what nathan said.

It would be pretty straightforward, but a fair bit of rototilling, to
change the drivers to check for a NULL return from the
interrupt-establishment API (and disable the device, and panic only if
this is not possible)..

given that we'll have to change the interrupt-establishment API
slightly for the MP case (so that MP-safe interrupt routines can be
identified and called from outside the "big lock" perimeter), it would
be reasonable to have that API not panic in the conflict case and make
drivers using the new api deal with the failure.

					- Bill