[ovs-dev] [RFC PATCH 2/5] odp-util: Enable communicating outer tunnel to/from the kernel.

[Jesse Gross]

On Fri, Sep 28, 2012 at 3:44 PM, Kyle Mestery (kmestery)
<kmestery at cisco.com> wrote:
> On Sep 28, 2012, at 5:40 PM, Jesse Gross wrote:
>> On Fri, Sep 28, 2012 at 11:59 AM, Kyle Mestery (kmestery)
>> <kmestery at cisco.com> wrote:
>>> On Sep 28, 2012, at 1:12 PM, Jesse Gross wrote:
>>>> Now that both the kernel and userspace can handle information about
>>>> the tunnel outer headers this adds userspace support for communicating
>>>> between the two.  At the moment userspace doesn't know how to do
>>>> anything with the extra information on receive and will only generate
>>>> actions containing tun_id.  However, both sides know how to ignore the
>>>> extra information.
>>>>
>>>> Signed-off-by: Jesse Gross <jesse at nicira.com>
>>>
>>>
>>> The patch I just submitted overlaps with this patch a little bit. I think there wasn't
>>> a really set delineation between what I was doing and what you were doing, so
>>> we may have to merge this patch with mine. I think we both pretty much made
>>> similar changes though, so it shouldn't be too bad. Acking this patch for now,
>>> assuming we'll work through the merge bits.
>>
>> I agree the changes are pretty similar.  My inclination is that it's a
>> little easier to use the version in my patch if that's OK with you for
>> a couple of reasons:
>> * I'm not sure that the change logically belongs in the second patch
>> of your series.  It could be moved into the first patch though.
>> * There isn't much benefit to userspace supporting both models of
>> kernel tunnel implementation because the kernel will reject any
>> actions that it doesn't understand.  As a result, once this patch goes
>> in we will immediately require a new version of the kernel module.
>> Ideally I'd like to push that requirement back as late as practical so
>> that we can start applying patches without worrying about breaking
>> compatibility more than necessary.  However, when we do put it in, we
>> might as well go all the way and drop support for the old mechanism in
>> userspace.
>
> That makes sense to me. Do you want me to rebase my first patch and fold this
> one into that? I can then rework my second patch based on this. Or even, we
> could move this patch as a second patch in my series, and I could just rebase
> my third patch.

I actually don't think that much needs to be done - this patch depends
on the previous patch to do all the userspace flow changes and we
don't really want to roll all of them together.  I would just drop the
odp-util.c changes from your second patch and then your patches fit in
right before my series.  That will avoid the conflict and shouldn't
have any dependency problems.

I haven't had a chance to review your patches that closely yet but my
impression is that the bulk of the second patch doesn't have any
changes visible to userspace.  Is that right?