|WARNING| [v20, 27/27] doc: update release notes with NXP DPAA changes

[dpdklab at iol.unh.edu]

Test-Label: ai-code-review
Test-Status: WARNING
https://dpdk.org/patch/170558

_AI Code Review_

=== Patch Review: series.patch (via Claude) ===
Review date: 2026-10-05

# Review of DPDK patch series (NXP DPAA improvements v20)

I have reviewed all 27 patches in this series against the AGENTS.md guidelines. Below is my consolidated review organized by severity.

---

## CORRECTNESS BUGS (ERRORS - highest priority)

### Patch 01/27: net/dpaa: fix double close and null deref on remove
**Good.** This patch correctly fixes a use-after-free and double-close bug. The flow now checks `eth_dev != NULL` and `state != UNUSED` before closing, and closes only once. This is a valid correctness fix.

### Patch 02/27: net/dpaa: fix probe success when FMC init fails
**Good.** Correctly propagates the error from `dpaa_port_fmc_init()` and zeros `num_rx_fqs` so the cleanup path does not use a negative count. This is a valid error propagation fix.

### Patch 03/27: net/dpaa: fix resource leak on probe failure
**Good.** Fixes a resource leak where the early return on SG pool creation failure bypassed cleanup. The new flow unwinds through `close_dev` and `release_port` labels, matching the existing error handling. This is a valid cleanup path fix.

### Patch 04/27: net/dpaa: fix free port resources on close
**Good.** Fixes a memory leak of `tx_conf_queues` that was never freed. The patch also ensures the offline/ONIC cleanup path does not skip the queue and congestion-group memory release. This is a valid resource leak fix.

### Patch 05/27: bus/dpaa: fix FQD dest wq channel decoding
**Good.** The old code extracted a 0-7 value as the channel and the real channel as the work queue due to incorrect bit shifting. The fix uses `qm_fqd_get_chan()` and `qm_fqd_get_wq()` helpers that correctly decode the 16-bit `dest_wq` field. This is a valid hardware register decoding bug fix.

### Patch 09/27: drivers: add process-type guards for secondary process
**Good.** The QDMA device allocates `struct fsl_qdma_engine` in shared memory but stores process-private mmap'd register bases in it. A secondary that probed the device would advertise a dmadev with unmapped MMIO pointers. The fix returns `-ENOTSUP` in the secondary until proper mapping is implemented. This is a valid secondary process bug fix.

For `rte_dpaa_remove()`, the fix prevents a secondary from decrementing the global device count and freeing the shared SG pool, which are primary-only resources. This is correct.

### Patch 16/27: mempool/dpaa: fix write after free on pool free
**Error.** `dpaa_mbuf_free_pool()` writes to `bp_info->bp` after freeing `bp_info`:
```c
rte_free(mp->pool_data);  // bp_info is mp->pool_data
bp_info->bp = NULL;       // use-after-free
```
The fix clears the field before freeing and frees `bp_info` directly rather than the alias. This is a valid use-after-free fix.

---

## STYLE AND PROCESS WARNINGS

### Patch 06/27: bus/dpaa: accept QDMA device name in devargs
**Warning.** The patch bounds the QDMA index by `RTE_DPAA_QDMA_DEVICES`, but the `#define` for this constant is not shown in the patch. Verify that `RTE_DPAA_QDMA_DEVICES` is defined and matches the number of QDMA devices created by `dpaa_create_device_list()`.

### Patch 18/27: dma/dpaa: add devargs for SG and errata workaround
**Style note (not an error).** The patch does not add data validation or the ERR050757 workaround; they already exist. The commit message correctly states this: "Scatter-gather batching, the post-transfer data validation helper and the ERR050757 read-transaction workaround already exist in the driver. This patch does not add them; it makes them configurable at runtime..."

The new devargs are documented in `doc/guides/dmadevs/dpaa.rst`.

