On Sun, 18 Jun 2017 15:21:07 +0800 Jiayu Hu <jiayu...@intel.com> wrote:
> +/** > + * This is the main reassembly API used in lightweight mode, which > + * merges numbers of packets at a time. After it returns, applications > + * can get GROed packets immediately. Applications don't need to > + * flush packets manually. In lightweight mode, applications just need > + * to tell the reassembly API what rules should be applied when merge > + * packets. Therefore, applications can perform GRO in very a simple > + * way. > + * > + * To process one packet, we find its corresponding reassembly table > + * according to the packet type. Then search for the reassembly table > + * to find one packet to merge. If find, chain the two packets together. > + * If not find, insert the inputted packet into the reassembly table. > + * Besides, to merge two packets is to chain them together. No > + * memory copy is needed. Before rte_gro_reassemble_burst returns, > + * header checksums of merged packets are re-calculated. > + * > + * @param pkts > + * a pointer array which points to the packets to reassemble. After > + * GRO, it is also used to keep GROed packets. > + * @param nb_pkts > + * the number of packets to reassemble. > + * @param param > + * Applications use param to tell rte_gro_reassemble_burst what rules > + * are demanded. > + * @return > + * the number of packets after GROed. > + */ > +uint16_t rte_gro_reassemble_burst(struct rte_mbuf **pkts __rte_unused, > + const uint16_t nb_pkts __rte_unused, > + const struct rte_gro_param param __rte_unused); I think the __rte_unused attribute should be on the function definition, not on the prototype. I think GCC ignores it on function prototypes.
