|WARNING| [v3] fib6: fix overflow of tbl8 indexes

[dpdklab at iol.unh.edu]

Test-Label: ai-code-review
Test-Status: WARNING
https://dpdk.org/patch/172355

_AI Code Review_

=== Patch Review: series.patch (via Claude) ===
Review date: 2026-11-03

# DPDK Patch Review

## Summary

This patch fixes overflow bugs in the FIB6 trie implementation when handling large routing tables with more than 2^23 or 2^24 tbl8 groups. The fixes are necessary and correct.

---

## ERRORS

None.

---

## WARNINGS

### Missing release notes

**File:** Not included in patch  
**Issue:** Patch fixes significant correctness bugs (crashes and incorrect routing behavior) but does not update release notes.

**Details:** The patch description documents two distinct failure modes:
1. SIGSEGV crash after ~640,000 routes when tbl8 group number reaches 2^23
2. Wrong routing entries written after 2^24 tbl8 groups, causing lookups to return default next hop

These are user-visible correctness bugs in a library API. According to guidelines, "Changes to API without release notes" and "Release notes are NOT required for: Internal API and helper functions (not exported to applications)" - since this is a library fix affecting application-visible behavior, release notes documenting the bug fix should be included.

**Suggested fix:** Add a "Bug Fixes" entry in `doc/guides/rel_notes/release_26_11.rst`:

```rst
* **fib: Fixed tbl8 index overflow in IPv6 FIB.**

  Fixed integer overflow and sign extension bugs that caused crashes or incorrect
  routing behavior in IPv6 FIB tables with more than 8 million tbl8 groups.
```

---

## INFO

### Clean correctness fix

The patch correctly identifies and fixes the root causes:
- Sign extension of negative int when tbl8 group >= 2^23 (line 330: `prev_idx = (uint64_t)tbl8_idx`)
- Truncation of 64-bit index to 32-bit return type in `get_idx()` when tbl8 group >= 2^24 (line 260-261: changed return type and `prev_idx` parameter to `uint64_t`)
- Integer overflow in shift before widening (lines 334, 365: cast to `uint64_t` before shifting)

All affected functions (`get_idx`, `build_common_root`, `install_to_dp`, `get_tbl_val_by_idx`, `get_tbl_pos`, `get_psd_idx`) now use `uint64_t` for table indices consistently.

The separate `tbl8_idx` variable (line 322) correctly remains signed int to hold the negative error return from `tbl8_alloc()`, and is cast to `uint64_t` only when used in the non-error path.

---

Review-Result: WARNING