app/test-pmd/csumonly.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2010-2014 Intel Corporation.
   3  * Copyright 2014 6WIND S.A.
   4  */
   5
   6 #include <stdarg.h>
   7 #include <stdio.h>
   8 #include <errno.h>
   9 #include <stdint.h>
  10 #include <unistd.h>
  11 #include <inttypes.h>
  12
  13 #include <sys/queue.h>
  14 #include <sys/stat.h>
  15
  16 #include <rte_common.h>
  17 #include <rte_byteorder.h>
  18 #include <rte_log.h>
  19 #include <rte_debug.h>
  20 #include <rte_cycles.h>
  21 #include <rte_memory.h>
  22 #include <rte_memcpy.h>
  23 #include <rte_launch.h>
  24 #include <rte_eal.h>
  25 #include <rte_per_lcore.h>
  26 #include <rte_lcore.h>
  27 #include <rte_atomic.h>
  28 #include <rte_branch_prediction.h>
  29 #include <rte_mempool.h>
  30 #include <rte_mbuf.h>
  31 #include <rte_interrupts.h>
  32 #include <rte_pci.h>
  33 #include <rte_ether.h>
  34 #include <rte_ethdev.h>
  35 #include <rte_ip.h>
  36 #include <rte_tcp.h>
  37 #include <rte_udp.h>
  38 #include <rte_vxlan.h>
  39 #include <rte_sctp.h>
  40 #include <rte_gtp.h>
  41 #include <rte_prefetch.h>
  42 #include <rte_string_fns.h>
  43 #include <rte_flow.h>
  44 #include <rte_gro.h>
  45 #include <rte_gso.h>
  46 #include <rte_geneve.h>
  47
  48 #include "testpmd.h"
  49
  50 #define IP_DEFTTL  64   /* from RFC 1340. */
  51
  52 #define GRE_CHECKSUM_PRESENT    0x8000
  53 #define GRE_KEY_PRESENT         0x2000
  54 #define GRE_SEQUENCE_PRESENT    0x1000
  55 #define GRE_EXT_LEN             4
  56 #define GRE_SUPPORTED_FIELDS    (GRE_CHECKSUM_PRESENT | GRE_KEY_PRESENT |\
  57                                  GRE_SEQUENCE_PRESENT)
  58
  59 /* We cannot use rte_cpu_to_be_16() on a constant in a switch/case */
  60 #if RTE_BYTE_ORDER == RTE_LITTLE_ENDIAN
  61 #define _htons(x) ((uint16_t)((((x) & 0x00ffU) << 8) | (((x) & 0xff00U) >> 8)))
  62 #else
  63 #define _htons(x) (x)
  64 #endif
  65
  66 uint16_t vxlan_gpe_udp_port = RTE_VXLAN_GPE_DEFAULT_PORT;
  67 uint16_t geneve_udp_port = RTE_GENEVE_DEFAULT_PORT;
  68
  69 /* structure that caches offload info for the current packet */
  70 struct testpmd_offload_info {
  71         uint16_t ethertype;
  72         uint8_t gso_enable;
  73         uint16_t l2_len;
  74         uint16_t l3_len;
  75         uint16_t l4_len;
  76         uint8_t l4_proto;
  77         uint8_t is_tunnel;
  78         uint16_t outer_ethertype;
  79         uint16_t outer_l2_len;
  80         uint16_t outer_l3_len;
  81         uint8_t outer_l4_proto;
  82         uint16_t tso_segsz;
  83         uint16_t tunnel_tso_segsz;
  84         uint32_t pkt_len;
  85 };
  86
  87 /* simplified GRE header */
  88 struct simple_gre_hdr {
  89         uint16_t flags;
  90         uint16_t proto;
  91 } __rte_packed;
  92
  93 static uint16_t
  94 get_udptcp_checksum(void *l3_hdr, void *l4_hdr, uint16_t ethertype)
  95 {
  96         if (ethertype == _htons(RTE_ETHER_TYPE_IPV4))
  97                 return rte_ipv4_udptcp_cksum(l3_hdr, l4_hdr);
  98         else /* assume ethertype == RTE_ETHER_TYPE_IPV6 */
  99                 return rte_ipv6_udptcp_cksum(l3_hdr, l4_hdr);
 100 }
 101
 102 /* Parse an IPv4 header to fill l3_len, l4_len, and l4_proto */
 103 static void
 104 parse_ipv4(struct rte_ipv4_hdr *ipv4_hdr, struct testpmd_offload_info *info)
 105 {
 106         struct rte_tcp_hdr *tcp_hdr;
 107
 108         info->l3_len = rte_ipv4_hdr_len(ipv4_hdr);
 109         info->l4_proto = ipv4_hdr->next_proto_id;
 110
 111         /* only fill l4_len for TCP, it's useful for TSO */
 112         if (info->l4_proto == IPPROTO_TCP) {
 113                 tcp_hdr = (struct rte_tcp_hdr *)
 114                         ((char *)ipv4_hdr + info->l3_len);
 115                 info->l4_len = (tcp_hdr->data_off & 0xf0) >> 2;
 116         } else if (info->l4_proto == IPPROTO_UDP)
 117                 info->l4_len = sizeof(struct rte_udp_hdr);
 118         else
 119                 info->l4_len = 0;
 120 }
 121
 122 /* Parse an IPv6 header to fill l3_len, l4_len, and l4_proto */
 123 static void
 124 parse_ipv6(struct rte_ipv6_hdr *ipv6_hdr, struct testpmd_offload_info *info)
 125 {
 126         struct rte_tcp_hdr *tcp_hdr;
 127
 128         info->l3_len = sizeof(struct rte_ipv6_hdr);
 129         info->l4_proto = ipv6_hdr->proto;
 130
 131         /* only fill l4_len for TCP, it's useful for TSO */
 132         if (info->l4_proto == IPPROTO_TCP) {
 133                 tcp_hdr = (struct rte_tcp_hdr *)
 134                         ((char *)ipv6_hdr + info->l3_len);
 135                 info->l4_len = (tcp_hdr->data_off & 0xf0) >> 2;
 136         } else if (info->l4_proto == IPPROTO_UDP)
 137                 info->l4_len = sizeof(struct rte_udp_hdr);
 138         else
 139                 info->l4_len = 0;
 140 }
 141
 142 /*
 143  * Parse an ethernet header to fill the ethertype, l2_len, l3_len and
 144  * ipproto. This function is able to recognize IPv4/IPv6 with optional VLAN
 145  * headers. The l4_len argument is only set in case of TCP (useful for TSO).
