|SUCCESS| [v12, 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/167928

_AI Code Review_

=== Patch Review: series.patch (via Claude) ===
Review date: 2026-08-18

# DPAA Patch Series Review

## Summary

This 26-patch series contains a mix of important bug fixes and new features for the DPAA driver. The series demonstrates good attention to correctness issues (resource leaks, error path cleanup, shutdown sequencing), though several patches could benefit from tighter scope and clearer commit messages.

---

## Patch-by-Patch Analysis

### Patch 01/26: net/dpaa: fix device remove

**Verdict:** Good fix, minor suggestion.

**Findings:**

*Info:*
- Correctly adds `eth_dev` NULL check before use
- Properly stores close return value in separate variable
- Logs close failure before proceeding with release

*Suggestions:*
- Consider making the `close_ret` variable declaration point-of-use style: `int close_ret = dpaa_eth_dev_close(eth_dev);` directly in the `if` condition block
- Minor: the warning message could include the port name (if available from `dpaa_dev->device.name`) for easier debugging in multi-port systems

---

### Patch 02/26: net/dpaa: fix Tx confirmation queue memory leak

**Verdict:** Correct fix for resource leak.

**Findings:**

*Correctness:*
- Adds missing `rte_free(dpaa_intf->tx_conf_queues)` in close path 
- Also frees in error path `free_tx:` 
- Sets pointer to NULL after free to avoid double-free 

*Observations:*
- The commit message is clear and helpful (explains the leak, the allocation site, and why `rte_eth_dev_release_port()` doesn't catch it)

---

### Patch 03/26: bus/dpaa: fix FQD dest wq channel decoding

**Verdict:** Correct bug fix, excellent commit message.

**Findings:**

*Correctness:*
- Fixes actual bit-field extraction bug: old code treated bits 2:0 as channel and bits 15:3 as work queue, reversed from the real hardware layout 
- Commit message explains the impact: pool-channel FQs were not drained correctly 

*Code quality:*
- Introduces inline helper functions `qm_fqd_get_chan()` and `qm_fqd_get_wq()` 
- Moves macros (`GENMASK`, `QM_FQID_MASK`, `qm_fqid_set`, `qm_fqid_get`) to the header for consistency 

*No issues found.*

---

### Patch 04/26: bus/dpaa: refine fman naming

**Verdict:** Good refactoring, naming is clearer.

**Findings:**

*Info:*
- Renames `ccsr_map` - `memac_map` (more specific: it's the MEMAC register space, not generic CCSR)
- Renames `bmi_map` - `rx_bmi_map` (distinguishes from Tx BMI)
- Changes are mechanical; no functional change

*No issues found.*

---

### Patch 05/26: bus/dpaa: scan max BPID from DTS

**Verdict:** Good feature, improves DTS-driven configuration.

**Findings:**

*Info:*
- Replaces hardcoded `bman_pool_max = 64` with value from `fsl,bpid-range` DTS property
- Falls back to default 64 if DTS property absent (with warning)
- Changes default hardware version to BMAN_REV21 for Layerscape devices
- Refactors `bman_global_init()` to handle errors from `bman_init_ccsr()`

*No issues found.*

---

### Patch 06/26: drivers: add process-type guards for secondary process

**Verdict:** Correct secondary process fix.

**Findings:**

*Correctness:*
- Adds `RTE_PROC_PRIMARY` check in `dpaa_qdma_init()` (secondary processes skip HW init) 
- Adds early return in `rte_dpaa_remove()` for secondary processes 

*No issues found.*

---

### Patch 07/26: drivers: shutdown DPAA FQ by fq descriptor

**Verdict:** Good fix, clarifies FQ shutdown API.

**Findings:**

*Info:*
- Changes `qman_shutdown_fq(u32 fqid)` - `qman_shutdown_fq(struct qman_fq *fq)` so that channel-affine portals can be accessed for push-mode Rx queues
- Adds `qman_shutdown_fq_by_fqid(fqid)` wrapper for callers that only have an FQID
- Commit message is clear about why this is needed

*No issues found.*

---

### Patch 08/26: drivers: add DPAA cgrid cleanup support

**Verdict:** Important correctness fix (CGR resource cleanup).

**Findings:**

*Correctness:*
- Adds `qman_pending_fq_by_cgrid()` to find FQs still attached to a CGR 
- Calls `dpaa_cgr_stale_fq_cleanup()` before `qman_delete_cgr()` to shut down any leftover FQs 
- Releases CGRID ranges in `dpaa_eth_dev_close()` 
- Cleans up CGRs in error paths of `dpaa_dev_init()` 

