this code makes equal sense for memory and I/O space, prefer to map
the PCI front end via memory space (conditionalized on a patchable kernel
variable), and do a bit of other random NetBSD-specific cleanup. (These
changes were sent to Justin Gibbs on March 28.)