page fault in kernel

[Tim Newsham]

finally got to the cpuburn.  Normal cpu burn for K7 didnt turn
anything up, but the burnMMX test which does a heavy memory
test actually did..  It wouldnt report errors for memory sizes
<= 256kb, but >= 512kb it would return code 254 after
a few seconds (when I ran it twice, once for each core).

So.. looks like memory after all (fits the pattern).  Using the
test I took the same ram and seated it in a different slot and
it looks like the problem went away.  When I tried adding in
the second ram stick I have (each 2G, for a total of 4G) the problem
came back, but I didnt diagnose it further to see if it was the
ram stick or the memory slot that was to blame (save that for
another day).

So anyway, going to try a big overnight build to see if that has
completely resolved the issue or not.

Here's my repro for my original crash:
  - boot single user
  - mount -o ro -a
  - (burnMMX P || echo $?) &
  - (burnMMX P || echo $?) &
  - ... wait ...

(useful program that burncpu package!).
will update tomorow if it fixed the original issue or not...

Tim

On Tue, Jul 26, 2011 at 8:39 PM, Tim Newsham <tim.newsham+fbsd at gmail.com> wrote:
> I have an amd64 box that is occasionally panic'ing with a kernel
> page fault while the system is under heavy load.  I tried swapping
> out ram and running ram tests and that hasn't seem to turn up
> anything.  I originally saw this with 8.1 but I recently updated to
> 8.2 to be sure and I'm still seeing the problem.  I'm up to date
> with "freebsd-update" and I verified that the core files all match
> the md5 of a clean 8.2 release (after it is also updated).  I also
> tried to disable various frills I had installed (like ip6 tunnel).
>
> The normal way I repro this issue is to start a "portupgrade -caP"
> going and I usually get a crash within a few hours.  Sometimes
> I see some unusual behavior instead of or right before a crash
> (ie. program segfaults that shouldn't normally).
>
> Are there any known kernel issues that sound similar to this
> problem?  I've been working under the assumption that this is
> probably hardware related, but I haven't yet ruled out software.
>
> I'll probably be trying some tests with cpuburn.  Any other advice
> for tracking down this issue would be appreciated.
>
> Tim
>