On Fri, 23 Oct 2026 23:57:41 +0800
Jiacheng Ye <[email protected]> wrote:
> This series fixes offload capability advertisement, cmdq/compact CQE
> configuration, and several defects found during review, and adds NIC
> type identification, FEC, CIR drop statistics, VF LAG/link state, COS
> mask/map, mbox recording, RSS hash configuration, compact CQE Rx/Tx
> paths, TSO segment expansion, Tx burst mode reporting, queue statistics,
> pseudo-header checksum, per-tile cmdq ops, SP560 NIC support, and device
> kvargs tunables.
AI review still sees some issues that need addressing before merge.
PS: it is ok to respond with "AI is hallucinating about XXX"
Subject: Re: [PATCH v4 00/27] net/hinic3: fixes and new features
Applies to main. Every commit builds with -Dwerror=true; bisect is
fixed. The v3 comments on copy offset, IPv6 l3_len, SP230 interrupt
masking, copy pool size and be32 proto are addressed.
Please include a changelog in the cover letter.
Errors
======
[PATCH 08/27] net/hinic3: fix pseudo-header checksum calculation
The fix adds hinic3_prep_pkts(), but nothing installs it until 21/27
sets tx_pkt_prepare. At 08 through 20, and in any stable backport of
08, the code is dead and checksums are still wrong. Wire it up in the
same patch:
eth_dev->rx_pkt_burst = hinic3_recv_pkts;
eth_dev->tx_pkt_burst = hinic3_xmit_pkts;
eth_dev->tx_pkt_prepare = hinic3_prep_pkts;
The commit message mentions SP560, which is not added until 19/27.
Say SP620 here.
Warnings
========
[PATCH 19/27] net/hinic3: add SP560 NIC type identification and
handling
hinic3.ini now says "Rx interrupt = Y", and the explanation of partial
support was dropped from the commit message, but the STN op still
does:
if (wqe_type == HINIC3_COMPACT_RQ_WQE && support_rq_sw_compact_cqe) {
rq_ctxt->ceq_attr |= RQ_CTXT_CEQ_ATTR_SET(1, EN);
In the same function EN is set from intr_disable. If EN=1 means
interrupt disabled, Rx interrupts never fire on SP560 with compact
CQE, and configure should reject intr_conf.rxq in that mode. If the
bit means something else when SW compact CQE is on, add a comment
saying so. Either way, the ini and the code must agree.
Info
====
[PATCH 23/27] The new tso_segsz == 0 check logs at ERR for every bad
packet in the Tx path. The drop is already counted in
offload_errors; drop the log or use a rate-limited data path log.
Series: a lot of v4 is splitting log format strings across lines and
reindenting. DPDK checkpatch ignores long string lines, and split
strings are harder to grep for. No need to do this.
