> > Another idea is that we sometime poke some invalid address in PCI space.
> > According to my 21285 data sheet that might not abort, but could well
> > stall for a long time.  If that is the case, then it's going to be a pig
> > to track down.
> 
> We might be, but I'd have hoped that we'd eventually come back from there. 

Not necessarily, particularly if there is an interrupt that isn't being 
cleared properly as a result.  It might be that the timeout leave so much 
in a mess (missed DMA transfers etc) that we never really get out of 
servicing interrupts.

R.