> 
> Oh, i forgott:
> 
> With cache disabled, it runs for days without problems.
> I currently compile a kernel with D-cache disabled and I-Cache enabled.
> If my theory (uncontrolled cacheing of controller status data) is right,
> this could help...
> 
Actually, this seems really strange. The cache on VAXen cannot be flushed
or invalidated or so, it just has a snoop cache, so the only thing that
can be done with the CPU cache is to turn it on or off.
It seems more likely that some cache invalid line in your Q-bus is 
broken, that could cause cache-related problems.

-- Ragge