I have made some changes today that should allow Hypersparc machines
to operate correctly in NetBSD/current.

It cost me some time to track down what was going on: Hypersparcs have
a VIPT cache, which means that to flush a page from the cache a
virtual-to-physical translation is done to do the cache tag comparison.
This will cause an entry to appear in the MMU's TLB.
So, when removing or changing mappings, doing a cache flush after
clearing entries from the TLB may cause spurious TLB entries
causing funny effects in your programs.  Some pmap functions
had the cache and TLB flush operation in the wrong order.


I will look into providing the necessary patches to enable hypersparc
support in NetBSD 1.4.1.

-pk