On Sun, Apr 09, 2006 at 07:58:41AM +0200, Ignatios Souvatzis wrote:
> On Fri, Apr 07, 2006 at 08:27:57PM +0200, Konstantin KABASSANOV wrote:
> 
> > I understand (I hope ;)) your viewpoint, but I'm not sure that it is normal
> > to accept packets with the same "unicast" source and destination Ethernet
> > addresses... Multicast and broadcast Ethernet traffic is not part of our
> > discussion...
> 
> Sure it is. Some Ethernet hardware don't provide a copy of their sent packet
> to the reception hardware, and rely on the receiver to pick it up from the
> cable in case it's needed for monitoring (we're talking about 10base-5/2
> here).

It is all a question of whether an ethernet interface should receive
packets that sends if it would receive them if another system sent them.
(this is close to a bikeshed requirement...)

On a traditional HDX ethernet link, the receive enable (RENA) signal
is required to be asserted by the transceiver back to the MAC unit
(part of the cable test).  The RX data is also very likely to be present,
and a chipset that is capable of FDX operation may receive the packet.

Some chipsets are known to RX the data at 10M, but ignore it at 100M,
even though no control registers need to be changed to switch between
the 2 speeds - indeed in can be difficult for the driver to determine
the speed.

If the OS expects transmitted frames to be receivable, it may have to
be done by the device driver.  For the chip above we had to do that
and filter out RX traffic that had the local source MAC address.

	David

-- 
David Laight: david@l8s.co.uk