> That's a good one.  I've seen that before, but never need it happen
> quite so regularly and severly...
> 
> The z8530 has a hardware settle time, and I'd always thought that that
> was taken care of in hardware on the alpha since i figured that if
> that _was_ a problem, it would have blown up more severely...

OK. 

> I've built a kernel which includes short delays when accessing the SCC
> registers.  it's now at:

This works just fine, thanks. There are some occasional silo overflows.

Chris sent me the patches in question via email (for those of you
out there w/ 3000/900s or perhaps those reading the archive :-)).

--jhawk