 146  */
 147 static void
 148 parse_ethernet(struct rte_ether_hdr *eth_hdr, struct testpmd_offload_info *info)
 149 {
 150         struct rte_ipv4_hdr *ipv4_hdr;
 151         struct rte_ipv6_hdr *ipv6_hdr;
 152         struct rte_vlan_hdr *vlan_hdr;
 153
 154         info->l2_len = sizeof(struct rte_ether_hdr);
 155         info->ethertype = eth_hdr->ether_type;
 156
 157         while (info->ethertype == _htons(RTE_ETHER_TYPE_VLAN) ||
 158                info->ethertype == _htons(RTE_ETHER_TYPE_QINQ)) {
 159                 vlan_hdr = (struct rte_vlan_hdr *)
 160                         ((char *)eth_hdr + info->l2_len);
 161                 info->l2_len  += sizeof(struct rte_vlan_hdr);
 162                 info->ethertype = vlan_hdr->eth_proto;
 163         }
 164
 165         switch (info->ethertype) {
 166         case _htons(RTE_ETHER_TYPE_IPV4):
 167                 ipv4_hdr = (struct rte_ipv4_hdr *)
 168                         ((char *)eth_hdr + info->l2_len);
 169                 parse_ipv4(ipv4_hdr, info);
 170                 break;
 171         case _htons(RTE_ETHER_TYPE_IPV6):
 172                 ipv6_hdr = (struct rte_ipv6_hdr *)
 173                         ((char *)eth_hdr + info->l2_len);
 174                 parse_ipv6(ipv6_hdr, info);
 175                 break;
 176         default:
 177                 info->l4_len = 0;
 178                 info->l3_len = 0;
 179                 info->l4_proto = 0;
 180                 break;
 181         }
 182 }
 183
 184 /* Fill in outer layers length */
 185 static void
 186 update_tunnel_outer(struct testpmd_offload_info *info)
 187 {
 188         info->is_tunnel = 1;
 189         info->outer_ethertype = info->ethertype;
 190         info->outer_l2_len = info->l2_len;
 191         info->outer_l3_len = info->l3_len;
 192         info->outer_l4_proto = info->l4_proto;
 193 }
 194
 195 /*
 196  * Parse a GTP protocol header.
 197  * No optional fields and next extension header type.
 198  */
 199 static void
 200 parse_gtp(struct rte_udp_hdr *udp_hdr,
 201           struct testpmd_offload_info *info)
 202 {
 203         struct rte_ipv4_hdr *ipv4_hdr;
 204         struct rte_ipv6_hdr *ipv6_hdr;
 205         struct rte_gtp_hdr *gtp_hdr;
 206         uint8_t gtp_len = sizeof(*gtp_hdr);
 207         uint8_t ip_ver;
 208
 209         /* Check udp destination port. */
 210         if (udp_hdr->dst_port != _htons(RTE_GTPC_UDP_PORT) &&
 211             udp_hdr->src_port != _htons(RTE_GTPC_UDP_PORT) &&
 212             udp_hdr->dst_port != _htons(RTE_GTPU_UDP_PORT))
 213                 return;
 214
 215         update_tunnel_outer(info);
 216         info->l2_len = 0;
 217
 218         gtp_hdr = (struct rte_gtp_hdr *)((char *)udp_hdr +
 219                   sizeof(struct rte_udp_hdr));
 220
 221         /*
 222          * Check message type. If message type is 0xff, it is
 223          * a GTP data packet. If not, it is a GTP control packet
 224          */
 225         if (gtp_hdr->msg_type == 0xff) {
 226                 ip_ver = *(uint8_t *)((char *)udp_hdr +
 227                          sizeof(struct rte_udp_hdr) +
 228                          sizeof(struct rte_gtp_hdr));
 229                 ip_ver = (ip_ver) & 0xf0;
 230
 231                 if (ip_ver == RTE_GTP_TYPE_IPV4) {
 232                         ipv4_hdr = (struct rte_ipv4_hdr *)((char *)gtp_hdr +
 233                                    gtp_len);
 234                         info->ethertype = _htons(RTE_ETHER_TYPE_IPV4);
 235                         parse_ipv4(ipv4_hdr, info);
 236                 } else if (ip_ver == RTE_GTP_TYPE_IPV6) {
 237                         ipv6_hdr = (struct rte_ipv6_hdr *)((char *)gtp_hdr +
 238                                    gtp_len);
 239                         info->ethertype = _htons(RTE_ETHER_TYPE_IPV6);
 240                         parse_ipv6(ipv6_hdr, info);
 241                 }
 242         } else {
 243                 info->ethertype = 0;
 244                 info->l4_len = 0;
 245                 info->l3_len = 0;
 246                 info->l4_proto = 0;
 247         }
 248
 249         info->l2_len += RTE_ETHER_GTP_HLEN;
 250 }
 251
 252 /* Parse a vxlan header */
 253 static void
 254 parse_vxlan(struct rte_udp_hdr *udp_hdr,
 255             struct testpmd_offload_info *info,
 256             uint32_t pkt_type)
 257 {
 258         struct rte_ether_hdr *eth_hdr;
 259
 260         /* check udp destination port, RTE_VXLAN_DEFAULT_PORT (4789) is the
 261          * default vxlan port (rfc7348) or that the rx offload flag is set
 262          * (i40e only currently)
 263          */
 264         if (udp_hdr->dst_port != _htons(RTE_VXLAN_DEFAULT_PORT) &&
 265                 RTE_ETH_IS_TUNNEL_PKT(pkt_type) == 0)
 266                 return;
 267
 268         update_tunnel_outer(info);
 269
 270         eth_hdr = (struct rte_ether_hdr *)((char *)udp_hdr +
 271                 sizeof(struct rte_udp_hdr) +
 272                 sizeof(struct rte_vxlan_hdr));
 273
 274         parse_ethernet(eth_hdr, info);
 275         info->l2_len += RTE_ETHER_VXLAN_HLEN; /* add udp + vxlan */
 276 }
 277
 278 /* Parse a vxlan-gpe header */
 279 static void
 280 parse_vxlan_gpe(struct rte_udp_hdr *udp_hdr,
 281             struct testpmd_offload_info *info)
 282 {
 283         struct rte_ether_hdr *eth_hdr;
 284         struct rte_ipv4_hdr *ipv4_hdr;
 285         struct rte_ipv6_hdr *ipv6_hdr;
 286         struct rte_vxlan_gpe_hdr *vxlan_gpe_hdr;
 287         uint8_t vxlan_gpe_len = sizeof(*vxlan_gpe_hdr);
 288
 289         /* Check udp destination port. */
 290         if (udp_hdr->dst_port != _htons(vxlan_gpe_udp_port))
 291                 return;
 292
 293         vxlan_gpe_hdr = (struct rte_vxlan_gpe_hdr *)((char *)udp_hdr +
 294                                 sizeof(struct rte_udp_hdr));
 295
 296         if (!vxlan_gpe_hdr->proto || vxlan_gpe_hdr->proto ==
 297             RTE_VXLAN_GPE_TYPE_IPV4) {
 298                 update_tunnel_outer(info);
 299
 300                 ipv4_hdr = (struct rte_ipv4_hdr *)((char *)vxlan_gpe_hdr +
 301                            vxlan_gpe_len);
 302
 303                 parse_ipv4(ipv4_hdr, info);
 304                 info->ethertype = _htons(RTE_ETHER_TYPE_IPV4);
