It turns out the bug in recent -current kernels is due to me
misunderstanding  how the MI tc autoconfiguration was intended to
work.

I've now hacked in a test that should stop a 3100 from trying to configure
a turbochannel bus, and then hanging.  I'm not going to commit this
until I've got the fix ``right''.

I have put a new  kernel on ftp.dsg.stanford.edu.  If someone could
try this out on a 3100 and let me know how much further it gets than
the previous kernel, that'd be cool :).

--Jonathan