Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Date:   Fri, 8 Apr 2022 13:48:08 +0200
From:   Michal Hocko <mhocko@suse.com>
To:     Nico Pache <npache@redhat.com>
Cc:     Thomas Gleixner <tglx@linutronix.de>,
        Peter Zijlstra <peterz@infradead.org>, linux-mm@kvack.org,
        linux-kernel@vger.kernel.org, Rafael Aquini <aquini@redhat.com>,
        Waiman Long <longman@redhat.com>, Baoquan He <bhe@redhat.com>,
        Christoph von Recklinghausen <crecklin@redhat.com>,
        Don Dutile <ddutile@redhat.com>,
        "Herton R . Krzesinski" <herton@redhat.com>,
        David Rientjes <rientjes@google.com>,
        Andrea Arcangeli <aarcange@redhat.com>,
        Andrew Morton <akpm@linux-foundation.org>,
        Davidlohr Bueso <dave@stgolabs.net>,
        Ingo Molnar <mingo@redhat.com>,
        Joel Savitz <jsavitz@redhat.com>,
        Darren Hart <dvhart@infradead.org>, stable@kernel.org
Subject: Re: [PATCH v8] oom_kill.c: futex: Don't OOM reap the VMA containing
 the robust_list_head
Message-ID: <YlAg+Pu7hIw6W+cc@dhcp22.suse.cz>
References: <20220408032809.3696798-1-npache@redhat.com>
 <20220408081549.GM2731@worktop.programming.kicks-ass.net>
 <87tub4j7hg.ffs@tglx>
 <ddf1755e-fe69-b60e-ee07-e78d663b11b2@redhat.com>
 <YlACJMp7AFaVa/Gt@dhcp22.suse.cz>
 <676fb197-d045-c537-c1f7-e18320a6d15f@redhat.com>
 <YlAHfU7VEI0wYAb/@dhcp22.suse.cz>
 <2293c547-3878-435a-ec1c-854c3181ad14@redhat.com>
 <YlATveUBFmWFY+SI@dhcp22.suse.cz>
 <465ab95b-3e71-5901-c184-812dc595af2f@redhat.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <465ab95b-3e71-5901-c184-812dc595af2f@redhat.com>
Precedence: bulk

On Fri 08-04-22 07:26:07, Nico Pache wrote:
[...]
> Ok so if i understand that correctly, delaying can have some ugly effects and
> kinda breaks the initial purpose of the OOM reaper?

No, not really. The primary objective of the oom_reaper is to _guaratee_
a forward progress. It is not really meant to be an optimization to
respond to the oom killer faster. The reason the oom_reaper is kicked
off right away is because that was the simplest implementation.

> I personally don't like the delay approach. Especially if we have a better one
> we know is working, and that doesnt add regressions.

Well, I would say that handling futex case more gracefully would be
preferable but my understanding is that this is not all that easy. I am
far from being a futex expert so I will leave that up to Thomas and Peter.

On the other hand delaying oom_reaper is rather straightforward and I do
not think there is a big risk of regressions. Any QoS during OOM is
simply out of the window and the main purpose of the reaper will be
preserved with a timeout as well. I also do agree with Thomas that this
would cover 99% of cases.

> If someone can prove to me the private lock case, I'd be more willing to bite.
> 
> Thanks for all the OOM context :)

Welcome. The oom handling is a maze and it is really easy to miss all
the subtlety and conflicting requirements that are applied here.
-- 
Michal Hocko
SUSE Labs