How about re-probing the ADB bus later on (from the booted system) when
no devices have been found?
I think the probe thread gets deadlocked...
I seem to recall that it's actually the IOP core itself that gets
confused. I don't have a Q950, and the code seems to be fine on a
IIfx (which is what was used in the original work). What we should
probably do is actually implement the IOP watchdog handling and
reset the thing if it crashes on us. There are some bits and pieces
of it already in the code.