> BTW, it surprised me that the dubious change got actually committed
> without prior discussion, and even pulled up to the 1.6 branch.
> Was that necessary?

 it was discussed on port-i386 for quite a while.  and it is clear that
 pci.c needs surgery, given that openbsd/freebsd boot without problem
 while netbsd panics during pci function probe.

> The change appears as not TRT, especially if
> the problem is actually not in pci probe code.

 send diff.

itojun