*No issues found.*

---

### Patch 09/26: bus/dpaa: improve FQ shutdown with channel validation

**Verdict:** Good fix, removes dead code, improves channel handling.

**Findings:**

*Correctness:*
- Replaces hardcoded pool-channel range check with DTS-derived values 
- Removes incorrect portal affinity check (the portal's dedicated channel never equals a pool channel, so the old check could never pass) 
- Any portal can subscribe to any pool channel via SDQCR; code now programs SDQCR unconditionally for both dedicated and pool channels 

*Info:*
- Adds `dpaa_get_qm_channel_pool_num()` to read pool-channel count from DTS
- Parses `pool-channel-range` property and stores start/count
- Logs warning if configured pool channel differs from default

*No issues found.*

---

### Patch 10/26: drivers: add BMI Tx statistics

**Verdict:** Good feature addition, handles missing register blocks correctly.

**Findings:**

*Info:*
- Adds Tx BMI counters (`fmbm_tfdc`, `fmbm_tfledc`, `fmbm_tfufdc`, `fmbm_tbdc`) to BMI statistics
- `fman_if_bmi_stats_enable/disable()` now handle both Rx and Tx BMI register blocks
- `fman_if_bmi_stats_get_all()` reports zero for Rx or Tx stats if the corresponding register block is not mapped (correct for offline/OH ports)

*No issues found.*

---

### Patch 11/26: net/dpaa: optimize FM deconfig

**Verdict:** Good cleanup, consolidates FM deconfig logic.

**Findings:**

*Info:*
- Moves FM deconfig to a single location in `dpaa_eth_dev_close()`, removing a duplicate call
- Removes redundant NULL check on `fc_conf` before free (the close path will free it anyway)
- For FMCLESS/shared-MAC mode, deconfig FM *before* FQ shutdown to direct ingress traffic to kernel
- Removes duplicate VSP cleanup and FM deconfig from the bottom of the close path

*No issues found.*

---

### Patch 12/26: net/dpaa: optimize FMC MAC type parsing

**Verdict:** Good simplification, clearer logic.

**Findings:**

*Info:*
- Replaces MAC-type-specific `if` chains with a single index lookup from the FMC port name
- Handles MAC9/MAC10 on ls104xa (which can be 10G, 2.5G, or 1G depending on serdes config)
- Returns `-ENODEV` (instead of `-EINVAL`) when the port is not a match, which is clearer for the scan loop

*No issues found.*

---

### Patch 13/26: drivers: release DPAA bpid on driver destructor

**Verdict:** Important resource leak fix, handles mempool cleanup.

**Findings:**

*Correctness:*
- Tracks allocated BPIDs in a static `s_dpaa_bpid_allocated_flag[]` table 
- Registers `dpaa_mpool_finish()` destructor (RTE_FINI_PRIO 104) to release all BPIDs still marked in-use at exit 
- Releases `rte_dpaa_bpid_info` array when no mempools remain 
- Adjusts mempool cache flush threshold to match hardware bulk release size (`FSL_BM_BURST_MAX`) 

*Info:*
- Changes acquire loop from hardcoded `8` to `FSL_BM_BURST_MAX` constant

*No issues found.*

---

### Patch 14/26: dma/dpaa: add SG data validation and ERR050757

**Verdict:** Good feature, ERR050757 workaround is well-gated.

**Findings:**

*Info:*
- Adds scatter-gather (SG) batching, enabled by default via `s_sg_enable = 1`
- Adds data validation mode (`s_data_validation`) for debugging, disabled by default
- Adds ERR050757 PCI read workaround (configures source stride when `RTE_DMA_DPAA_ERRATA_ERR050757` is defined)
- Adds `dpaa_dma_sg_disable`, `dpaa_dma_data_validation`, and `dpaa_dma_pci_read_disable` device arguments
- Documents all devargs in `doc/guides/dmadevs/dpaa.rst`

*No issues found.*

---

### Patch 15/26: net/dpaa: support Rx/Tx taildrop threshold devarg

**Verdict:** Good feature, clear documentation.

**Findings:**

*Info:*
- Adds `drv_rx_taildrop` and `drv_tx_taildrop` device arguments to set per-port frame queue taildrop thresholds
- Falls back to existing `DPAA_TX_TAILDROP_THRESHOLD` environment variable if the `drv_tx_taildrop` devarg is absent
- Introduces `dpaa_get_devargs_int()` helper for parsing integer devargs
- Caps values at `UINT16_MAX` to avoid overflow

