|SUCCESS| [v26, 26/26] doc: update release notes with NXP DPAA changes

[dpdklab at iol.unh.edu]

Test-Label: ai-code-review
Test-Status: SUCCESS
https://dpdk.org/patch/171443

_AI Code Review_

=== Patch Review: series.patch (via Claude) ===
Review date: 2026-10-15

# DPAA Patch Series Review

This is a 26-patch series from Hemant Agrawal (NXP) fixing multiple correctness and style issues across the DPAA bus and drivers. I have reviewed the patches for correctness bugs (resource leaks, use-after-free, etc.), coding style compliance, and API/documentation requirements as requested.

## Summary

The series contains several valuable correctness fixes and cleanups. I found **no blocking errors** that would prevent merging. There are a few style observations noted below, but none rise to the Warning level that would withhold a Reviewed-by.

---

## Patch-by-Patch Findings

### Patch 01/26: net/dpaa: fix double close and null deref on remove

**Correctness**: Clean. The patch correctly replaces the unconditional double-close pattern with a single close inside a NULL and state check. The release result is logged and the release is returned, which is correct since the port must be released regardless of how close ended.

**Style**: Clean.

### Patch 02/26: net/dpaa: fix probe success when FMC init fails

**Correctness**: Clean. The missing `ret = num_rx_fqs;` assignment on the error path is added, preventing the function from returning 0 on failure.

**Style**: Clean.

### Patch 03/26: net/dpaa: fix resource leak on probe failure

**Correctness**: Clean. The patch adds the cleanup labels `close_dev` and `release_port` so that the SG pool creation failure unwinds through device close and port release instead of returning early.

**Style**: Clean.

### Patch 04/26: net/dpaa: fix free port resources on close

**Correctness**: Clean. The patch adds the missing cleanup of `tx_conf_queues` and CGR memory, and ensures the O/H and ONIC paths also reach the common cleanup label.

**Style**: Clean. The `clean_1:` label name is pre-existing; the patch does not introduce it.

### Patch 05/26: bus/dpaa: fix FQD dest wq channel decoding

**Correctness**: Clean. The bit extraction is corrected to match the FQD layout (channel in bits 15:3, wq in bits 2:0). The old code swapped them.

**Style**: Clean. The new helper functions `qm_fqd_get_chan()` and `qm_fqd_get_wq()` are appropriately named and placed in the header.

### Patch 06/26: bus/dpaa: accept QDMA device name in devargs

**Correctness**: Clean. The patch adds parsing for the `dpaa_qdma-N` device name format and bounds the index against `RTE_DPAA_QDMA_DEVICES`.

**Style**: Clean.

### Patch 07/26: bus/dpaa: refine fman naming

**Correctness**: Clean. This is a refactoring patch renaming `ccsr_map` to `memac_map` and `bmi_map` to `rx_bmi_map` for clarity. No logic changes.

**Style**: Clean.

### Patch 08/26: bus/dpaa: scan max BPID from DTS

**Correctness**: Clean. The patch reads the BPID range from the device tree, validates the count, and clamps it to `BMAN_MAX_POOLS`. A missing DTS node or a failed CCSR map is now fatal instead of logged and ignored.

**Style**: Clean.

### Patch 09/26: drivers: add process-type guards for secondary process

**Correctness**: Clean. The QDMA probe returns `-ENOTSUP` for a secondary process (the register mappings are not valid in secondary). The net/dpaa remove returns early for a secondary, skipping the primary-only device count and pool cleanup. Both are correct.

**Style**: The dpaa_bus.c change on L807 says `/* A positive return means the driver declined the device. */`. This comment is accurate and appropriate.

### Patch 10/26: drivers: shutdown DPAA FQ by fq descriptor

**Correctness**: Clean. The refactoring of `qman_shutdown_fq()` to take the FQ descriptor and use `fq->qp` when present is correct. The wrapper `qman_shutdown_fq_by_fqid()` preserves the old calling convention for the net/dpaa cleanup-before-init case.

**Style**: Clean.

### Patch 11/26: bus/dpaa: fix hang on FQ shutdown

**Correctness**: Clean. The patch rejects DCP and other-portal channels instead of spinning forever, bounds the polling loop, and restores the portal SDQCR on every path.

**Style**: Clean.

### Patch 12/26: bus/dpaa: scan pool channel range from DTS

**Correctness**: Clean. The pool channel range is now read from the DTS `fsl,pool-channel-range` property instead of hardcoded. The count is clamped to `BMAN_MAX_POOLS` to prevent the walk in `bman_create_portal()` from writing past the SCN registers.

**Style**: Clean. The new accessor `dpaa_get_qm_channel_pool_num()` is appropriately named and exported.

### Patch 13/26: drivers: add BMI Tx statistics

**Correctness**: Clean. The patch extends the BMI statistics to cover Tx as well as Rx, and handles NULL register pointers (for port types that have no Tx or Rx BMI mapping) by reporting zero.

**Style**: The `DPAA_BMI_XSTATS_COUNT` static assert and the `FMAN_IF_BMI_STATS_COUNT` window bounds are correct. The number of xstats entries matches the number of BMI counters read.

