FreeBSD 5.3-R, postfix smtp dying with sig11

[Bryan Fullerton]

On Mon, 10 Jan 2005 20:47:44 -0800, Kris Kennaway <kris at obsecurity.org> wrote:
> On Mon, Jan 10, 2005 at 10:47:24PM -0500, Bryan Fullerton wrote:
> 
> > So... what could cause this? Is the sig 11 causing the failure, or is
> > master killing off smtp as a result of some other issue? Anything I
> > can poke at?
> 
> Usually hardware failure.  Consult the mailing list archives
> or google for extensive discussion.

If it was hardware failure I'd expect to see sig11's elsewhere -- I've
never seen anything else sig11 on the box except postfix smtp
processes, including repeatedly building world and kernel and dozens
of other ports (it's a jail system, so I build package updates on the
host OS and deploy them to the jails).

I suspected yesterday that the problem might be controller write cache
and softupdates combining to return success on writes before the data
actually hit the drives, which could cause a short read of the on-disk
queue file if it was accessed before the write completed. But I
disabled the controller cache using the tw_cli command and I'm still
seeing the same thing (unless tw_cli isn't really disabling all the
write caching and I need to do something in the controller BIOS).

I've also tried running a non-SMP kernel (it's a HTT system), but that
didn't help. I'll try disabling APIC/ACPI today (which will of course
also kill SMP) and see if that helps.

Thanks,
Bryan