I feel much less helpless now. I found my troubles with the SCSI_TEST
kernel and ps, et. al.

It seems that kvm_mkdb was not making a new database for the libkvm stuff
to do its thing. Removing the database and running kvm_mkdb again forced
the right thing to happen.

Why the testdb() routine was failing to detect that I was running a
different kernel is not entirely obvious to me, but I have a simple
solution; I will add a test for which is more recent: the kernel or the
database, before the rest of the validity tests. stat(2) is your friend.

Send-pr with diffs to follow.

Erik Fair