
	Joonsoo Kim <iamjoonsoo.kim@lge.com>,
	Vlastimil Babka <vbabka@suse.cz>, Roman Gushchin <guro@fb.com>,
	linux-kernel@vger.kernel.org, cgroups@vger.kernel.org,
	linux-mm@kvack.org, Shakeel Butt <shakeelb@google.com>,
	Muchun Song <songmuchun@bytedance.com>,
	Alex Shi <alex.shi@linux.alibaba.com>,
	Chris Down <chris@chrisdown.name>,
	Yafang Shao <laoar.shao@gmail.com>,
	Wei Yang <richard.weiyang@gmail.com>,
	Masayoshi Mizuma <msys.mizuma@gmail.com>,
	Xing Zhengjun <zhengjun.xing@linux.intel.com>,
	Matthew Wilcox <willy@infradead.org>
Subject: Re: [PATCH v4 1/5] mm/memcg: Move mod_objcg_state() to memcontrol.c
Date: Mon, 19 Apr 2021 17:11:37 -0400	[thread overview]
Message-ID: <YH3yCZn9EeSPKKGY@cmpxchg.org> (raw)
In-Reply-To: <d1c36f26-b958-49e0-ae44-1cf6334fa4c5@redhat.com>

On Mon, Apr 19, 2021 at 01:26:29PM -0400, Waiman Long wrote:
> On 4/19/21 1:19 PM, Waiman Long wrote:
> > On 4/19/21 1:13 PM, Johannes Weiner wrote:
> > > The CONFIG_MEMCG_KMEM has become sort of useless now. It used to be
> > > configurable, but now it just means CONFIG_MEMCG && !CONFIG_SLOB. And
> > > even that doesn't make sense because while slob doesn't support slab
> > > object tracking, we can still do all the other stuff we do under
> > > KMEM. I have a patch in the works to remove the symbol and ifdefs.
> > > 
> > > With that in mind, it's better to group the functions based on what
> > > they do rather than based on CONFIG_MEMCG_KMEM. It's easier to just
> > > remove another ifdef later than it is to reorder the functions.
> > > 
> > OK, I will make the move then. Thanks for the explanation.

Thanks!

> BTW, have you ever thought of moving the cgroup-v1 specific functions out
> into a separate memcontrol-v1.c file just like kernel/cgroup/cgroup-v1.c?
> 
> I thought of that before, but memcontrol.c is a frequently changed file and
> so a bit hard to do.

I haven't looked too deeply at it so far, but I think it would make
sense to try.

There are indeed many of the entry paths from the MM code that are
shared between cgroup1 and cgroup2, with smaller branches here and
there to adjust behavior. Those would throw conflicts, but those we
should probably keep in the main memcontrol.c for readability anyway.

But there is also plenty of code that is exclusively about cgroup1,
and which actually doesn't change much in a long time. Moving that
elsewhere shouldn't create difficult conflicts - maybe a few line
offset warnings or fuzz in the diff context of unrelated changes:

- the soft limit tree and soft limit reclaim

- the threshold and oom event notification stuff

- the charge moving code

- remaining v1 interface files, as well as their helper functions

From a quick scan, this adds up to ~2,500 lines of old code with no
actual dependencies from the common code or from v2, and which could
be moved out of the way without disrupting ongoing development much.