 305                 info->l2_len = 0;
 306
 307         } else if (vxlan_gpe_hdr->proto == RTE_VXLAN_GPE_TYPE_IPV6) {
 308                 update_tunnel_outer(info);
 309
 310                 ipv6_hdr = (struct rte_ipv6_hdr *)((char *)vxlan_gpe_hdr +
 311                            vxlan_gpe_len);
 312
 313                 info->ethertype = _htons(RTE_ETHER_TYPE_IPV6);
 314                 parse_ipv6(ipv6_hdr, info);
 315                 info->l2_len = 0;
 316
 317         } else if (vxlan_gpe_hdr->proto == RTE_VXLAN_GPE_TYPE_ETH) {
 318                 update_tunnel_outer(info);
 319
 320                 eth_hdr = (struct rte_ether_hdr *)((char *)vxlan_gpe_hdr +
 321                           vxlan_gpe_len);
 322
 323                 parse_ethernet(eth_hdr, info);
 324         } else
 325                 return;
 326
 327         info->l2_len += RTE_ETHER_VXLAN_GPE_HLEN;
 328 }
 329
 330 /* Parse a geneve header */
 331 static void
 332 parse_geneve(struct rte_udp_hdr *udp_hdr,
 333             struct testpmd_offload_info *info)
 334 {
 335         struct rte_ether_hdr *eth_hdr;
 336         struct rte_ipv4_hdr *ipv4_hdr;
 337         struct rte_ipv6_hdr *ipv6_hdr;
 338         struct rte_geneve_hdr *geneve_hdr;
 339         uint16_t geneve_len;
 340
 341         /* Check udp destination port. */
 342         if (udp_hdr->dst_port != _htons(geneve_udp_port))
 343                 return;
 344
 345         geneve_hdr = (struct rte_geneve_hdr *)((char *)udp_hdr +
 346                                 sizeof(struct rte_udp_hdr));
 347         geneve_len = sizeof(struct rte_geneve_hdr) + geneve_hdr->opt_len * 4;
 348         if (!geneve_hdr->proto || geneve_hdr->proto ==
 349             _htons(RTE_ETHER_TYPE_IPV4)) {
 350                 update_tunnel_outer(info);
 351                 ipv4_hdr = (struct rte_ipv4_hdr *)((char *)geneve_hdr +
 352                            geneve_len);
 353                 parse_ipv4(ipv4_hdr, info);
 354                 info->ethertype = _htons(RTE_ETHER_TYPE_IPV4);
 355                 info->l2_len = 0;
 356         } else if (geneve_hdr->proto == _htons(RTE_ETHER_TYPE_IPV6)) {
 357                 update_tunnel_outer(info);
 358                 ipv6_hdr = (struct rte_ipv6_hdr *)((char *)geneve_hdr +
 359                            geneve_len);
 360                 info->ethertype = _htons(RTE_ETHER_TYPE_IPV6);
 361                 parse_ipv6(ipv6_hdr, info);
 362                 info->l2_len = 0;
 363
 364         } else if (geneve_hdr->proto == _htons(RTE_GENEVE_TYPE_ETH)) {
 365                 update_tunnel_outer(info);
 366                 eth_hdr = (struct rte_ether_hdr *)((char *)geneve_hdr +
 367                           geneve_len);
 368                 parse_ethernet(eth_hdr, info);
 369         } else
 370                 return;
 371
 372         info->l2_len +=
 373                 (sizeof(struct rte_udp_hdr) + sizeof(struct rte_geneve_hdr) +
 374                 ((struct rte_geneve_hdr *)geneve_hdr)->opt_len * 4);
 375 }
 376
 377 /* Parse a gre header */
 378 static void
 379 parse_gre(struct simple_gre_hdr *gre_hdr, struct testpmd_offload_info *info)
 380 {
 381         struct rte_ether_hdr *eth_hdr;
 382         struct rte_ipv4_hdr *ipv4_hdr;
 383         struct rte_ipv6_hdr *ipv6_hdr;
 384         uint8_t gre_len = 0;
 385
 386         gre_len += sizeof(struct simple_gre_hdr);
 387
 388         if (gre_hdr->flags & _htons(GRE_KEY_PRESENT))
 389                 gre_len += GRE_EXT_LEN;
 390         if (gre_hdr->flags & _htons(GRE_SEQUENCE_PRESENT))
 391                 gre_len += GRE_EXT_LEN;
 392         if (gre_hdr->flags & _htons(GRE_CHECKSUM_PRESENT))
 393                 gre_len += GRE_EXT_LEN;
 394
 395         if (gre_hdr->proto == _htons(RTE_ETHER_TYPE_IPV4)) {
 396                 update_tunnel_outer(info);
 397
 398                 ipv4_hdr = (struct rte_ipv4_hdr *)((char *)gre_hdr + gre_len);
 399
 400                 parse_ipv4(ipv4_hdr, info);
 401                 info->ethertype = _htons(RTE_ETHER_TYPE_IPV4);
 402                 info->l2_len = 0;
 403
 404         } else if (gre_hdr->proto == _htons(RTE_ETHER_TYPE_IPV6)) {
 405                 update_tunnel_outer(info);
 406
 407                 ipv6_hdr = (struct rte_ipv6_hdr *)((char *)gre_hdr + gre_len);
 408
 409                 info->ethertype = _htons(RTE_ETHER_TYPE_IPV6);
 410                 parse_ipv6(ipv6_hdr, info);
 411                 info->l2_len = 0;
 412
 413         } else if (gre_hdr->proto == _htons(RTE_ETHER_TYPE_TEB)) {
 414                 update_tunnel_outer(info);
 415
 416                 eth_hdr = (struct rte_ether_hdr *)((char *)gre_hdr + gre_len);
 417
 418                 parse_ethernet(eth_hdr, info);
 419         } else
 420                 return;
 421
 422         info->l2_len += gre_len;
 423 }
 424
 425
 426 /* Parse an encapsulated ip or ipv6 header */
 427 static void
 428 parse_encap_ip(void *encap_ip, struct testpmd_offload_info *info)
 429 {
 430         struct rte_ipv4_hdr *ipv4_hdr = encap_ip;
 431         struct rte_ipv6_hdr *ipv6_hdr = encap_ip;
 432         uint8_t ip_version;
 433
 434         ip_version = (ipv4_hdr->version_ihl & 0xf0) >> 4;
 435
 436         if (ip_version != 4 && ip_version != 6)
 437                 return;
 438
 439         info->is_tunnel = 1;
 440         info->outer_ethertype = info->ethertype;
 441         info->outer_l2_len = info->l2_len;
 442         info->outer_l3_len = info->l3_len;
 443
 444         if (ip_version == 4) {
 445                 parse_ipv4(ipv4_hdr, info);
 446                 info->ethertype = _htons(RTE_ETHER_TYPE_IPV4);
 447         } else {
 448                 parse_ipv6(ipv6_hdr, info);
 449                 info->ethertype = _htons(RTE_ETHER_TYPE_IPV6);
 450         }
 451         info->l2_len = 0;
 452 }
 453
 454 /* if possible, calculate the checksum of a packet in hw or sw,
 455  * depending on the testpmd command line configuration */
 456 static uint64_t
 457 process_inner_cksums(void *l3_hdr, const struct testpmd_offload_info *info,
 458         uint64_t tx_offloads)
 459 {
 460         struct rte_ipv4_hdr *ipv4_hdr = l3_hdr;
 461         struct rte_udp_hdr *udp_hdr;
