|WARNING| net/mlx5: fix double free in vectorized Rx recovery

[dpdklab at iol.unh.edu]

Test-Label: ai-code-review
Test-Status: WARNING
https://dpdk.org/patch/165609

_AI Code Review_

=== Patch Review: series.patch (via Claude) ===
Review date: 2026-06-17

# DPDK Patch Review

## Summary
This patch fixes a critical double-free/use-after-free bug in the mlx5 vectorized Rx error recovery path. The fix is correct and addresses a serious memory safety issue.

---

## Errors

None. The patch correctly fixes the bug it describes.

---

## Warnings

### Missing test coverage
The patch fixes a bug in an error path (mempool exhaustion during Rx queue recovery) but does not add a test case to verify the fix or prevent regression. Consider adding a unit test or functional test that exercises the error recovery path with a depleted mempool.

---

## Info

### Excellent commit message
The commit message clearly explains:
1. The bug mechanism (wrong bitmask in rollback loop)
2. Why it's wrong (isolates a single bit instead of wrapping index)
3. The consequences (leak new mbufs, double-free live mbuf and fake_mbuf)
4. The fix (use existing e_mask variable)

This is a model commit message for a correctness bug fix.

### Code review: the fix is correct
- The forward allocation loop uses `(elts_ci + i) & e_mask` (line not shown but implied by patch context)
- The rollback loop previously used `(elts_ci + i) & elts_n` -- **wrong**: `elts_n` is a power of 2, so this yields only 0 or `elts_n`
- The patch corrects it to `(elts_ci + i) & e_mask` where `e_mask = elts_n - 1` (correct power-of-2 ring buffer index masking)
- This matches the pattern used elsewhere in the function

### Appropriate for stable backport
The patch includes `Cc: stable at dpdk.org` and references a commit from 2019, making it a good candidate for LTS backports.

---

## Conclusion

**LGTM with suggestion to add test coverage.**

The patch is correct and fixes a serious memory safety bug (use-after-free + double-free + leak). No code style or API issues. The only improvement would be adding a test to verify the fix, but that is not a blocking issue for merging this important fix.