examples/ipsec-secgw: allow larger burst size for vectors
[dpdk.git] / examples / ipsec-secgw / ipsec-secgw.c
index 865f87e..88984a6 100644 (file)
@@ -47,6 +47,7 @@
 #include <rte_ip.h>
 #include <rte_ip_frag.h>
 #include <rte_alarm.h>
+#include <rte_telemetry.h>
 
 #include "event_helper.h"
 #include "flow.h"
@@ -69,11 +70,6 @@ volatile bool force_quit;
 
 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
 
-/* Configure how many packets ahead to prefetch, when reading packets */
-#define PREFETCH_OFFSET        3
-
-#define MAX_RX_QUEUE_PER_LCORE 16
-
 #define MAX_LCORE_PARAMS 1024
 
 /*
@@ -114,6 +110,11 @@ struct flow_info flow_info_tbl[RTE_MAX_ETHPORTS];
 #define CMD_LINE_OPT_REASSEMBLE                "reassemble"
 #define CMD_LINE_OPT_MTU               "mtu"
 #define CMD_LINE_OPT_FRAG_TTL          "frag-ttl"
+#define CMD_LINE_OPT_EVENT_VECTOR      "event-vector"
+#define CMD_LINE_OPT_VECTOR_SIZE       "vector-size"
+#define CMD_LINE_OPT_VECTOR_TIMEOUT    "vector-tmo"
+#define CMD_LINE_OPT_VECTOR_POOL_SZ    "vector-pool-sz"
+#define CMD_LINE_OPT_PER_PORT_POOL     "per-port-pool"
 
 #define CMD_LINE_ARG_EVENT     "event"
 #define CMD_LINE_ARG_POLL      "poll"
@@ -138,6 +139,11 @@ enum {
        CMD_LINE_OPT_REASSEMBLE_NUM,
        CMD_LINE_OPT_MTU_NUM,
        CMD_LINE_OPT_FRAG_TTL_NUM,
+       CMD_LINE_OPT_EVENT_VECTOR_NUM,
+       CMD_LINE_OPT_VECTOR_SIZE_NUM,
+       CMD_LINE_OPT_VECTOR_TIMEOUT_NUM,
+       CMD_LINE_OPT_VECTOR_POOL_SZ_NUM,
+       CMD_LINE_OPT_PER_PORT_POOL_NUM,
 };
 
 static const struct option lgopts[] = {
@@ -151,6 +157,11 @@ static const struct option lgopts[] = {
        {CMD_LINE_OPT_REASSEMBLE, 1, 0, CMD_LINE_OPT_REASSEMBLE_NUM},
        {CMD_LINE_OPT_MTU, 1, 0, CMD_LINE_OPT_MTU_NUM},
        {CMD_LINE_OPT_FRAG_TTL, 1, 0, CMD_LINE_OPT_FRAG_TTL_NUM},
+       {CMD_LINE_OPT_EVENT_VECTOR, 0, 0, CMD_LINE_OPT_EVENT_VECTOR_NUM},
+       {CMD_LINE_OPT_VECTOR_SIZE, 1, 0, CMD_LINE_OPT_VECTOR_SIZE_NUM},
+       {CMD_LINE_OPT_VECTOR_TIMEOUT, 1, 0, CMD_LINE_OPT_VECTOR_TIMEOUT_NUM},
+       {CMD_LINE_OPT_VECTOR_POOL_SZ, 1, 0, CMD_LINE_OPT_VECTOR_POOL_SZ_NUM},
+       {CMD_LINE_OPT_PER_PORT_POOL, 0, 0, CMD_LINE_OPT_PER_PORT_POOL_NUM},
        {NULL, 0, 0, 0}
 };
 
@@ -163,7 +174,7 @@ static int32_t promiscuous_on = 1;
 static int32_t numa_on = 1; /**< NUMA is enabled by default. */
 static uint32_t nb_lcores;
 static uint32_t single_sa;
-static uint32_t nb_bufs_in_pool;
+uint32_t nb_bufs_in_pool;
 
 /*
  * RX/TX HW offload capabilities to enable/use on ethernet ports.
@@ -175,9 +186,9 @@ static uint64_t dev_tx_offload = UINT64_MAX;
 /*
  * global values that determine multi-seg policy
  */
-static uint32_t frag_tbl_sz;
+uint32_t frag_tbl_sz;
 static uint32_t frame_buf_size = RTE_MBUF_DEFAULT_BUF_SIZE;
-static uint32_t mtu_size = RTE_ETHER_MTU;
+uint32_t mtu_size = RTE_ETHER_MTU;
 static uint64_t frag_ttl_ns = MAX_FRAG_TTL_NS;
 static uint32_t stats_interval;
 
@@ -189,11 +200,6 @@ struct app_sa_prm app_sa_prm = {
                };
 static const char *cfgfile;
 
-struct lcore_rx_queue {
-       uint16_t port_id;
-       uint8_t queue_id;
-} __rte_cache_aligned;
-
 struct lcore_params {
        uint16_t port_id;
        uint8_t queue_id;
@@ -208,29 +214,7 @@ static uint16_t nb_lcore_params;
 static struct rte_hash *cdev_map_in;
 static struct rte_hash *cdev_map_out;
 
-struct buffer {
-       uint16_t len;
-       struct rte_mbuf *m_table[MAX_PKT_BURST] __rte_aligned(sizeof(void *));
-};
-
-struct lcore_conf {
-       uint16_t nb_rx_queue;
-       struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
-       uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
-       struct buffer tx_mbufs[RTE_MAX_ETHPORTS];
-       struct ipsec_ctx inbound;
-       struct ipsec_ctx outbound;
-       struct rt_ctx *rt4_ctx;
-       struct rt_ctx *rt6_ctx;
-       struct {
-               struct rte_ip_frag_tbl *tbl;
-               struct rte_mempool *pool_dir;
-               struct rte_mempool *pool_indir;
-               struct rte_ip_frag_death_row dr;
-       } frag;
-} __rte_cache_aligned;
-
-static struct lcore_conf lcore_conf[RTE_MAX_LCORE];
+struct lcore_conf lcore_conf[RTE_MAX_LCORE];
 
 static struct rte_eth_conf port_conf = {
        .rxmode = {
@@ -252,10 +236,12 @@ static struct rte_eth_conf port_conf = {
 
 struct socket_ctx socket_ctx[NB_SOCKETS];
 
+bool per_port_pool;
+
 /*
  * Determine is multi-segment support required:
  *  - either frame buffer size is smaller then mtu
- *  - or reassmeble support is requested
+ *  - or reassemble support is requested
  */
 static int
 multi_seg_required(void)
@@ -264,32 +250,6 @@ multi_seg_required(void)
                frame_buf_size || frag_tbl_sz != 0);
 }
 
-static inline void
-adjust_ipv4_pktlen(struct rte_mbuf *m, const struct rte_ipv4_hdr *iph,
-       uint32_t l2_len)
-{
-       uint32_t plen, trim;
-
-       plen = rte_be_to_cpu_16(iph->total_length) + l2_len;
-       if (plen < m->pkt_len) {
-               trim = m->pkt_len - plen;
-               rte_pktmbuf_trim(m, trim);
-       }
-}
-
-static inline void
-adjust_ipv6_pktlen(struct rte_mbuf *m, const struct rte_ipv6_hdr *iph,
-       uint32_t l2_len)
-{
-       uint32_t plen, trim;
-
-       plen = rte_be_to_cpu_16(iph->payload_len) + sizeof(*iph) + l2_len;
-       if (plen < m->pkt_len) {
-               trim = m->pkt_len - plen;
-               rte_pktmbuf_trim(m, trim);
-       }
-}
-
 
 struct ipsec_core_statistics core_statistics[RTE_MAX_LCORE];
 
@@ -354,366 +314,6 @@ print_stats_cb(__rte_unused void *param)
        rte_eal_alarm_set(stats_interval * US_PER_S, print_stats_cb, NULL);
 }
 
