> > I thought of creating one "dummy netdev" per function which will represent > > each function in open vswitch. These ports can be treated like logical > > connection to the functions. Whenever there is flow-miss in the device, it > will > > punt the packet to the linux kernel via respective "dummy netdev" and get > > the flow entry via control path. > > Yeah, this would be a good approach, it would also need a send and receive > function > allow sending packets to the slow path. >
You might start by looking at the Rocker switch implementation which creates multiple netdev devices for each port of the switch. In your case you'd likely create these dummy netdev definitions within you PF driver. My assumption is that ALL packets would start as a flow-miss when the device comes up, even broadcast/multicast traffic, and the packets would be forwarded to OVS for handling. I don't believe OVS has any awareness of switchdev yet so no offload is currently possible using ndo_fdb_add/del. Dave _______________________________________________ dev mailing list dev@openvswitch.org http://openvswitch.org/mailman/listinfo/dev
