Hi Jiyau,

Thanks for the quick review comments. Will address the review comments. Require clarification in one of the comments. Please find it inline.

On Thu, Jun 8, 2023 at 9:35 AM Hu, Jiayu <jiayu.hu@intel.com> wrote:
Hi Kumara,

Please see replies inline.

In addition, you need to update the programmer guide in generic_receive_offload_lib.rst,
and release note release_23_07.rst.

Thanks,
Jiayu

> -----Original Message-----
> From: Kumara Parameshwaran <kumaraparamesh92@gmail.com>
> Sent: Tuesday, June 6, 2023 10:58 PM
> To: Hu, Jiayu <jiayu.hu@intel.com>
> Cc: dev@dpdk.org; Kumara Parameshwaran
> <kumaraparamesh92@gmail.com>
> Subject: [PATCH v4] gro : ipv6 changes to support GRO for TCP/ipv6
>
> The patch adds GRO support for TCP/ipv6 packets. This does not include the
> support for vxlan, udp ipv6 packets.
>
> Signed-off-by: Kumara Parameshwaran <kumaraparamesh92@gmail.com>>
> diff --git a/lib/gro/gro_tcp.c b/lib/gro/gro_tcp.c new file mode 100644 index
> 0000000000..02a7d0f8c5
> --- /dev/null
> +++ b/lib/gro/gro_tcp.c

For gro_tcp.c and gro_tcp.h, it's better to add "_internal" in the file name.

> @@ -0,0 +1,128 @@
> +/* SPDX-License-Identifier: BSD-3-Clause
> + * Copyright(c) 2017 Intel Corporation
> + */
> +#include <rte_malloc.h>
> +#include <rte_mbuf.h>
> +#include <rte_ethdev.h>
> +
> +#include "gro_tcp.h"
> +
> +static inline uint32_t
> +find_an_empty_item(struct gro_tcp_item *items,
> +     uint32_t max_item_num)
> +{
> +     uint32_t i;
> +
> +     for (i = 0; i < max_item_num; i++)
> +             if (items[i].firstseg == NULL)
> +                     return i;
> +     return INVALID_ARRAY_INDEX;
> +}
> +
> +uint32_t
> +insert_new_tcp_item(struct rte_mbuf *pkt,
> +             struct gro_tcp_item *items,
> +             uint32_t *item_num,
> +             uint32_t max_item_num,
> +             uint64_t start_time,
> +             uint32_t prev_idx,
> +             uint32_t sent_seq,
> +             uint16_t ip_id,
> +             uint8_t is_atomic)

This function can be inline.

> +{
> +     uint32_t item_idx;
> +
> +     item_idx = find_an_empty_item(items, max_item_num);
> +     if (item_idx == INVALID_ARRAY_INDEX)
> +             return INVALID_ARRAY_INDEX;
> +
> +     items[item_idx].firstseg = pkt;
> +     items[item_idx].lastseg = rte_pktmbuf_lastseg(pkt);
> +     items[item_idx].start_time = start_time;
> +     items[item_idx].next_pkt_idx = INVALID_ARRAY_INDEX;
> +     items[item_idx].sent_seq = sent_seq;
> +     items[item_idx].ip_id = ip_id;
> +     items[item_idx].nb_merged = 1;
> +     items[item_idx].is_atomic = is_atomic;
> +     (*item_num) += 1;
> +
> +     /* if the previous packet exists, chain them together. */
> +     if (prev_idx != INVALID_ARRAY_INDEX) {
> +             items[item_idx].next_pkt_idx =
> +                     items[prev_idx].next_pkt_idx;
> +             items[prev_idx].next_pkt_idx = item_idx;
> +     }
> +
> +     return item_idx;
> +}
> +
> +uint32_t
> +delete_tcp_item(struct gro_tcp_item *items, uint32_t item_idx,
> +             uint32_t *item_num,
> +             uint32_t prev_item_idx)
> +{
> +     uint32_t next_idx = items[item_idx].next_pkt_idx;
> +
> +     /* NULL indicates an empty item */
> +     items[item_idx].firstseg = NULL;
> +     (*item_num) -= 1;
> +     if (prev_item_idx != INVALID_ARRAY_INDEX)
> +             items[prev_item_idx].next_pkt_idx = next_idx;
> +
> +     return next_idx;
> +}

This function can be inline.

