Back in 1998, thorpej did this change to sys/arch/i386/pci/pci_machdep.c:

http://cvsweb.netbsd.org/bsdweb.cgi/syssrc/sys/arch/i386/pci/pci_machdep.c.diff?r1=1.29&r2=1.30

What it does is to disable memory mapped access to PCI devices on hosts
that has a SiS 85C496 host-PCI bridge. I'm the lucky (?) owner of such a
motherboard, and I'd like to use memory mapped access on it. I knew it
works on Windows, so I wondered what lied behind this piece of code.

There does'nt seem to be a PR on it. And by searching Google, I found only
one reference to this in Linux, some frame grabber driver makes a notice
about this problem if the chipset is detected.

I have commented the code out and intend to try a PCI graphics card and
memory mapped access with it - since I am adventurous of nature :-)

But I'd still like to know what facts are behind this. Also it seems that
more than two years after this code was put in, it still only concerns
that single chipset. Does'nt NetBSD have a 'knowledge' base, where these
things could be documented. It occurs to me that the kernel is not the
right place for that, since the world will always produce broken
chipsets, meaning that if nobody does anything, then the entropy of the
kernel will just rise and rise (thus following the second law of
thermodynamics) :-)

-- 
Anders Dinsen
anders@dinsen.net
http://dinsen.net/anders/