Re: [PATCH 0/2] mm->owner to mm->memcg fixes

From: Michal Hocko <mhocko@kernel.org>
To: Andrew Morton <akpm@linux-foundation.org>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>,
	Johannes Weiner <hannes@cmpxchg.org>,
	Kirill Tkhai <ktkhai@virtuozzo.com>,
	peterz@infradead.org, viro@zeniv.linux.org.uk, mingo@kernel.org,
	paulmck@linux.vnet.ibm.com, keescook@chromium.org,
	riel@redhat.com, tglx@linutronix.de,
	kirill.shutemov@linux.intel.com, marcos.souza.org@gmail.com,
	hoeun.ryu@gmail.com, pasha.tatashin@oracle.com,
	gs051095@gmail.com, dhowells@redhat.com, rppt@linux.vnet.ibm.com,
	linux-kernel@vger.kernel.org,
	Balbir Singh <balbir@linux.vnet.ibm.com>,
	Tejun Heo <tj@kernel.org>, Oleg Nesterov <oleg@redhat.com>
Subject: Re: [PATCH 0/2] mm->owner to mm->memcg fixes
Date: Wed, 30 May 2018 14:17:21 +0200	[thread overview]
Message-ID: <20180530121721.GD27180@dhcp22.suse.cz> (raw)
In-Reply-To: <20180524141635.c99b7025a73a709e179f92a2@linux-foundation.org>

On Thu 24-05-18 14:16:35, Andrew Morton wrote:
> On Thu, 24 May 2018 13:10:02 +0200 Michal Hocko <mhocko@kernel.org> wrote:
> 
> > I would really prefer and appreciate a repost with all the fixes folded
> > in.
> 
> [1/2]

Thanks Andrew and sorry it took so long! This seems to be missing the
fix for the issue I've mentioned in http://lkml.kernel.org/r/20180510121838.GE5325@dhcp22.suse.cz
and Eric wanted to handle by http://lkml.kernel.org/r/87wovu889o.fsf@xmission.com.
I do not think that this fix is really correct one though. I will
comment in a reply to that email.

In any case I really think this should better be reposted in one patch
series and restart the discussion. I strongly suggest revisiting my
previous attempt http://lkml.kernel.org/r/1436358472-29137-8-git-send-email-mhocko@kernel.org
including the follow up discussion regarding the unfortunate CLONE_VM
outside of thread group case and potential solution for that.

> -static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
> +static void mm_init_memcg(struct mm_struct *mm)
>  {
>  #ifdef CONFIG_MEMCG
> -	mm->owner = p;
> +	struct cgroup_subsys_state *css;
> +
> +	/* Ensure mm->memcg is initialized */
> +	mm->memcg = NULL;
> +
> +	rcu_read_lock();
> +	css = task_css(current, memory_cgrp_id);
> +	if (css && css_tryget(css))
> +		mm_update_memcg(mm, mem_cgroup_from_css(css));
> +	rcu_read_unlock();

Is it possible that css_tryget ever fails here? I remember I used to do
plain css_get in my patch.

> @@ -4850,15 +4836,16 @@ static int mem_cgroup_can_attach(struct
>  	if (!move_flags)
>  		return 0;
>  
> -	from = mem_cgroup_from_task(p);
> +	from = mem_cgroup_from_css(task_css(p, memory_cgrp_id));
>  
>  	VM_BUG_ON(from == memcg);
>  
>  	mm = get_task_mm(p);
>  	if (!mm)
>  		return 0;
> -	/* We move charges only when we move a owner of the mm */
> -	if (mm->owner == p) {
> +
> +	/* We move charges except for creative uses of CLONE_VM */
> +	if (mm->memcg == from) {

I must be missing something here but how do you prevent those cases?
AFAICS processes sharing the mm will simply allow to migrate.

>  		VM_BUG_ON(mc.from);
>  		VM_BUG_ON(mc.to);
>  		VM_BUG_ON(mc.precharge);

Other than that the patch makes sense to me.
-- 
Michal Hocko
SUSE Labs