unionfs and kernel core dumps

[Taylor R Campbell <campbell+netbsd%mumble.net@localhost>]

(I am not subscribed to this list, so please cc me in replies.)

I would like to use unionfs for various purposes.  Example: making
light-weight pkg_comp chroots by creating one and then unionfsing the
rest, to substantially reduce the time spent in tar's disk I/O and to
enable parallel package builds.  Unfortunately, I have already run
into several panics with unionfs.  The first is PR kern/43560; after I
worked around that by making the device nodes in the other layer of
the union, and started trying to install packages, the second panic
was a simple `panic: trap' during `cd /usr/pkgsrc/editors/emacs &&
make bin-install'; here's what's in dmesg:

   uvm_fault(0xcf1d8930, 0, 1) -> 0xe
   fatal page fault in supervisor mode
   trap type 6 code 0 eip c053b6b8 cs 8 eflags 10282 cr2 4 ilevel 0
   panic: trap
   Begin traceback...
   End traceback...

   dumping to dev 0,1 offset 1049975
   dump succeeded

Also unfortunately, savecore just reports `(null): Bad address' and
never actually saves a core image in /var/crash:

   Jul  2 22:11:36 ... savecore: reboot after panic: panic: trap
   Jul  2 22:11:36 ... savecore: (null): Bad address

It behaves similarly even if I pass -f.

It would be nice if unionfs worked, or, failing that, if savecore
worked well enough to get a backtrace or anything.  Is this paucity of
information worth submitting a PR over, about either unionfs or
savecore?  Am I doing something wrong with savecore?  Do I need to
tweak some sysctl knob to persuade NetBSD to actually dump a core?
This is on NetBSD/i386 5.1_RC2, by the way.