 462         struct rte_tcp_hdr *tcp_hdr;
 463         struct rte_sctp_hdr *sctp_hdr;
 464         uint64_t ol_flags = 0;
 465         uint32_t max_pkt_len, tso_segsz = 0;
 466
 467         /* ensure packet is large enough to require tso */
 468         if (!info->is_tunnel) {
 469                 max_pkt_len = info->l2_len + info->l3_len + info->l4_len +
 470                         info->tso_segsz;
 471                 if (info->tso_segsz != 0 && info->pkt_len > max_pkt_len)
 472                         tso_segsz = info->tso_segsz;
 473         } else {
 474                 max_pkt_len = info->outer_l2_len + info->outer_l3_len +
 475                         info->l2_len + info->l3_len + info->l4_len +
 476                         info->tunnel_tso_segsz;
 477                 if (info->tunnel_tso_segsz != 0 && info->pkt_len > max_pkt_len)
 478                         tso_segsz = info->tunnel_tso_segsz;
 479         }
 480
 481         if (info->ethertype == _htons(RTE_ETHER_TYPE_IPV4)) {
 482                 ipv4_hdr = l3_hdr;
 483                 ipv4_hdr->hdr_checksum = 0;
 484
 485                 ol_flags |= PKT_TX_IPV4;
 486                 if (info->l4_proto == IPPROTO_TCP && tso_segsz) {
 487                         ol_flags |= PKT_TX_IP_CKSUM;
 488                 } else {
 489                         if (tx_offloads & DEV_TX_OFFLOAD_IPV4_CKSUM)
 490                                 ol_flags |= PKT_TX_IP_CKSUM;
 491                         else
 492                                 ipv4_hdr->hdr_checksum =
 493                                         rte_ipv4_cksum(ipv4_hdr);
 494                 }
 495         } else if (info->ethertype == _htons(RTE_ETHER_TYPE_IPV6))
 496                 ol_flags |= PKT_TX_IPV6;
 497         else
 498                 return 0; /* packet type not supported, nothing to do */
 499
 500         if (info->l4_proto == IPPROTO_UDP) {
 501                 udp_hdr = (struct rte_udp_hdr *)((char *)l3_hdr + info->l3_len);
 502                 /* do not recalculate udp cksum if it was 0 */
 503                 if (udp_hdr->dgram_cksum != 0) {
 504                         udp_hdr->dgram_cksum = 0;
 505                         if (tx_offloads & DEV_TX_OFFLOAD_UDP_CKSUM)
 506                                 ol_flags |= PKT_TX_UDP_CKSUM;
 507                         else {
 508                                 udp_hdr->dgram_cksum =
 509                                         get_udptcp_checksum(l3_hdr, udp_hdr,
 510                                                 info->ethertype);
 511                         }
 512                 }
 513                 if (info->gso_enable)
 514                         ol_flags |= PKT_TX_UDP_SEG;
 515         } else if (info->l4_proto == IPPROTO_TCP) {
 516                 tcp_hdr = (struct rte_tcp_hdr *)((char *)l3_hdr + info->l3_len);
 517                 tcp_hdr->cksum = 0;
 518                 if (tso_segsz)
 519                         ol_flags |= PKT_TX_TCP_SEG;
 520                 else if (tx_offloads & DEV_TX_OFFLOAD_TCP_CKSUM)
 521                         ol_flags |= PKT_TX_TCP_CKSUM;
 522                 else {
 523                         tcp_hdr->cksum =
 524                                 get_udptcp_checksum(l3_hdr, tcp_hdr,
 525                                         info->ethertype);
 526                 }
 527                 if (info->gso_enable)
 528                         ol_flags |= PKT_TX_TCP_SEG;
 529         } else if (info->l4_proto == IPPROTO_SCTP) {
 530                 sctp_hdr = (struct rte_sctp_hdr *)
 531                         ((char *)l3_hdr + info->l3_len);
 532                 sctp_hdr->cksum = 0;
 533                 /* sctp payload must be a multiple of 4 to be
 534                  * offloaded */
 535                 if ((tx_offloads & DEV_TX_OFFLOAD_SCTP_CKSUM) &&
 536                         ((ipv4_hdr->total_length & 0x3) == 0)) {
 537                         ol_flags |= PKT_TX_SCTP_CKSUM;
 538                 } else {
 539                         /* XXX implement CRC32c, example available in
 540                          * RFC3309 */
 541                 }
 542         }
 543
 544         return ol_flags;
 545 }
 546
 547 /* Calculate the checksum of outer header */
 548 static uint64_t
 549 process_outer_cksums(void *outer_l3_hdr, struct testpmd_offload_info *info,
 550         uint64_t tx_offloads, int tso_enabled)
 551 {
 552         struct rte_ipv4_hdr *ipv4_hdr = outer_l3_hdr;
 553         struct rte_ipv6_hdr *ipv6_hdr = outer_l3_hdr;
 554         struct rte_udp_hdr *udp_hdr;
 555         uint64_t ol_flags = 0;
 556
 557         if (info->outer_ethertype == _htons(RTE_ETHER_TYPE_IPV4)) {
 558                 ipv4_hdr->hdr_checksum = 0;
 559                 ol_flags |= PKT_TX_OUTER_IPV4;
 560
 561                 if (tx_offloads & DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM)
 562                         ol_flags |= PKT_TX_OUTER_IP_CKSUM;
 563                 else
 564                         ipv4_hdr->hdr_checksum = rte_ipv4_cksum(ipv4_hdr);
 565         } else
 566                 ol_flags |= PKT_TX_OUTER_IPV6;
 567
 568         if (info->outer_l4_proto != IPPROTO_UDP)
 569                 return ol_flags;
 570
 571         udp_hdr = (struct rte_udp_hdr *)
 572                 ((char *)outer_l3_hdr + info->outer_l3_len);
 573
 574         if (tso_enabled)
 575                 ol_flags |= PKT_TX_TCP_SEG;
 576
 577         /* Skip SW outer UDP checksum generation if HW supports it */
 578         if (tx_offloads & DEV_TX_OFFLOAD_OUTER_UDP_CKSUM) {
 579                 if (info->outer_ethertype == _htons(RTE_ETHER_TYPE_IPV4))
 580                         udp_hdr->dgram_cksum
 581                                 = rte_ipv4_phdr_cksum(ipv4_hdr, ol_flags);
 582                 else
 583                         udp_hdr->dgram_cksum
 584                                 = rte_ipv6_phdr_cksum(ipv6_hdr, ol_flags);
 585
 586                 ol_flags |= PKT_TX_OUTER_UDP_CKSUM;
 587                 return ol_flags;
 588         }
 589
 590         /* outer UDP checksum is done in software. In the other side, for
 591          * UDP tunneling, like VXLAN or Geneve, outer UDP checksum can be
 592          * set to zero.
