[ovs-dev] [PATCH 0/6 v3] OXM VLAN VID and VLAN PCP

[Simon Horman]

On Thu, Jul 19, 2012 at 04:49:29PM +0900, Simon Horman wrote:
> On Thu, Jul 19, 2012 at 12:25:30AM -0700, Ben Pfaff wrote:
> > On Wed, Jul 18, 2012 at 10:47:55AM +0900, Simon Horman wrote:
> > > the purpose of this series is to:
> > > 
> > > * Add encoding of the OXM VLAN VID and VLAN PCP matches and;
> > > * Correct the decoding of OXM VLAN VID and VLAN PCP matches
> > >   without breaking ovs-ofctl add-flows, which uses some of the
> > >   same code paths.
> > > 
> > > This series differs to v2 in that internally the CFI bit is now
> > > honoured which in turn allows it to be honoured for the OXM VLAN VID match.
> > > 
> > > Backwards compatibility for ovs-ofcfl add-flows dl_vlan and dl_vlan_pcp
> > > is provided by the second patch in this series.
> > 
> > I may be nuts, but I'm still worried about regressions, so I've sent out
> > another patch that adds even more tests:
> >         http://openvswitch.org/pipermail/dev/2012-July/019147.html
> 
> The more the merrier :)

FWIW, this series does not seem to cause the tests at the link above to
fail.

> > I'll try to review your patches against that tomorrow morning.
> 
> FWIW, I am not entirely convinced that honouring the CFI bit is
> the best idea. It implies that an OXM match should set the CFI bit
> if it wants to match packet with an 802.1Q header with the CFI bit set,
> which as I believe is generally the case if a VID is set.
> 
> However, the OpenFlow1.2 spec doesn't specify that a match needs to set the
> CFI bit in such cases. (Actually, frustratingly, it doesn't specify what to
> do with the CFI but except in the special cases of OFPVID_NONE and
> OFPVID_PRESENT).