From: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
To: mhocko@kernel.org, linux-mm@kvack.org
Cc: rientjes@google.com, hannes@cmpxchg.org,
akpm@linux-foundation.org, linux-kernel@vger.kernel.org,
mhocko@suse.com, oleg@redhat.com, viro@zeniv.linux.org.uk
Subject: Re: [RFC 3/4] mm, oom: do not rely on TIF_MEMDIE for exit_oom_victim
Date: Sun, 4 Sep 2016 10:50:02 +0900 [thread overview]
Message-ID: <201609041050.BFG65134.OHVFQJOOSLMtFF@I-love.SAKURA.ne.jp> (raw)
In-Reply-To: <1472723464-22866-4-git-send-email-mhocko@kernel.org>
Michal Hocko wrote:
> mark_oom_victim and exit_oom_victim are used for oom_killer_disable
> which should block as long as there any any oom victims alive. Up to now
> we have relied on TIF_MEMDIE task flag to count how many oom victim
> we have. This is not optimal because only one thread receives this flag
> at the time while the whole process (thread group) is killed and should
> die. As a result we do not thaw the whole thread group and so a multi
> threaded process can leave some threads behind in the fridge. We really
> want to thaw all the threads.
>
> This is not all that easy because there is no reliable way to count
> threads in the process as the oom killer might race with copy_process.
What is wrong with racing with copy_process()? Threads doing copy_process()
are not frozen and thus we don't need to thaw such threads. Also, being
OOM-killed implies receiving SIGKILL. Thus, newly created thread will also
enter do_exit().
> So marking all threads with TIF_MEMDIE and increment oom_victims
> accordingly is not safe. Also TIF_MEMDIE flag should just die so
> we should better come up with a different approach.
>
> All we need to guarantee is that exit_oom_victim is called at the time
> when no further access to (possibly suspended) devices or generate other
> IO (which would clobber suspended image and only once per process)
> is possible. It seems we can rely on exit_notify for that because we
> already have to detect the last thread to do a cleanup. Let's propagate
> that information up to do_exit and only call exit_oom_victim for such
> a thread. With this in place we can safely increment oom_victims only
> once per thread group and thaw all the threads from the process.
> freezing_slow_path can also rely on tsk_is_oom_victim as well now.
If marking all threads which belong to tsk thread group with TIF_MEMDIE
is not safe (due to possible race with copy_process()), how can
rcu_read_lock();
for_each_thread(tsk, t)
__thaw_task(t);
rcu_read_unlock();
in mark_oom_victim() guarantee that all threads which belong to tsk
thread group are thawed?
Unless all threads which belong to tsk thread group in __refrigerator()
are guaranteed to be thawed, they might fail to leave __refrigerator()
in order to enter do_exit() which means that exit_oom_victim() won't be
called.
Do we want to thaw OOM victims from the beginning? If the freezer
depends on CONFIG_MMU=y , we don't need to thaw OOM victims.
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>
next prev parent reply other threads:[~2016-09-04 1:50 UTC|newest]
Thread overview: 21+ messages / expand[flat|nested] mbox.gz Atom feed top
2016-09-01 9:51 [RFC 0/4] mm, oom: get rid of TIF_MEMDIE Michal Hocko
2016-09-01 9:51 ` [RFC 1/4] mm, oom: do not rely on TIF_MEMDIE for memory reserves access Michal Hocko
2016-09-04 1:49 ` Tetsuo Handa
2016-09-09 14:00 ` Michal Hocko
2016-09-01 9:51 ` [RFC 2/4] mm: replace TIF_MEMDIE checks by tsk_is_oom_victim Michal Hocko
2016-09-04 1:49 ` Tetsuo Handa
2016-09-09 14:05 ` Michal Hocko
2016-09-01 9:51 ` [RFC 3/4] mm, oom: do not rely on TIF_MEMDIE for exit_oom_victim Michal Hocko
2016-09-04 1:50 ` Tetsuo Handa [this message]
2016-09-09 14:08 ` Michal Hocko
2016-09-10 6:29 ` Tetsuo Handa
2016-09-10 12:55 ` Tetsuo Handa
2016-09-12 9:11 ` Michal Hocko
2016-09-13 6:25 ` Tetsuo Handa
2016-09-13 7:21 ` Michal Hocko
2016-09-14 13:50 ` Michal Hocko
2016-09-01 9:51 ` [RFC 4/4] arch: get rid of TIF_MEMDIE Michal Hocko
2016-09-15 14:41 ` [RFC 0/4] mm, oom: " Johannes Weiner
2016-09-16 7:15 ` Michal Hocko
2016-09-19 16:18 ` Johannes Weiner
2016-09-19 19:02 ` Michal Hocko
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=201609041050.BFG65134.OHVFQJOOSLMtFF@I-love.SAKURA.ne.jp \
--to=penguin-kernel@i-love.sakura.ne.jp \
--cc=akpm@linux-foundation.org \
--cc=hannes@cmpxchg.org \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-mm@kvack.org \
--cc=mhocko@kernel.org \
--cc=mhocko@suse.com \
--cc=oleg@redhat.com \
--cc=rientjes@google.com \
--cc=viro@zeniv.linux.org.uk \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).