> On Jun 6, 2014, at 10:03 PM, Daniele Di Proietto <ddiproie...@vmware.com> > wrote: > >> - Related to the above, I don't see why cloning an ofpbuf to the dpif_packet >> should be necessary. Maybe later patches get rid of the copying? > > It is necessary, for example in dpif_netdev_execute: we get an ofpbuf and we > have to pass a dpif_packet to the dpif provider, somehow. The copy would have > happened most of the times anyways, since dp_netdev_execute was called with > ‘may_steal’ set to false (it is true now).
I did not check this now, but it used to be that a normally received packet that gets a single output action, and no userspace action, need not the cloned, but the output device can consume the original packet. Jarno _______________________________________________ dev mailing list dev@openvswitch.org http://openvswitch.org/mailman/listinfo/dev
