Date: Mon, 7 Dec 2009 14:59:37 +0000
From: Mel Gorman <mel@csn.ul.ie>
To: David John <davidjon@xenontk.org>
Cc: Christoph Lameter <cl@linux-foundation.org>, linux-kernel@vger.kernel.org,
       Jonathan Miles <jon@cybus.co.uk>, Pekka Enberg <penberg@cs.helsinki.fi>
Subject: Re: OOM kernel behaviour
Message-ID: <20091207145937.GB14743@csn.ul.ie>
References: <4B1402FC.80307@cybus.co.uk> <4B1537CA.7020107@xenontk.org> <alpine.DEB.2.00.0912011007360.8841@router.home> <4B154E10.7050300@xenontk.org> <alpine.DEB.2.00.0912011124490.10426@router.home> <20091202155537.GH1457@csn.ul.ie> <4B1C93E0.4070806@xenontk.org>
MIME-Version: 1.0
Content-Type: text/plain; charset=iso-8859-15
Content-Disposition: inline
In-Reply-To: <4B1C93E0.4070806@xenontk.org>
User-Agent: Mutt/1.5.17+20080114 (2008-01-14)
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 2641
Lines: 61

On Mon, Dec 07, 2009 at 11:04:24AM +0530, David John wrote:
> On 12/02/2009 09:25 PM, Mel Gorman wrote:
> > On Tue, Dec 01, 2009 at 11:26:37AM -0600, Christoph Lameter wrote:
> >> On Tue, 1 Dec 2009, David John wrote:
> >>
> >>> Here are three logs from three days. Log3.txt is today's log and the OOM
> >>> killer murdered Thunderbird as I was attempting to write this message.
> >>> The kernel config is also attached.
> >>
> >> Hmmm... This is all caused by the HIGHMEM zone freecount going beyond min
> >> which then triggers reclaim which for some reason fails (should not there
> >> is sufficient material there to reclaim). There is enough memory in the
> >> NORMAL zone. Wonder if something broke in 2.6.31 in reclaim? Mel?
> >>
> > 
> > I'm not aware of breakage of that level, nor do I believe the page
> > allocator problems are related to this bug.
> > 
> > However, I just took a look at the logs from the three days and I see
> > things like
> > 
> > Nov 25 23:58:53 avalanche kernel: Free swap  = 0kB
> > Nov 25 23:58:53 avalanche kernel: Total swap = 2048248kB
> > 
> > 
> > Something on that system is leaking badly. Do something like
> > 
> > ps aux --sort vsz
> > 
> > and see what process has gone mental and is consuming all of swap. It's
> > possible that the OOM killer is triggering too easily but it's possible
> > that a delayed triggering of the OOM killer would have been just that -
> > a delay. Eventually all memory and all swap would be used.
> > 
> 
> It is a leak in Compiz. Killing and restarting Compiz frees up the swap.
> The issue is better in 2.6.32 for some reason. The funny thing is I've
> been using Compiz with 2.6.31 for a couple of months now, with no
> updates to either, so I'm not sure what triggered this problem.
> 

This is a total stab in the dark. Is it possible there was a change in
DRM between 2.6.31 and 2.6.32 that means resources (like textures) are
no longer been freed properly? This might be particularly the case if
you were not using KMS before but you are now.

If something like that has changed, it should probably be brought to the
attention of David Airlie.

If nothing in that regard has changed, I don't have a better alternative
theory as to why it's leaking now.

-- 
Mel Gorman
Part-time Phd Student                          Linux Technology Center
University of Limerick                         IBM Dublin Software Lab
--
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/