Simplify length computation for INT report encap headers

[ccascone]

In the current implementation, we compute the INT encap headers length (report_ipv4.total_len and report_udp.len) by using the frame length (intrinsic metadata), and by subtracting from it every possible header that we strip in the INT mirror parser, such as MPLS, VLAN, GTP-U, etc. As the list of stripped headers grows (adding more in #224 and #251) we try to simplify the logic.

In this PR we propose an alternative approach: instead of subtracting from the frame length, we add to the inner ipv4 length. Different from before, the addends are constant and do not depend on the parsed headers (i.e., no more stripped_* metadata). This is based o the assumption that we always strip all headers between Ethernet and IPv4 (inner one in case of GTP-U).

Moreover, we revisit the initialization of report headers to slightly reduce PHV utilization at the cost of T-PHV:

• Set all constant fields in the INT mirror parser to allow T-PHV allocation;
• Set variable fields in the pipe (to avoid redundant PHV initialization in the parser;
• Avoid using hdr = {...} which sets validity bit to 1 and creates some unexpected behaviors in the pipe (likely a compiler bug).

[ccascone]

test.FabricIntLocalReportTest fails with is_device_spine=True and send_report_to_spine=True. There are some issues with popping the MPLS label for the report packet after recirculation. fixed.

[codecov]

Codecov Report

Merging #253 (a28a1ca) into main (5dcbfa6) will not change coverage.
The diff coverage is n/a.

❗ Current head a28a1ca differs from pull request most recent head 7d56850. Consider uploading reports for the commit 7d56850 to get more accurate results

@@            Coverage Diff            @@
##               main     #253   +/-   ##
=========================================
  Coverage     74.11%   74.11%           
  Complexity      273      273           
=========================================
  Files            18       18           
  Lines          1866     1866           
  Branches        153      153           
=========================================
  Hits           1383     1383           
  Misses          399      399           
  Partials         84       84           

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 5dcbfa6...7d56850. Read the comment docs.

[Yi-Tseng]

LGTM overall, only one comment

[osinstom]

LGTM, but I agree with @Yi-Tseng to consider adding @padding for the pad field. The other concern I have is that we'll have much code related to the IPv6 support commented out, and it may make code less "clean". I'd suggest adding FIXME to each occurrence of IPv6-related code and even add a keyword to correlate these comments (e.g. FIXME (ipv6-support)).

[ccascone]

I removed the commented out ipv6 code, there's not much value in keeping it there. As for #227, support for generating INT reports for IPv6 packets was untested and without a clear use case. If we need it, we can add the parser logic again together with PTF tests.