Re: [patch net-next v2 00/10] introduce rocker switch driver with hardware accelerated datapath api - phase 1: bridge fdb offload

[John Fastabend <john.fastabend@gmail.com>]

On 11/09/2014 08:58 PM, Simon Horman wrote:
> On Sun, Nov 09, 2014 at 11:03:40PM -0500, Jamal Hadi Salim wrote:
>> Hi Simon,
>>
>> On 11/09/14 22:46, Simon Horman wrote:
>>> Hi Jamal, Hi Jiri,
>>>
>>> On a somewhat related note I am also wondering what if any progress has
>>> been made regarding discussions of (and code for) the following:
>>>
>>> 1. Exposing flow tables to user-space
>>>     - I realise that this is Open vSwitch specific to some extent
>>>       but I am in no way implying that it should be done instead of
>>>       non-Open vSwitch specific work.
>>>     - Jiri, IIRC this was part ~v2 of your earlier offload patchset
>>>
>>
>> I dont know what Rocker crowd is doing; however, I know
>> John F. has been doing some work which i have stared at
>> and I was hoping to join in with Ben's effort and show tc flow
>> offload on the realtek chip in my infinite spare time unles.
>> (for both Linux bridge and ports).
>> The priority is to merge the obvious bits first.
>
> Merging the obvious bits first is quite fine my me.
>

+1

>>> 2. Describing Switch Hardware
>>>     - I see John Fastabend moving forwards on this in his git repository
>>>       https://github.com/jrfastab/flow-net-next
>>>
>>> The way that I see things is that both of the above could be exposed via
>>> netlink. And that the first at least could be backed by NDOs.  As such I
>>> see this work as complementary and perhaps applying on top of this
>>> patchset. If I am mistaken in this regards it would be good to know :)
>>>
>>
>> You are correct - I will let John speak on his work, but
>> that is the intent.
>> The challenge is there are many schools of thoughts and i am hoping
>> it is not an arms race.
>
> That is also my hope.

My intent is to submit the Flow API bits once the base rocker switch
gets committed. I've implemented the Flow API against ixgbe and a
sadly a proprietary SDK. I'll implement it against the rocker switch
as well.

>
>>> I am of course also interested to know if the above are moving forwards.
>>> To be clear I am very interested in being able to use these APIs to
>>> perform Open vSwitch offloads and I am very happy to help.
>>> (Jamal: I'm also interested in non-Open vSwitch offloads :)
>>>
>>

I think they are moving forward. I have some code cleanup to do on
the flow API, but its mostly in place. Then I want to implement
an example on Rocker switch so we could experiment with something
why we wait for a real hardware driver. From my side assuming I at
least got it close to correct should be doable in the next few week.

After that I want to work with Jesse/Jamal and look at integrating
with OVS and other stacks. I thought a bit about the OVS integration
path but I'll hold that discussion for the moment.

Simon, if your feeling adventurous any feedback on the repo link
would be great. I still need to smash the commit log into something
coherent though at the moment you can see all the errors and rewrites,
etc as I made them.

>> Hey, OVS should be able to use these APIs; i am just interested in making
>> sure they are not just for OVS or OF. Then we are all happy;->
>
> I think we are all happy :)

I'm happy :) Also my intent is the flow API is more general then
just OVS. My view is OVS should be one user of the API.

.John

> --
> To unsubscribe from this list: send the line "unsubscribe netdev" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>


-- 
John Fastabend         Intel Corporation