Just so everyone on the list knows...

I tracked down the `hanging mount' problem earlier today.  It's due to
an `optimization' in the pmap module which causes it to fail to flush
a TLB entry in some cases when changing the permissions on a mapping.
This bug technically predates the recent pmap change, but that change
causes it to occur in more cases.

As Ross points out, the bug probably appeared to be machine-specific
because the TLB differs between models.

A fix should be available by SUP/FTP/etc. tomorrow morning.

Thanks to Bernd for providing me with crash dumps and other
information.  No thanks to Matt for providing inaccurate information
and generally being a pest.