-static inline void
-prepare_one_packet(struct rte_mbuf *pkt, struct ipsec_traffic *t)
-{
-       const struct rte_ether_hdr *eth;
-       const struct rte_ipv4_hdr *iph4;
-       const struct rte_ipv6_hdr *iph6;
-       const struct rte_udp_hdr *udp;
-       uint16_t ip4_hdr_len;
-       uint16_t nat_port;
-
-       eth = rte_pktmbuf_mtod(pkt, const struct rte_ether_hdr *);
-       if (eth->ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4)) {
-
-               iph4 = (const struct rte_ipv4_hdr *)rte_pktmbuf_adj(pkt,
-                       RTE_ETHER_HDR_LEN);
-               adjust_ipv4_pktlen(pkt, iph4, 0);
-
-               switch (iph4->next_proto_id) {
-               case IPPROTO_ESP:
-                       t->ipsec.pkts[(t->ipsec.num)++] = pkt;
-                       break;
-               case IPPROTO_UDP:
-                       if (app_sa_prm.udp_encap == 1) {
-                               ip4_hdr_len = ((iph4->version_ihl &
-                                       RTE_IPV4_HDR_IHL_MASK) *
-                                       RTE_IPV4_IHL_MULTIPLIER);
-                               udp = rte_pktmbuf_mtod_offset(pkt,
-                                       struct rte_udp_hdr *, ip4_hdr_len);
-                               nat_port = rte_cpu_to_be_16(IPSEC_NAT_T_PORT);
-                               if (udp->src_port == nat_port ||
-                                       udp->dst_port == nat_port){
-                                       t->ipsec.pkts[(t->ipsec.num)++] = pkt;
-                                       pkt->packet_type |=
-                                               MBUF_PTYPE_TUNNEL_ESP_IN_UDP;
-                                       break;
-                               }
-                       }
-               /* Fall through */
-               default:
-                       t->ip4.data[t->ip4.num] = &iph4->next_proto_id;
-                       t->ip4.pkts[(t->ip4.num)++] = pkt;
-               }
-               pkt->l2_len = 0;
-               pkt->l3_len = sizeof(*iph4);
-               pkt->packet_type |= RTE_PTYPE_L3_IPV4;
-               if  (pkt->packet_type & RTE_PTYPE_L4_TCP)
-                       pkt->l4_len = sizeof(struct rte_tcp_hdr);
-               else if (pkt->packet_type & RTE_PTYPE_L4_UDP)
-                       pkt->l4_len = sizeof(struct rte_udp_hdr);
-       } else if (eth->ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6)) {
-               int next_proto;
-               size_t l3len, ext_len;
-               uint8_t *p;
-
-               /* get protocol type */
-               iph6 = (const struct rte_ipv6_hdr *)rte_pktmbuf_adj(pkt,
-                       RTE_ETHER_HDR_LEN);
-               adjust_ipv6_pktlen(pkt, iph6, 0);
-
-               next_proto = iph6->proto;
-
-               /* determine l3 header size up to ESP extension */
-               l3len = sizeof(struct ip6_hdr);
-               p = rte_pktmbuf_mtod(pkt, uint8_t *);
-               while (next_proto != IPPROTO_ESP && l3len < pkt->data_len &&
-                       (next_proto = rte_ipv6_get_next_ext(p + l3len,
-                                               next_proto, &ext_len)) >= 0)
-                       l3len += ext_len;
-
-               /* drop packet when IPv6 header exceeds first segment length */
-               if (unlikely(l3len > pkt->data_len)) {
-                       free_pkts(&pkt, 1);
-                       return;
-               }
-
-               switch (next_proto) {
-               case IPPROTO_ESP:
-                       t->ipsec.pkts[(t->ipsec.num)++] = pkt;
-                       break;
-               case IPPROTO_UDP:
-                       if (app_sa_prm.udp_encap == 1) {
-                               udp = rte_pktmbuf_mtod_offset(pkt,
-                                       struct rte_udp_hdr *, l3len);
-                               nat_port = rte_cpu_to_be_16(IPSEC_NAT_T_PORT);
-                               if (udp->src_port == nat_port ||
-                                       udp->dst_port == nat_port){
-                                       t->ipsec.pkts[(t->ipsec.num)++] = pkt;
-                                       pkt->packet_type |=
-                                               MBUF_PTYPE_TUNNEL_ESP_IN_UDP;
-                                       break;
-                               }
-                       }
-               /* Fall through */
-               default:
-                       t->ip6.data[t->ip6.num] = &iph6->proto;
-                       t->ip6.pkts[(t->ip6.num)++] = pkt;
-               }
-               pkt->l2_len = 0;
-               pkt->l3_len = l3len;
-               pkt->packet_type |= RTE_PTYPE_L3_IPV6;
-       } else {
-               /* Unknown/Unsupported type, drop the packet */
-               RTE_LOG(ERR, IPSEC, "Unsupported packet type 0x%x\n",
-                       rte_be_to_cpu_16(eth->ether_type));
-               free_pkts(&pkt, 1);
-               return;
-       }
-
-       /* Check if the packet has been processed inline. For inline protocol
-        * processed packets, the metadata in the mbuf can be used to identify
-        * the security processing done on the packet. The metadata will be
-        * used to retrieve the application registered userdata associated
-        * with the security session.
-        */
-
-       if (pkt->ol_flags & RTE_MBUF_F_RX_SEC_OFFLOAD &&
-                       rte_security_dynfield_is_registered()) {
-               struct ipsec_sa *sa;
-               struct ipsec_mbuf_metadata *priv;
-               struct rte_security_ctx *ctx = (struct rte_security_ctx *)
-                                               rte_eth_dev_get_sec_ctx(
-                                               pkt->port);
-
-               /* Retrieve the userdata registered. Here, the userdata
-                * registered is the SA pointer.
-                */
-               sa = (struct ipsec_sa *)rte_security_get_userdata(ctx,
-                               *rte_security_dynfield(pkt));
-               if (sa == NULL) {
-                       /* userdata could not be retrieved */
-                       return;
-               }
-
-               /* Save SA as priv member in mbuf. This will be used in the
-                * IPsec selector(SP-SA) check.
-                */
-
-               priv = get_priv(pkt);
-               priv->sa = sa;
-       }
-}
-
-static inline void
-prepare_traffic(struct rte_mbuf **pkts, struct ipsec_traffic *t,
-               uint16_t nb_pkts)
-{
-       int32_t i;
-
-       t->ipsec.num = 0;
-       t->ip4.num = 0;
-       t->ip6.num = 0;
-
-       for (i = 0; i < (nb_pkts - PREFETCH_OFFSET); i++) {
-               rte_prefetch0(rte_pktmbuf_mtod(pkts[i + PREFETCH_OFFSET],
-                                       void *));
-               prepare_one_packet(pkts[i], t);
-       }
-       /* Process left packets */
-       for (; i < nb_pkts; i++)
-               prepare_one_packet(pkts[i], t);
-}
-
-static inline void
-prepare_tx_pkt(struct rte_mbuf *pkt, uint16_t port,
-               const struct lcore_conf *qconf)
-{
-       struct ip *ip;
-       struct rte_ether_hdr *ethhdr;
-
-       ip = rte_pktmbuf_mtod(pkt, struct ip *);
-
-       ethhdr = (struct rte_ether_hdr *)
-               rte_pktmbuf_prepend(pkt, RTE_ETHER_HDR_LEN);
-
-       if (ip->ip_v == IPVERSION) {
-               pkt->ol_flags |= qconf->outbound.ipv4_offloads;
-               pkt->l3_len = sizeof(struct ip);
-               pkt->l2_len = RTE_ETHER_HDR_LEN;
-
-               ip->ip_sum = 0;
-
-               /* calculate IPv4 cksum in SW */
-               if ((pkt->ol_flags & RTE_MBUF_F_TX_IP_CKSUM) == 0)
-                       ip->ip_sum = rte_ipv4_cksum((struct rte_ipv4_hdr *)ip);
-
-               ethhdr->ether_type = rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4);
-       } else {
-               pkt->ol_flags |= qconf->outbound.ipv6_offloads;
-               pkt->l3_len = sizeof(struct ip6_hdr);
-               pkt->l2_len = RTE_ETHER_HDR_LEN;
-
-               ethhdr->ether_type = rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6);
-       }
-
-       memcpy(&ethhdr->src_addr, &ethaddr_tbl[port].src,
-                       sizeof(struct rte_ether_addr));
-       memcpy(&ethhdr->dst_addr, &ethaddr_tbl[port].dst,
-                       sizeof(struct rte_ether_addr));
-}
-
-static inline void
-prepare_tx_burst(struct rte_mbuf *pkts[], uint16_t nb_pkts, uint16_t port,
-               const struct lcore_conf *qconf)
-{
-       int32_t i;
-       const int32_t prefetch_offset = 2;
-
-       for (i = 0; i < (nb_pkts - prefetch_offset); i++) {
-               rte_mbuf_prefetch_part2(pkts[i + prefetch_offset]);
-               prepare_tx_pkt(pkts[i], port, qconf);
-       }
-       /* Process left packets */
-       for (; i < nb_pkts; i++)
-               prepare_tx_pkt(pkts[i], port, qconf);
-}
-
-/* Send burst of packets on an output interface */
-static inline int32_t
-send_burst(struct lcore_conf *qconf, uint16_t n, uint16_t port)
-{
-       struct rte_mbuf **m_table;
-       int32_t ret;
-       uint16_t queueid;
-
-       queueid = qconf->tx_queue_id[port];
-       m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
-
-       prepare_tx_burst(m_table, n, port, qconf);
-
-       ret = rte_eth_tx_burst(port, queueid, m_table, n);
-
-       core_stats_update_tx(ret);
-
-       if (unlikely(ret < n)) {
-               do {
-                       free_pkts(&m_table[ret], 1);
-               } while (++ret < n);
-       }
-
-       return 0;
-}
-
-/*
- * Helper function to fragment and queue for TX one packet.
- */
-static inline uint32_t
-send_fragment_packet(struct lcore_conf *qconf, struct rte_mbuf *m,
-       uint16_t port, uint8_t proto)
-{
-       struct buffer *tbl;
-       uint32_t len, n;
-       int32_t rc;
-
-       tbl =  qconf->tx_mbufs + port;
-       len = tbl->len;
-
-       /* free space for new fragments */
-       if (len + RTE_LIBRTE_IP_FRAG_MAX_FRAG >=  RTE_DIM(tbl->m_table)) {
-               send_burst(qconf, len, port);
-               len = 0;
-       }
-
-       n = RTE_DIM(tbl->m_table) - len;
-
-       if (proto == IPPROTO_IP)
-               rc = rte_ipv4_fragment_packet(m, tbl->m_table + len,
-                       n, mtu_size, qconf->frag.pool_dir,
-                       qconf->frag.pool_indir);
-       else
-               rc = rte_ipv6_fragment_packet(m, tbl->m_table + len,
-                       n, mtu_size, qconf->frag.pool_dir,
-                       qconf->frag.pool_indir);
-
-       if (rc >= 0)
-               len += rc;
-       else
-               RTE_LOG(ERR, IPSEC,
-                       "%s: failed to fragment packet with size %u, "
-                       "error code: %d\n",
-                       __func__, m->pkt_len, rte_errno);
-
-       free_pkts(&m, 1);
-       return len;
-}
-
-/* Enqueue a single packet, and send burst if queue is filled */
-static inline int32_t
-send_single_packet(struct rte_mbuf *m, uint16_t port, uint8_t proto)
-{
-       uint32_t lcore_id;
-       uint16_t len;
-       struct lcore_conf *qconf;
-
-       lcore_id = rte_lcore_id();
-
-       qconf = &lcore_conf[lcore_id];
-       len = qconf->tx_mbufs[port].len;
-
-       if (m->pkt_len <= mtu_size) {
-               qconf->tx_mbufs[port].m_table[len] = m;
-               len++;
-
-       /* need to fragment the packet */
-       } else if (frag_tbl_sz > 0)
-               len = send_fragment_packet(qconf, m, port, proto);
-       else
-               free_pkts(&m, 1);
-
-       /* enough pkts to be sent */
-       if (unlikely(len == MAX_PKT_BURST)) {
-               send_burst(qconf, MAX_PKT_BURST, port);
-               len = 0;
-       }
-
-       qconf->tx_mbufs[port].len = len;
-       return 0;
-}
-
-static inline void
-inbound_sp_sa(struct sp_ctx *sp, struct sa_ctx *sa, struct traffic_type *ip,
-               uint16_t lim)
-{
-       struct rte_mbuf *m;
-       uint32_t i, j, res, sa_idx;
-
-       if (ip->num == 0 || sp == NULL)
-               return;
-
-       rte_acl_classify((struct rte_acl_ctx *)sp, ip->data, ip->res,
-                       ip->num, DEFAULT_MAX_CATEGORIES);
-
-       j = 0;
-       for (i = 0; i < ip->num; i++) {
-               m = ip->pkts[i];
-               res = ip->res[i];
-               if (res == BYPASS) {
-                       ip->pkts[j++] = m;
-                       continue;
-               }
-               if (res == DISCARD) {
-                       free_pkts(&m, 1);
-                       continue;
-               }
-
-               /* Only check SPI match for processed IPSec packets */
-               if (i < lim && ((m->ol_flags & RTE_MBUF_F_RX_SEC_OFFLOAD) == 0)) {
-                       free_pkts(&m, 1);
-                       continue;
-               }
-
-               sa_idx = res - 1;
-               if (!inbound_sa_check(sa, m, sa_idx)) {
-                       free_pkts(&m, 1);
-                       continue;
-               }
-               ip->pkts[j++] = m;
-       }
-       ip->num = j;
-}
-
 static void
 split46_traffic(struct ipsec_traffic *trf, struct rte_mbuf *mb[], uint32_t num)
 {
@@ -753,6 +353,7 @@ static inline void
 process_pkts_inbound(struct ipsec_ctx *ipsec_ctx,
                struct ipsec_traffic *traffic)
 {
+       unsigned int lcoreid = rte_lcore_id();
        uint16_t nb_pkts_in, n_ip4, n_ip6;
 
