|WARNING| net/vhost: preserve Tx ownership on VLAN failure

[dpdklab at iol.unh.edu]

Test-Label: ai-code-review
Test-Status: WARNING
https://dpdk.org/patch/166940

_AI Code Review_

=== Patch Review: series.patch (via Claude) ===
Review date: 2026-07-26

# DPDK Patch Review: net/vhost: preserve Tx ownership on VLAN failure

## Errors

### 1. Incorrect missed packet accounting
**Issue:** The missed packet count calculation is wrong.

The line `nb_missed = nb_send;` assigns the total number of packets prepared for sending to the missed count, rather than the number that failed to be enqueued by vhost.

**Why it matters:** This produces incorrect statistics -- packets successfully sent to vhost will be reported as missed.

**Fix:**
```c
nb_missed = nb_send - nb_tx;
```

The missed count should be the packets that were prepared (`nb_send`) minus those actually accepted by vhost (`nb_tx`).

---

## Warnings

### 1. Missing release notes entry
**Issue:** The patch fixes a user-visible bug in Tx burst semantics but does not update the release notes.

**Fix:** Add an entry to `doc/guides/rel_notes/release_26_07.rst` (or the appropriate current release file) under "Fixed Issues in net/vhost PMD" describing the ownership and statistics fixes.

### 2. Potential correctness verification needed
**Issue:** The change from `bufs[nb_send] = m` to `bufs[i] = m` assumes the preprocessed mbufs should remain in-place rather than being compacted.

**Context:** The original code compacted the array by incrementing `nb_send` only for successfully preprocessed packets. The new code keeps preprocessed packets at their original indices but stops at the first failure.

**Verification:** Confirm that leaving gaps in the array (if VLAN insertion fails on packet 0 but succeeds on later packets in a future modification) would not break the vhost enqueue loop. The current patch is safe because it breaks immediately on first failure, but the in-place write pattern differs from the original intent.

This is acceptable given the stated goal of preserving caller ownership, but worth noting for future maintainers.

---

## Summary

The patch correctly addresses the ownership issue described in the commit message: when VLAN insertion fails, the mbuf and all subsequent mbufs now remain owned by the caller, allowing proper cleanup.

However, the missed packet accounting is incorrect -- it counts all prepared packets as missed instead of only those not accepted by vhost. This must be fixed before merge.