*No issues found.*

---

### Patch 16/26: net/dpaa: add Tx rate limiting API

**Verdict:** Good API addition, clear documentation.

**Findings:**

*Info:*
- Adds `rte_pmd_dpaa_port_set_rate_limit(port_id, burst, rate)` to configure FMAN port rate limiter
- Validates port is backed by DPAA PMD before dereferencing private data
- Opens FMAN handle, programs rate limit via `fm_port_set_rate_limit()`, closes handle
- Allows disabling rate limit by passing `burst=0` or `rate=0` (calls `fm_port_delete_rate_limit()`)
- Marked `__rte_experimental` with version `26.11` in `RTE_EXPORT_EXPERIMENTAL_SYMBOL`

*No issues found.*

---

### Patch 17/26: bus/dpaa: orp queue create and burst enqueue

**Verdict:** Good ORP feature, clean API.

**Findings:**

*Info:*
- Adds `qman_enqueue_multi_orp()` to enqueue multiple FDs with ORP (order restoration point) sequence numbers
- Falls back to standard `qman_enqueue_multi()` if `!orp || !orp_seqnum || fq->force_ooo`
- Adds `force_ooo` flag to `struct qman_fq` to allow per-FQ override of ORP
- Defines ORP window size, auto-advance, and late-arrival enums
- Does NOT self-censor: the implementation is straightforward and the patch is focused

*No issues found.*

---

### Patch 18/26: net/dpaa: support fmcless rxq number as devargs

**Verdict:** Good usability improvement.

**Findings:**

*Info:*
- In FMCLESS mode, changes default Rx queue count from `rte_lcore_count()` to `DPAA_MAX_NUM_PCD_QUEUES` (because multiple queues may be processed on the same core)
- Adds `drv_fmcless_rxq` device argument to override the default
- Parses the argument via `dpaa_get_devargs_int()`
- Caps the value at `DPAA_MAX_NUM_PCD_QUEUES` and returns `-EINVAL` if `< 1`

*No issues found.*

---

### Patch 19/26: net/dpaa: support non fmX-macY type of shared Ethernet name

**Verdict:** Good usability feature, helpful for shared MACs.

**Findings:**

*Info:*
- Adds `drv_sh_if_name` device argument to provide the kernel Linux interface name when it differs from the `fmX-macY` style used by DPDK
- Uses the provided name in `dpaa_eth_dev_configure()` when calling `ioctl(SIOCGIFMTU)` on the shared interface
- Falls back to `dpaa_intf->name` if the devarg is absent or empty

*No issues found.*

---

### Patch 20/26: bus/dpaa: optimize DPAA multi-entry buffer pool operations

**Verdict:** Good optimization, clearer code.

**Findings:**

*Info:*
- Replaces hardcoded `8` with `FSL_BM_BURST_MAX` when acquiring buffers
- Uses a single `bm_hw_buf_desc` structure for HW descriptor initialization, then copies it to remaining entries (ensures consistent HW state)
- Renames local variable `bm_bufs` type from `struct bm_hw_buf_desc[]` to `struct bm_buffer[]` for consistency with the union type

*No issues found.*

---

### Patch 21/26: bus/dpaa: improve log macro and fix bus detection

**Verdict:** Good cleanup, no functional change.

**Findings:**

*Info:*
- Replaces `DPAA_BUS_LOG(LEVEL, ...)` with shorthand macros (`DPAA_BUS_INFO`, `DPAA_BUS_ERR`, `DPAA_BUS_WARN`, `DPAA_BUS_DEBUG`) for consistency
- Removes `-1` magic return values in error paths, replacing them with `-ENOMEM` or other appropriate errno values
- No change to actual error-handling logic, just clearer return codes

*No issues found.*

---

### Patch 22/26: net/dpaa: enhance VSP port support

**Verdict:** Good VSP refactoring, fixes ONIC port type handling.

**Findings:**

*Correctness:*
- Adds `fman_onic` handling to `get_rx_port_type()` so that ONIC and offline-internal ports map to `e_FM_PORT_TYPE_OH_OFFLINE_PARSING` 
- Removes `fif` parameter from `dpaa_port_vsp_cleanup()` (not used) 

*Info:*
- Introduces `struct dpaa_if_vsp` to hold VSP handle, buffer pool pointers, and pool count
- Changes `dpaa_intf->vsp_handle[]` and `vsp_bpid[]` to a single `vsp[]` array of `struct dpaa_if_vsp`
- Simplifies `dpaa_port_vsp_update()` and `dpaa_port_vsp_cleanup()` by working with the new structure

*No issues found.*

---

### Patch 23/26: drivers: add offline (O/H) port device support

**Verdict:** Large feature, correctness looks good.

**Findings:**

*Correctness:*
- Adds offline (O/H) port support via `drv_oldev` bus device argument 
- Registers ioctl control device (`/dev/cdx_ctrl`) for communicating FQIDs, BPIDs, and channel I