 593          *
 594          * If a packet will be TSOed into small packets by NIC, we cannot
 595          * set/calculate a non-zero checksum, because it will be a wrong
 596          * value after the packet be split into several small packets.
 597          */
 598         if (tso_enabled)
 599                 udp_hdr->dgram_cksum = 0;
 600
 601         /* do not recalculate udp cksum if it was 0 */
 602         if (udp_hdr->dgram_cksum != 0) {
 603                 udp_hdr->dgram_cksum = 0;
 604                 if (info->outer_ethertype == _htons(RTE_ETHER_TYPE_IPV4))
 605                         udp_hdr->dgram_cksum =
 606                                 rte_ipv4_udptcp_cksum(ipv4_hdr, udp_hdr);
 607                 else
 608                         udp_hdr->dgram_cksum =
 609                                 rte_ipv6_udptcp_cksum(ipv6_hdr, udp_hdr);
 610         }
 611
 612         return ol_flags;
 613 }
 614
 615 /*
 616  * Helper function.
 617  * Performs actual copying.
 618  * Returns number of segments in the destination mbuf on success,
 619  * or negative error code on failure.
 620  */
 621 static int
 622 mbuf_copy_split(const struct rte_mbuf *ms, struct rte_mbuf *md[],
 623         uint16_t seglen[], uint8_t nb_seg)
 624 {
 625         uint32_t dlen, slen, tlen;
 626         uint32_t i, len;
 627         const struct rte_mbuf *m;
 628         const uint8_t *src;
 629         uint8_t *dst;
 630
 631         dlen = 0;
 632         slen = 0;
 633         tlen = 0;
 634
 635         dst = NULL;
 636         src = NULL;
 637
 638         m = ms;
 639         i = 0;
 640         while (ms != NULL && i != nb_seg) {
 641
 642                 if (slen == 0) {
 643                         slen = rte_pktmbuf_data_len(ms);
 644                         src = rte_pktmbuf_mtod(ms, const uint8_t *);
 645                 }
 646
 647                 if (dlen == 0) {
 648                         dlen = RTE_MIN(seglen[i], slen);
 649                         md[i]->data_len = dlen;
 650                         md[i]->next = (i + 1 == nb_seg) ? NULL : md[i + 1];
 651                         dst = rte_pktmbuf_mtod(md[i], uint8_t *);
 652                 }
 653
 654                 len = RTE_MIN(slen, dlen);
 655                 memcpy(dst, src, len);
 656                 tlen += len;
 657                 slen -= len;
 658                 dlen -= len;
 659                 src += len;
 660                 dst += len;
 661
 662                 if (slen == 0)
 663                         ms = ms->next;
 664                 if (dlen == 0)
 665                         i++;
 666         }
 667
 668         if (ms != NULL)
 669                 return -ENOBUFS;
 670         else if (tlen != m->pkt_len)
 671                 return -EINVAL;
 672
 673         md[0]->nb_segs = nb_seg;
 674         md[0]->pkt_len = tlen;
 675         md[0]->vlan_tci = m->vlan_tci;
 676         md[0]->vlan_tci_outer = m->vlan_tci_outer;
 677         md[0]->ol_flags = m->ol_flags;
 678         md[0]->tx_offload = m->tx_offload;
 679
 680         return nb_seg;
 681 }
 682
 683 /*
 684  * Allocate a new mbuf with up to tx_pkt_nb_segs segments.
 685  * Copy packet contents and offload information into the new segmented mbuf.