> +
> +int32_t
> +gro_process_tcp_item(struct rte_mbuf *pkt,
> +     struct rte_tcp_hdr *tcp_hdr,
> +     int32_t tcp_dl,
> +     struct gro_tcp_item *items,
> +     uint32_t item_idx,
> +     uint32_t *item_num,
> +     uint32_t max_item_num,
> +     uint16_t ip_id,
> +     uint8_t is_atomic,
> +     uint64_t start_time)

It is for internal use, so it's better to remove "gro_" from the function name.

> +{
> +     uint32_t cur_idx;
> +     uint32_t prev_idx;
> +     int cmp;
> +     uint32_t sent_seq;
> +
> +     sent_seq = rte_be_to_cpu_32(tcp_hdr->sent_seq);
> +     /*
> +      * Check all packets in the flow and try to find a neighbor for
> +      * the input packet.
> +      */
> +     cur_idx = item_idx;
> +     prev_idx = cur_idx;
> +     do {
> +             cmp = check_seq_option(&items[cur_idx], tcp_hdr,
> +                             sent_seq, ip_id, pkt->l4_len, tcp_dl, 0,
> +                             is_atomic);
> +             if (cmp) {
> +                     if (merge_two_tcp_packets(&items[cur_idx],
> +                                             pkt, cmp, sent_seq, ip_id, 0))
> +                             return 1;
> +                     /*
> +                      * Fail to merge the two packets, as the packet
> +                      * length is greater than the max value. Store
> +                      * the packet into the flow.
> +                      */
> +                     if (insert_new_tcp_item(pkt, items, item_num,
> max_item_num, start_time, cur_idx,
> +                                             sent_seq, ip_id, is_atomic) ==
> +                                     INVALID_ARRAY_INDEX)
> +                             return -1;
> +                     return 0;
> +             }
> +             prev_idx = cur_idx;
> +             cur_idx = items[cur_idx].next_pkt_idx;
> +     } while (cur_idx != INVALID_ARRAY_INDEX);
> +
> +     /* Fail to find a neighbor, so store the packet into the flow. */
> +     if (insert_new_tcp_item(pkt, items, item_num, max_item_num,
> start_time, prev_idx, sent_seq,
> +                             ip_id, is_atomic) == INVALID_ARRAY_INDEX)
> +             return -1;
> +
> +     return 0;
> +}
> diff --git a/lib/gro/gro_tcp.h b/lib/gro/gro_tcp.h new file mode 100644 index
> 0000000000..4b5b4eda9c
> --- /dev/null
> +++ b/lib/gro/gro_tcp.h
> @@ -0,0 +1,209 @@
> +#ifndef _GRO_TCP_H_
> +#define _GRO_TCP_H_
> +
> +#define INVALID_ARRAY_INDEX 0xffffffffUL
> +
> +#include <rte_tcp.h>
> +
> +/*
> + * The max length of a IPv4 packet, which includes the length of the L3
> + * header, the L4 header and the data payload.
> + */
> +#define MAX_IP_PKT_LENGTH UINT16_MAX
> +
> +/* The maximum TCP header length */
> +#define MAX_TCP_HLEN 60
> +#define INVALID_TCP_HDRLEN(len) \
> +     (((len) < sizeof(struct rte_tcp_hdr)) || ((len) > MAX_TCP_HLEN))
> +
> +struct gro_tcp_flow {

This structure name is confusing. In the upper layer, tcp4 and tcp6 have gro_tcp4_flow
and gro_tcp6_flow, which represent a flow. Inside gro_tcp4/6_flow, there are keys,
represented by struct tcp4/6_flow_key. But inside struct tcp4/6_flow_key, there is
struct gro_tcp_flow. Need to rename struct gro_tcp_flow, like common_tcp_flow_key.

> +     struct rte_ether_addr eth_saddr;
> +     struct rte_ether_addr eth_daddr;
> +     uint32_t recv_ack;
> +     uint16_t src_port;
> +     uint16_t dst_port;
> +};
> +
> +#define ASSIGN_TCP_FLOW_KEY(k1, k2) \

Ditto. The macro needs rename, like ASSIGN_COMMON_TCP_FLOW_KEY.

