> > If (2), then let me know so I can resign myself to MacOS.
> 
> Maybe, for now anyway if the above doesn't work.  What we really need is
> someone with time, some kernel hacking knowledge (or the willingness to
> learn) and a PowerBook who can figure out how the internal video mapping
> works for these machines....

That project was done during INTVIDTEST0. :) The conclusion I came to was that
it actually doesn't have anything to do with the mappings. We know them
(if they've been lost I have a sheet of paper with them). There's something
bizzaro going on with the way the TT* registers are used on the PowerBooks.

If you look closely at pmap_bootstrap.c you'll notice that the memory map
is faked up from netbsd during a serial boot, thus the working powerbooks
on serial boot. However, during a normal boot we try and figure out how
the memory is mapped. The goofy TT* register stuff messes up this process
and the machine comes to a halt. My original intvidtest patches simply
skipped the mapping probe and fooled the machine into thinking it was
serial booting for the purposes of memory mapping.

I've mentioned this to Scott and did a Send-PR some time ago. I think there's
an attempt at a fix in the -current as you mentioned.