[ofa-general] [PATCH 2/4] ipoib: fix loss of connectivity after bonding failover on both sides

[Roland Dreier]

So I'm finally understanding this patch.  And I finally see that it is
adding a 16-byte memcpy to the data path for every packet we send.  Is
the overhead of this really negligible?  Can we think of a better way to
handle this rare failure (double failover that causes an ARP to be lost)
in a way that doesn't penalize the common datapath?

 - R.