> +     do {\
> +             rte_ether_addr_copy(&(k1->eth_saddr), &(k2->eth_saddr)); \
> +             rte_ether_addr_copy(&(k1->eth_daddr), &(k2->eth_daddr));
> \
> +             k2->recv_ack = k1->recv_ack; \
> +             k2->src_port = k1->src_port; \
> +             k2->dst_port = k1->dst_port; \
> +     } while (0)
> +
> +struct gro_tcp_item {
> +     /*
> +      * The first MBUF segment of the packet. If the value
> +      * is NULL, it means the item is empty.
> +      */
> +     struct rte_mbuf *firstseg;
> +     /* The last MBUF segment of the packet */
> +     struct rte_mbuf *lastseg;
> +     /*
> +      * The time when the first packet is inserted into the table.
> +      * This value won't be updated, even if the packet is merged
> +      * with other packets.
> +      */
> +     uint64_t start_time;
> +     /*
> +      * next_pkt_idx is used to chain the packets that
> +      * are in the same flow but can't be merged together
> +      * (e.g. caused by packet reordering).
> +      */
> +     uint32_t next_pkt_idx;
> +     /* TCP sequence number of the packet */
> +     uint32_t sent_seq;
> +     /* IPv4 ID of the packet */
> +     uint16_t ip_id;

The ip_id field is not used by tcp6. It's better to use an union to include ip_id for IPv4 and
an useless member for IPv6 with some comments to avoid confusing.

