Kernel oops - help interpreting partial trace info

Kernel oops - help interpreting partial trace info

[Samson Yeung]

[-- Attachment #1: Type: text/plain, Size: 1354 bytes --]

Hello,

(Sent to the output of ./scripts/get_maintainer.pl -f
kernel/sched_fair.c - please CC me as I am not on-list.)


Some of our customer's systems running a 2.6.32.11 kernel.org kernel
crash, leaving the attached partial oops on screen. I doubt it is a
hardware problem since we've already replaced one system and the
problem came back.

I run the values after "Code:" from the bottom of the oops through
linux-2.6/scripts/decode/code as described in
Documentation/oops-tracing.txt, and this is the first few lines of
output:

Code: 00 44 8b 15 23 76 68 00 45 85 d2 74 32 48 8b 50 08 8b 5a 18 48
8b 90 90 06 00 00 48 8b 4a 50 48 85 c9 74 1b 48 63 db 48 8b 51 20 <48>
03 14 dd 00 fb 74 81 4c 01 32 48 8b 49 58 48 85 c9 75 e8 48
All code
========
   0:   00 44 8b 15             add    %al,0x15(%rbx,%rcx,4)
   4:   23 76 68                and    0x68(%rsi),%esi
   7:   00 45 85                add    %al,-0x7b(%rbp)
   a:   d2                      (bad)
...

Should (bad) ever be output by decodecode?

I dump the region that RIP indicates using gdb against the mactching
kernel and get different values.

What could cause this?

I'm working on getting a full dump via the serial port, but the
customers are on different continents than I.

Is there any more information I can get from the screenshot other than
probable stack corruption?


-Samson

[-- Attachment #2: dump2.jpg --]
[-- Type: image/jpeg, Size: 84656 bytes --]

Re: Kernel oops - help interpreting partial trace info

[Peter Zijlstra]

On Thu, 2010-12-30 at 08:35 -0800, Samson Yeung wrote:
> Code: 00 44 8b 15 23 76 68 00 45 85 d2 74 32 48 8b 50 08 8b 5a 18 48
> 8b 90 90 06 00 00 48 8b 4a 50 48 85 c9 74 1b 48 63 db 48 8b 51 20 <48>
> 03 14 dd 00 fb 74 81 4c 01 32 48 8b 49 58 48 85 c9 75 e8 48
> All code
> ========
>    0:   00 44 8b 15             add    %al,0x15(%rbx,%rcx,4)
>    4:   23 76 68                and    0x68(%rsi),%esi
>    7:   00 45 85                add    %al,-0x7b(%rbp)
>    a:   d2                      (bad)
> ...
> 
> Should (bad) ever be output by decodecode? 

Don't think so, my guess is the decoder started at a non ins boundary
and got confused. "85 d2" is "test %edx, %edx" and is something that
occurs quite frequently in the code. The following 74 is "je", which
makes sense after a "test".

"8b 15 xx xx xx xx" is "movl xx,%edx"

That however does leave us with a 45 unexplained, that's usually part of
a mem argument, its x(%ebp) like stuff.

Dunno really..