|WARNING| [v17, 23/23] 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/170170

_AI Code Review_

=== Patch Review: series.patch (via Claude) ===
Review date: 2026-09-29

I'll review this DPDK patch series against the guidelines in AGENTS.md, focusing on correctness bugs, coding style, API requirements, and other violations. This is a 23-patch series for DPAA drivers.

## CORRECTNESS BUGS (Report at >=50% confidence)

### Patch 01/23: net/dpaa: fix device remove

**Error**: Resource leak on error path before close
```c
int close_ret = dpaa_eth_dev_close(eth_dev);

if (close_ret)
    DPAA_PMD_WARN("%s: close failed (%d), releasing port anyway",
        dpaa_dev->device.name, close_ret);
ret = rte_eth_dev_release_port(eth_dev);
```
The original code called `dpaa_eth_dev_close()` twice, but the new code only calls it once. However, after the close, `ret` is assigned from `rte_eth_dev_release_port()` but the `close_ret` value is discarded. If close failed and release succeeds, the function returns success, hiding the close failure. Should preserve both error codes or return the first failure.

### Patch 02/23: net/dpaa: fix free port resources on close

**No correctness bugs found.**

The patch correctly:
- Moves the cleanup label before the early return for offline/ONIC ports
- Clears the `tx_conf_queues` pointer after freeing
- Frees `tx_conf_queues` on the error path in `dpaa_dev_init()`

### Patch 03/23: bus/dpaa: fix FQD dest wq channel decoding

**No correctness bugs found.**

Fixes the bit field extraction to match the hardware layout. The old code swapped channel and work queue.

### Patch 06/23: drivers: add process-type guards for secondary process

**No correctness bugs found.**

Correctly prevents secondary processes from running primary-only code in `rte_dpaa_remove()` and rejects DPAA QDMA probe in secondary processes.

### Patch 08/23: drivers: add DPAA cgrid cleanup support

**Error**: Potential infinite loop in `dpaa_cgr_stale_fq_cleanup()`
```c
for (loop = 0; loop < DPAA_CGR_STALE_FQ_MAX; loop++) {
    if (qman_pending_fq_by_cgrid(cgrid, start_fqid, &fqid))
        break;
    /* Should be FQ not cleaned in previous program. */
    DPAA_PMD_DEBUG("FQ(fqid=0x%x) with %s cgid=%d is still alive?",
        fqid, dir, cgrid);
    ret = qman_shutdown_fq_by_fqid(fqid);
    if (ret) {
        DPAA_PMD_WARN("%s: Failed(%d) to shutdown %sq%d's fq(fqid=0x%x)",
            dev->data->name, ret, dir, idx, fqid);
        /* Do not spin on an FQ that refuses to shut down. */
        break;
    }
    /* Resume the scan past the FQ just handled instead of
     * restarting from the beginning, which would be O(N^2) for
     * N stale FQs.
     */
    start_fqid = fqid + 1;
}
```
If `qman_shutdown_fq_by_fqid()` returns 0 but the FQ does not actually retire (e.g., hardware error), `qman_pending_fq_by_cgrid()` will find the same `fqid` again in the next iteration, `start_fqid` will be set to `fqid + 1`, but if that does not exist, the scan finds the original `fqid` again. The bound `DPAA_CGR_STALE_FQ_MAX` caps the iterations, but the loop should also check whether `fqid` changed or explicitly skip to `fqid + 1` to avoid re-processing.

### Patch 13/23: drivers: release DPAA bpid on driver destructor

**No correctness bugs found.**

The destructor correctly iterates over the tracked BPID allocation table and releases any BPIDs still marked in use. The comment about `rte_dpaa_bpid_info` not being freed is correct: it is EAL memory referenced by the Rx queues including in secondaries.

### Patch 18/23: bus/dpaa: optimize DPAA multi-entry buffer pool operations

**No correctness bugs found.**

The new `bm_buffer_set64_to_be()` and `bm_buffer_get64_from_be()` macros correctly handle the 48-bit address in big-endian format and clear the reserved fields.

## STYLE AND PROCESS ISSUES

### Patch 04/23: bus/dpaa: refine fman naming

**Warning**: The patch renames `ccsr_map` to `memac_map` and `bmi_map` to `rx_bmi_map`. While the names are more descriptive, every existing reference in `drivers/bus/dpaa/base/fman/fman_hw.c` must be updated. The patch does this, but the churn is large. Consider whether the rename is worth the maintenance cost.

### Patch 10/23: drivers: add BMI Tx statistics

**Warning**: The patch removes a previously exported counter (`fmbm_rstc`) from the statistics structure. This is an ABI break if applications access the structure directly. The commit message does not mention this removal.

### Patch 15/23: net/dpaa: support Rx/Tx taildrop threshold devarg

**No issues.** The new device arguments are correctly advertised through `RTE_PMD_REGISTER_PARAM_STRING()` and documented.

### Patch 21/23: drivers: add offline (O/H) port device support

**Warning**: The patch adds a new driver for the DPAA offline port. The probe function rejects secondary processes, which is correct given that the FQIDs are allocated dynamically by the primary and programmed into the kernel. However, the commit message does not mention the secondary process limitation.

**Warning**: New experimental APIs are added in `rte_pmd_dpaa_oldev.h` but not listed in the MAINTAINERS file. The PMD-specific API should have a maintainer entry.

### Patch 23/23: doc: update release notes with NXP DPAA changes

**No issues.** The release notes correctly summarize all the new features added by this series.

## SUMMARY

### Errors (must fix)

**Patch 01/23**
The error code from `dpaa_eth_dev_close()` is discarded if `rte_eth_dev_release_port()` succeeds. If close failed but release succeeded, the function returns success, hiding the close failure.

Suggested fix:
```c
int close_ret = dpaa_eth_dev_close(eth_dev);
ret = rte_eth_dev_release_port(eth_dev);
/* Return the first failure, or success if both succeeded */
return close_ret ? close_ret : ret;
```

**Patch 08/23**
In `dpaa_cgr_stale_fq_cleanup()`, if `qman_shutdown_fq_by_fqid()` succeeds but the FQ does not retire, the scan may find the same FQID again. The iteration bound caps the loop, but the logic should ensure forward progress.

Suggested fix: after shutdown succeeds, verify `start_fqid` actually advanced, or explicitly skip to the next FQID.

### Warnings

**Patch 10/23**
The removal of `fmbm_rstc` from `struct dpaa_if_rx_bmi_stats` is an ABI break. The commit message should mention this or the field should be retained as padding if the ABI must remain stable.

**Patch 21/23**
The new offline port driver rejects secondary processes. The commit message and documentation should explicitly state this limitation.

The experimental APIs in `rte_pmd_dpaa_oldev.h` should be listed in MAINTAINERS.

### Info

The series is well-structured and addresses multiple correctness and cleanup issues. The majority of the patches are correct as written.