> +     /* the number of merged packets */
> +     uint16_t nb_merged;
> +     /* Indicate if IPv4 ID can be ignored */
> +     uint8_t is_atomic;
> +};
> +
> +uint32_t
> +insert_new_tcp_item(struct rte_mbuf *pkt,
> +             struct gro_tcp_item *items,
> +             uint32_t *item_num,
> +             uint32_t table_size,
> +             uint64_t start_time,
> +             uint32_t prev_idx,
> +             uint32_t sent_seq,
> +             uint16_t ip_id,
> +             uint8_t is_atomic);
> +
> +uint32_t
> +delete_tcp_item(struct gro_tcp_item *items,
> +             uint32_t item_idx,
> +             uint32_t *item_num,
> +             uint32_t prev_item_idx);
> +
> +int32_t
> +gro_process_tcp_item(struct rte_mbuf *pkt,
> +     struct rte_tcp_hdr *tcp_hdr,
> +     int32_t tcp_dl,
> +     struct gro_tcp_item *items,
> +     uint32_t item_idx,
> +     uint32_t *item_num,
> +     uint32_t table_size,
> +     uint16_t ip_id,
> +     uint8_t is_atomic,
> +     uint64_t start_time);
> +
> +/*
> + * Merge two TCP packets without updating checksums.
> + * If cmp is larger than 0, append the new packet to the
> + * original packet. Otherwise, pre-pend the new packet to
> + * the original packet.
> + */
> +static inline int
> +merge_two_tcp_packets(struct gro_tcp_item *item,
> +             struct rte_mbuf *pkt,
> +             int cmp,
> +             uint32_t sent_seq,
> +             uint16_t ip_id,
> +             uint16_t l2_offset)
> +{
> +     struct rte_mbuf *pkt_head, *pkt_tail, *lastseg;
> +     uint16_t hdr_len, l2_len;
> +
> +     if (cmp > 0) {
> +             pkt_head = item->firstseg;
> +             pkt_tail = pkt;
> +     } else {
> +             pkt_head = pkt;
> +             pkt_tail = item->firstseg;
> +     }
> +
> +     /* check if the IPv4 packet length is greater than the max value */
> +     hdr_len = l2_offset + pkt_head->l2_len + pkt_head->l3_len +
> +             pkt_head->l4_len;
> +     l2_len = l2_offset > 0 ? pkt_head->outer_l2_len : pkt_head->l2_len;
> +     if (unlikely(pkt_head->pkt_len - l2_len + pkt_tail->pkt_len -
> +                             hdr_len > MAX_IP_PKT_LENGTH))
> +             return 0;
> +
> +     /* remove the packet header for the tail packet */
> +     rte_pktmbuf_adj(pkt_tail, hdr_len);
> +
> +     /* chain two packets together */
> +     if (cmp > 0) {
> +             item->lastseg->next = pkt;
> +             item->lastseg = rte_pktmbuf_lastseg(pkt);
> +             /* update IP ID to the larger value */
> +             item->ip_id = ip_id;
> +     } else {
> +             lastseg = rte_pktmbuf_lastseg(pkt);
> +             lastseg->next = item->firstseg;
> +             item->firstseg = pkt;
> +             /* update sent_seq to the smaller value */
> +             item->sent_seq = sent_seq;
> +             item->ip_id = ip_id;
> +     }
> +     item->nb_merged++;
> +
> +     /* update MBUF metadata for the merged packet */
> +     pkt_head->nb_segs += pkt_tail->nb_segs;
> +     pkt_head->pkt_len += pkt_tail->pkt_len;
> +
> +     return 1;
> +}
> +
> +/*
> + * Check if two TCP/IPv4 packets are neighbors.
> + */
> +static inline int
> +check_seq_option(struct gro_tcp_item *item,
> +             struct rte_tcp_hdr *tcph,
> +             uint32_t sent_seq,
> +             uint16_t ip_id,
> +             uint16_t tcp_hl,
> +             uint16_t tcp_dl,
> +             uint16_t l2_offset,
> +             uint8_t is_atomic)
> +{
> +     struct rte_mbuf *pkt_orig = item->firstseg;
> +     char *iph_orig;
> +     struct rte_tcp_hdr *tcph_orig;
> +     uint16_t len, tcp_hl_orig;
> +
> +     iph_orig = (char *)(rte_pktmbuf_mtod(pkt_orig, char *) +
> +                     l2_offset + pkt_orig->l2_len);
> +     tcph_orig = (struct rte_tcp_hdr *)(iph_orig + pkt_orig->l3_len);
> +     tcp_hl_orig = pkt_orig->l4_len;
> +
> +     /* Check if TCP option fields equal */
> +     len = RTE_MAX(tcp_hl, tcp_hl_orig) - sizeof(struct rte_tcp_hdr);
> +     if ((tcp_hl != tcp_hl_orig) || ((len > 0) &&
> +                             (memcmp(tcph + 1, tcph_orig + 1,
> +                                     len) != 0)))
> +             return 0;
> +
> +     /* Don't merge packets whose DF bits are different */
> +     if (unlikely(item->is_atomic ^ is_atomic))
> +             return 0;
> +
> +     /* check if the two packets are neighbors */
> +     len = pkt_orig->pkt_len - l2_offset - pkt_orig->l2_len -
> +             pkt_orig->l3_len - tcp_hl_orig;
> +     if ((sent_seq == item->sent_seq + len) && (is_atomic ||
> +                             (ip_id == item->ip_id + 1)))
> +             /* append the new packet */
> +             return 1;
> +     else if ((sent_seq + tcp_dl == item->sent_seq) && (is_atomic ||
> +                             (ip_id + item->nb_merged == item->ip_id)))
> +             /* pre-pend the new packet */
> +             return -1;
> +
> +     return 0;
> +}
> +
> +static inline int
> +is_same_tcp_flow(struct gro_tcp_flow *k1, struct gro_tcp_flow *k2) {
> +     return (!memcmp(k1, k2, sizeof(struct gro_tcp_flow))); }

I think this function needs rename, as the result of this function cannot identify if they are
same TCP flow.

