On Sun, Dec 09, 2001 at 08:47:33PM -0500, Geoff Adams wrote:
> > Jason Wright says that he also had some stability issues with OpenBSD on
> > Ultra 30 machines and added the following to the bottom of 
> > psycho_attach()
> > to solve the problem:
> >
> >         /*
> >          * XXX Linux magic, helps U30
> >          * "PROM sets the IRQ retry value too low, increase it."
> >          */
> >         sc->sc_regs->intr_retry_timer = 0xff;
> >
> > I have not added this since I really don't like to add untested code, 
> > but
> > I'll be happy to do so if it fixes your problem.
> 
> I'm quite pleased to report that this does, in fact, correct the 
> problem! I recommend this change be committed. Out of curiosity, what, 
> exactly does this mean? Are interrupts frequently missed on PCI busses? 
> That sounds bad, although I admit that I wouldn't know if this was 
> standard operating procedure. Or does this mean something else?
> 
I have no documentation for psycho other than linux source code (yick!),
but my impression of the problem is the same as yours: that PCI
interrupts are sometimes missed.  I don't like committing code I don't
fully understand, but this seemed to fix a very annoying problem =)

--Jason L. Wright