        n_ip4 = traffic->ip4.num;
@@ -768,16 +369,20 @@ process_pkts_inbound(struct ipsec_ctx *ipsec_ctx,
                ipsec_process(ipsec_ctx, traffic);
        }
 
-       inbound_sp_sa(ipsec_ctx->sp4_ctx, ipsec_ctx->sa_ctx, &traffic->ip4,
-                       n_ip4);
+       inbound_sp_sa(ipsec_ctx->sp4_ctx,
+               ipsec_ctx->sa_ctx, &traffic->ip4, n_ip4,
+               &core_statistics[lcoreid].inbound.spd4);
 
-       inbound_sp_sa(ipsec_ctx->sp6_ctx, ipsec_ctx->sa_ctx, &traffic->ip6,
-                       n_ip6);
+       inbound_sp_sa(ipsec_ctx->sp6_ctx,
+               ipsec_ctx->sa_ctx, &traffic->ip6, n_ip6,
+               &core_statistics[lcoreid].inbound.spd6);
 }
 
 static inline void
-outbound_sp(struct sp_ctx *sp, struct traffic_type *ip,
-               struct traffic_type *ipsec)
+outbound_spd_lookup(struct sp_ctx *sp,
+               struct traffic_type *ip,
+               struct traffic_type *ipsec,
+               struct ipsec_spd_stats *stats)
 {
        struct rte_mbuf *m;
        uint32_t i, j, sa_idx;
@@ -788,17 +393,23 @@ outbound_sp(struct sp_ctx *sp, struct traffic_type *ip,
        rte_acl_classify((struct rte_acl_ctx *)sp, ip->data, ip->res,
                        ip->num, DEFAULT_MAX_CATEGORIES);
 
-       j = 0;
-       for (i = 0; i < ip->num; i++) {
+       for (i = 0, j = 0; i < ip->num; i++) {
                m = ip->pkts[i];
                sa_idx = ip->res[i] - 1;
-               if (ip->res[i] == DISCARD)
+
+               if (unlikely(ip->res[i] == DISCARD)) {
                        free_pkts(&m, 1);
-               else if (ip->res[i] == BYPASS)
+
+                       stats->discard++;
+               } else if (unlikely(ip->res[i] == BYPASS)) {
                        ip->pkts[j++] = m;
-               else {
+
+                       stats->bypass++;
+               } else {
                        ipsec->res[ipsec->num] = sa_idx;
                        ipsec->pkts[ipsec->num++] = m;
+
+                       stats->protect++;
                }
        }
        ip->num = j;
@@ -810,15 +421,20 @@ process_pkts_outbound(struct ipsec_ctx *ipsec_ctx,
 {
        struct rte_mbuf *m;
        uint16_t idx, nb_pkts_out, i;
+       unsigned int lcoreid = rte_lcore_id();
 
        /* Drop any IPsec traffic from protected ports */
        free_pkts(traffic->ipsec.pkts, traffic->ipsec.num);
 
        traffic->ipsec.num = 0;
 
-       outbound_sp(ipsec_ctx->sp4_ctx, &traffic->ip4, &traffic->ipsec);
+       outbound_spd_lookup(ipsec_ctx->sp4_ctx,
+               &traffic->ip4, &traffic->ipsec,
+               &core_statistics[lcoreid].outbound.spd4);
 
-       outbound_sp(ipsec_ctx->sp6_ctx, &traffic->ip6, &traffic->ipsec);
+       outbound_spd_lookup(ipsec_ctx->sp6_ctx,
+               &traffic->ip6, &traffic->ipsec,
+               &core_statistics[lcoreid].outbound.spd6);
 
        if (app_sa_prm.enable == 0) {
 
@@ -926,136 +542,6 @@ process_pkts_outbound_nosp(struct ipsec_ctx *ipsec_ctx,
        }
 }
 
