From mboxrd@z Thu Jan  1 00:00:00 1970
From: Neil Brown <neilb@suse.de>
Subject: Re: Deadlock in md barrier code? / RAID1 / LVM CoW snapshot + ext3
 / Debian 5.0 - lenny 2.6.26 kernel
Date: Wed, 22 Sep 2010 08:30:39 +1000
Message-ID: <20100922083039.283ccdfd@notabene>
References: <4C938103.1010304@seoss.co.uk>
	<20100918085925.5fee83ee@notabene>
	<4C97BD21.1040405@seoss.co.uk>
	<4C991D61.1040400@seoss.co.uk>
Mime-Version: 1.0
Content-Type: text/plain; charset=US-ASCII
Content-Transfer-Encoding: 7bit
Return-path: <linux-raid-owner@vger.kernel.org>
In-Reply-To: <4C991D61.1040400@seoss.co.uk>
Sender: linux-raid-owner@vger.kernel.org
To: Tim Small <tim@seoss.co.uk>
Cc: "linux-raid@vger.kernel.org" <linux-raid@vger.kernel.org>
List-Id: linux-raid.ids

On Tue, 21 Sep 2010 22:02:25 +0100
Tim Small <tim@seoss.co.uk> wrote:

> Tim Small wrote:
> > http://buttersideup.com/files/md-raid1-lockup-lvm-snapshot/iodeadlock-sysrq-t.txt
> >
> > ... this was soon after the io to md2 stopped - md0 seems fine...
> >
> > oldshoreham:~# cat /proc/mdstat
> > Personalities : [raid1]
> > md2 : active raid1 sda6[0] sdb6[1]
> >       404600128 blocks [2/2] [UU]
> >       [>....................]  resync =  0.1% (437056/404600128)
> > finish=343321.2min speed=19K/sec
> >
> >
> > ... I also tried an older Debian 5.0.x kernel from Mar 2009, which is a
> > less-patched 2.6.26, and got the same results.  2.6.32 hasn't deadlocked
> > after 10 minutes (2.6.26 usually does within a minute of boot-up), so
> > I'll leave it re-syncing overnight...
> >   
> 
> 2.6.32 resynced to completion, but if possible I'd really like to get
> 2.6.26 running on this box, as it uses openvz, and the 2.6.32 openvz
> patches are still pretty green, I think.  Is their anything in the
> sysrq-t output in that link of any use?  Are there any patches I should
> try, or would it be better to start bisecting between 2.6.26 and 2.6.32
> (assuming I can reproduce the problem without the load pattern which
> openvz is producing - which I probably can).
> 
> Cheers,
> 
> Tim.
> 
> 

It is odd that 2.6.32 works and 2.6.26 doesn't as I cannot find any change
between the two that could be related.
There were some deadlock issues if a read-error was detected during resync
but I don't think you are getting read errors are you?

A bisect would of course be easier for me, though not particularly easy for
you....

Maybe if you can get me a full sysrq-T list that might help.

NeilBrown