This is good stuff. A couple of thoughts:

- The device namespace is supposed to be the same as the isapnp
  namespace. Would it make sense to use the device names/descriptions
  from isapnpdevs? Then instead of output like this (from my laptop):

  com3 at pnpbios0 (PNP0501), io 3f8-3ff, irq 4
  ...
  PNP0510 (io 2e8-2ef, irq 3) at pnpbios0 ignored

  we could have the more informative:

  com3 at pnpbios0 (Generic 16550A), io 3f8-3ff, irq 4
  ...
  Generic IRDA-compatible device (io 2e8-2ef, irq 3) at pnpbios ignored

  (seems like that should be "not configured" for consistency with
  other busses).


- This is clearly good information for finding out what resources to
  avoid for dynamic device configuration. However, there are some
  things found by this that aren't explicitly part of the autoconf
  mechanisim - the ISA DMA controller and real-time clock, for
  example. We need some way to keep dynamic devices away from these
  resources without causing problems for the code that actually does
  map them.

- I suspect that some of this stuff is going to get us back into
  needing more sophisticated config_defer() mechanics. 

Thanks for doing this work!

        - Nathan