> +
> +#endif
> diff --git a/lib/gro/gro_tcp4.c b/lib/gro/gro_tcp4.c index
> 0014096e63..ffc33747c4 100644
> --- a/lib/gro/gro_tcp4.c
> +++ b/lib/gro/gro_tcp4.c
> @@ -7,6 +7,7 @@
>  #include <rte_ethdev.h>
>
>  #include "gro_tcp4.h"
> +#include "gro_tcp.h"
>
>  void *
>  gro_tcp4_tbl_create(uint16_t socket_id, @@ -30,7 +31,7 @@
> gro_tcp4_tbl_create(uint16_t socket_id,
>       if (tbl == NULL)
>               return NULL;
>
> -     size = sizeof(struct gro_tcp4_item) * entries_num;
> +     size = sizeof(struct gro_tcp_item) * entries_num;
>       tbl->items = rte_zmalloc_socket(__func__,
>                       size,
>                       RTE_CACHE_LINE_SIZE,
> @@ -71,18 +72,6 @@ gro_tcp4_tbl_destroy(void *tbl)
>       rte_free(tcp_tbl);
>  }
>
> -static inline uint32_t
> -find_an_empty_item(struct gro_tcp4_tbl *tbl) -{
> -     uint32_t i;
> -     uint32_t max_item_num = tbl->max_item_num;
> -
> -     for (i = 0; i < max_item_num; i++)
> -             if (tbl->items[i].firstseg == NULL)
> -                     return i;
> -     return INVALID_ARRAY_INDEX;
> -}
> -
>  static inline uint32_t
>  find_an_empty_flow(struct gro_tcp4_tbl *tbl)  { @@ -95,56 +84,6 @@
> find_an_empty_flow(struct gro_tcp4_tbl *tbl)
>       return INVALID_ARRAY_INDEX;
>  }
>
> -static inline uint32_t
> -insert_new_item(struct gro_tcp4_tbl *tbl,
> -             struct rte_mbuf *pkt,
> -             uint64_t start_time,
> -             uint32_t prev_idx,
> -             uint32_t sent_seq,
> -             uint16_t ip_id,
> -             uint8_t is_atomic)
> -{
> -     uint32_t item_idx;
> -
> -     item_idx = find_an_empty_item(tbl);
> -     if (item_idx == INVALID_ARRAY_INDEX)
> -             return INVALID_ARRAY_INDEX;
> -
> -     tbl->items[item_idx].firstseg = pkt;
> -     tbl->items[item_idx].lastseg = rte_pktmbuf_lastseg(pkt);
> -     tbl->items[item_idx].start_time = start_time;
> -     tbl->items[item_idx].next_pkt_idx = INVALID_ARRAY_INDEX;
> -     tbl->items[item_idx].sent_seq = sent_seq;
> -     tbl->items[item_idx].ip_id = ip_id;
> -     tbl->items[item_idx].nb_merged = 1;
> -     tbl->items[item_idx].is_atomic = is_atomic;
> -     tbl->item_num++;
> -
> -     /* if the previous packet exists, chain them together. */
> -     if (prev_idx != INVALID_ARRAY_INDEX) {
> -             tbl->items[item_idx].next_pkt_idx =
> -                     tbl->items[prev_idx].next_pkt_idx;
> -             tbl->items[prev_idx].next_pkt_idx = item_idx;
> -     }
> -
> -     return item_idx;
> -}
> -
> -static inline uint32_t
> -delete_item(struct gro_tcp4_tbl *tbl, uint32_t item_idx,
> -             uint32_t prev_item_idx)
> -{
> -     uint32_t next_idx = tbl->items[item_idx].next_pkt_idx;
> -
> -     /* NULL indicates an empty item */
> -     tbl->items[item_idx].firstseg = NULL;
> -     tbl->item_num--;
> -     if (prev_item_idx != INVALID_ARRAY_INDEX)
> -             tbl->items[prev_item_idx].next_pkt_idx = next_idx;
> -
> -     return next_idx;
> -}
> -
>  static inline uint32_t
>  insert_new_flow(struct gro_tcp4_tbl *tbl,
>               struct tcp4_flow_key *src,
> @@ -159,13 +98,10 @@ insert_new_flow(struct gro_tcp4_tbl *tbl,
>
>       dst = &(tbl->flows[flow_idx].key);
>
> -     rte_ether_addr_copy(&(src->eth_saddr), &(dst->eth_saddr));
> -     rte_ether_addr_copy(&(src->eth_daddr), &(dst->eth_daddr));
> +     ASSIGN_TCP_FLOW_KEY((&src->tcp_flow), (&dst->tcp_flow));
> +
>       dst->ip_src_addr = src->ip_src_addr;
>       dst->ip_dst_addr = src->ip_dst_addr;
> -     dst->recv_ack = src->recv_ack;
> -     dst->src_port = src->src_port;
> -     dst->dst_port = src->dst_port;
>
>       tbl->flows[flow_idx].start_index = item_idx;
>       tbl->flow_num++;
> @@ -173,21 +109,6 @@ insert_new_flow(struct gro_tcp4_tbl *tbl,
>       return flow_idx;
>  }
>
> -/*
> - * update the packet length for the flushed packet.
> - */
> -static inline void
> -update_header(struct gro_tcp4_item *item) -{
> -     struct rte_ipv4_hdr *ipv4_hdr;
> -     struct rte_mbuf *pkt = item->firstseg;
> -
> -     ipv4_hdr = (struct rte_ipv4_hdr *)(rte_pktmbuf_mtod(pkt, char *) +
> -                     pkt->l2_len);
> -     ipv4_hdr->total_length = rte_cpu_to_be_16(pkt->pkt_len -
> -                     pkt->l2_len);
> -}
> -
>  int32_t
>  gro_tcp4_reassemble(struct rte_mbuf *pkt,
>               struct gro_tcp4_tbl *tbl,
> @@ -195,16 +116,15 @@ gro_tcp4_reassemble(struct rte_mbuf *pkt,  {
>       struct rte_ether_hdr *eth_hdr;
>       struct rte_ipv4_hdr *ipv4_hdr;
> -     struct rte_tcp_hdr *tcp_hdr;
> -     uint32_t sent_seq;
>       int32_t tcp_dl;
> +     struct rte_tcp_hdr *tcp_hdr;
>       uint16_t ip_id, hdr_len, frag_off, ip_tlen;
>       uint8_t is_atomic;
> +     uint32_t sent_seq;

