Taking things somewhat out of order:

Adam Glass says:
> rarp support is used to support nfs diskless operation.

Indeed, the nfs_boot.c file is a client of the kernel rarp code.

As long as its there, do you think it would be worthwhile to add
support for sun-style autorevarp address configuration from ifconfig?
(I'm not sure, given that I've never found real use for it on a sun,
but it might make "boot this and it installs" disks easier to build in
some environments.

> > Also, in the brand new rarp support, why doesn't the
> > kernel put immediately when it hits revarpinput and there is no rarp
> > request in progress?

That should have said "punt" rather than "put". I still wonder why it
doesn't punt but instead elaborately goes through checks to see if
there is an arp request in the rarp packet (I know of no
implementations that do that!). I guess the harm isn't too great,
though, given that rarp packets are very infrequent.

Incidently, has anyone ever checked if the bpf path invoked on every
packet when you are running a rarpd on a machine causes enough
overhead to warrant another mechanism to punt rarp request packets
upstairs by a more direct mechanism?

Enough idle curiosity for today...

Perry