Re: [PATCH] mm: skip zombie in OOM-killer

[Andrew Vagin <avagin@gmail.com>]

[-- Attachment #1: Type: text/plain, Size: 2172 bytes --]

2011/3/7 David Rientjes <rientjes@google.com>:
> On Sun, 6 Mar 2011, KOSAKI Motohiro wrote:
>
>> > When we check that task has flag TIF_MEMDIE, we forgot check that
>> > it has mm. A task may be zombie and a parent may wait a memor.
>> >
>> > v2: Check that task doesn't have mm one time and skip it immediately
>> >
>> > Signed-off-by: Andrey Vagin <avagin@openvz.org>
>>
>> This seems incorrect. Do you have a reprodusable testcasae?
>> Your patch only care thread group leader state, but current code
>> care all thread in the process. Please look at oom_badness() and
>> find_lock_task_mm().
>>
>
> That's all irrelevant, the test for TIF_MEMDIE specifically makes the oom
> killer a complete no-op when an eligible task is found to have been oom
> killed to prevent needlessly killing additional tasks.  oom_badness() and
> find_lock_task_mm() have nothing to do with that check to return
> ERR_PTR(-1UL) from select_bad_process().
>
> Andrey is patching the case where an eligible TIF_MEMDIE process is found
> but it has already detached its ->mm.  In combination with the patch
> posted to linux-mm, oom: prevent unnecessary oom kills or kernel panics,
> which makes select_bad_process() iterate over all threads, it is an
> effective solution.

Probably you said about the first version of my patch.
This version is incorrect because of
http://git.kernel.org/?p=linux/kernel/git/torvalds/linux-2.6.git;a=commit;h=dd8e8f405ca386c7ce7cbb996ccd985d283b0e03

but my first patch is correct and it has a simple reproducer(I
attached it). You can execute it and your kernel hangs up, because the
parent doesn't wait children, but the one child (zombie) will have
flag TIF_MEMDIE, oom_killer will kill nobody


The link on the first patch:
http://groups.google.com/group/linux.kernel/browse_thread/thread/b9c6ddf34d1671ab/2941e1877ca4f626?lnk=raot&pli=1
>
> Thanks.
> --
> 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/
>

[-- Attachment #2: memeater_killer.py --]
[-- Type: application/octet-stream, Size: 780 bytes --]

import sys, time, mmap, os
from subprocess import Popen, PIPE
import random

global mem_size

def info(msg):
	pid = os.getpid()
	print >> sys.stderr, "%s: %s" % (pid, msg)
	sys.stderr.flush()



def memory_loop(cmd = "a"):
	"""
	cmd may be:
		c: check memory
		else: touch memory
	"""
	c = 0
	for j in xrange(0, mem_size):
		if cmd == "c":
			if f[j<<12] != chr(j % 255):
				info("Data corruption")
				sys.exit(1)
		else:
			f[j<<12] = chr(j % 255)
for i in xrange(20):
	pid = os.fork()
	time.sleep(1)
	if (pid == 0):
		sys.stdout.write("mmap\n")
		sys.stdout.flush()
		mem_size = 400 * 1024
		f = mmap.mmap(-1, mem_size << 12, mmap.MAP_ANONYMOUS|mmap.MAP_PRIVATE)
		memory_loop()
		time.sleep(100)
		f.close()
		sys.stdout.write("ummap\n")
		sys.stdout.flush()
time.sleep(100)