I've made some fixes today to the MI PCI framework which improve the
i386 MP interrupt routing situation significantly.  

I fixed two bugs:

 - primary pci busses which were not bus #0 mistakenly had their
interrupts swizzled.

 - there was no MI mechanism to do interrupt routing based on
unswizzled interrupt pins.

Partial credit for this fix should go to Jason Thorpe, who put
enough of a fix in place to make today's work relatively trivial.

Folks may want to try the test kernels in:

	ftp.netbsd.org:/pub/NetBSD/arch/i386/mptest/test-20020518/

If previous kernels have not worked for you, send reports of how far
these manage to boot on your systems to sommerfeld@netbsd.org and/or
tech-smp@netbsd.org.

dmesg output from a successful boot (or an unsuccessful boot, if you
can rig up a serial console or otherwise capture the output) would be
appreciated.  GENERIC.MPDEBUG output will contain more verbose
information on interupt routing and so is preferred.

					- Bill