**Do NOT flag** the lack of release notes for test-only changes or internal API.

### Patch 19/27: net/dpaa: support Rx/Tx taildrop threshold devarg
**Good.** The patch adds `drv_rx_taildrop` and `drv_tx_taildrop` device arguments with documentation in `doc/guides/nics/dpaa.rst` and advertises them through `RTE_PMD_REGISTER_PARAM_STRING()`. The environment variable `DPAA_TX_TAILDROP_THRESHOLD` is still honored when the devarg is absent for backward compatibility.

### Patch 20/27: net/dpaa: support fmcless rxq number as devargs
**Good.** Adds `drv_fmcless_rxq` device argument with documentation in `doc/guides/nics/dpaa.rst` and advertises it through `RTE_PMD_REGISTER_PARAM_STRING()`.

### Patch 21/27: net/dpaa: support non fmX-macY type of shared Ethernet name
**Good.** Adds `drv_sh_if_name` device argument with documentation in `doc/guides/nics/dpaa.rst` and advertises it through `RTE_PMD_REGISTER_PARAM_STRING()`.

### Patch 25/27: drivers: add offline (O/H) port device support
**Warning - Documentation.** The patch adds four new experimental functions declared in `rte_pmd_dpaa_oldev.h`:
- `rte_pmd_dpaa_ol_set_classif_info()`
- `rte_pmd_dpaa_ol_reset_classif_info()`
- `rte_pmd_dpaa_ol_set_lgw_info()`
- `rte_pmd_dpaa_ol_reset_lgw_info()`

The functions are marked `__rte_experimental` and have `RTE_EXPORT_EXPERIMENTAL_SYMBOL()` in the `.c` file.

However, the header declares `struct rte_pmd_dpaa_uplink_cls_info_s` and `struct rte_pmd_dpaa_lgw_info_s` without Doxygen comments. Each function's Doxygen says "@param cls_info Pointer to the classification information to program. Must not be NULL." but does not describe the structure fields or their expected ranges.

**Suggestion:** Add Doxygen documentation for the two structures, describing each field and any constraints (e.g., valid ranges for `num_ports`, `num_subnets`, the meaning of the flags constants).

### Patch 27/27: doc: update release notes with NXP DPAA changes
**Good.** The patch adds a single consolidated "Updated NXP DPAA drivers" entry with sub-bullets for the new features and devargs. This is the correct approach for a multi-patch feature series.

**Do NOT flag** the single entry covering multiple patches; the contributor guide allows grouping related changes under one note.

---

## ITEMS THAT ARE CORRECT (do NOT flag)

### Patch 07/27: bus/dpaa: refine fman naming
**Good.** Renames `ccsr_map` to `memac_map` and `bmi_map` to `rx_bmi_map` for clarity. This is a refactoring patch with no correctness issues.

### Patch 08/27: bus/dpaa: scan max BPID from DTS
**Good.** Dynamically calculates the maximum BPID from the device tree instead of hardcoding it. The patch bounds `bman_pool_max` to `BMAN_MAX_POOLS` to prevent overflow. This is correct.

### Patch 10/27: drivers: shutdown DPAA FQ by fq descriptor
**Good.** Changes `qman_shutdown_fq()` to take the FQ descriptor instead of just the FQID, allowing it to use `fq->qp` when available. This is a valid API refactoring.

### Patch 11/27: drivers: add DPAA cgrid cleanup support
**Good.** The patch adds `qman_pending_fq_by_cgrid_range()` to find FQs still attached to a CGR range, and uses it in `dpaa_cgr_stale_fq_cleanup()` to shut down stale FQs before deleting the CGR. This is correct.

The loop bound is `cgrid_num * DPAA_CGR_STALE_FQ_MAX` to avoid spinning forever on stale FQs; the resume logic (`start_fqid = fqid + 1`) is O(N) instead of O(N2). This is correct.