-static inline int32_t
-get_hop_for_offload_pkt(struct rte_mbuf *pkt, int is_ipv6)
-{
-       struct ipsec_mbuf_metadata *priv;
-       struct ipsec_sa *sa;
-
-       priv = get_priv(pkt);
-
-       sa = priv->sa;
-       if (unlikely(sa == NULL)) {
-               RTE_LOG(ERR, IPSEC, "SA not saved in private data\n");
-               goto fail;
-       }
-
-       if (is_ipv6)
-               return sa->portid;
-
-       /* else */
-       return (sa->portid | RTE_LPM_LOOKUP_SUCCESS);
-
-fail:
-       if (is_ipv6)
-               return -1;
-
-       /* else */
-       return 0;
-}
-
-static inline void
-route4_pkts(struct rt_ctx *rt_ctx, struct rte_mbuf *pkts[], uint8_t nb_pkts)
-{
-       uint32_t hop[MAX_PKT_BURST * 2];
-       uint32_t dst_ip[MAX_PKT_BURST * 2];
-       int32_t pkt_hop = 0;
-       uint16_t i, offset;
-       uint16_t lpm_pkts = 0;
-
-       if (nb_pkts == 0)
-               return;
-
-       /* Need to do an LPM lookup for non-inline packets. Inline packets will
-        * have port ID in the SA
-        */
-
-       for (i = 0; i < nb_pkts; i++) {
-               if (!(pkts[i]->ol_flags & RTE_MBUF_F_TX_SEC_OFFLOAD)) {
-                       /* Security offload not enabled. So an LPM lookup is
-                        * required to get the hop
-                        */
-                       offset = offsetof(struct ip, ip_dst);
-                       dst_ip[lpm_pkts] = *rte_pktmbuf_mtod_offset(pkts[i],
-                                       uint32_t *, offset);
-                       dst_ip[lpm_pkts] = rte_be_to_cpu_32(dst_ip[lpm_pkts]);
-                       lpm_pkts++;
-               }
-       }
-
-       rte_lpm_lookup_bulk((struct rte_lpm *)rt_ctx, dst_ip, hop, lpm_pkts);
-
-       lpm_pkts = 0;
-
-       for (i = 0; i < nb_pkts; i++) {
-               if (pkts[i]->ol_flags & RTE_MBUF_F_TX_SEC_OFFLOAD) {
-                       /* Read hop from the SA */
-                       pkt_hop = get_hop_for_offload_pkt(pkts[i], 0);
-               } else {
-                       /* Need to use hop returned by lookup */
-                       pkt_hop = hop[lpm_pkts++];
-               }
-
-               if ((pkt_hop & RTE_LPM_LOOKUP_SUCCESS) == 0) {
-                       free_pkts(&pkts[i], 1);
-                       continue;
-               }
-               send_single_packet(pkts[i], pkt_hop & 0xff, IPPROTO_IP);
-       }
-}
-
-static inline void
-route6_pkts(struct rt_ctx *rt_ctx, struct rte_mbuf *pkts[], uint8_t nb_pkts)
-{
-       int32_t hop[MAX_PKT_BURST * 2];
-       uint8_t dst_ip[MAX_PKT_BURST * 2][16];
-       uint8_t *ip6_dst;
-       int32_t pkt_hop = 0;
-       uint16_t i, offset;
-       uint16_t lpm_pkts = 0;
-
-       if (nb_pkts == 0)
-               return;
-
-       /* Need to do an LPM lookup for non-inline packets. Inline packets will
-        * have port ID in the SA
-        */
-
-       for (i = 0; i < nb_pkts; i++) {
-               if (!(pkts[i]->ol_flags & RTE_MBUF_F_TX_SEC_OFFLOAD)) {
-                       /* Security offload not enabled. So an LPM lookup is
-                        * required to get the hop
-                        */
-                       offset = offsetof(struct ip6_hdr, ip6_dst);
-                       ip6_dst = rte_pktmbuf_mtod_offset(pkts[i], uint8_t *,
-                                       offset);
-                       memcpy(&dst_ip[lpm_pkts][0], ip6_dst, 16);
-                       lpm_pkts++;
-               }
-       }
-
-       rte_lpm6_lookup_bulk_func((struct rte_lpm6 *)rt_ctx, dst_ip, hop,
-                       lpm_pkts);
-
-       lpm_pkts = 0;
-
-       for (i = 0; i < nb_pkts; i++) {
-               if (pkts[i]->ol_flags & RTE_MBUF_F_TX_SEC_OFFLOAD) {
-                       /* Read hop from the SA */
-                       pkt_hop = get_hop_for_offload_pkt(pkts[i], 1);
-               } else {
-                       /* Need to use hop returned by lookup */
-                       pkt_hop = hop[lpm_pkts++];
-               }
-
-               if (pkt_hop == -1) {
-                       free_pkts(&pkts[i], 1);
-                       continue;
-               }
-               send_single_packet(pkts[i], pkt_hop & 0xff, IPPROTO_IPV6);
-       }
-}
-
 static inline void
 process_pkts(struct lcore_conf *qconf, struct rte_mbuf **pkts,
                uint8_t nb_pkts, uint16_t portid)
@@ -1080,21 +566,6 @@ process_pkts(struct lcore_conf *qconf, struct rte_mbuf **pkts,
        route6_pkts(qconf->rt6_ctx, traffic.ip6.pkts, traffic.ip6.num);
 }
 
-static inline void
-drain_tx_buffers(struct lcore_conf *qconf)
-{
-       struct buffer *buf;
-       uint32_t portid;
-
-       for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
-               buf = &qconf->tx_mbufs[portid];
-               if (buf->len == 0)
-                       continue;
-               send_burst(qconf, buf->len, portid);
-               buf->len = 0;
-       }
-}
-
 static inline void
 drain_crypto_buffers(struct lcore_conf *qconf)
 {
@@ -1122,6 +593,7 @@ drain_inbound_crypto_queues(const struct lcore_conf *qconf,
 {
        uint32_t n;
        struct ipsec_traffic trf;
+       unsigned int lcoreid = rte_lcore_id();
 
        if (app_sa_prm.enable == 0) {
 
@@ -1139,13 +611,15 @@ drain_inbound_crypto_queues(const struct lcore_conf *qconf,
 
        /* process ipv4 packets */
        if (trf.ip4.num != 0) {
-               inbound_sp_sa(ctx->sp4_ctx, ctx->sa_ctx, &trf.ip4, 0);
+               inbound_sp_sa(ctx->sp4_ctx, ctx->sa_ctx, &trf.ip4, 0,
+                       &core_statistics[lcoreid].inbound.spd4);
                route4_pkts(qconf->rt4_ctx, trf.ip4.pkts, trf.ip4.num);
        }
 
        /* process ipv6 packets */
        if (trf.ip6.num != 0) {
-               inbound_sp_sa(ctx->sp6_ctx, ctx->sa_ctx, &trf.ip6, 0);
+               inbound_sp_sa(ctx->sp6_ctx, ctx->sa_ctx, &trf.ip6, 0,
+                       &core_statistics[lcoreid].inbound.spd6);
                route6_pkts(qconf->rt6_ctx, trf.ip6.pkts, trf.ip6.num);
        }
 }
@@ -1218,7 +692,6 @@ ipsec_poll_mode_worker(void)
        qconf->outbound.session_pool = socket_ctx[socket_id].session_pool;
        qconf->outbound.session_priv_pool =
                        socket_ctx[socket_id].session_priv_pool;
-       qconf->frag.pool_dir = socket_ctx[socket_id].mbuf_pool;
        qconf->frag.pool_indir = socket_ctx[socket_id].mbuf_pool_indir;
 
        rc = ipsec_sad_lcore_cache_init(app_sa_prm.cache_sz);
@@ -1411,6 +884,9 @@ print_usage(const char *prgname)
                " [--" CMD_LINE_OPT_TX_OFFLOAD " TX_OFFLOAD_MASK]"
                " [--" CMD_LINE_OPT_REASSEMBLE " REASSEMBLE_TABLE_SIZE]"
                " [--" CMD_LINE_OPT_MTU " MTU]"
+               " [--event-vector]"
+               " [--vector-size SIZE]"
+               " [--vector-tmo TIMEOUT in ns]"
                "\n\n"
                "  -p PORTMASK: Hexadecimal bitmask of ports to configure\n"
                "  -P : Enable promiscuous mode\n"
@@ -1466,6 +942,13 @@ print_usage(const char *prgname)
                "  --" CMD_LINE_OPT_FRAG_TTL " FRAG_TTL_NS"
                ": fragments lifetime in nanoseconds, default\n"
                "    and maximum value is 10.000.000.000 ns (10 s)\n"
+               "  --event-vector enables event vectorization\n"
+               "  --vector-size Max vector size (default value: 16)\n"
+               "  --vector-tmo Max vector timeout in nanoseconds"
+               "    (default value: 102400)\n"
+               "  --" CMD_LINE_OPT_PER_PORT_POOL " Enable per port mbuf pool\n"
+               "  --" CMD_LINE_OPT_VECTOR_POOL_SZ " Vector pool size\n"
+               "                    (default value is based on mbuf count)\n"
                "\n",
                prgname);
 }
@@ -1632,6 +1115,7 @@ parse_args(int32_t argc, char **argv, struct eh_conf *eh_conf)
        int32_t option_index;
        char *prgname = argv[0];
        int32_t f_present = 0;
+       struct eventmode_conf *em_conf = NULL;
 
        argvopt = argv;
 
