Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > linux.kernel > #1299211 > unrolled thread

[PATCH] mm,oom: Always sleep before retrying.

Back to article view | Back to linux.kernel

Contents

  [PATCH] mm,oom: Always sleep before retrying. Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp> - 2015-12-30 03:10 +0100
    Re: [PATCH] mm,oom: Always sleep before retrying. Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp> - 2016-01-01 09:00 +0100

#1299211 — [PATCH] mm,oom: Always sleep before retrying.

>From c0b5820c594343e06239f15afb35d23b4b8ac0d0 Mon Sep 17 00:00:00 2001
From: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Date: Wed, 30 Dec 2015 10:55:59 +0900
Subject: [PATCH] mm,oom: Always sleep before retrying.

When we entered into "Reclaim has failed us, start killing things"
state, sleep function is called only when mutex_trylock(&oom_lock)
in __alloc_pages_may_oom() failed or immediately after returning from
oom_kill_process() in out_of_memory(). This may be insufficient for
giving other tasks a chance to run because mutex_trylock(&oom_lock)
will not fail under non-preemptive UP kernel.

If it is a !__GFP_FS && !__GFP_NOFAIL allocation request,
__alloc_pages_may_oom() will return without sleeping, and
__alloc_pages_slowpath() will retry without sleeping.
As a result, other tasks will never acquire a chance to run.

If it is a __GFP_FS || __GFP_NOFAIL allocation request, out_of_memory()
will be called. But if the OOM victim failed to terminate before
schedule_timeout_killable(1) returns, the victim will never acquire
a chance to run again because the task which called out_of_memory()
will not sleep again.

We should not rely on mutex_trylock(&oom_lock) for a sleep. This patch
makes sure everybody sleeps before __alloc_pages_slowpath() retries.

Signed-off-by: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
---
 mm/page_alloc.c | 7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)

diff --git a/mm/page_alloc.c b/mm/page_alloc.c
index 2565154..6f7f786 100644
--- a/mm/page_alloc.c
+++ b/mm/page_alloc.c
@@ -2734,7 +2734,6 @@ __alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
 	 */
 	if (!mutex_trylock(&oom_lock)) {
 		*did_some_progress = 1;
-		schedule_timeout_uninterruptible(1);
 		return NULL;
 	}
 
@@ -3282,6 +3281,12 @@ retry:
 	/* Retry as long as the OOM killer is making progress */
 	if (did_some_progress) {
 		no_progress_loops = 0;
+		/*
+		 * Make sure that other tasks (e.g. OOM victims, workqueue
+		 * items) are given a chance to run.
+		 */
+		if (!test_thread_flag(TIF_MEMDIE))
+			schedule_timeout_uninterruptible(1);
 		goto retry;
 	}
 
-- 
1.8.3.1
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

[toc] | [next] | [standalone]

#1299852

Tetsuo Handa wrote:
> When we entered into "Reclaim has failed us, start killing things"
> state, sleep function is called only when mutex_trylock(&oom_lock)
> in __alloc_pages_may_oom() failed or immediately after returning from
> oom_kill_process() in out_of_memory(). This may be insufficient for
> giving other tasks a chance to run because mutex_trylock(&oom_lock)
> will not fail under non-preemptive UP kernel.

My misunderstanding. I thought cond_resched() is a no-op under
non-preemptive UP kernel.

Calling schedule_timeout_uninterruptible(1) will allow other pending
workqueue items a chance to run if current thread is kworker thread.
But if current thread is one of threads which the OOM victim depends
on, calling it merely delays termination of the OOM victim. Therefore,
nobody can judge whether calling it will help the OOM victim and its
dependent threads to make use of CPU cycles for making progress.
Although always sleeping helps saving CPU cycles under OOM livelock,
we need to give up waiting for the OOM victim at some point (i.e.
trigger kernel panic like panic_on_oom_timeout or choose subsequent
OOM victims).
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

[toc] | [prev] | [standalone]

Back to top | Article view | linux.kernel

csiph-web