No need to change tcp_hdr and sent_seq here.
The flow matching is done in the function and if the flow is not found insert_new_tcp_item is invoked from this function itself. Did you mean to move that to the process_tcp_item as well? If that is the case we should pass the start_idx as INVALID_ARRAY_INDEX and in process_tcp_item check if INVALID_ARRAY_INDEX do a insert_new_tcp_item and return, do not do the sequnce number checks etc.
>
>       struct tcp4_flow_key key;
> -     uint32_t cur_idx, prev_idx, item_idx;
> +     uint32_t item_idx;
>       uint32_t i, max_flow_num, remaining_flow_num;
> -     int cmp;
>       uint8_t find;
>
>       /*
> @@ -216,7 +136,7 @@ gro_tcp4_reassemble(struct rte_mbuf *pkt,
>
>       eth_hdr = rte_pktmbuf_mtod(pkt, struct rte_ether_hdr *);
>       ipv4_hdr = (struct rte_ipv4_hdr *)((char *)eth_hdr + pkt->l2_len);
> -     tcp_hdr = (struct rte_tcp_hdr *)((char *)ipv4_hdr + pkt->l3_len);
> +     tcp_hdr = rte_pktmbuf_mtod_offset(pkt, struct rte_tcp_hdr *,
> +pkt->l2_len + pkt->l3_len);
>       hdr_len = pkt->l2_len + pkt->l3_len + pkt->l4_len;
>
>       /*
> @@ -230,7 +150,6 @@ gro_tcp4_reassemble(struct rte_mbuf *pkt,
>       ip_tlen = rte_be_to_cpu_16(ipv4_hdr->total_length);
>       if (pkt->pkt_len > (uint32_t)(ip_tlen + pkt->l2_len))
>               rte_pktmbuf_trim(pkt, pkt->pkt_len - ip_tlen - pkt->l2_len);
> -
>       /*
>        * Don't process the packet whose payload length is less than or
>        * equal to 0.
> @@ -239,6 +158,13 @@ gro_tcp4_reassemble(struct rte_mbuf *pkt,
>       if (tcp_dl <= 0)
>               return -1;
>
> +     rte_ether_addr_copy(&(eth_hdr->src_addr),
> &(key.tcp_flow.eth_saddr));
> +     rte_ether_addr_copy(&(eth_hdr->dst_addr),
> &(key.tcp_flow.eth_daddr));
> +     key.ip_src_addr = ipv4_hdr->src_addr;
> +     key.ip_dst_addr = ipv4_hdr->dst_addr;
> +     key.tcp_flow.src_port = tcp_hdr->src_port;
> +     key.tcp_flow.dst_port = tcp_hdr->dst_port;
> +     key.tcp_flow.recv_ack = tcp_hdr->recv_ack;
>       /*
>        * Save IPv4 ID for the packet whose DF bit is 0. For the packet
>        * whose DF bit is 1, IPv4 ID is ignored.
> @@ -246,15 +172,6 @@ gro_tcp4_reassemble(struct rte_mbuf *pkt,
>       frag_off = rte_be_to_cpu_16(ipv4_hdr->fragment_offset);
>       is_atomic = (frag_off & RTE_IPV4_HDR_DF_FLAG) ==
> RTE_IPV4_HDR_DF_FLAG;
>       ip_id = is_atomic ? 0 : rte_be_to_cpu_16(ipv4_hdr->packet_id);
> -     sent_seq = rte_be_to_cpu_32(tcp_hdr->sent_seq);
> -
> -     rte_ether_addr_copy(&(eth_hdr->src_addr), &(key.eth_saddr));
> -     rte_ether_addr_copy(&(eth_hdr->dst_addr), &(key.eth_daddr));
> -     key.ip_src_addr = ipv4_hdr->src_addr;
> -     key.ip_dst_addr = ipv4_hdr->dst_addr;
> -     key.src_port = tcp_hdr->src_port;
> -     key.dst_port = tcp_hdr->dst_port;
> -     key.recv_ack = tcp_hdr->recv_ack;
>
>       /* Search for a matched flow. */
>       max_flow_num = tbl->max_flow_num;
> @@ -270,63 +187,44 @@ gro_tcp4_reassemble(struct rte_mbuf *pkt,
>               }
>       }
>
> -     /*
> -      * Fail to find a matched flow. Insert a new flow and store the
> -      * packet into the flow.
> -      */
>       if (find == 0) {
> -             item_idx = insert_new_item(tbl, pkt, start_time,
> -                             INVALID_ARRAY_INDEX, sent_seq, ip_id,
> -                             is_atomic);
> +             sent_seq = rte_be_to_cpu_32(tcp_hdr->sent_seq);
> +             item_idx = insert_new_tcp_item(pkt, tbl->items, &tbl-
> >item_num, tbl->max_item_num, start_time,
> +                                             INVALID_ARRAY_INDEX,
> sent_seq, ip_id,
> +                                             is_atomic);
>               if (item_idx == INVALID_ARRAY_INDEX)
>                       return -1;
>               if (insert_new_flow(tbl, &key, item_idx) ==
> -                             INVALID_ARRAY_INDEX) {
> +                     INVALID_ARRAY_INDEX) {
>                       /*
>                        * Fail to insert a new flow, so delete the
>                        * stored packet.
> -                      */
> -                     delete_item(tbl, item_idx, INVALID_ARRAY_INDEX);
> +                     */
> +                     delete_tcp_item(tbl->items, item_idx, &tbl-
> >item_num,
> +INVALID_ARRAY_INDEX);
>                       return -1;
>               }
>               return 0;
>       }
> +     item_idx = tbl->flows[i].start_index;

