[ovs-dev] [PATCH V3 00/12] netdev datapath offload: Support IPv6 and VXLAN encap

[Ilya Maximets]

On 6/24/20 4:16 PM, Ilya Maximets wrote:
> On 6/24/20 12:47 PM, Eli Britstein wrote:
>>
>> On 6/24/2020 1:37 PM, Ilya Maximets wrote:
>>> On 6/21/20 1:19 PM, Eli Britstein wrote:
>>>> This patch set includes additional offloads - IPv6 and VXLAN encap, and
>>>> enhanced logging to increase debugability.
>>>>
>>>> Patches #1-#8:   Add support for offloads of IPv6 patterns, partial
>>>>                   TCP/UDP ports, set IPv6 and encap actions
>>>>                   (clone/output).
>>>> Patch #9:        Bug fix of partial offloads.
>>>> Patches #10-#11: Enhance log prints for debugability.
>>>> Patch #12:       Fix Ethernet matching for type only.
>>>>
>>>> v2-v1:
>>>> - Removed redundant out label.
>>>> - Added a patch to fix dl_type match only.
>>>> v3-v2:
>>>> - Rebased, and more elaboration in #7 commit message.
>>> Hi.
>>>
>>> I noticed that you didn't include Acked-by tags from Harsha.
>>> Was it intentional? i.e. if there was any significant changes during
>>> rebase process?
>>>
>>> I'd like to have these patches acked by Harsha before applying, so
>>> my question if tags from v2 are valid for v3 and can I use them?
>>> Harsha?  Eli?
>>
>> No. It was by mistake. Sorry. The rebase had only one minor conflict in
>>
>> [PATCH V3 06/12] netdev-offload: Use dpif type instead of class.
>>
>> The conflict was due to previous commit [1].
>>
>> As I will have to rebase again for sure to add testpmd prints for VLANs, as [2] was merged, will it be OK I'll add the Acked-by then?
> 
> Sure.  I think it should be OK for patches without any functional
> changes.  Current patches doesn't apply starting from the second one
> due to merged patch [2], so they need a rebase anyway.
> 
> Regarding the testpmd related patch itself, I'm not sure about it.
> It might be hard to support in terms that we will need to track
> testpmd changes and it's also hard to tell for which testpmd version
> these cmdlines should be applicable.
> 
> Another point is that it's actually printing of the same information
> twice.  And also lots of extra printing code.
> Maybe, if you need these cmdlines, it's better to write a helper script
> to convert existing output to testpmd cmdlines?  Or even parse offloaded
> and non-offloaded flows from the output of dump-flows.

Anoher option is to just drop current format entirely and use format
of testpmd cmdlines instead.  Current format is not standardized and
doesn't comply with OVS flow dumps format anyway.

What do you think?

> 
>>
>> [1] 191536574e3b ("netdev-offload: Implement terse dump support").
>>
>> [2] 029273855939 ("netdev-offload-dpdk: Support offload of VLAN PUSH/POP actions.")
>>
>>>
>>> Best regards, Ilya Maximets.
>>>
>>>> Travis:
>>>> v1: https://eur03.safelinks.protection.outlook.com/?url=https%3A%2F%2Ftravis-ci.org%2Fgithub%2Felibritstein%2FOVS%2Fbuilds%2F688413350&data=02%7C01%7Celibr%40mellanox.com%7C6f3ba17b2ba0437c191708d8182aa79e%7Ca652971c7d2e4d9ba6a4d149256f461b%7C0%7C0%7C637285918771531067&sdata=PjzU%2Fiwtc7nhzTJ99oELQhi4lDyZqDwHwGuId27W4dc%3D&reserved=0
>>>> v2: https://eur03.safelinks.protection.outlook.com/?url=https%3A%2F%2Ftravis-ci.org%2Fgithub%2Felibritstein%2FOVS%2Fbuilds%2F691375847&data=02%7C01%7Celibr%40mellanox.com%7C6f3ba17b2ba0437c191708d8182aa79e%7Ca652971c7d2e4d9ba6a4d149256f461b%7C0%7C0%7C637285918771531067&sdata=SoYpbcTbkl872Mz9q%2Fg%2FrBLBOv37qCYwEXXJ%2BcXblj4%3D&reserved=0
>>>> v3: https://eur03.safelinks.protection.outlook.com/?url=https%3A%2F%2Ftravis-ci.org%2Fgithub%2Felibritstein%2FOVS%2Fbuilds%2F700534550&data=02%7C01%7Celibr%40mellanox.com%7C6f3ba17b2ba0437c191708d8182aa79e%7Ca652971c7d2e4d9ba6a4d149256f461b%7C0%7C0%7C637285918771531067&sdata=2ZIo6xz56YdJWFxuQO9iqTygcLnwPuhbGo28m3y8EC0%3D&reserved=0
>>>>
>>>>
>>>> Eli Britstein (10):
>>>>    netdev-offload-dpdk: Remove pre-validate of patterns function
>>>>    netdev-offload-dpdk: Add IPv6 pattern matching
>>>>    netdev-offload-dpdk: Support offload of set IPv6 actions
>>>>    netdev-offload-dpdk: Support partial TCP/UDP port matching
>>>>    netdev-offload-dpdk: Support offload of clone tnl_push/output actions
>>>>    netdev-offload-dpdk: Support tnl/push using vxlan encap attribute
>>>>    dpif-netdev: Don't use zero flow mark
>>>>    dpif-netdev: Add mega ufid in flow add log
>>>>    netdev-offload-dpdk: Add testpmd log commands
>>>>    netdev-offload-dpdk: Fix Ethernet matching for type only
>>>>
>>>> Ilya Maximets (2):
>>>>    netdev: Allow storing dpif type into netdev structure.
>>>>    netdev-offload: Use dpif type instead of class.
>>>>
>>>>   Documentation/howto/dpdk.rst  |   4 +-
>>>>   NEWS                          |   3 +
>>>>   lib/dpif-netdev.c             |  26 +-
>>>>   lib/dpif-netlink.c            |  23 +-
>>>>   lib/dpif.c                    |  21 +-
>>>>   lib/netdev-offload-dpdk.c     | 649 +++++++++++++++++++++++++++++++++++-------
>>>>   lib/netdev-offload-tc.c       |   3 +-
>>>>   lib/netdev-offload.c          |  52 ++--
>>>>   lib/netdev-offload.h          |  16 +-
>>>>   lib/netdev-provider.h         |   3 +-
>>>>   lib/netdev.c                  |  16 ++
>>>>   lib/netdev.h                  |   2 +
>>>>   ofproto/ofproto-dpif-upcall.c |   5 +-
>>>>   tests/dpif-netdev.at          |  20 +-
>>>>   tests/ofproto-macros.at       |   3 +-
>>>>   15 files changed, 657 insertions(+), 189 deletions(-)
>>>>
>