Linux Virtualization - May 2018 - [PATCH net-next v9 3/4] virtio_net: Extend virtio to use VF datapath when available

Mon, Apr 30, 2018 at 09:26:34PM CEST, sridhar.samudrala at intel.com
wrote:
>On 4/30/2018 12:12 AM, Jiri Pirko wrote:
>> Mon, Apr 30, 2018 at 05:00:33AM CEST, sridhar.samudrala at intel.com
wrote:
>> > On 4/28/2018 1:24 AM, Jiri Pirko wrote:
>> > > Fri, Apr 27, 2018 at 07:06:59PM CEST, sridhar.samudrala at
intel.com wrote:
>> > > > This patch enables virtio_net to switch over to a VF
datapath when a VF
>> > > > netdev is present with the same MAC address. It allows
live migration
>> > > > of a VM with a direct attached VF without the need to
setup a bond/team
>> > > > between a VF and virtio net device in the guest.
>> > > > 
>> > > > The hypervisor needs to enable only one datapath at any
time so that
>> > > > packets don't get looped back to the VM over the
other datapath. When a VF
>> > > Why? Both datapaths could be enabled at a time. Why the loop
on
>> > > hypervisor side would be a problem. This in not an issue for
>> > > bonding/team as well.
>> > Somehow the hypervisor needs to make sure that the
broadcasts/multicasts from the VM
>> > sent over the VF datapath don't get looped back to the VM via
the virtio-net datapth.
>> Why? Please see below.
>> 
>> 
>> > This can happen if both datapaths are enabled at the same time.
>> > 
>> > I would think this is an issue even with bonding/team as well when
virtio-net and
>> > VF are backed by the same PF.
>> > 
>> > 
>> I believe that the scenario is the same as on an ordinary nic/swich
>> network:
>> 
>> ...................
>> 
>>    host
>>         bond0
>>        /     \
>>      eth0   eth1
>>       |       |
>> ...................
>>       |       |
>>       p1      p2
>> 
>>    switch
>> 
>> ...................
>> 
>> It is perfectly valid to p1 and p2 be up and "bridged"
together. Bond
>> has to cope with loop-backed frames. "Failover driver" should
too,
>> it's the same scenario.
>
>OK. So looks like we should be able to handle this by returning
RX_HANDLER_EXACT
>for frames received on standby device when primary is present.
Yep.