On Thu, Jan 19, 2017 at 03:12:20PM -0800, Joe Stringer wrote:
> On 19 January 2017 at 12:17, Pravin Shelar <[email protected]> wrote:
> > On Wed, Jan 18, 2017 at 4:53 PM, Yang, Yi <[email protected]> wrote:
> >> On Wed, Jan 18, 2017 at 01:29:14PM -0800, Joe Stringer wrote:
> >>> On 18 January 2017 at 11:54, Eric Garver <[email protected]> wrote:
> >>> > On Tue, Jan 17, 2017 at 12:37:19AM +0000, Yang, Yi Y wrote:
> >>> >> What userspace do "802.1ad patches" depend on? Per Pravin's statement, 
> >>> >> we just backport 802.1ad patches to ovs, then the below patch can be 
> >>> >> applied to ovs.
> >>> >
> >>> > Userspace does not yet have support for 802.1ad. I'm still working on
> >>> > it. You can check the list archives for a recent RFC version.
> >>> >
> >>> > I don't know if it's acceptable to backport the datapath (kernel module)
> >>> > support before the userspace support is accepted. If not, you'll have to
> >>> > wait on the userspace.
> >>> > Perhaps Pravin can answer.
> >>>
> >>> IMO the general method of:
> >>> 1) Add support upstream
> >>> 2) Add userspace support
> >>> 3) Add backport
> >>>
> >>> ...works nicely because we get feedback for all interested parties for
> >>> the APIs in (1), (2) can add tests and be easily tested against a
> >>> version that works (upstream kernel) and a version that doesn't
> >>> (version in tree) to ensure both cases are handled in a reasonable
> >>> way, then (3) allows people on older kernels to gain access to the
> >>> newer features.
> >>>
> >>> That said, if other people are blocking on (3) then I think that piece
> >>> should be expedited. Let's say (2) and (3) were swapped, it just means
> >>> we need to be a bit more careful when reviewing/testing to check that
> >>> the newer userspace still handles old kernels (that lack support)
> >>> fine.
> >>>
> >>> The nice thing about getting the backport earlier is, the closer to
> >>> upstream we are, the sooner we may find issues that affect the latest
> >>> code.
> >>
> >> If so, I think the below patch is the best solution to current
> >> situation, it will be automatically overwritten once userspace part and
> >> 802.1ad backport are ready to merge.
> >>
> > It is much easier to backport and review patches when it is done in
> > same sequence as applied on upstream kernel. It is not just about this
> > patch. by changing backport order the implementation and review of
> > both L3 patch series and 802.1AD patch series becomes harder than
> > necessary.
> 
> Sorry, yes, my earlier explanation was regarding a particular feature
> and ordering of userspace implementation vs. backport patches.
> 
> With regards to the overall kernel backport, I agree the upstream
> patches should be backported in sequence according to upstream. Then
> we know we haven't missed anything. Also, dependencies from one patch
> to the next are less likely to cause trouble while rebasing, so it
> makes the backporter's life easier.

Sounds good to me. Yi is welcome to try the 802.1ad datapath backport.
Otherwise I'll get to when I can.
Thanks everyone.
_______________________________________________
dev mailing list
[email protected]
https://mail.openvswitch.org/mailman/listinfo/ovs-dev

Reply via email to