@@ -1826,6 +1310,37 @@ parse_args(int32_t argc, char **argv, struct eh_conf *eh_conf)
                        }
                        frag_ttl_ns = ret;
                        break;
+               case CMD_LINE_OPT_EVENT_VECTOR_NUM:
+                       em_conf = eh_conf->mode_params;
+                       em_conf->ext_params.event_vector = 1;
+                       break;
+               case CMD_LINE_OPT_VECTOR_SIZE_NUM:
+                       ret = parse_decimal(optarg);
+
+                       if (ret > MAX_PKT_BURST_VEC) {
+                               printf("Invalid argument for \'%s\': %s\n",
+                                       CMD_LINE_OPT_VECTOR_SIZE, optarg);
+                               print_usage(prgname);
+                               return -1;
+                       }
+                       em_conf = eh_conf->mode_params;
+                       em_conf->ext_params.vector_size = ret;
+                       break;
+               case CMD_LINE_OPT_VECTOR_TIMEOUT_NUM:
+                       ret = parse_decimal(optarg);
+
+                       em_conf = eh_conf->mode_params;
+                       em_conf->vector_tmo_ns = ret;
+                       break;
+               case CMD_LINE_OPT_VECTOR_POOL_SZ_NUM:
+                       ret = parse_decimal(optarg);
+
+                       em_conf = eh_conf->mode_params;
+                       em_conf->vector_pool_sz = ret;
+                       break;
+               case CMD_LINE_OPT_PER_PORT_POOL_NUM:
+                       per_port_pool = 1;
+                       break;
                default:
                        print_usage(prgname);
                        return -1;
@@ -1982,7 +1497,7 @@ add_mapping(struct rte_hash *map, const char *str, uint16_t cdev_id,
 
        ret = rte_hash_add_key_data(map, &key, (void *)i);
        if (ret < 0) {
-               printf("Faled to insert cdev mapping for (lcore %u, "
+               printf("Failed to insert cdev mapping for (lcore %u, "
                                "cdev %u, qp %u), errno %d\n",
                                key.lcore_id, ipsec_ctx->tbl[i].id,
                                ipsec_ctx->tbl[i].qp, ret);
@@ -2015,7 +1530,7 @@ add_cdev_mapping(struct rte_cryptodev_info *dev_info, uint16_t cdev_id,
                str = "Inbound";
        }
 
-       /* Required cryptodevs with operation chainning */
+       /* Required cryptodevs with operation chaining */
        if (!(dev_info->feature_flags &
                                RTE_CRYPTODEV_FF_SYM_OPERATION_CHAINING))
                return ret;
@@ -2165,6 +1680,147 @@ cryptodevs_init(uint16_t req_queue_num)
        return total_nb_qps;
 }
 
+static int
+check_ptype(int portid)
+{
+       int l3_ipv4 = 0, l3_ipv6 = 0, l4_udp = 0, tunnel_esp = 0;
+       int i, nb_ptypes;
+       uint32_t mask;
+
+       mask = (RTE_PTYPE_L3_MASK | RTE_PTYPE_L4_MASK |
+                     RTE_PTYPE_TUNNEL_MASK);
+
+       nb_ptypes = rte_eth_dev_get_supported_ptypes(portid, mask, NULL, 0);
+       if (nb_ptypes <= 0)
+               return 0;
+
+       uint32_t ptypes[nb_ptypes];
+
+       nb_ptypes = rte_eth_dev_get_supported_ptypes(portid, mask, ptypes, nb_ptypes);
+       for (i = 0; i < nb_ptypes; ++i) {
+               if (RTE_ETH_IS_IPV4_HDR(ptypes[i]))
+                       l3_ipv4 = 1;
+               if (RTE_ETH_IS_IPV6_HDR(ptypes[i]))
+                       l3_ipv6 = 1;
+               if ((ptypes[i] & RTE_PTYPE_TUNNEL_MASK) == RTE_PTYPE_TUNNEL_ESP)
+                       tunnel_esp = 1;
+               if ((ptypes[i] & RTE_PTYPE_L4_MASK) == RTE_PTYPE_L4_UDP)
+                       l4_udp = 1;
+       }
+
+       if (l3_ipv4 == 0)
+               printf("port %d cannot parse RTE_PTYPE_L3_IPV4\n", portid);
+
+       if (l3_ipv6 == 0)
+               printf("port %d cannot parse RTE_PTYPE_L3_IPV6\n", portid);
+
+       if (l4_udp == 0)
+               printf("port %d cannot parse RTE_PTYPE_L4_UDP\n", portid);
+
+       if (tunnel_esp == 0)
+               printf("port %d cannot parse RTE_PTYPE_TUNNEL_ESP\n", portid);
+
+       if (l3_ipv4 && l3_ipv6 && l4_udp && tunnel_esp)
+               return 1;
+
+       return 0;
+
+}
+
+static inline void
+parse_ptype(struct rte_mbuf *m)
+{
+       uint32_t packet_type = RTE_PTYPE_UNKNOWN;
+       const struct rte_ipv4_hdr *iph4;
+       const struct rte_ipv6_hdr *iph6;
+       const struct rte_ether_hdr *eth;
+       const struct rte_udp_hdr *udp;
+       uint16_t nat_port, ether_type;
+       int next_proto = 0;
+       size_t ext_len = 0;
+       const uint8_t *p;
+       uint32_t l3len;
+
+       eth = rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
+       ether_type = eth->ether_type;
+
+       if (ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4)) {
+               iph4 = (const struct rte_ipv4_hdr *)(eth + 1);
+               l3len = ((iph4->version_ihl & RTE_IPV4_HDR_IHL_MASK) *
+                              RTE_IPV4_IHL_MULTIPLIER);
+
+               if (l3len == sizeof(struct rte_ipv4_hdr))
+                       packet_type |= RTE_PTYPE_L3_IPV4;
+               else
+                       packet_type |= RTE_PTYPE_L3_IPV4_EXT_UNKNOWN;
+
+               next_proto = iph4->next_proto_id;
+               p = (const uint8_t *)iph4;
+       } else if (ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6)) {
+               iph6 = (const struct rte_ipv6_hdr *)(eth + 1);
+               l3len = sizeof(struct ip6_hdr);
+
+               /* determine l3 header size up to ESP extension */
+               next_proto = iph6->proto;
+               p = (const uint8_t *)iph6;
+               while (next_proto != IPPROTO_ESP && l3len < m->data_len &&
+                       (next_proto = rte_ipv6_get_next_ext(p + l3len,
+                                               next_proto, &ext_len)) >= 0)
+                       l3len += ext_len;
+
+               /* Skip IPv6 header exceeds first segment length */
+               if (unlikely(l3len + RTE_ETHER_HDR_LEN > m->data_len))
+                       goto exit;
+
+               if (l3len == sizeof(struct ip6_hdr))
+                       packet_type |= RTE_PTYPE_L3_IPV6;
+               else
+                       packet_type |= RTE_PTYPE_L3_IPV6_EXT;
+       }
+
+       switch (next_proto) {
+       case IPPROTO_ESP:
+               packet_type |= RTE_PTYPE_TUNNEL_ESP;
+               break;
+       case IPPROTO_UDP:
+               if (app_sa_prm.udp_encap == 1) {
+                       udp = (const struct rte_udp_hdr *)(p + l3len);
+                       nat_port = rte_cpu_to_be_16(IPSEC_NAT_T_PORT);
+                       if (udp->src_port == nat_port ||
+                           udp->dst_port == nat_port)
+                               packet_type |=
+                                       MBUF_PTYPE_TUNNEL_ESP_IN_UDP;
+               }
+               break;
+       default:
+               break;
+       }
+exit:
+       m->packet_type = packet_type;
+}
+
+static uint16_t
+parse_ptype_cb(uint16_t port __rte_unused, uint16_t queue __rte_unused,
+              struct rte_mbuf *pkts[], uint16_t nb_pkts,
+              uint16_t max_pkts __rte_unused,
+              void *user_param __rte_unused)
+{
+       uint32_t i;
+
+       if (unlikely(nb_pkts == 0))
+               return nb_pkts;
+
+       rte_prefetch0(rte_pktmbuf_mtod(pkts[0], struct ether_hdr *));
+       for (i = 0; i < (unsigned int) (nb_pkts - 1); ++i) {
+               rte_prefetch0(rte_pktmbuf_mtod(pkts[i+1],
+                       struct ether_hdr *));
+               parse_ptype(pkts[i]);
+       }
+       parse_ptype(pkts[i]);
+
+       return nb_pkts;
+}
+
 static void
 port_init(uint16_t portid, uint64_t req_rx_offloads, uint64_t req_tx_offloads)
 {
@@ -2176,6 +1832,7 @@ port_init(uint16_t portid, uint64_t req_rx_offloads, uint64_t req_tx_offloads)
        struct lcore_conf *qconf;
        struct rte_ether_addr ethaddr;
        struct rte_eth_conf local_port_conf = port_conf;
+       int ptype_supported;
 
