On Thu, 21 Apr 2005, Andy Ruhl wrote:

> On 4/21/05, Jeff Johnson <jeff.johnson@wsm.com> wrote:
> > Nathan J. Williams wrote:
> > 
> > >Jeff Johnson <jeff.johnson@wsm.com> writes:
> > >
> > >
> > >
> > >>    I have installed NetBSD 2.0.2/amd64 port on to a Newisys 4300 (4
> > >>way Opteron 848). I have run into a problem where bringing up bge0
> > >>will cause a GPF and crash the system. I am able to being up bge1
> > >>without any errors or GPF. Both bge0 and bge1 are Broadcom 5703
> > >>controllers. I tried several media options from auto to hard setting
> > >>100baseT, 1000baseT, full and half duplex, etc. Anytime bge0 is
> > >>brought up the system augers in. I am running netbsd.GENERIC.MP kernel
> > >>that comes with the 2.0.2 release.
> > >>
> > >>    I saw a similar bug report in a search. Has this been fixed yet?
> > >>If not I can provide access to the system (via ssh) if any developers
> > >>wish to take a look under the hood.
> > >>
> > >>
> > >
> > >I experienced a similar problem on a system with a 3ware RAID
> > >controller, driven by twe(4). I noticed that the problem only occurred
> > >when the twe card was in the first PCI slot; if it was out, or if it
> > >was in another slot, everything worked fine. I haven't had a chance to
> > >dig much deeper than that, but that and the location of the crash
> > >makes me suspect a problem around the apic programming and dispatch.
> > >
> > >        - Nathan
> > >
> > >
> > I hope it does not depend on PCI slot placement as these controllers are
> > onboard devices. I guess I could use a dremel tool to move them but I
> > bet they would really not work then.
> 
> Or maybe you can remove or change configuration in the bios?
> 
> Andy
> 

I looked into that. I can disable the ports in the bios but I am not able 
to move them or change their position in the pci tree. I find it 
interesting that bge0 will cause a GPF/crash but bge1 functions perfectly. 
They look identical by comparison. The only difference would be their pci 
addresses and the ordering in which they appear to the kernel. It would be 
interesting if I could somehow reverse their order so the controller/phy 
that is now bge1 was seen as bge0 by the kernel and vice-versa. That way 
being able to see the GPF/crash follow the physical chip or the ordinal 
position.

-Jeff