> Apparently the DELAY() in draining the PROM
> output is slightly too short on a r4400: the PROM is not printing the
> priority before the chip gets clobbered by the kernel driver.

Uh, oh.  That's the reason why one garbage character popped off in his
case, I see.

DELAY()/delay() should be neutral from effective CPU speed.  Many
NetBSDs try to calibrate delay constant as a part of MD startup
procedures; hp300, mvme68k, atari, amiga and alpha do such things. 
i386 seems to ultilize (free-running?) clock timer on the fly.  Is it
possible for NetBSD/pmax to have such a delay calibration mech?

Tohru Nishimura
Nara Institute of Science and Technology