[tulip-bug] PCI fatal bus error

[Donald Becker]

On 6 Dec 2001, Glenn Ammons wrote:

> Donald Becker <becker@scyld.com> writes:
> > It looks as if you are getting parity errors.
> 
> > > The machine is an old Gateway P90.  The documentation for the card
> > > says that it requires a PCI 2.1 compliant motherboard and BIOS, and
> > > that the card should be inserted in a bus mastering PCI slot.  I have
> > > no idea if my motherboard and BIOS are up to snuff or if the slot is
> > > bus mastering (how would I tell?).
> > 
> > If the board works at all, it's in a bus master slot.  However some
> > motherboards cheat and share bus master signals -- that will cause
> > similar problems.
> 
> OK.  I've tried all the slots, and they all work the same, so I guess
> they're all bus master slots.  By the way, `cat
> /proc/pci' says
> 
>   PCI devices found:
...
>     Bus  0, device  12, function  0:
>       Ethernet controller: Unknown vendor Unknown device (rev 17).
>         Vendor id=1317. Device id=985.
>         Medium devsel.  Fast back-to-back capable.  IRQ 11.  Master Capable.  Late
>   ncy=66.  Min Gnt=255.Max Lat=255.

Hmm, the grant and latency request times are curious.  But I don't think
that's the problem.

> There are two master-capable devices listed.  Does that mean that it
> could be a bus master signal-sharing problem, assuming that my
> motherboard is also the cheating kind?

If you tried different slots, this isn't the problem.

> It seems that I have bad hardware.  Are there any tulip settings that
> I could use to mitigate this problem, or am I stuck replacing the
> ethernet card or motherboard?

You can try the module option  csr0=0x00A04800 to reduce the PCI burst
length, but if it's really a parity error you have a hardware problem
that can't effectively be worked around.

Donald Becker				becker@scyld.com
Scyld Computing Corporation		http://www.scyld.com
410 Severn Ave. Suite 210		Second Generation Beowulf Clusters
Annapolis MD 21403			410-990-9993