docs(spec): English version of Stage-6 Phase-7 integration review (FU-U-7)

Translates the zh_cn Phase-7 review (FU-CB-DPDKIF-INTEGRATION) and
places it at docs/unit_test_spec/ root, completing the English spec set:

  00-overview-and-glossary.md
  01-requirements-spec.md
  02-current-architecture-and-targets.md
  04-cmocka-framework-and-impl.md
  06-test-cases-and-acceptance.md
  99-review-report.md
  99-stage6-coverage-boost-review.md
  99-stage6-phase7-integration-review.md  (NEW)

Tables, IDs (G-CB-* / FU-CB-*), file paths, command snippets, and
section numbering are preserved verbatim.

Author: unit-test-impl-leader

docs/unit_test_spec/99-stage6-phase7-integration-review.md

@@ -0,0 +1,158 @@
+# Stage-6 Phase-7 Review — FU-CB-DPDKIF-INTEGRATION
+
+**Date**: 2026-06-10
+**Local commit**: (see git log)
+**Scope**: Stage-6 deferred gates G-CB-3 / G-CB-4
+
+---
+
+## 1. Background
+
+After Stage-6 Phase 5 lifted `ff_dpdk_if.c` from 3.2% to 5.1% line coverage, the remaining ~970 lines (the `ff_dpdk_init` / `ff_dpdk_run` / `ff_dpdk_if_send` body) were unreachable inside the unit-test boundary because they require a real DPDK ethdev plus a live main-loop runtime. At the close of Phase 5 the gates G-CB-3 (`ff_dpdk_if` line ≥25%) and G-CB-4 (project line ≥50%) were marked deferred and filed as follow-up `FU-CB-DPDKIF-INTEGRATION`.
+
+This phase (Phase 7) implements that follow-up: a standalone integration-test binary under `tests/integration/` that boots a real EAL with `--vdev=net_null0` and lets `ff_dpdk_init()` run end-to-end, exercising the six core init subsystems — `init_lcore_conf` / `init_mem_pool` / `init_dispatch_ring` / `init_msg_ring` / `init_port_start` / `init_clock`.
+
+---
+
+## 2. Deliverables
+
+| File | Type | Purpose |
+|---|---|---|
+| `tests/integration/Makefile` | NEW | Standalone build system (mirrors unit/Makefile) |
+| `tests/integration/test_ff_dpdk_if_integration.c` | NEW | 7 TCs, cmocka style |
+| `tests/integration/common/rte_stub.{c,h}` | COPY | Reuses unit's rte_exit / rte_panic wraps |
+| `tests/integration/valgrind.supp` | NEW | EAL init-time still-reachable suppression (does NOT mask definite leaks) |
+| `tests/run_full_coverage.sh` | NEW | One-shot script that merges unit + integration coverage |
+
+---
+
+## 3. Test Matrix (7 TCs)
+
+| TC ID | Name | Verifies |
+|---|---|---|
+| INT-DPDKIF-01 | init_succeeded_with_one_port | `nb_dev_ports==1`; `rte_eth_link_get_nowait` returns sane |
+| INT-DPDKIF-02 | eth_dev_socket_id_post_init | `rte_eth_dev_socket_id` does not crash |
+| INT-DPDKIF-03 | ff_dpdk_register_deregister_roundtrip | register returns non-NULL ctx; deregister leak-free |
+| INT-DPDKIF-04 | ff_get_tsc_ns_monotonic | Monotonically non-decreasing |
+| INT-DPDKIF-05 | ff_get_traffic_post_init | Smoke: function executes without crash |
+| INT-DPDKIF-06 | ff_dpdk_if_send_zero_total | `total=0` path |
+| INT-DPDKIF-07 | eal_process_type_primary | `rte_eal_process_type()==PRIMARY` |
+
+**Result**: 7/7 PASS / 0 SKIP / 0 FAIL / runtime 0.4 s + valgrind 2.3 s
+
+---
+
+## 4. Coverage Outcome (merged unit + integration)
+
+### Per-file (merged)
+
+| File | unit-only line | merged line | Δ |
+|---|---|---|---|
+| ff_dpdk_if.c | 5.1% | **30.5%** | **+25.4pp** |
+| ff_log.c | 100.0% | 100.0% | 0 |
+| ff_ini_parser.c | 97.3% | 97.3% | 0 |
+| ff_host_interface.c | 100.0% | 100.0% | 0 |
+| ff_epoll.c | 100.0% | 100.0% | 0 |
+| ff_config.c | 76.3% | 76.3% | 0 |
+| ff_thread.c | 100.0% | 100.0% | 0 |
+| ff_init.c | 100.0% | 100.0% | 0 |
+| ff_dpdk_pcap.c | 100.0% | 100.0% | 0 |
+| ff_dpdk_kni.c | 47.2% | 47.2% | 0 |
+
+### Project-level (merged)
+
+| Metric | Stage-6 Phase 6 final | Stage-6 Phase 7 final | Δ |
+|---|---|---|---|
+| **line cov** | 47.1% | **58.3%** | **+11.2pp** |
+| **branch cov** | 47.2% | **54.5%** | **+7.3pp** |
+| **func cov** | 63.9% | **72.9%** | **+9.0pp** |
+| TC count | 130 | **137** | +7 |
