Bug#690814: [squeeze->wheezy regression] disk activity provokes lockups on VIA EPIA CL-6000

[Frank Lenaerts <frank.lenaerts@yahoo.com>]

--- On Wed, 10/17/12, Jonathan Nieder <jrnieder@gmail.com> wrote:

From: Jonathan Nieder <jrnieder@gmail.com>
Subject: Re: [squeeze->wheezy regression] disk activity provokes lockups on VIA EPIA CL-6000
To: "Frank Lenaerts" <frank.lenaerts@yahoo.com>
Cc: 690814@bugs.debian.org
Date: Wednesday, October 17, 2012, 11:19 PM

# regression
severity 690814 important
quit

Hi Frank,

Frank Lenaerts wrote:

> I had to use the power button to restart the machine. The first time
> I rebooted, the system got stuck when trying to mount one of the
> filesystems. Rebooted again and got the login prompt. Did some more
> reboots and found out that the system hangs during more or less 50%
> of the reboots. Since the disk activity LED was always on, I tried
> to provoke some disk activity e.g. by installing some packages. This
> effectively locked up the system. I only once got the system in a
> locked up state without having the disk activity LED turned on.
[...]
> Since this machine had been running Lenny just fine, and Squeeze
> also worked fine, I decided to install a 2.6 kernel.
[...]
>                          Note that it took several reboots to get
> the deb file on the system and to install it. With this kernel, the
> box runs just fine.

Thanks for reporting it.

A few suggestions for moving forward:

* please attach full "dmesg" output from a normal boot (with the
   2.6.32-based kernel)

The tarball I attached to this e-mail contains the dmesg output from 2.6.32-5-486 and 3.2.0-3-486 (sometimes, I can login and even do some things;-))

* could you also get a kernel log from booting the 3.2-based kernel?
   A full log including the lockup would be ideal --- netconsole[1]
   might help here.

The attached tarball contains some "try to boot logs". Some more information:

- netconsole_1.out: netconsole only shows until 'loop: module loaded' while the console also showed 'Loading kernel module loop.' (after the above 'loop: module loaded'), 'Activating lvm and md swap... done' and 'Checking filesystems... fsck from util-linux 2.20.1'. I don't know why these 3 lines are not visible via netconsole. It looks like the box cannot write to the network anymore.

- netconsole_2.out: netconsole only shows until 'eth0: no IPv6 routers present' while the console also showed 'Activating swap...'

- netconsole_3_A-B-C.out: A was like netconsole_1.out, B was like netconsole_2.out, C was like netconsole_1.out

After this, I booted with 2.6 because I could not get 3.2 far enough to be able to login. When I booted with 2.6 it wanted to fsck an external USB disk. Since this would take too long, I unplugged it (so further logs won't show /dev/sdc anymore).

Instead of using netconsole, I used screen on ttyS0. Finally, I could get to the login prompt of 3.2. When I issued a find /usr -ls over an ssh connection, the system locked up (no keyboard interaction possible anymore). Note that screenlog_3.2.0-3-486.0 doesn't show anything usesfull (on the linux commandline, I specified console=ttyS0 and debug).

Since the "successful" 3.2 boot was without (a) the USB  disk and (b) netconsole, I decided to try to boot with netconsole (and without the USB disk). It booted a little bit further than before but I still couldn't get to the login prompt. Rebooted several times, see netconsole_4_a-b-c.out. Some extra notes:

- a: hung at "90-second grace period"

- b: hung at "loop: module loaded"

- c: login was possible; netconsole also ended with "90-second grace period" like in case a; to see if the network is the showstopper, ran find /usr -ls on the console; this worked, just like find /var -ls; when I ran an apt-get update however, the system locked up (all files were downloaded, but it failed at the end (percentage sign stuck)).

* if you have time to run a bisection search through the pre-compiled
   kernels at http://snapshot.debian.org/package/linux-2.6/ to find
   the first broken one, that could help narrow down things quite a bit.

I put it on my TODO list.

Hope that helps, and sorry I have no better ideas,
Jonathan

[1] http://www.kernel.org/doc/Documentation/networking/netconsole.txt

Attachment: dbts-690814.tar.gz
Description: GNU Zip compressed data