From: Thomas Gleixner <tglx@linutronix.de>
To: Andrea Arcangeli <andrea@suse.de>
Cc: Andrew Morton <akpm@osdl.org>,
marcelo.tosatti@cyclades.com, LKML <linux-kernel@vger.kernel.org>,
Nick Piggin <nickpiggin@yahoo.com.au>
Subject: Re: [PATCH] oom killer (Core)
Date: Thu, 02 Dec 2004 12:09:19 +0100 [thread overview]
Message-ID: <1101985759.13353.102.camel@tglx.tec.linutronix.de> (raw)
In-Reply-To: <20041202033619.GA32635@dualathlon.random>
On Thu, 2004-12-02 at 04:36 +0100, Andrea Arcangeli wrote:
> Could you please test this? Perhaps it won't be as effective as the
> hacks I nuked, but it fixed basic things for me in a quick test so I
> like it more. It's important to use pages_high to avoid livelocks.
>
Sorry man. The behaviour is worse than before, but I expected that :)
It kills sshd, the application, the controlling terminal and some other
innocent processes like hotplug.
Take a decent UP machine enable PREEMPT and start # hackbench NN , where
NN is a number of processes, which exhausts the memory of the machine
and see yourself.
If there is the point where memory is exhausted you have to kill a
process _and_ wait until the memory is freed before even thinking about
another invocation of oom_kill().
There is usually more than one process on a machine which requests
memory. So as long as the first kill did not proceed to the point where
the memory is actually freed, you will call out_of_memory() from
different processes or kernel functions and make the machine less usable
than neccecary.
The oom_kill is invoked about 100 times. It ends up killing hotplug when
already 112MB memory are available. See below.
I agree, that the timed hack stuff in out_of_memory is ugly, but we need
some mechanism to avoid
1. reentrancy into oom_kill
2. invocation of oom_kill until the already killed process
has finally released memory.
Further we need a better selection of whom to kill.
tglx
<SNIP>
Free pages: 112360kB (0kB HighMem)
<SNIP>
[<c0132151>] out_of_memory+0x21/0x30
[<c0132f96>] __alloc_pages+0x2f6/0x380
[<c013303f>] __get_free_pages+0x1f/0x40
[<c013665f>] kmem_getpages+0x1f/0xd0
[<c0137339>] cache_grow+0xb9/0x180
[<c0137573>] cache_alloc_refill+0x173/0x220
[<c01378d4>] __kmalloc+0x74/0x80
[<c02534d7>] alloc_skb+0x47/0xf0
[<c0252717>] sock_alloc_send_pskb+0xc7/0x1d0
[<c025284d>] sock_alloc_send_skb+0x2d/0x40
[<c02ad466>] unix_stream_sendmsg+0x196/0x400
[<c012f987>] filemap_nopage+0x207/0x3a0
[<c024fa66>] sock_aio_write+0xf6/0x120
[<c014c967>] do_sync_write+0xb7/0xf0
[<c016048b>] do_pollfd+0x5b/0xa0
[<c0126f90>] autoremove_wake_function+0x0/0x60
[<c016053a>] do_poll+0x6a/0xd0
[<c014ca9f>] vfs_write+0xff/0x130
[<c014cba1>] sys_write+0x51/0x80
[<c01027cb>] syscall_call+0x7/0xb
Out of Memory: Killed process 3037 (hotplug).
next prev parent reply other threads:[~2004-12-02 11:09 UTC|newest]
Thread overview: 66+ messages / expand[flat|nested] mbox.gz Atom feed top
2004-12-01 9:49 tglx
2004-12-01 21:16 ` Andrea Arcangeli
2004-12-01 22:06 ` Thomas Gleixner
2004-12-01 22:33 ` Andrea Arcangeli
2004-12-02 3:36 ` Andrea Arcangeli
2004-12-02 11:09 ` Thomas Gleixner [this message]
2004-12-02 13:48 ` Thomas Gleixner
2004-12-02 16:47 ` Andrea Arcangeli
2004-12-02 16:55 ` Andrew Morton
2004-12-02 11:18 ` Marcelo Tosatti
2004-12-02 17:17 ` Thomas Gleixner
2004-12-02 17:27 ` Andrew Morton
2004-12-02 18:08 ` Andrea Arcangeli
2004-12-02 18:29 ` Andrew Morton
2004-12-02 19:01 ` Thomas Gleixner
2004-12-02 18:55 ` Thomas Gleixner
2004-12-02 19:07 ` Andrew Morton
2004-12-02 19:08 ` Thomas Gleixner
2004-12-02 19:22 ` Andrew Morton
2004-12-02 19:24 ` Thomas Gleixner
2004-12-02 20:11 ` Andre Tomt
2004-12-03 22:45 ` Thomas Gleixner
2004-12-02 23:47 ` Andrea Arcangeli
2004-12-03 14:41 ` Helge Hafting
2004-12-03 21:20 ` Thomas Gleixner
2004-12-05 21:14 ` Helge Hafting
2004-12-02 23:35 ` Andrea Arcangeli
2004-12-03 2:28 ` Andrea Arcangeli
2004-12-03 22:37 ` Thomas Gleixner
2004-12-03 22:51 ` Thomas Gleixner
2004-12-03 23:08 ` Andrea Arcangeli
2004-12-10 16:36 ` William Lee Irwin III
2004-12-10 17:35 ` Andrea Arcangeli
2004-12-10 17:43 ` William Lee Irwin III
2004-12-10 17:55 ` Andrea Arcangeli
2004-12-10 18:00 ` William Lee Irwin III
2004-12-10 18:15 ` Andrea Arcangeli
2004-12-10 18:19 ` William Lee Irwin III
2004-12-10 19:05 ` Andrea Arcangeli
2004-12-10 16:51 ` William Lee Irwin III
2004-12-03 21:10 ` Thomas Gleixner
2004-12-03 22:21 ` Andrea Arcangeli
2004-12-05 2:52 ` William Lee Irwin III
2004-12-05 13:38 ` Thomas Gleixner
2004-12-05 15:22 ` Andrea Arcangeli
2004-12-10 16:32 ` William Lee Irwin III
2004-12-10 16:52 ` Thomas Gleixner
2004-12-10 17:43 ` William Lee Irwin III
2004-12-10 17:47 ` William Lee Irwin III
2004-12-10 17:49 ` Andrea Arcangeli
2004-12-10 17:57 ` William Lee Irwin III
2004-12-12 0:12 ` William Lee Irwin III
2004-12-24 1:18 ` Andrea Arcangeli
2004-12-01 10:21 tvrtko.ursulin
2004-12-04 7:00 Voluspa
2004-12-04 8:08 ` Andrea Arcangeli
2004-12-04 12:42 Voluspa
2004-12-04 16:43 ` Andrea Arcangeli
2004-12-04 18:33 ` Thomas Gleixner
2004-12-04 21:02 ` Thomas Gleixner
2004-12-05 0:27 ` Andrea Arcangeli
2004-12-05 14:55 ` Thomas Gleixner
2004-12-05 15:34 ` Andrea Arcangeli
2004-12-05 16:29 ` Thomas Gleixner
2004-12-05 2:22 Voluspa
2004-12-05 8:32 Voluspa
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=1101985759.13353.102.camel@tglx.tec.linutronix.de \
--to=tglx@linutronix.de \
--cc=akpm@osdl.org \
--cc=andrea@suse.de \
--cc=linux-kernel@vger.kernel.org \
--cc=marcelo.tosatti@cyclades.com \
--cc=nickpiggin@yahoo.com.au \
/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
all inboxes | Powered by JetHome®