On Wed, Oct 12, 2005 at 03:13:27PM +0100, Matthias Scheler wrote:
> > > http://mail-index.netbsd.org/source-changes/2005/10/02/0011.html
> > > http://mail-index.netbsd.org/source-changes/2005/10/02/0012.html
> > yup, those are the ones.
> 
> I've backed out both changes in my source tree (I don't have time for
> a more complete analysis at the moment) and the resulting kernel
> seems to work stable. It survived running "./build.sh -j 2" and
> a build of "kdepim3" at the same time without crashing.

I've been able to reproduce the problem myself now, though it's very
inconsistent.  once it happened within a minute of starting the stress test,
but mostly it can run for hours without a problem.  I'll run the series of
tests that I asked for earlier and see which change is responsible.

-Chuck