[OpenAFS] Stack overflows on Red Hat 2.4.20 kernels

Derrick J Brashear shadow@dementia.org
Mon, 15 Sep 2003 14:28:05 -0400 (EDT)

On Thu, 4 Sep 2003, Steve Snodgrass wrote:

> On Wed, Aug 27, 2003 at 02:00:05PM -0400, Derrick J Brashear wrote:
> > Can you build a module with --enable-debug-kernel so the stack trace is
> > vageuly more correct?
> OK, here's a dump with the debug module, though I can't see too much
> difference.  How can you tell if debug was used or not?
> This is from a Dell PowerEdge 2550 running the Red Hat 7.3 2.4.20-19.7smp
> kernel.

The do_softirq in there bothers me, we had a problem before where softirqs
could trigger in any thread instead of just in the kernel softirq threads,
and they can use too much stack given where they might trigger.

And it's nothing we (afs) do causing it, the kernel tiggers them when it
feels like it; RedHat had an errant patch triggering it when it shouldn't
have been, and IIRC they fixed it at the time.

You're right, it looks not much different. Did -fomit-frame-pointer appear
in the arguments to cc when you were building the kernel module?