fix(ibverbs): fix integration tests failing on InfiniBand link layer

[adamcavendish]

Summary

• Fix GID filter in tests to accept GidType::InfiniBand (previously only matched RoCE GIDs, causing all IB tests to panic on unwrap())
• Add PortAttr::lid() accessor and replace hardcoded setup_dest_lid(1) with the actual port LID from query_port()
• Fixes test_qp, test_post_send, test_query_qp, and test_post_recv_errors on InfiniBand fabrics

Root Cause

The GID filter used in all integration tests:

.find(|&&gid| !gid.gid().is_unicast_link_local() || gid.gid_type() == GidType::RoceV1)
.unwrap();

This filter accepts GIDs that are either:

1. Not link-local (i.e., routable RoCEv2 GIDs), OR
2. RoCEv1 type

On InfiniBand, all GIDs are fe80:: link-local with type GidType::InfiniBand. Neither condition matches, so find() returns None and unwrap() panics.

Additionally, setup_dest_lid(1) was hardcoded, but on IB fabrics the subnet manager assigns LIDs dynamically (e.g., port_lid: 3471 on our test node).

Failure Logs (before fix)

Tested on A800 node with 4x mlx5 ConnectX-6 InfiniBand HCAs (FW 20.39.3560):

=== test_qp ===
running 1 test
qp pointer is BasicQueuePair { qp: 0x7f47ac013018 }

thread 'main' panicked at tests/test_qp.rs:66:14:
called `Option::unwrap()` on a `None` value
test main ... FAILED

test result: FAILED. 0 passed; 1 failed; 0 ignored; 0 measured; 0 filtered out

=== test_post_send ===
running 4 tests

thread 'main::case_3' panicked at tests/test_post_send.rs:108:14:
called `Option::unwrap()` on a `None` value
thread 'main::case_2' panicked at tests/test_post_send.rs:108:14:
called `Option::unwrap()` on a `None` value
thread 'main::case_1' panicked at tests/test_post_send.rs:108:14:
called `Option::unwrap()` on a `None` value
thread 'main::case_4' panicked at tests/test_post_send.rs:108:14:
called `Option::unwrap()` on a `None` value

test result: FAILED. 0 passed; 4 failed; 0 ignored; 0 measured; 0 filtered out

Unit tests in queue_pair.rs also failed at the same filter:

test ibverbs::queue_pair::tests::test_query_qp ... FAILED
test ibverbs::queue_pair::tests::test_post_recv_errors ... FAILED

thread 'ibverbs::queue_pair::tests::test_query_qp' panicked at src/ibverbs/queue_pair.rs:2610:22:
called `Option::unwrap()` on a `None` value
thread 'ibverbs::queue_pair::tests::test_post_recv_errors' panicked at src/ibverbs/queue_pair.rs:2699:22:
called `Option::unwrap()` on a `None` value

Fix

.find(|&&gid| {
    !gid.gid().is_unicast_link_local()
        || gid.gid_type() == GidType::RoceV1
        || gid.gid_type() == GidType::InfiniBand  // IB GIDs are link-local but valid
})

And replace setup_dest_lid(1) with:

let port_attr = ctx.query_port(1).unwrap();
ah_attr.setup_dest_lid(port_attr.lid())

After fix — all tests pass

=== Unit Tests ===
running 55 tests
...
test ibverbs::queue_pair::tests::test_post_recv_errors ... ok
test ibverbs::queue_pair::tests::test_query_qp ... ok

=== test_qp ===
test main ... ok
test result: ok. 1 passed; 0 failed

=== test_post_send ===
test main::case_1 ... ok
test main::case_4 ... ok
test main::case_2 ... ok
test main::case_3 ... ok
test result: ok. 4 passed; 0 failed

=== compiletest ===
test compile_test ... ok
test result: ok. 1 passed; 0 failed
=== ALL DONE ===

Test plan

• All 55 unit tests pass on mlx5 ConnectX-6 InfiniBand HCAs
• test_qp passes (QP state transitions Init→RTR→RTS)
• test_post_send all 4 cases pass (loopback send/write/inline with both basic and extended QP)
• test_cq and test_mr_and_cq still pass
• compiletest (compile-fail ownership tests) still pass

[gemini-code-assist]

Code Review

This pull request introduces a lid() method to the PortAttr struct and updates test cases in queue_pair.rs, test_post_send.rs, and test_qp.rs to use dynamic LID values and include InfiniBand GIDs in the selection criteria. The review feedback suggests further refining the GID selection logic in these tests by explicitly filtering for the correct port number to ensure reliability on multi-port systems.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

Files with missing lines	Coverage Δ
src/ibverbs/device_context.rs	81.79% <100.00%> (+0.12%)	⬆️
src/ibverbs/queue_pair.rs	88.30% <100.00%> (-0.05%)	⬇️

... and 2 files with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.