Add XDP hardware offload support for certain Mizar functions with Netronome NIC

[yangpenger]

What type of PR is this?

/kind design

What this PR does / why we need it:

This PR can offload some functions in trn_transit_xdp.c on Netronome NIC.

Which issue(s) this PR fixes:

Not fixes any issue.

Special notes for your reviewer:

The deployment of offloaded XDP only works when detecting a supported interface. Otherwise, offloaded XDP will not be deployed and Mizar works as normal. To make no effect on Mizar, we only add codes and do not change any logic of Mizar.

Does this PR introduce a user-facing change?:

In deploy.mizar.yaml, we support one env variable named "OFFLOAD_XDP" for offloaded XDP on and off.

[vinaykul]

Please look into the unit test failures. https://github.com/CentaurusInfra/mizar/runs/7871759634?check_suite_focus=true

You can run 'make unittest' and then run './build/tests/test_dmn' to repro the issue and investigate failures.

[yangpenger]

It is our first time contributing to an open source project, so some stupid mistakes may exist in our codes. Anyway, thanks for the review @vinaykul . I have addressed some problems, but I think we need discussions for some details in Mizar's Weekly Open Source Meeting if necessary.

[vinaykul]

Thanks for the updates. As discussed in the community meeting yesterday, I'll look forward to fixes and updates that will hopefully be able to get rid of the new CLI extensions and use the flag Phu has added to load the new XDP offload program on Netronome NIC while running the current transit XDP program on the same NIC in generic mode. I'll review the code from the top once we have those changes.

[vinaykul]

It would also be helpful to capture in the PR description the different functions that are being offloaded. It is not clear whether bouncer or divider or both of those are being offloaded. Could you also please add the reasons for the prefix length in the offloaded code used in the offload map lookup keys? Could you also please add a list of all the limitations we encountered with Netronome in the issue description?

I think we also need an E2E test for this code.

[vinaykul]

Could we also get performance numbers of the offloaded eBPF code vs generic eBPF code?

[vinaykul]

/hold

[vinaykul]

LGTM We can merge this pending Phu's LGTM

[xieus]

@yangpenger Can you update your PR to address Phu's comments, mostly minor? We would like to move this forward quickly. Thanks.

[xieus]

@vinaykul @phudtran Pls take another quick look and merge if all good :-)

[phudtran]

Thanks LGTM.