 686  */
 687 static struct rte_mbuf *
 688 pkt_copy_split(const struct rte_mbuf *pkt)
 689 {
 690         int32_t n, rc;
 691         uint32_t i, len, nb_seg;
 692         struct rte_mempool *mp;
 693         uint16_t seglen[RTE_MAX_SEGS_PER_PKT];
 694         struct rte_mbuf *p, *md[RTE_MAX_SEGS_PER_PKT];
 695
 696         mp = current_fwd_lcore()->mbp;
 697
 698         if (tx_pkt_split == TX_PKT_SPLIT_RND)
 699                 nb_seg = rte_rand() % tx_pkt_nb_segs + 1;
 700         else
 701                 nb_seg = tx_pkt_nb_segs;
 702
 703         memcpy(seglen, tx_pkt_seg_lengths, nb_seg * sizeof(seglen[0]));
 704
 705         /* calculate number of segments to use and their length. */
 706         len = 0;
 707         for (i = 0; i != nb_seg && len < pkt->pkt_len; i++) {
 708                 len += seglen[i];
 709                 md[i] = NULL;
 710         }
 711
 712         n = pkt->pkt_len - len;
 713
 714         /* update size of the last segment to fit rest of the packet */
 715         if (n >= 0) {
 716                 seglen[i - 1] += n;
 717                 len += n;
 718         }
 719
 720         nb_seg = i;
 721         while (i != 0) {
 722                 p = rte_pktmbuf_alloc(mp);
 723                 if (p == NULL) {
 724                         TESTPMD_LOG(ERR,
 725                                 "failed to allocate %u-th of %u mbuf "
 726                                 "from mempool: %s\n",
 727                                 nb_seg - i, nb_seg, mp->name);
 728                         break;
 729                 }
 730
 731                 md[--i] = p;
 732                 if (rte_pktmbuf_tailroom(md[i]) < seglen[i]) {
 733                         TESTPMD_LOG(ERR, "mempool %s, %u-th segment: "
 734                                 "expected seglen: %u, "
 735                                 "actual mbuf tailroom: %u\n",
 736                                 mp->name, i, seglen[i],
 737                                 rte_pktmbuf_tailroom(md[i]));
 738                         break;
 739                 }
 740         }
 741
 742         /* all mbufs successfully allocated, do copy */
 743         if (i == 0) {
 744                 rc = mbuf_copy_split(pkt, md, seglen, nb_seg);
 745                 if (rc < 0)
 746                         TESTPMD_LOG(ERR,
 747                                 "mbuf_copy_split for %p(len=%u, nb_seg=%u) "
 748                                 "into %u segments failed with error code: %d\n",
 749                                 pkt, pkt->pkt_len, pkt->nb_segs, nb_seg, rc);
 750
 751                 /* figure out how many mbufs to free. */
 752                 i = RTE_MAX(rc, 0);
 753         }
 754
 755         /* free unused mbufs */
 756         for (; i != nb_seg; i++) {
 757                 rte_pktmbuf_free_seg(md[i]);
 758                 md[i] = NULL;
 759         }
 760
 761         return md[0];
 762 }
 763
 764 /*
 765  * Receive a burst of packets, and for each packet:
 766  *  - parse packet, and try to recognize a supported packet type (1)
 767  *  - if it's not a supported packet type, don't touch the packet, else:
 768  *  - reprocess the checksum of all supported layers. This is done in SW
 769  *    or HW, depending on testpmd command line configuration
 770  *  - if TSO is enabled in testpmd command line, also flag the mbuf for TCP
 771  *    segmentation offload (this implies HW TCP checksum)
 772  * Then transmit packets on the output port.
 773  *
 774  * (1) Supported packets are:
 775  *   Ether / (vlan) / IP|IP6 / UDP|TCP|SCTP .
 776  *   Ether / (vlan) / outer IP|IP6 / outer UDP / VxLAN / Ether / IP|IP6 /
 777  *           UDP|TCP|SCTP
 778  *   Ether / (vlan) / outer IP|IP6 / outer UDP / VXLAN-GPE / Ether / IP|IP6 /
 779  *           UDP|TCP|SCTP
 780  *   Ether / (vlan) / outer IP|IP6 / outer UDP / VXLAN-GPE / IP|IP6 /
 781  *           UDP|TCP|SCTP
 782  *   Ether / (vlan) / outer IP / outer UDP / GTP / IP|IP6 / UDP|TCP|SCTP
 783  *   Ether / (vlan) / outer IP|IP6 / GRE / Ether / IP|IP6 / UDP|TCP|SCTP
 784  *   Ether / (vlan) / outer IP|IP6 / GRE / IP|IP6 / UDP|TCP|SCTP
 785  *   Ether / (vlan) / outer IP|IP6 / IP|IP6 / UDP|TCP|SCTP
 786  *
 787  * The testpmd command line for this forward engine sets the flags
 788  * TESTPMD_TX_OFFLOAD_* in ports[tx_port].tx_ol_flags. They control
 789  * wether a checksum must be calculated in software or in hardware. The
 790  * IP, UDP, TCP and SCTP flags always concern the inner layer. The
 791  * OUTER_IP is only useful for tunnel packets.
 792  */
 793 static void
 794 pkt_burst_checksum_forward(struct fwd_stream *fs)
 795 {
 796         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
 797         struct rte_mbuf *gso_segments[GSO_MAX_PKT_BURST];
 798         struct rte_gso_ctx *gso_ctx;
 799         struct rte_mbuf **tx_pkts_burst;
 800         struct rte_port *txp;
 801         struct rte_mbuf *m, *p;
 802         struct rte_ether_hdr *eth_hdr;
 803         void *l3_hdr = NULL, *outer_l3_hdr = NULL; /* can be IPv4 or IPv6 */
 804         void **gro_ctx;
 805         uint16_t gro_pkts_num;
 806         uint8_t gro_enable;
 807         uint16_t nb_rx;
 808         uint16_t nb_tx;
 809         uint16_t nb_prep;
 810         uint16_t i;
 811         uint64_t rx_ol_flags, tx_ol_flags;
 812         uint64_t tx_offloads;
 813         uint32_t retry;
 814         uint32_t rx_bad_ip_csum;
 815         uint32_t rx_bad_l4_csum;
 816         uint32_t rx_bad_outer_l4_csum;
 817         uint32_t rx_bad_outer_ip_csum;
 818         struct testpmd_offload_info info;
 819         uint16_t nb_segments = 0;
 820         int ret;
 821
 822         uint64_t start_tsc = 0;
 823
 824         get_start_cycles(&start_tsc);
 825
 826         /* receive a burst of packet */
 827         nb_rx = rte_eth_rx_burst(fs->rx_port, fs->rx_queue, pkts_burst,
 828                                  nb_pkt_per_burst);
 829         inc_rx_burst_stats(fs, nb_rx);
 830         if (unlikely(nb_rx == 0))
 831                 return;
 832
 833         fs->rx_packets += nb_rx;
 834         rx_bad_ip_csum = 0;
 835         rx_bad_l4_csum = 0;
 836         rx_bad_outer_l4_csum = 0;
 837         rx_bad_outer_ip_csum = 0;
 838         gro_enable = gro_ports[fs->rx_port].enable;
 839
 840         txp = &ports[fs->tx_port];
 841         tx_offloads = txp->dev_conf.txmode.offloads;
 842         memset(&info, 0, sizeof(info));
 843         info.tso_segsz = txp->tso_segsz;
 844         info.tunnel_tso_segsz = txp->tunnel_tso_segsz;
 845         if (gso_ports[fs->tx_port].enable)
 846                 info.gso_enable = 1;
 847
 848         for (i = 0; i < nb_rx; i++) {
 849                 if (likely(i < nb_rx - 1))
 850                         rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[i + 1],
 851                                                        void *));
 852
 853                 m = pkts_burst[i];
 854                 info.is_tunnel = 0;
 855                 info.pkt_len = rte_pktmbuf_pkt_len(m);
 856                 tx_ol_flags = m->ol_flags &
 857                               (IND_ATTACHED_MBUF | EXT_ATTACHED_MBUF);
 858                 rx_ol_flags = m->ol_flags;
 859
 860                 /* Update the L3/L4 checksum error packet statistics */
 861                 if ((rx_ol_flags & PKT_RX_IP_CKSUM_MASK) == PKT_RX_IP_CKSUM_BAD)