        ret = rte_eth_dev_info_get(portid, &dev_info);
        if (ret != 0)
@@ -2183,7 +1840,7 @@ port_init(uint16_t portid, uint64_t req_rx_offloads, uint64_t req_tx_offloads)
                        "Error during getting device (port %u) info: %s\n",
                        portid, strerror(-ret));
 
-       /* limit allowed HW offloafs, as user requested */
+       /* limit allowed HW offloads, as user requested */
        dev_info.rx_offload_capa &= dev_rx_offload;
        dev_info.tx_offload_capa &= dev_tx_offload;
 
@@ -2230,7 +1887,7 @@ port_init(uint16_t portid, uint64_t req_rx_offloads, uint64_t req_tx_offloads)
                        local_port_conf.rxmode.offloads)
                rte_exit(EXIT_FAILURE,
                        "Error: port %u required RX offloads: 0x%" PRIx64
-                       ", avaialbe RX offloads: 0x%" PRIx64 "\n",
+                       ", available RX offloads: 0x%" PRIx64 "\n",
                        portid, local_port_conf.rxmode.offloads,
                        dev_info.rx_offload_capa);
 
@@ -2238,7 +1895,7 @@ port_init(uint16_t portid, uint64_t req_rx_offloads, uint64_t req_tx_offloads)
                        local_port_conf.txmode.offloads)
                rte_exit(EXIT_FAILURE,
                        "Error: port %u required TX offloads: 0x%" PRIx64
-                       ", avaialbe TX offloads: 0x%" PRIx64 "\n",
+                       ", available TX offloads: 0x%" PRIx64 "\n",
                        portid, local_port_conf.txmode.offloads,
                        dev_info.tx_offload_capa);
 
@@ -2246,10 +1903,7 @@ port_init(uint16_t portid, uint64_t req_rx_offloads, uint64_t req_tx_offloads)
                local_port_conf.txmode.offloads |=
                        RTE_ETH_TX_OFFLOAD_MBUF_FAST_FREE;
 
-       if (dev_info.tx_offload_capa & RTE_ETH_TX_OFFLOAD_IPV4_CKSUM)
-               local_port_conf.txmode.offloads |= RTE_ETH_TX_OFFLOAD_IPV4_CKSUM;
-
-       printf("port %u configurng rx_offloads=0x%" PRIx64
+       printf("port %u configuring rx_offloads=0x%" PRIx64
                ", tx_offloads=0x%" PRIx64 "\n",
                portid, local_port_conf.rxmode.offloads,
                local_port_conf.txmode.offloads);
@@ -2276,6 +1930,11 @@ port_init(uint16_t portid, uint64_t req_rx_offloads, uint64_t req_tx_offloads)
                rte_exit(EXIT_FAILURE, "Cannot adjust number of descriptors: "
                                "err=%d, port=%d\n", ret, portid);
 
+       /* Check if required ptypes are supported */
+       ptype_supported = check_ptype(portid);
+       if (!ptype_supported)
+               printf("Port %d: softly parse packet type info\n", portid);
+
        /* init one TX queue per lcore */
        tx_queueid = 0;
        for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
@@ -2313,6 +1972,7 @@ port_init(uint16_t portid, uint64_t req_rx_offloads, uint64_t req_tx_offloads)
                /* init RX queues */
                for (queue = 0; queue < qconf->nb_rx_queue; ++queue) {
                        struct rte_eth_rxconf rxq_conf;
+                       struct rte_mempool *pool;
 
                        if (portid != qconf->rx_queue_list[queue].port_id)
                                continue;
@@ -2324,13 +1984,28 @@ port_init(uint16_t portid, uint64_t req_rx_offloads, uint64_t req_tx_offloads)
 
                        rxq_conf = dev_info.default_rxconf;
                        rxq_conf.offloads = local_port_conf.rxmode.offloads;
+
+                       if (per_port_pool)
+                               pool = socket_ctx[socket_id].mbuf_pool[portid];
+                       else
+                               pool = socket_ctx[socket_id].mbuf_pool[0];
+
                        ret = rte_eth_rx_queue_setup(portid, rx_queueid,
-                                       nb_rxd, socket_id, &rxq_conf,
-                                       socket_ctx[socket_id].mbuf_pool);
+                                       nb_rxd, socket_id, &rxq_conf, pool);
                        if (ret < 0)
                                rte_exit(EXIT_FAILURE,
                                        "rte_eth_rx_queue_setup: err=%d, "
                                        "port=%d\n", ret, portid);
+
+                       /* Register Rx callback if ptypes are not supported */
+                       if (!ptype_supported &&
+                           !rte_eth_add_rx_callback(portid, queue,
+                                                    parse_ptype_cb, NULL)) {
+                               printf("Failed to add rx callback: port=%d, "
+                                      "queue=%d\n", portid, queue);
+                       }
+
+
                }
        }
        printf("\n");
@@ -2439,28 +2114,37 @@ session_priv_pool_init(struct socket_ctx *ctx, int32_t socket_id,
 }
 
 static void
-pool_init(struct socket_ctx *ctx, int32_t socket_id, uint32_t nb_mbuf)
+pool_init(struct socket_ctx *ctx, int32_t socket_id, int portid,
+         uint32_t nb_mbuf)
 {
        char s[64];
        int32_t ms;
 
-       snprintf(s, sizeof(s), "mbuf_pool_%d", socket_id);
-       ctx->mbuf_pool = rte_pktmbuf_pool_create(s, nb_mbuf,
-                       MEMPOOL_CACHE_SIZE, ipsec_metadata_size(),
-                       frame_buf_size, socket_id);
+
+       /* mbuf_pool is initialised by the pool_init() function*/
+       if (socket_ctx[socket_id].mbuf_pool[portid])
+               return;
+
+       snprintf(s, sizeof(s), "mbuf_pool_%d_%d", socket_id, portid);
+       ctx->mbuf_pool[portid] = rte_pktmbuf_pool_create(s, nb_mbuf,
+                                                        MEMPOOL_CACHE_SIZE,
+                                                        ipsec_metadata_size(),
+                                                        frame_buf_size,
+                                                        socket_id);
 
        /*
         * if multi-segment support is enabled, then create a pool
-        * for indirect mbufs.
+        * for indirect mbufs. This is not per-port but global.
         */
        ms = multi_seg_required();
-       if (ms != 0) {
+       if (ms != 0 && !ctx->mbuf_pool_indir) {
                snprintf(s, sizeof(s), "mbuf_pool_indir_%d", socket_id);
                ctx->mbuf_pool_indir = rte_pktmbuf_pool_create(s, nb_mbuf,
                        MEMPOOL_CACHE_SIZE, 0, 0, socket_id);
        }
 
-       if (ctx->mbuf_pool == NULL || (ms != 0 && ctx->mbuf_pool_indir == NULL))
+       if (ctx->mbuf_pool[portid] == NULL ||
+           (ms != 0 && ctx->mbuf_pool_indir == NULL))
                rte_exit(EXIT_FAILURE, "Cannot init mbuf pool on socket %d\n",
                                socket_id);
        else
@@ -2522,6 +2206,17 @@ inline_ipsec_event_callback(uint16_t port_id, enum rte_eth_event_type type,
        return -1;
 }
 
