[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <BANLkTi=fk3DUT9cYd2gAzC98c69F6HXX7g@mail.gmail.com>
Date: Thu, 12 May 2011 07:54:26 -0400
From: Andrew Lutomirski <luto@....edu>
To: Andi Kleen <andi@...stfloor.org>
Cc: linux-kernel@...r.kernel.org
Subject: Re: Kernel falls apart under light memory pressure (i.e. linking vmlinux)
On Thu, May 12, 2011 at 1:46 AM, Andi Kleen <andi@...stfloor.org> wrote:
>> Here's a nice picture of alt-sysrq-m with lots of memory free but the
>> system mostly hung. I can still switch VTs.
>
> Would rather need backtraces. Try setting up netconsole or crashdump
> first.
Here are some logs for two different failure mores.
incorrect_oom_kill.txt is an OOM kill when there was lots of available
swap to use. AFAICT the kernel should not have OOM killed at all.
stuck_xyz is when the system is wedged with plenty (~300MB) free
memory but no swap. The sysrq files are self-explanatory.
stuck-sysrq-f.txt is after the others so that it won't have corrupted
the output. After taking all that data, I waited awhile and started
getting soft lockup messges.
I'm having trouble reproducing the "stuck" failure mode on my
lockdep-enabled kernel right now (the OOM kill is easy), so no lock
state trace. But I got one yesterday and IIRC it showed a few tty
locks and either kworker or kcryptd holding (kqueue) and
((&io->work)).
I compressed the larger files.
--Andy
>
> -Andi
>
> --
> ak@...ux.intel.com -- Speaking for myself only.
>
View attachment "stuck-sysrq-m.txt" of type "text/plain" (3349 bytes)
Download attachment "incorrect_oom_kill.txt.xz" of type "application/x-xz" (25112 bytes)
Download attachment "stuck-sysrq-t.txt.xz" of type "application/x-xz" (31008 bytes)
View attachment "stuck-sysrq-w.txt" of type "text/plain" (28818 bytes)
View attachment "stuck-sysrq-f.txt" of type "text/plain" (9768 bytes)
View attachment "stuck-softlockup.txt" of type "text/plain" (25678 bytes)
Powered by blists - more mailing lists