>> 	isn't it better to make it to always behave as IFF_SIMPLEX, by
>> 	dropping multicast packet from itself?  see smc91cxx.c.
>
>I'm not sure. It would add an ether_cmp() to a criticial path, but unlike
>the smc91cxx.c case not only in promiscuous mode. Maybe it should only check
>that if we are in half-duplex mode?
>
>But on the other hand, the current solution does not cause any performance
>hit and works - what's the problem with it?

	the problem is that !IFF_SIMPLEX interfaces are not really supported in
	L3 logic.  we need to look at that (IPv4/v6 multicast, IPv6 ND, and
	such).  i don't think i can make IPv6 DAD work with !IFF_SIMPLEX
	interfaces, due to the way IPv6 DAD is specified.

itojun