[ovs-dev] Support for in_phy_port

Ben Pfaff blp at nicira.com
Wed Dec 4 16:18:22 UTC 2013


On Wed, Dec 04, 2013 at 06:13:58PM +0900, Simon Horman wrote:
> On Wed, Dec 04, 2013 at 06:08:17PM +0900, Simon Horman wrote:
> > On Mon, Dec 02, 2013 at 01:21:28PM -0800, Ben Pfaff wrote:
> > > On Mon, Dec 02, 2013 at 11:41:11AM +0900, Simon Horman wrote:
> > > > I am wondering if you would like me to add support for matching on
> > > > in_phy_port, which appears to be optional.  I am quite happy to do so, and
> > > > indeed I have most of the pieces in place to do so. However if not I wonder
> > > > if there is anything much to be done at all as the spec states that
> > > > in_phy_port may be omitted if it is the same as phy_port: our current plan
> > > > is for that to always be the case.
> > > 
> > > Hmm.
> > > 
> > > I guess there is nothing to do for OF1.2+ then.
> > > 
> > > I checked what we did for OF1.1, and the answer was that we had never
> > > implemented packet-in at all for OF1.1 (!).  I sent out a pair of
> > > patches to fix that:
> > >         http://openvswitch.org/pipermail/dev/2013-December/034461.html
> > >         http://openvswitch.org/pipermail/dev/2013-December/034462.html
> > > 
> > > I guess no one has tested OF1.1 support yet.  Not too surprising,
> > > since it's pretty recent in OVS and hardly used at all in the wild.
> > 
> > No not surprising, other than that no one noticed.
> > 
> > What is your feeling on what is required for in_phy_port on OF1.1?
> > 
> > Mine is that OF1.1 stipulates that in_port must be the port used for matching
> > and must be available to OpenFlow processing. My reading is that means
> > that we don't have to make in_phy_port available as a match and
> > moreover we probably don't want to at this stage. So it seems to me
> > that if/when OF1.1 packet_in support is implemented it will
> > be sufficient just to set in_phy_port to in_port when creating
> > the OF1.1 packet_in message: I imagine this will be trivial
> > a trivial part of the implementation.
> 
> ... I now see that you did that. So I think we can remove
> the in_phy_port item in OPENFLOW-1.1+ if your patches are merged.
> Or at the very least update it to say that there is a minimal
> implementation in place and something more sophisticated seems
> to be of dubious value.

I think we can just remove it.  I folded that into the second patch,
thanks for reminding me.



More information about the dev mailing list