2.6.7 SMP kernel crash under load.

[James Wilkinson]

Greg Trounson wrote:
> I have a dual Athlon 2600+ system running Fedora Core 2, updated to 
> kernel 2.6.7.
> Recently the machine has locked up every few days (3/9/04, 9/9/04, 
> 11/9/04). Every time it has been just after 4am, the time that 
> cron.daily is set to activate. In the case below, it crashed while 
> running updatedb, but the actual process is not the same every time.
> 
> This coincides with /etc/cron.daily, which consists of:
> 00-logwatch 0anacron inn-cron-expire makewhatis.cron rpm slrnpull-expire 
> tetex.cron
> 00webalizer cyrus-imapd logrotate prelink slocate.cron squirrelmail.cron 
> tmpwatch
> 
> I have run memtest86 on this machine for several days and found no problems.
> 
> Relevant log message follows:
> 
> Sep 3 04:08:03 vector kernel: Unable to handle kernel paging request at 
> virtual address ee67494e
> Sep 3 04:08:03 vector kernel: printing eip:
<large snip>
> 'Tainted' is due to the Nvidia driver being installed.
> 
> any ideas what could be causing this? Kernel bug?

Yes... user space should not be able to make the computer crash.

There are exceptions, such as if the program is granted write access to
/dev/kmem and scribbles on it, but, in general, a user space program
should not be able to provoke a kernel crash.

(Having said that, I suspect that X drivers still have to have enough
hardware access that they can theoretically crash the machine. I haven't
been keeping up with those issues).

The programs that you're wondering about don't run a GUI, and so don't
need X at all. Can you temporarily comment out the Nvidia modules from
/etc/modprobe.conf and boot the computer into runlevel 3? (It might be
a good idea to disable rhgb, as well: you can do all this from the
kernel command line, or through /boot/grub/grub.conf).

Once you've made sure that the kernel isn't tainted, see if you can
reproduce the error. If you can reproduce the error against a
non-tainted kernel, put the results in Fedora bugzilla. If you can't,
then it depends on where you got those modules: complain to Nvidia or
the packagers.

If you can't get any joy out of them, and you still want to use the
Nvidia closed-source drivers, try compiling your own kernel, applying
the drivers, and seeing if that makes a change.

Good luck!

James.

-- 
E-mail address: james | Banana in disk drive error
@westexe.demon.co.uk  |