On Wed, Feb 25, 2004 at 04:04:11PM +0100, Frank van der Linden wrote:
 [...]
> Hmm.. well, RealTeks are known to be crappy, so watchdog timeouts aren't
> that unusual for some versions of them.

Not as crappy as they seem to be with IOAPIC.

I would get them *inevitably* with IOAPIC, followed by near lockup of the
card.  Once the watchdog tiemrs start to hit, the card is practically
useless until you reboot the machine.

All identical hardware, turn off IOAPIC, and I can't remember the
last time that I saw a watchdog timer on this rtk card.


> Do you have another type of network card lying aroung?
>
> If a different card has problems too, you may have a more general
> interrupt problem somehow, which may indeed be an APIC problem. Which
> would explain the IDE issue as well.
> 
> You can try to not use APICs by disabling the ioapic and MPBIOS lines
> in the GENERIC kernel config, and compiling that. However, I do
> consider that to be somewhat of an unsupported configuration.. but
> it should still work.

Speaking for myself:

It seems to work for me.  It seems to be practically required.

If I don't disable it, the system is very unhappy, and some PCI
devices don't work or randomly stop working.

If the choice is between "supported" or "usable", I'll take "usable".
"Support" for a volunteer project like NetBSD tends to be descriptive
rather than prescriptive.  (^&

...if anyone has any suggestions to make for patches to improve
IOAPIC on nForce3 boards, I'm willing to build and test kernels, though.
I don't have the knowlege (nor the present danger of acquiring said
knowledge) to debug this myself.


-- 
  "I probably don't know what I'm talking about."  http://www.olib.org/~rkr/