Compile a new kernel with pv_va zero'd in pv_unlink() for the cleanup case.
This is the original "fix" I tried when I had the problem on my LX (sun4m)
system; it worked just fine, but a different solution was ultimately
committed to the source tree.

Note, I have *no* idea why that worked for me - I was simply trying to
follow good program hygiene when I did that (always free things you're done
with, etc), and I got lucky. I ran with that patch for a month before I
changed over to a kernel that had the official change, and I have not had
the problem since. If I had to guess, I'd say we have a race condition in
the MMU code somewhere - an interrupt routine is doing something when it
shouldn't.

Just out of curiosity, do you make much use of SunOS or Solaris binaries
and the NetBSD emulation thereof?

	Erik <fair@clock.org>