### Patch 12/27: bus/dpaa: improve FQ shutdown with channel validation
**Good.** Reads the pool-channel range from DTS and validates FQ channels against it. Rejects FQs scheduled on another portal's dedicated channel with `-EBUSY` instead of trying to drain them. Bounds the FQRN wait by `QMAN_FQRN_WAIT_MAX` iterations. All of these are correct.

### Patch 13/27: drivers: add BMI Tx statistics
**Good.** Adds BMI Tx statistics support, mapping the Tx BMI registers and exposing them through xstats. Reports zero for register blocks not mapped for a given port type. This is correct.

### Patch 14/27: net/dpaa: optimize FM deconfig
**Good.** Consolidates FM deconfiguration to avoid duplicate calls. Moves `fm_deconfig()` after FQ shutdown and before releasing the port. This is correct.

### Patch 15/27: net/dpaa: optimize FMC MAC type parsing
**Good.** Replaces the MAC type + port number parsing with a direct port name index extraction using `dpaa_port_fmc_get_idx_from_name()`. This handles LS104xA MAC9/10 correctly. This is correct.

### Patch 17/27: drivers: release DPAA bpid on driver destructor
**Good.** Tracks allocated BPIDs in a static per-BPID table and releases them from a driver destructor if the application exits without calling `rte_mempool_free()`. Uses `bman_free_bpid()` to release the BPID from flags alone without touching the `bman_pool` object (which may already be gone when the destructor runs). This is correct.

### Patch 22/27: bus/dpaa: optimize DPAA multi-entry buffer pool operations
**Good.** Replaces hardcoded `8` with `FSL_BM_BURST_MAX` and adds `struct bm_hw_buf_desc` overlay with helpers `bm_buffer_set64_to_be()` and `bm_buffer_get64_from_be()` to build/parse descriptors in one 8-byte access. This is correct.

The commit message explains the memcpy optimization: entry 0 is written directly, entries 1..n-1 are copied in one `memcpy()`. This is correct.

### Patch 23/27: bus/dpaa: improve log macro usages
**Good.** Replaces `DPAA_BUS_LOG(LEVEL, ...)` with shorthand macros. This is a refactoring patch with no correctness issues.

### Patch 24/27: net/dpaa: enhance VSP port support
**Good.** Adds `fman_onic` handling to `get_rx_port_type()`, replaces parallel `vsp_handle[]` and `vsp_bpid[]` arrays with a single `struct dpaa_if_vsp` per profile, bounds the VSP ID in `dpaa_eth_rx_queue_bp_check()`, and skips Rx disable in `dpaa_eth_dev_stop()` for ONIC ports. All of these are correct.

### Patch 26/27: crypto/dpaa_sec: improve crypto fq resource handling
**Good.** Allocates multiple FQIDs in a single `qman_alloc_fqid_range()` call instead of one at a time, and adds cleanup on the error paths. This is correct.

---

## SUMMARY

**Errors (must fix):**
1. **Patch 16/27** - Use-after-free in `dpaa_mbuf_free_pool()` (already fixed in the patch itself; this is a review of the fix, which is correct).

**Warnings (should fix):**
1. **Patch 06/27** - Verify `RTE_DPAA_QDMA_DEVICES` is defined and consistent.
2. **Patch 25/27** - Add Doxygen documentation for `struct rte_pmd_dpaa_uplink_cls_info_s` and `struct rte_pmd_dpaa_lgw_info_s`.

**All other patches are correct.**

---

## FINAL VERDICT

This series contains **valid correctness bug fixes** (patches 01-05, 09, 16) and **well-implemented new features** (patches 18-21, 25-27). The refactoring patches (07, 10, 13-15, 17, 22-24, 26) are clean and do not introduce regressions.

The only issues are:
- **Patch 06** should verify the `RTE_DPAA_QDMA_DEVICES` definition.
- **Patch 25** should add structure field documentation.

The series is otherwise in good shape and ready for merge after addressing the two warnings.