> The bmac driver in Linux indeed uses some of the special "FCR" (i.e.
> Feature Control Register) bits in mac-io (ohare, heathrow or paddington).

I don't think the bug is in the bmac driver itself: NetBSD's bmac
initialization code is pretty much identical to both Linux's and
Darwin's. I suspect the bug is somewhere else in the macppc dev stuff
(either in the PCI code since obio attaches to pci, or in the more
generic code), since I've seen basically the same failure mode (device
fails to work if/unless OF touches it) in pciide, bmac, and ahc.

-- Dan