No need to update item_idx, and you can directly pass tbl->flows[i].start_index to
gro_process_tcp_item(). And same in gro_tcp6_reassemble().

>
> -     /*
> -      * Check all packets in the flow and try to find a neighbor for
> -      * the input packet.
> -      */
> -     cur_idx = tbl->flows[i].start_index;
> -     prev_idx = cur_idx;
> -     do {
> -             cmp = check_seq_option(&(tbl->items[cur_idx]), tcp_hdr,
> -                             sent_seq, ip_id, pkt->l4_len, tcp_dl, 0,
> -                             is_atomic);
> -             if (cmp) {
> -                     if (merge_two_tcp4_packets(&(tbl->items[cur_idx]),
> -                                             pkt, cmp, sent_seq, ip_id, 0))
> -                             return 1;
> -                     /*
> -                      * Fail to merge the two packets, as the packet
> -                      * length is greater than the max value. Store
> -                      * the packet into the flow.
> -                      */
> -                     if (insert_new_item(tbl, pkt, start_time, cur_idx,
> -                                             sent_seq, ip_id, is_atomic) ==
> -                                     INVALID_ARRAY_INDEX)
> -                             return -1;
> -                     return 0;
> -             }
> -             prev_idx = cur_idx;
> -             cur_idx = tbl->items[cur_idx].next_pkt_idx;
> -     } while (cur_idx != INVALID_ARRAY_INDEX);
> +     return gro_process_tcp_item(pkt, tcp_hdr, tcp_dl, tbl->items,
> item_idx,
> +                                             &tbl->item_num, tbl-
> >max_item_num,
> +                                             ip_id, is_atomic, start_time);
> +}
>
> -     /* Fail to find a neighbor, so store the packet into the flow. */
> -     if (insert_new_item(tbl, pkt, start_time, prev_idx, sent_seq,
> -                             ip_id, is_atomic) == INVALID_ARRAY_INDEX)
> -             return -1;