itojun@iijlab.net writes:

> 	here's openbsd's dmesg.  it apparently touches function 1 of
> 	SC1100 device, but does not panic like netbsd does.

Digging deeper into the SC1100 manual, I found an interesting
register. Page 169 of SC1100.pdf from National Semiconductor:

"Table 5-29. F0: PCI Header and Bridge Configuration for GPIO and LPC
Support

Index 41h       PCI Function Control Register 2 (R/W) Reset Value: 00h

Bit     Description
...
1       Power Management Configuration Trap.
        If this bit is set to 1 and an access occurs to one of the
        configuration resisters in PCI function 1 (F1) register space,
        an SMI is generated."

(other bits exist to trap accesses to IDE (F2), "XpressAUDIO" (F3),
and legacy ISA I/O space).

Given the existence of this register, it seems possible that NetBSD is
setting (or OpenBSD and FreeBSD are clearing) that bit, or perhaps are
set up to handle SMI differently. A little code in
sys/arch/i386/pci/pchb.c or, if it's booted, "pcictl /dev/pci0 dump -d
0 -f 0" would be informative.

        - Nathan