 862                         rx_bad_ip_csum += 1;
 863                 if ((rx_ol_flags & PKT_RX_L4_CKSUM_MASK) == PKT_RX_L4_CKSUM_BAD)
 864                         rx_bad_l4_csum += 1;
 865                 if (rx_ol_flags & PKT_RX_OUTER_L4_CKSUM_BAD)
 866                         rx_bad_outer_l4_csum += 1;
 867                 if (rx_ol_flags & PKT_RX_OUTER_IP_CKSUM_BAD)
 868                         rx_bad_outer_ip_csum += 1;
 869
 870                 /* step 1: dissect packet, parsing optional vlan, ip4/ip6, vxlan
 871                  * and inner headers */
 872
 873                 eth_hdr = rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
 874                 rte_ether_addr_copy(&peer_eth_addrs[fs->peer_addr],
 875                                 &eth_hdr->d_addr);
 876                 rte_ether_addr_copy(&ports[fs->tx_port].eth_addr,
 877                                 &eth_hdr->s_addr);
 878                 parse_ethernet(eth_hdr, &info);
 879                 l3_hdr = (char *)eth_hdr + info.l2_len;
 880
 881                 /* check if it's a supported tunnel */
 882                 if (txp->parse_tunnel) {
 883                         if (info.l4_proto == IPPROTO_UDP) {
 884                                 struct rte_udp_hdr *udp_hdr;
 885
 886                                 udp_hdr = (struct rte_udp_hdr *)
 887                                         ((char *)l3_hdr + info.l3_len);
 888                                 parse_gtp(udp_hdr, &info);
 889                                 if (info.is_tunnel) {
 890                                         tx_ol_flags |= PKT_TX_TUNNEL_GTP;
 891                                         goto tunnel_update;
 892                                 }
 893                                 parse_vxlan_gpe(udp_hdr, &info);
 894                                 if (info.is_tunnel) {
 895                                         tx_ol_flags |=
 896                                                 PKT_TX_TUNNEL_VXLAN_GPE;
 897                                         goto tunnel_update;
 898                                 }
 899                                 parse_vxlan(udp_hdr, &info,
 900                                             m->packet_type);
 901                                 if (info.is_tunnel) {
 902                                         tx_ol_flags |=
 903                                                 PKT_TX_TUNNEL_VXLAN;
 904                                         goto tunnel_update;
 905                                 }
 906                                 parse_geneve(udp_hdr, &info);
 907                                 if (info.is_tunnel) {
 908                                         tx_ol_flags |=
 909                                                 PKT_TX_TUNNEL_GENEVE;
 910                                         goto tunnel_update;
 911                                 }
 912                         } else if (info.l4_proto == IPPROTO_GRE) {
 913                                 struct simple_gre_hdr *gre_hdr;
 914
 915                                 gre_hdr = (struct simple_gre_hdr *)
 916                                         ((char *)l3_hdr + info.l3_len);
 917                                 parse_gre(gre_hdr, &info);
 918                                 if (info.is_tunnel)
 919                                         tx_ol_flags |= PKT_TX_TUNNEL_GRE;
 920                         } else if (info.l4_proto == IPPROTO_IPIP) {
 921                                 void *encap_ip_hdr;
 922
 923                                 encap_ip_hdr = (char *)l3_hdr + info.l3_len;
 924                                 parse_encap_ip(encap_ip_hdr, &info);
 925                                 if (info.is_tunnel)
 926                                         tx_ol_flags |= PKT_TX_TUNNEL_IPIP;
 927                         }
 928                 }
 929
 930 tunnel_update:
 931                 /* update l3_hdr and outer_l3_hdr if a tunnel was parsed */
 932                 if (info.is_tunnel) {
 933                         outer_l3_hdr = l3_hdr;
 934                         l3_hdr = (char *)l3_hdr + info.outer_l3_len + info.l2_len;
 935                 }
 936
 937                 /* step 2: depending on user command line configuration,
 938                  * recompute checksum either in software or flag the
 939                  * mbuf to offload the calculation to the NIC. If TSO
 940                  * is configured, prepare the mbuf for TCP segmentation. */
 941
 942                 /* process checksums of inner headers first */
 943                 tx_ol_flags |= process_inner_cksums(l3_hdr, &info,
 944                         tx_offloads);
 945
 946                 /* Then process outer headers if any. Note that the software
 947                  * checksum will be wrong if one of the inner checksums is
 948                  * processed in hardware. */
 949                 if (info.is_tunnel == 1) {
 950                         tx_ol_flags |= process_outer_cksums(outer_l3_hdr, &info,
 951                                         tx_offloads,
 952                                         !!(tx_ol_flags & PKT_TX_TCP_SEG));
 953                 }
 954
 955                 /* step 3: fill the mbuf meta data (flags and header lengths) */
 956
 957                 m->tx_offload = 0;
 958                 if (info.is_tunnel == 1) {
 959                         if (info.tunnel_tso_segsz ||
 960                             (tx_offloads &
 961                              DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM) ||
 962                             (tx_offloads &
 963                              DEV_TX_OFFLOAD_OUTER_UDP_CKSUM) ||
 964                             (tx_ol_flags & PKT_TX_OUTER_IPV6)) {
 965                                 m->outer_l2_len = info.outer_l2_len;
 966                                 m->outer_l3_len = info.outer_l3_len;
 967                                 m->l2_len = info.l2_len;
 968                                 m->l3_len = info.l3_len;
 969                                 m->l4_len = info.l4_len;
 970                                 m->tso_segsz = info.tunnel_tso_segsz;
 971                         }
 972                         else {
 973                                 /* if there is a outer UDP cksum
 974                                    processed in sw and the inner in hw,
 975                                    the outer checksum will be wrong as
 976                                    the payload will be modified by the
 977                                    hardware */
 978                                 m->l2_len = info.outer_l2_len +
 979                                         info.outer_l3_len + info.l2_len;
 980                                 m->l3_len = info.l3_len;
 981                                 m->l4_len = info.l4_len;
 982                         }
 983                 } else {
 984                         /* this is only useful if an offload flag is
 985                          * set, but it does not hurt to fill it in any
 986                          * case */
 987                         m->l2_len = info.l2_len;
 988                         m->l3_len = info.l3_len;
 989                         m->l4_len = info.l4_len;
 990                         m->tso_segsz = info.tso_segsz;
 991                 }
