Charles M. Hannum wrote:
> From an architectural point of view, your patch seriously violates the
> spl abstraction and is totally unacceptable.

I know.
It's necessary to block interrupts (not necessarily this way) on my PS/2's
pckbc which uses level-triggered interrupts. I though the fix for PS/2
controller would be worth try for you too.
 
> Furthermore, your explanation makes no sense; it was NOT the case that
> the system was wedged for me, which is what would happen if the
> interrupt was stuck.  Only the keyboard was wedged.  And since it's an
> edge-triggered interrupt, it's not even possible for it to get stuck.

OK. Do you still have that machine available? It would be worth
try to e.g. try to remove the code calling pckbcintr() in
dev/ic/pckbc.c:pckbc_set_poll().
BTW, does the problem happen with pccons too?

Jaromir