Re: Kernel panics -> possible memory corruption?

["Alex Hudson" <nospam [at] york.ac.uk>]

First off, what in heck is an eml attachment? (That's what this thing says
it is!) I've never seen one of those in my life ;))

Anyway, onto the OOPsen... for those who can't read oopses, what's happened
is a null pointer dereference, far and away the most common oops. This isn't
a raw OOPS, it's been passed though ksymoops which cleans it up a little and
gives us some more information.

We can find out exactly where the program was when it OOP'sed - EIP, the x86
instruction register is set to sock_poll (the function), and it was at
offset 7 (length 36). Unfortuantely, I'm in Win at the moment, or I would be
able to look this up and we would find out exactly where the kernel was when
it failed. Function calls are matched by ksymoops - if you've ever wondered
what $(SRC)/linux/System.map is for, this is it - mapping function calls to
addresses. Of course, this only works for statically linked functions -
modules will go in anywhere, at any time, and the equivilent file for
modules is /proc/ksyms (if insmod has been passed the -m parameter).

We know the failing process was xmms, so it's probably a valid guess that
there was some hardware access that caused the kernel to fail. I can't say
for sure; unless the CODE: section was disassembled, although having a look
in the function that caused the problem would give some clues. However, the
prime suspect is probably the hard-drive: I can't believe that memtest would
pass faulty memory, unless it was some sort of thermal problem (there are a
lot of papers written about testing memory, checking for stuck bits, etc.,
usually the subject of such papers are more efficient algorithms, etc.).

Jose, if you like, you can send me a batch of OOPs, I'll bob them on a disk
and see what my linux box says about them, although you'll have to make sure
you ksymoops them all first, 'cos I can't do that (for obvious reasons!) It
should be possible to pin-point the fault exactly, which should make clear
what has happened.

Of course, the other option is that it is a bug, but who ever heard of those
in Linux ? ;))

Cheers,

Alex.

---------------------------------------------------------------------
Sheffield Linux User's Group - http://www.sheflug.co.uk
To unsubscribe from this list send mail to
- <sheflug-request [at] vuw.ac.nz> - with the word 
 "unsubscribe" in the body of the message. 

  GNU the choice of a complete generation.