OK, i have a patch for the PCI code to use the OFW tree for probing
PCI busses (and also providing a "pa_node" member in pci_attach_args),
all dependant on `#ifdef __PCI_OFW_BINDING'.  this should fix the
probe lossage you saw last... i hope :-)

i will upload a GENERIC kernel once it has finished building and i
have tested it on my U5.  that will be tomorrow (later today).


one useful side effect of this change is that onboard devices are
now probed *first*, so perhaps my concerns about `hme0' changing
will not longer be relevant because it won't happen :-)


.mrg.