On Thu, Jul 09, 1998 at 02:27:09PM -0400, Perry E. Metzger wrote:
> 
> "Nathan J. Williams" writes:
> > >In general, this problem is bad enough that I'm thinking that install
> > >kernels ought to automatically exclude "common" IRQ assignments from
> > >consideration -- 3,4,5 & 7 spring to mind immediately. Anyone have
> > >suggestions on what the mask ought to be?
> > 
> > 	Excluding 5 and 7 seems like a fine idea. But why 3 and 4?
> > Are there cases where INSTALL kernels haven't found existing serial
> > ports there? 

I have seen several portables, which often have a com device and
an (extended com) infrared device available for use with IRQs 3 and,
that have broken default soft device configurations (for
whatever reason) where only one of those two devices are
enabled, leading to the above-mentioned problem.

Although I think such turned-off devices usually really are
turned off, I think it would still be best to try to avoid using
those interrupts.

> Just being careful.

Since I cannot think of a real, general solution to this problem,
and the lowest common denominator IRQ mask for pcics is with
time going to shrink to zero, perhaps a better "mask" would
a prioritized list of all non-"built-in" interrupts (ie. not
IRQ 8 and such), beginning with safe ones like IRQ 15, which
is very likely to be either free or accounted for by
PCI or an IDE controller which will be found, etc.


-- 
Soren