+static int
+ethdev_reset_event_callback(uint16_t port_id,
+               enum rte_eth_event_type type,
+                void *param __rte_unused, void *ret_param __rte_unused)
+{
+       printf("Reset Event on port id %d type %d\n", port_id, type);
+       printf("Force quit application");
+       force_quit = true;
+       return 0;
+}
+
 static uint16_t
 rx_callback(__rte_unused uint16_t port, __rte_unused uint16_t queue,
        struct rte_mbuf *pkt[], uint16_t nb_pkts,
@@ -2568,7 +2263,7 @@ rx_callback(__rte_unused uint16_t port, __rte_unused uint16_t queue,
                                rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6)) {
 
                        struct rte_ipv6_hdr *iph;
-                       struct ipv6_extension_fragment *fh;
+                       struct rte_ipv6_fragment_ext *fh;
 
                        iph = (struct rte_ipv6_hdr *)(eth + 1);
                        fh = rte_ipv6_frag_get_ipv6_fragment_header(iph);
@@ -2835,6 +2530,333 @@ calculate_nb_mbufs(uint16_t nb_ports, uint16_t nb_crypto_qp, uint32_t nb_rxq,
                       8192U);
 }
 
+
+static int
+handle_telemetry_cmd_ipsec_secgw_stats(const char *cmd __rte_unused,
+               const char *params, struct rte_tel_data *data)
+{
+       uint64_t total_pkts_dropped = 0, total_pkts_tx = 0, total_pkts_rx = 0;
+       unsigned int coreid;
+
+       rte_tel_data_start_dict(data);
+
+       if (params) {
+               coreid = (uint32_t)atoi(params);
+               if (rte_lcore_is_enabled(coreid) == 0)
+                       return -EINVAL;
+
+               total_pkts_dropped = core_statistics[coreid].dropped;
+               total_pkts_tx = core_statistics[coreid].tx;
+               total_pkts_rx = core_statistics[coreid].rx;
+
+       } else {
+               for (coreid = 0; coreid < RTE_MAX_LCORE; coreid++) {
+
+                       /* skip disabled cores */
+                       if (rte_lcore_is_enabled(coreid) == 0)
+                               continue;
+
+                       total_pkts_dropped += core_statistics[coreid].dropped;
+                       total_pkts_tx += core_statistics[coreid].tx;
+                       total_pkts_rx += core_statistics[coreid].rx;
+               }
+       }
+
+       /* add telemetry key/values pairs */
+       rte_tel_data_add_dict_u64(data, "packets received",
+                               total_pkts_rx);
+
+       rte_tel_data_add_dict_u64(data, "packets transmitted",
+                               total_pkts_tx);
+
+       rte_tel_data_add_dict_u64(data, "packets dropped",
+                               total_pkts_dropped);
+
+
+       return 0;
+}
+
+static void
+update_lcore_statistics(struct ipsec_core_statistics *total, uint32_t coreid)
+{
+       struct ipsec_core_statistics *lcore_stats;
+
+       /* skip disabled cores */
+       if (rte_lcore_is_enabled(coreid) == 0)
+               return;
+
+       lcore_stats = &core_statistics[coreid];
+
+       total->rx = lcore_stats->rx;
+       total->dropped = lcore_stats->dropped;
+       total->tx = lcore_stats->tx;
+
+       /* outbound stats */
+       total->outbound.spd6.protect += lcore_stats->outbound.spd6.protect;
+       total->outbound.spd6.bypass += lcore_stats->outbound.spd6.bypass;
+       total->outbound.spd6.discard += lcore_stats->outbound.spd6.discard;
+
+       total->outbound.spd4.protect += lcore_stats->outbound.spd4.protect;
+       total->outbound.spd4.bypass += lcore_stats->outbound.spd4.bypass;
+       total->outbound.spd4.discard += lcore_stats->outbound.spd4.discard;
+
+       total->outbound.sad.miss += lcore_stats->outbound.sad.miss;
+
+       /* inbound stats */
+       total->inbound.spd6.protect += lcore_stats->inbound.spd6.protect;
+       total->inbound.spd6.bypass += lcore_stats->inbound.spd6.bypass;
+       total->inbound.spd6.discard += lcore_stats->inbound.spd6.discard;
+
+       total->inbound.spd4.protect += lcore_stats->inbound.spd4.protect;
+       total->inbound.spd4.bypass += lcore_stats->inbound.spd4.bypass;
+       total->inbound.spd4.discard += lcore_stats->inbound.spd4.discard;
+
+       total->inbound.sad.miss += lcore_stats->inbound.sad.miss;
+
+
+       /* routing stats */
+       total->lpm4.miss += lcore_stats->lpm4.miss;
+       total->lpm6.miss += lcore_stats->lpm6.miss;
+}
+
+static void
+update_statistics(struct ipsec_core_statistics *total, uint32_t coreid)
+{
+       memset(total, 0, sizeof(*total));
+
+       if (coreid != UINT32_MAX) {
+               update_lcore_statistics(total, coreid);
+       } else {
+               for (coreid = 0; coreid < RTE_MAX_LCORE; coreid++)
+                       update_lcore_statistics(total, coreid);
+       }
+}
+
+static int
+handle_telemetry_cmd_ipsec_secgw_stats_outbound(const char *cmd __rte_unused,
+               const char *params, struct rte_tel_data *data)
+{
+       struct ipsec_core_statistics total_stats;
+
+       struct rte_tel_data *spd4_data = rte_tel_data_alloc();
+       struct rte_tel_data *spd6_data = rte_tel_data_alloc();
+       struct rte_tel_data *sad_data = rte_tel_data_alloc();
+       unsigned int coreid = UINT32_MAX;
+       int rc = 0;
+
+       /* verify allocated telemetry data structures */
+       if (!spd4_data || !spd6_data || !sad_data) {
+               rc = -ENOMEM;
+               goto exit;
+       }
+
+       /* initialize telemetry data structs as dicts */
+       rte_tel_data_start_dict(data);
+
+       rte_tel_data_start_dict(spd4_data);
+       rte_tel_data_start_dict(spd6_data);
+       rte_tel_data_start_dict(sad_data);
+
+       if (params) {
+               coreid = (uint32_t)atoi(params);
+               if (rte_lcore_is_enabled(coreid) == 0) {
+                       rc = -EINVAL;
+                       goto exit;
+               }
+       }
+
+       update_statistics(&total_stats, coreid);
+
+       /* add spd 4 telemetry key/values pairs */
+
+       rte_tel_data_add_dict_u64(spd4_data, "protect",
+               total_stats.outbound.spd4.protect);
+       rte_tel_data_add_dict_u64(spd4_data, "bypass",
+               total_stats.outbound.spd4.bypass);
+       rte_tel_data_add_dict_u64(spd4_data, "discard",
+               total_stats.outbound.spd4.discard);
+
+       rte_tel_data_add_dict_container(data, "spd4", spd4_data, 0);
+
+       /* add spd 6 telemetry key/values pairs */
+
+       rte_tel_data_add_dict_u64(spd6_data, "protect",
+               total_stats.outbound.spd6.protect);
+       rte_tel_data_add_dict_u64(spd6_data, "bypass",
+               total_stats.outbound.spd6.bypass);
+       rte_tel_data_add_dict_u64(spd6_data, "discard",
+               total_stats.outbound.spd6.discard);
+
+       rte_tel_data_add_dict_container(data, "spd6", spd6_data, 0);
+
+       /* add sad telemetry key/values pairs */
+
+       rte_tel_data_add_dict_u64(sad_data, "miss",
+               total_stats.outbound.sad.miss);
+
+       rte_tel_data_add_dict_container(data, "sad", sad_data, 0);
+
+exit:
+       if (rc) {
+               rte_tel_data_free(spd4_data);
+               rte_tel_data_free(spd6_data);
+               rte_tel_data_free(sad_data);
+       }
+       return rc;
+}
+
+static int
+handle_telemetry_cmd_ipsec_secgw_stats_inbound(const char *cmd __rte_unused,
+               const char *params, struct rte_tel_data *data)
+{
+       struct ipsec_core_statistics total_stats;
+
+       struct rte_tel_data *spd4_data = rte_tel_data_alloc();
+       struct rte_tel_data *spd6_data = rte_tel_data_alloc();
+       struct rte_tel_data *sad_data = rte_tel_data_alloc();
+       unsigned int coreid = UINT32_MAX;
+       int rc = 0;
+
+       /* verify allocated telemetry data structures */
+       if (!spd4_data || !spd6_data || !sad_data) {
+               rc = -ENOMEM;
+               goto exit;
+       }
+
+       /* initialize telemetry data structs as dicts */
+       rte_tel_data_start_dict(data);
+       rte_tel_data_start_dict(spd4_data);
+       rte_tel_data_start_dict(spd6_data);
+       rte_tel_data_start_dict(sad_data);
+
+       /* add children dicts to parent dict */
+
+       if (params) {
+               coreid = (uint32_t)atoi(params);
+               if (rte_lcore_is_enabled(coreid) == 0) {
+                       rc = -EINVAL;
+                       goto exit;
+               }
+       }
+
+       update_statistics(&total_stats, coreid);
+
+       /* add sad telemetry key/values pairs */
+
+       rte_tel_data_add_dict_u64(sad_data, "miss",
+               total_stats.inbound.sad.miss);
+
+       rte_tel_data_add_dict_container(data, "sad", sad_data, 0);
+
+       /* add spd 4 telemetry key/values pairs */
+
+       rte_tel_data_add_dict_u64(spd4_data, "protect",
+               total_stats.inbound.spd4.protect);
+       rte_tel_data_add_dict_u64(spd4_data, "bypass",
+               total_stats.inbound.spd4.bypass);
+       rte_tel_data_add_dict_u64(spd4_data, "discard",
+               total_stats.inbound.spd4.discard);
+
+       rte_tel_data_add_dict_container(data, "spd4", spd4_data, 0);
+
+       /* add spd 6 telemetry key/values pairs */
+
+       rte_tel_data_add_dict_u64(spd6_data, "protect",
+               total_stats.inbound.spd6.protect);
+       rte_tel_data_add_dict_u64(spd6_data, "bypass",
+               total_stats.inbound.spd6.bypass);
+       rte_tel_data_add_dict_u64(spd6_data, "discard",
+               total_stats.inbound.spd6.discard);
+
+       rte_tel_data_add_dict_container(data, "spd6", spd6_data, 0);
+
+exit:
+       if (rc) {
+               rte_tel_data_free(spd4_data);
+               rte_tel_data_free(spd6_data);
+               rte_tel_data_free(sad_data);
+       }
+       return rc;
+}
+
+static int
+handle_telemetry_cmd_ipsec_secgw_stats_routing(const char *cmd __rte_unused,
+               const char *params, struct rte_tel_data *data)
+{
+       struct ipsec_core_statistics total_stats;
+
+       struct rte_tel_data *lpm4_data = rte_tel_data_alloc();
+       struct rte_tel_data *lpm6_data = rte_tel_data_alloc();
+       unsigned int coreid = UINT32_MAX;
+       int rc = 0;
+
+       /* verify allocated telemetry data structures */
+       if (!lpm4_data || !lpm6_data) {
+               rc = -ENOMEM;
+               goto exit;
+       }
+
+       /* initialize telemetry data structs as dicts */
+       rte_tel_data_start_dict(data);
+       rte_tel_data_start_dict(lpm4_data);
+       rte_tel_data_start_dict(lpm6_data);
+
+
+       if (params) {
+               coreid = (uint32_t)atoi(params);
+               if (rte_lcore_is_enabled(coreid) == 0) {
+                       rc = -EINVAL;
+                       goto exit;
+               }
+       }
+
+       update_statistics(&total_stats, coreid);
+
+       /* add lpm 4 telemetry key/values pairs */
+       rte_tel_data_add_dict_u64(lpm4_data, "miss",
+               total_stats.lpm4.miss);
+
+       rte_tel_data_add_dict_container(data, "IPv4 LPM", lpm4_data, 0);
+
+       /* add lpm 6 telemetry key/values pairs */
+       rte_tel_data_add_dict_u64(lpm6_data, "miss",
+               total_stats.lpm6.miss);
+
+       rte_tel_data_add_dict_container(data, "IPv6 LPM", lpm6_data, 0);
+
+exit:
+       if (rc) {
+               rte_tel_data_free(lpm4_data);
+               rte_tel_data_free(lpm6_data);
+       }
+       return rc;
+}
+
+static void
+ipsec_secgw_telemetry_init(void)
+{
+       rte_telemetry_register_cmd("/examples/ipsec-secgw/stats",
+               handle_telemetry_cmd_ipsec_secgw_stats,
+               "Returns global stats. "
+               "Optional Parameters: int <logical core id>");
+
+       rte_telemetry_register_cmd("/examples/ipsec-secgw/stats/outbound",
+               handle_telemetry_cmd_ipsec_secgw_stats_outbound,
+               "Returns outbound global stats. "
+               "Optional Parameters: int <logical core id>");
+
+       rte_telemetry_register_cmd("/examples/ipsec-secgw/stats/inbound",
+               handle_telemetry_cmd_ipsec_secgw_stats_inbound,
+               "Returns inbound global stats. "
+               "Optional Parameters: int <logical core id>");
+
+       rte_telemetry_register_cmd("/examples/ipsec-secgw/stats/routing",
+               handle_telemetry_cmd_ipsec_secgw_stats_routing,
+               "Returns routing stats. "
+               "Optional Parameters: int <logical core id>");
+}
+
+
 int32_t
 main(int32_t argc, char **argv)
 {
@@ -2872,6 +2894,8 @@ main(int32_t argc, char **argv)
        if (ret < 0)
                rte_exit(EXIT_FAILURE, "Invalid parameters\n");
 
