The "dvma_mapin: null page frame" panic keeps popping up on microsparc II
machines. The fix suggested last time - turning off the Branch Predict logic -
may have had some effect, but it does not cure it completely.

When this happens, one or more of the local registers (%l0 - %l7) in
dvma_mapin's register window have values most odd..

If I turn off the cache, none of this happens (to me, at least).

Now, it might be there's an obscure bug in the cache handling code that
somehow other processor models never trip over. On the other hand, the
microsparcII might need a workaround for some cache-related hardware
flaw.  Can anybody out there tell me more?

-pk