We're lucky in the Sun arena; our "machine" exactly corresponds to a
specific processor:

sun3	mc68020
sun3x	mc68030

Have I missed something, or are we not telling the compiler to generate
instructions specific to those processors when building for those platforms?

Of course, right now, kernel aside, a binary compiled for one will run on
the other, because we're not generating instructions more specifically than
"mc68000". Are we trying to maintain binary compatability across the entire
m68k NetBSD world? If we're trying to avoid the MxN problem for
distribution purposes, then shouldn't we have one userland for all m68k?

I'd like to argue for turning on the processor specific instruction
generation flags for the sun3 and sun3x distribution, and documenting that
we have done so (which means that if you serve stuff up via NFS (or
whatever) that you can't expect a sun3x binary to run on a sun3).

	comments?

	Erik <fair@clock.org>