On 2/9/2012 4:39 PM, Sridhar Samudrala wrote:

On Thu, 2012-02-09 at 12:30 -0800, John Fastabend wrote:

quoted

On 2/9/2012 10:14 AM, Sridhar Samudrala wrote:

quoted

On Wed, 2012-02-08 at 19:22 -0800, John Fastabend wrote:

quoted

Propagate software FDB table into hardware uc, mc lists when
the NETIF_F_HW_FDB is set.

This resolves the case below where an embedded switch is used
in hardware to do inter-VF or VF-PF switching. This patch
pushes the FDB entry (specifically the MAC address) into the
embedded switch with dev_add_uc and dev_add_mc so the switch
"learns" about the software bridge.


          veth0  veth2
            |      |
          ------------
          |  bridge0 |   <---- software bridging
          ------------
               /
               /
  ethx.y      ethx
    VF         PF
     \         \          <---- propagate FDB entries to HW
     \         \
  --------------------
  |  Embedded Bridge |    <---- hardware offloaded switching
  --------------------

This scenario works now as adding an interface to a bridge puts it in
promiscuous mode. So adding a PF to a software bridge should not be
a problem as it supports promiscuous mode. But adding a VF will not
work.

It shouldn't work because the embedded bridge will lookup the address
in its FDB and when it doesn't match any unicast filters it will forward
the packet onto the wire. Because the veth0 and veth2 above never get
inserted into the embedded brdige's FDB the packets will _never_ get
routed there.

That said the current 'ixgbe' driver is doing something broken in that
it is always setting the unicast hash table and mirroring bits to 1. So
if you think this is working your seeing a bug where packets are being
sent onto the wire AND upto the PF. Packets with destination addresses
matching veth1 should not end up on the wire and vice versa. This is
specific to ixgbe and is not the case for other SR-IOV devices.

OK. Is this behavior going to be fixed.

Only after we have a mechanism to either configure the NIC FDB directly
or have it stay in sync with the SW switch. Flooding traffic seems better
than being unable to send traffic to the virtual device altogether. This
behavior is driver specific some devices just fail outright.

I'm thinking over Jamal's comment now.

quoted

This causes some issues (a) has some very real performance implications,
(b) at this point you have some strange behavior from my point of view.
The embedded bridge is not a learning bridge nor is it acting like an
802.1Q VEB or VEPA.

quoted

Are you trying to avoid the requirement of having to put the interface 
in promiscuous mode when adding to a bridge?

I think the bridge being in promiscuous mode is correct.

The interface that is added to the bridge is put in promiscuous mode,
not the bridge itself.  In this example, i assumed that setting
promiscuous on PF is putting the embedded bridge in learning mode.

Yes I misspoke I mean the PF. The embedded bridge in this case does not
support learning. Also I'm not aware of any SR-IOV NICs that do support
learning.

Thanks
Sridhar

quoted

Hope that helps sorry its a bit long winded.
John

--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html