[cooker-amd64] Re: kernel oops followed shortly after (10 seconds) by hard lock

[Marc Bevand]

Jeremy Barnes wrote:
> This was on a dual Opteron 246 with 6GB ECC ram (full specs are here: 
> http://archives.mandrakelinux.com/cooker-amd64/2003-11/msg00174.php)
>
> kernel: kernel-smp-2.4.22.27mdk-1-1mdk.amd64.rpm, directly from updates.
>
> The machine was moderately loaded, with a load of about 4.  There were 3 batch 
> processes running with about 1.5GB of memory usage each, plus I was editing a 
> file in emacs.  I noticed the following message on the terminal:
>
> Message from syslogd@opteron at Mon Feb  2 23:06:25 2004 ...
> opteron kernel: Oops: 0000
>
> Message from syslogd@opteron at Mon Feb  2 23:06:25 2004 ...
> opteron kernel: RIP [kmem_cache_alloc_batch+86/320] RSP <00000100f9309e68>
[...]

The AMD64 architecture specifies that RSP must always be 16-bytes
aligned. It is not the case in your logs (RSP is <00000100f9309e68>).
This has naturally generated a CPU fault and a kernel oops.

This is very likely a bug of the compiler used by Mandrake to generate
their kernel (kernel-smp-2.4.22.27mdk-1-1mdk.amd64.rpm). Try another
kernel, or compile one yourself.

--
Marc Bevand