According to Jonathan Stone:
> 
> Yes, its mandated by the original blue-book DIX spec.  It dates back
> to XNS, routing at the Ethernet layer, and the school that says all
> hosts should have one network address: if you have multiple MACs you
> reprogram them so that they all have the *same* MAC address.
> 
That sounds interesting.  I guess a lot of such information lurks where 
novices are unlikely to look.  I knew there was a reason for lurking on 
this list :-)

But wouldn't such a requirement throw a spanner in the works of BOOTP 
and DHCP, if a single server spans multiple networks?  Ted Lemon, your 
views?

> SunOS just used SIOSIFADDR with AF_UNSPEC to set and get raw MAC
> addresses.  However we do it, it'd be nice to be able to retrieve the
> in-ROM (or `hardware') address after setting the physical address, in
> case you ever want to set it back to the hardware, address.  (those
> are the DIX names; dunno what 802 uses.)  That means another
> ether_addr in the struct used for ethernet macs, and another ioctl
> type (or overloading the DLI stuff, maybe?).
> 
I'd suggest having a "reset/restore" function to put the firmware 
address back, and implementing the functionality in the driver, at the 
cost of some kernel bloat.  It does mean that different boards with 
similar chips may be able to share driver code, doesn't it?

++L