Re: make buildworld: Signal 11; Illegal instruction

On Fri, Aug 01, 2003 at 01:04:16PM +0200, Tobias Roth wrote:
> On Thu, Jul 31, 2003 at 09:52:08PM +0100, Bruce Cran wrote:
> > On Thu, Jul 31, 2003 at 03:03:01PM -0400, Chris Shenton wrote:
> > > Chris Shenton <chris_at_shenton.org> writes:
> > > 
> > > >   *** Signal 11
> > > >... 
> > > >   Illegal instruction (core dumped)
> > > >   *** Error code 132
> > > 
> > > Also seeing
> > > 
> > > *** Signal 4
> > > 
> > > if it matters.  This sounds way too flakey to be SW.
> > 
> > I'm seeing the same symptoms.   I got a signal 4 when running 'clean'
> in the 
> > pam authentication directory, and I've just had a signal 11 running 
> > 'rm -f libradius.so'.  This is an install from a snapshot I built
> today - 
> > during the install I had panics in _mtx_init_ and a backtrace traced
> through 
> > vfs and ffs functions, and I only managed to install successfully when
> I 
> > had the CPU throttled to 30%.  This is the same computer which ran
> memtest86
> > for 8 hours without a single fault last night, so I doubt the
> hardware's 
> > faulty, at least not the memory or the CPU.
> 
> memtest86 does not always catch memory errors. sig11 and sig4 at varying
> locations during buildworld are a sure indicator for a hardware problem.
> most likely a memory or overheating issue, though other hardware related
> causes are possible.
> 
> if you still are not convinced that this is a hardware issue, run build-
> world on a -stable system.
> 
> more and more latest generation laptops from different manufacturers
> show
> these symptoms during hot days. my guess is that mobile pentium 4
> systems
> are just not as stable as they should. let's hope things get better with
> the pentium m chips. are the manufaturers deploy better quality control
> to
> catch the numerous faulty systems.

My stock Dell Optiplex GX260, P4 based with 256 MB RAM, running -current,
would spit signal 4,10 and 11 (and also 6, don't remember) all over the place
during buildworld when not having these kernel options:

options         DISABLE_PSE
options         DISABLE_PG_G

Search the -current archive, it's due to a processor bug but there is
no detailed public information about it and hence no 'official' fix.

You might try and see if it helps for you. memtest86 and other hardware
testers won't notice anything because it's in the CPU and officially
unknown.

But yes, also keep in mind that there might be overheating issues if 
the wheather is hot; yesterday my -stable machine at home rebooted during
a port build: turned out to be a flatcable being too close to the CPU fan...

Karel.