Hello Dariusz, Thanks for the review.
On Fri, 11 Sept 2026 at 11:00, Dariusz Sosnowski <[email protected]> wrote: > > @@ -1963,17 +1989,15 @@ mlx5_dev_spawn(struct rte_device *dpdk_dev, > > mlx5_flex_item_port_cleanup(eth_dev); > > mlx5_free(priv->ext_rxqs); > > mlx5_free(priv->ext_txqs); > > + mlx5_free(priv->mac); > > + eth_dev->data->mac_addrs = NULL; > > This can segfault because eth_dev is allocated later than priv. > Previous version of the error rollback accounted for that > (where mac_addrs was reset only if eth_dev != NULL). > The same logic applies in Windows code. > > Could you please revert that? Well, the code before looked fishy to me. Resetting mac_addrs if (eth_dev != NULL) outside of the if (priv != NULL) block seems incorrect. > > + mlx5_free(priv->mac_own); > > mlx5_free(priv); > > if (eth_dev != NULL) > > eth_dev->data->dev_private = NULL; I would rather reset mac_addrs to NULL along the dev_private reset in the if (eth_dev != NULL) block right after. The comment about mac_addrs (see below) being in dev_private can also be removed. WDYT? > > } > > - if (eth_dev != NULL) { > > - /* mac_addrs must not be freed alone because part of > > - * dev_private > > - **/ > > - eth_dev->data->mac_addrs = NULL; > > + if (eth_dev != NULL) > > rte_eth_dev_release_port(eth_dev); > > - } > > if (sh) > > mlx5_free_shared_dev_ctx(sh); > > if (nl_rdma >= 0) -- David Marchand