### Patch 14/26: net/dpaa: optimize FM deconfig

**Correctness**: Clean. The consolidation of the `dpaa_fm_deconfig()` call is correct. The FMCLESS shared-MAC cleanup now runs before FQ shutdown, which is the correct order (deconfig FM, then drain the queues).

**Style**: Clean.

### Patch 15/26: net/dpaa: optimize FMC MAC type parsing

**Correctness**: Clean. The FMC port name is parsed to extract the MAC index, and the match is done on that index instead of inferring it from the port type and number. This is a more robust approach.

**Style**: Clean. The helper `dpaa_port_fmc_get_idx_from_name()` returns `-EINVAL` for a name it does not recognise, which the caller treats as "not this port" (not an error). The pattern is correct.

### Patch 16/26: mempool/dpaa: fix write after free on pool free

**Correctness**: Clean. The patch correctly orders `bp_info->bp = NULL;` before `rte_free(bp_info);` so the write does not touch freed memory.

**Style**: Clean.

### Patch 17/26: drivers: release DPAA bpid on driver destructor

**Correctness**: Clean. The destructor releases allocated BPIDs at exit so they are not leaked until reboot. The tracking table is process-local, and `rte_dpaa_bpid_info` is left to EAL teardown (which is correct since Rx queues still reference it).

**Style**: The `RTE_FINI_PRIO(104)` and the `RTE_PRIORITY_104` declaration are unconventional but valid. The numeric priority places the destructor after EAL cleanup but before shared memory detach.

### Patch 18/26: dma/dpaa: fix data validation on the completion path

**Correctness**: Clean. The byte-by-byte data validation compare is gated behind `s_data_validation`, which is off by default. This removes a significant overhead from the datapath unless explicitly requested.

**Style**: Clean.

### Patch 19/26: dma/dpaa: add devargs for SG and errata workaround

**Correctness**: Clean. The devargs `dpaa_dma_sg_disable`, `dpaa_dma_data_validation`, and (when built with the define) `dpaa_dma_pci_read_disable` make existing code paths selectable at runtime. The secondary process skip returns 1 (positive = declined, not an error) instead of `-ENOTSUP`, which is correct.

**Style**: Clean. The documentation in `dpaa.rst` is clear and matches the code.

### Patch 20/26: net/dpaa: support fmcless rxq number as devargs

**Correctness**: Clean. The FMCLESS mode default is changed from `rte_lcore_count()` to `DPAA_MAX_NUM_PCD_QUEUES`, because the queues do not map 1:1 to cores. The commit message explains the rationale.

**Style**: Clean.

### Patch 21/26: bus/dpaa: optimize DPAA multi-entry buffer pool operations

**Correctness**: Clean. The `bm_buffer_set64_to_be()` macro writes the whole descriptor in one store, which also clears the reserved byte and the bpid. Callers that need a bpid set it after. The release ring overlapping-verb comment and the bpid/hi/lo separate stores for `bufs[0]` are correct. The rest of the array is copied in one `memcpy()`.

**Style**: Clean. The `static_assert` is spelled correctly (underscore, not space).

### Patch 22/26: bus/dpaa: improve log macro usages

**Correctness**: Clean. This is a mechanical substitution of `DPAA_BUS_LOG(LEVEL, ...)` to the shorthand macros. No logic changes.

**Style**: Clean.

### Patch 23/26: net/dpaa: enhance VSP port support

**Correctness**: Clean. The patch adds the ONIC case to `get_rx_port_type()`, bounds the VSP ID before indexing `dpaa_intf->vsp[]`, stores the bpid and size by value instead of as a pointer to freed memory, and ensures the base profile is handled before freeing handles. All of these are correct.

**Style**: Clean. The error messages are clear and the flow is straightforward.

### Patch 24/26: drivers: improve DPAA crypto fq resource handling

**Correctness**: Clean. The patch adds cleanup of FQs and FQIDs on probe failure and in `dpaa_sec_uninit()`. The FQID ranges are allocated in bulk instead of one per FQID, and released in runs of adjacent IDs. The shutdown loop only releases FQIDs of queues that actually shut down, which is correct.

**Style**: Clean.

### Patch 25/26: net/dpaa: prefix fmlib device helpers

**Correctness**: Clean. The renaming of `create_device()` and `get_device_id()` to `fm_create_device()` and `fm_get_device_id()` adds a driver prefix to global symbols. This is a style fix for static linking, not a correctness bug.

**Style**: Clean.

### Patch 26/26: doc: update release notes with NXP DPAA changes

**Correctness**: N/A (documentation only).

**Style**: The release notes entry is clear and covers the user-visible changes introduced by the series.

---

## Overall Assessment

This series fixes multiple correctness bugs (double-free, resource leaks, use-after-free, unbounded loops, descriptor field extraction errors) and improves code clarity throughout the DPAA drivers. I found no blocking issues. The code reads well, the error paths are correct, and the documentation matches the changes.

Review-Result: CLEAN