To answer my own question, I dug out the source.  It seems the XFree86
scanpci uses KDENABIO, which seems to be present only if pcvt is
configured; I wasn't using pcvt.  That would explain the coredump.

I'd submit a patch to Xfree86 to use i386_iopl(), but I don't know how
far back i386_iopl() goes.  Was it available in 1.1?  If not, is it
better to support non-pcvt 1.2 configurations, or to continue using
KDENABIO and KDDISABIO?   And, if the latter is preferable, can we
move the definitions out of arch/i386/isa/pcvt,  to somewhere where
they'll get installed in /usr/include?

Lastly, I think non-x86 NetBSD ports (e..,g Alpha) use Xfree86, and
have PCI busses.  Should we pull in some more machine-independent
version of, or replacement for, scanpci into the NetBSD tree?  The
current PCI drivers don't report on PCI parameters like MIN_GNT and
MAX_GNT, and the (1.2) PCI chipset drivers don't seem to care about
the pci/host bridge latency, let alone say what it is...


thanks
--Jonathan