On Thu, 11 Jun 2026 19:50:08 +0530 Rahul Bhansali <[email protected]> wrote:
> As per the requirement of rte_mbuf_raw_reset_bulk(), the mbuf's > 'next' and 'nb_segs' fields are required to be reset. > This reset these field for multi-segment mbufs on cn9k platform. > > Signed-off-by: Rahul Bhansali <[email protected]> Please put a cover letter on large multi-patch series in future. The CI AI review doesn't look at original source and uses cost optimized model (ie. is not that smart). Did a UI based review and it saw: Reviewed the v2 series. Three issues, rest look good. [06/22] net/cnxk: reserve memory for lookup mem at probe Error: error path returns success. At the new check rc is still 0 from the prior successful roc_nix_dev_init(). When cnxk_nix_fastpath_lookup_mem_get() returns NULL, "goto dev_fini" falls through to "return rc" with rc == 0, so dev init reports success and the rc=%d message prints 0. Set rc before the jump: if (!cnxk_nix_fastpath_lookup_mem_get()) { plt_err("Failed to reserve lookup memory"); rc = -ENOMEM; goto dev_fini; } [07/22] drivers: add support for devargs skip size Warning: shared memzone freed while other ports still use it. SKIP_SIZE_PKIND_MEMZONE is a single global memzone created once (lookup- guarded) in roc_npc_init(), but roc_npc_fini() frees it unconditionally. On a multi-port device the first port closed tears down the table the other ports still read in roc_npc_skip_size_pkind_get() during inbound SA creation; the lookup then returns NULL and skip-size pkind selection silently stops working for the surviving ports. Refcount the memzone or tie its lifetime to the common/inline layer instead of per-NPC fini. [19/22] net/cnxk: add FEC get set and capability ops Warning: feature not reflected in the features matrix. features.rst maps the FEC feature to fec_get_capability/fec_get/fec_set, which this patch implements, but doc/guides/nics/features/cnxk.ini is not updated with "FEC = Y". The ops return NOTSUP on VF/SDP, so cnxk_vf.ini is correct as- is. Add the matrix entry. Note on [16/22]: the changes are good. Moving cpt_cq_ena inside the "if (idev && idev->nix_inl_dev)" block fixes a NULL deref of inl_dev, and cpt_cq_ena is initialized to 0 so the fall-through default is correct. The roc_dev.c / roc_ree.c error-path rework fixes real leaks and wrong- success returns. Other patches reviewed with no issues.