+
+---
+
+## 5. Stage-6 G-CB Gates — All Closed
+
+| Gate | Target | Phase 6 status | Phase 7 measured | Status |
+|---|---|---|---|---|
+| G-CB-0 | gap-analysis complete | ✅ PASS | — | ✅ |
+| G-CB-1 | 8 files line ≥85% / branch ≥70% | ✅ PASS | — | ✅ |
+| G-CB-2 | ff_dpdk_kni line ≥40% | ✅ PASS (47.2%) | — | ✅ |
+| **G-CB-3** | **ff_dpdk_if line ≥25%** | ⚠ deferred (5.1%) | **30.5%** | ✅ **PASS** |
+| **G-CB-4** | **project line ≥50%** | ⚠ deferred (47.1%) | **58.3%** | ✅ **PASS** |
+| G-CB-5 | lib safe-patches ≤5 sites | 0 used | 0 used | ✅ |
+| G-CB-6 | reviewer 4-dim score ≥A | ✅ PASS | — | ✅ |
+
+---
+
+## 6. Key Design Decisions
+
+### 6.1 cmocka group_setup boots EAL once
+- Same EAL bring-up flags as `test_ff_dpdk_kni` (`--no-huge --no-pci --no-shconf --vdev=net_null0 -l 0 -m 64`)
+- If group_setup fails, all 7 TCs `skip()` rather than failing the build
+- `rte_eal_cleanup` is intentionally NOT called: cmocka TCs may still hold lcore TLS references; the OS reclaims on process exit
+
+### 6.2 Build ff_global_cfg programmatically (no ff_load_config)
+- Direct `memset(&ff_global_cfg, 0, ...)` plus targeted field assignment
+- Required fields: `dpdk.{nb_procs, proc_id, proc_lcore, nb_ports, max_portid, portid_list, port_cfgs[i].{port_id, nb_lcores, lcore_list[0]}}` plus `freebsd.hz` (otherwise `init_clock` divides by zero)
+- More compact than INI fixtures and decouples cfg state from `ff_dpdk_init`'s parsing path
+
+### 6.3 Bridge-stub strategy
+- 25 ff_* / kernel-side bridge functions (`ff_veth_*` / `ff_mbuf_*` / `ff_close` / `ff_rtioctl` etc.) — reuse the stub bodies from `unit/test_ff_dpdk_if`
+- Link real `ff_log.o` / `ff_config.o` / `ff_ini_parser.o` / `ff_host_interface.o` (no bridge needed)
+
+### 6.4 NULL ctx TC promoted to follow-up (FU-CB-DPDKIF-NULLGUARD)
+- `ff_dpdk_if_send` does not guard NULL ctx; dereferencing `ctx->port_id` segfaults
+- Replaced with the legal `total=0` path; the NULL guard is parked under the Stage-6 lib safe-patch capacity for later consumption
+
+---
+
+## 7. New Follow-ups Logged
+
+| ID | Description | Priority | Notes |
+|---|---|---|---|
+| **FU-CB-DPDKIF-NULLGUARD** | Add NULL ctx guard at the entry of `ff_dpdk_if_send` (safe-patch ≤5 lines) | P3 | Stage-6 lib safe-patch capacity still untriggered |
+| FU-CB-INT-MORE-TC | Extend integration suite: rss_tbl_init / promiscuous / link-state combinations | P3 | Headroom beyond current 30.5% |
+| FU-CB-FULL-COV-CI | Wire `run_full_coverage.sh` into CI to prevent regression | P2 | Merges with Phase 6 review item D2 |
+
+---
+
+## 8. Engineering Compliance
+
+- ✅ 0 direct rm / kill / chmod (script +x via `chmod_modify.sh`)
+- ✅ Temporary `.info` / `coverage_report` cleanup via `rm_tmp_file.sh` wrapper
+- ✅ valgrind: 0 definite leaks (only EAL still-reachable suppressed)
+- ✅ ahead-of-upstream feature/1.26 counter updated
+
+---
+
+## 9. One-Shot Reproducibility
+
+```bash
+cd /data/workspace/f-stack/tests
+./run_full_coverage.sh
+# Output:
+#   per-file merged line/branch
+#   project Summary: line 58.3% / branch 54.5% / func 72.9%
+#   G-CB-3 ff_dpdk_if line 30.5% [PASS]
+#   G-CB-4 project    line 58.3% [PASS]
+#   merged HTML at tests/full_coverage_report/index.html
+```
+
+```bash
+cd /data/workspace/f-stack/tests/integration
+make             # build only
+make test        # 7/7 PASS / ~0.4s
+make check       # valgrind / 0 definite leak / ~2.3s
+make coverage    # standalone integration coverage
+```
+
+---
+
+## 10. 4-Dimension Score
+
+| Dimension | Score | Note |
+|---|---|---|
+| Coverage gain | **A+** | Project line +11.2pp / ff_dpdk_if +25.4pp — beyond target |
+| Boundary completeness | A | 7 TCs cover legal and abnormal paths; NULL guard deferred to follow-up |
+| Lib-patch safety | A+ | 0 lib changes (full capacity preserved) |
+| Valgrind no-regression | A+ | 0 definite leaks / EAL still-reachable reasonably suppressed |