+       ipsec_secgw_telemetry_init();
+
        /* parse configuration file */
        if (parse_cfg_file(cfgfile) < 0) {
                printf("parsing file \"%s\" failed\n",
@@ -2885,6 +2909,9 @@ main(int32_t argc, char **argv)
                rte_exit(EXIT_FAILURE, "Invalid unprotected portmask 0x%x\n",
                                unprotected_port_mask);
 
+       if (unprotected_port_mask && !nb_sa_in)
+               rte_exit(EXIT_FAILURE, "Cannot use unprotected portmask without configured SA inbound\n");
+
        if (check_poll_mode_params(eh_conf) < 0)
                rte_exit(EXIT_FAILURE, "check_poll_mode_params failed\n");
 
@@ -2933,11 +2960,22 @@ main(int32_t argc, char **argv)
                else
                        socket_id = 0;
 
-               /* mbuf_pool is initialised by the pool_init() function*/
-               if (socket_ctx[socket_id].mbuf_pool)
+               if (per_port_pool) {
+                       RTE_ETH_FOREACH_DEV(portid) {
+                               if ((enabled_port_mask & (1 << portid)) == 0)
+                                       continue;
+
+                               pool_init(&socket_ctx[socket_id], socket_id,
+                                         portid, nb_bufs_in_pool);
+                       }
+               } else {
+                       pool_init(&socket_ctx[socket_id], socket_id, 0,
+                                 nb_bufs_in_pool);
+               }
+
+               if (socket_ctx[socket_id].session_pool)
                        continue;
 
-               pool_init(&socket_ctx[socket_id], socket_id, nb_bufs_in_pool);
                session_pool_init(&socket_ctx[socket_id], socket_id, sess_sz);
                session_priv_pool_init(&socket_ctx[socket_id], socket_id,
                        sess_sz);
@@ -2971,13 +3009,14 @@ main(int32_t argc, char **argv)
                if ((enabled_port_mask & (1 << portid)) == 0)
                        continue;
 
-               /* Create flow before starting the device */
-               create_default_ipsec_flow(portid, req_rx_offloads[portid]);
-
                ret = rte_eth_dev_start(portid);
                if (ret < 0)
                        rte_exit(EXIT_FAILURE, "rte_eth_dev_start: "
                                        "err=%d, port=%d\n", ret, portid);
+
+               /* Create flow after starting the device */
+               create_default_ipsec_flow(portid, req_rx_offloads[portid]);
+
                /*
                 * If enabled, put device in promiscuous mode.
                 * This allows IO forwarding mode to forward packets
@@ -2992,6 +3031,9 @@ main(int32_t argc, char **argv)
                                        rte_strerror(-ret), portid);
                }
 
+               rte_eth_dev_callback_register(portid, RTE_ETH_EVENT_INTR_RESET,
+                       ethdev_reset_event_callback, NULL);
+
                rte_eth_dev_callback_register(portid,
                        RTE_ETH_EVENT_IPSEC, inline_ipsec_event_callback, NULL);
        }
@@ -3006,7 +3048,7 @@ main(int32_t argc, char **argv)
        /* Replicate each context per socket */
        for (i = 0; i < NB_SOCKETS && i < rte_socket_count(); i++) {
                socket_id = rte_socket_id_by_idx(i);
-               if ((socket_ctx[socket_id].mbuf_pool != NULL) &&
+               if ((socket_ctx[socket_id].session_pool != NULL) &&
                        (socket_ctx[socket_id].sa_in == NULL) &&
                        (socket_ctx[socket_id].sa_out == NULL)) {
                        sa_init(&socket_ctx[socket_id], socket_id);