 992                 m->ol_flags = tx_ol_flags;
 993
 994                 /* Do split & copy for the packet. */
 995                 if (tx_pkt_split != TX_PKT_SPLIT_OFF) {
 996                         p = pkt_copy_split(m);
 997                         if (p != NULL) {
 998                                 rte_pktmbuf_free(m);
 999                                 m = p;
1000                                 pkts_burst[i] = m;
1001                         }
1002                 }
1003
1004                 /* if verbose mode is enabled, dump debug info */
1005                 if (verbose_level > 0) {
1006                         char buf[256];
1007
1008                         printf("-----------------\n");
1009                         printf("port=%u, mbuf=%p, pkt_len=%u, nb_segs=%u:\n",
1010                                 fs->rx_port, m, m->pkt_len, m->nb_segs);
1011                         /* dump rx parsed packet info */
1012                         rte_get_rx_ol_flag_list(rx_ol_flags, buf, sizeof(buf));
1013                         printf("rx: l2_len=%d ethertype=%x l3_len=%d "
1014                                 "l4_proto=%d l4_len=%d flags=%s\n",
1015                                 info.l2_len, rte_be_to_cpu_16(info.ethertype),
1016                                 info.l3_len, info.l4_proto, info.l4_len, buf);
1017                         if (rx_ol_flags & PKT_RX_LRO)
1018                                 printf("rx: m->lro_segsz=%u\n", m->tso_segsz);
1019                         if (info.is_tunnel == 1)
1020                                 printf("rx: outer_l2_len=%d outer_ethertype=%x "
1021                                         "outer_l3_len=%d\n", info.outer_l2_len,
1022                                         rte_be_to_cpu_16(info.outer_ethertype),
1023                                         info.outer_l3_len);
1024                         /* dump tx packet info */
1025                         if ((tx_offloads & (DEV_TX_OFFLOAD_IPV4_CKSUM |
1026                                             DEV_TX_OFFLOAD_UDP_CKSUM |
1027                                             DEV_TX_OFFLOAD_TCP_CKSUM |
1028                                             DEV_TX_OFFLOAD_SCTP_CKSUM)) ||
1029                                 info.tso_segsz != 0)
1030                                 printf("tx: m->l2_len=%d m->l3_len=%d "
1031                                         "m->l4_len=%d\n",
1032                                         m->l2_len, m->l3_len, m->l4_len);
1033                         if (info.is_tunnel == 1) {
1034                                 if ((tx_offloads &
1035                                     DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM) ||
1036                                     (tx_offloads &
1037                                     DEV_TX_OFFLOAD_OUTER_UDP_CKSUM) ||
1038                                     (tx_ol_flags & PKT_TX_OUTER_IPV6))
1039                                         printf("tx: m->outer_l2_len=%d "
1040                                                 "m->outer_l3_len=%d\n",
1041                                                 m->outer_l2_len,
1042                                                 m->outer_l3_len);
1043                                 if (info.tunnel_tso_segsz != 0 &&
1044                                                 (m->ol_flags & PKT_TX_TCP_SEG))
1045                                         printf("tx: m->tso_segsz=%d\n",
1046                                                 m->tso_segsz);
1047                         } else if (info.tso_segsz != 0 &&
1048                                         (m->ol_flags & PKT_TX_TCP_SEG))
1049                                 printf("tx: m->tso_segsz=%d\n", m->tso_segsz);
1050                         rte_get_tx_ol_flag_list(m->ol_flags, buf, sizeof(buf));
1051                         printf("tx: flags=%s", buf);
1052                         printf("\n");
1053                 }
1054         }
1055
1056         if (unlikely(gro_enable)) {
1057                 if (gro_flush_cycles == GRO_DEFAULT_FLUSH_CYCLES) {
1058                         nb_rx = rte_gro_reassemble_burst(pkts_burst, nb_rx,
1059                                         &(gro_ports[fs->rx_port].param));
1060                 } else {
1061                         gro_ctx = current_fwd_lcore()->gro_ctx;
1062                         nb_rx = rte_gro_reassemble(pkts_burst, nb_rx, gro_ctx);
1063
1064                         if (++fs->gro_times >= gro_flush_cycles) {
1065                                 gro_pkts_num = rte_gro_get_pkt_count(gro_ctx);
1066                                 if (gro_pkts_num > MAX_PKT_BURST - nb_rx)
1067                                         gro_pkts_num = MAX_PKT_BURST - nb_rx;
1068
1069                                 nb_rx += rte_gro_timeout_flush(gro_ctx, 0,
1070                                                 RTE_GRO_TCP_IPV4,
1071                                                 &pkts_burst[nb_rx],
1072                                                 gro_pkts_num);
1073                                 fs->gro_times = 0;
1074                         }
1075                 }
1076         }
1077
1078         if (gso_ports[fs->tx_port].enable == 0)
1079                 tx_pkts_burst = pkts_burst;
1080         else {
1081                 gso_ctx = &(current_fwd_lcore()->gso_ctx);
1082                 gso_ctx->gso_size = gso_max_segment_size;
1083                 for (i = 0; i < nb_rx; i++) {
1084                         ret = rte_gso_segment(pkts_burst[i], gso_ctx,
1085                                         &gso_segments[nb_segments],
1086                                         GSO_MAX_PKT_BURST - nb_segments);
1087                         if (ret >= 1) {
1088                                 /* pkts_burst[i] can be freed safely here. */
1089                                 rte_pktmbuf_free(pkts_burst[i]);
1090                                 nb_segments += ret;
1091                         } else if (ret == 0) {
1092                                 /* 0 means it can be transmitted directly
1093                                  * without gso.
1094                                  */
1095                                 gso_segments[nb_segments] = pkts_burst[i];
1096                                 nb_segments += 1;
1097                         } else {
1098                                 TESTPMD_LOG(DEBUG, "Unable to segment packet");
1099                                 rte_pktmbuf_free(pkts_burst[i]);
1100                         }
1101                 }
1102
1103                 tx_pkts_burst = gso_segments;
1104                 nb_rx = nb_segments;
1105         }
1106
1107         nb_prep = rte_eth_tx_prepare(fs->tx_port, fs->tx_queue,
1108                         tx_pkts_burst, nb_rx);
1109         if (nb_prep != nb_rx)
1110                 printf("Preparing packet burst to transmit failed: %s\n",
1111                                 rte_strerror(rte_errno));
1112
1113         nb_tx = rte_eth_tx_burst(fs->tx_port, fs->tx_queue, tx_pkts_burst,
1114                         nb_prep);
1115
1116         /*
1117          * Retry if necessary
1118          */
1119         if (unlikely(nb_tx < nb_rx) && fs->retry_enabled) {
1120                 retry = 0;
1121                 while (nb_tx < nb_rx && retry++ < burst_tx_retry_num) {
1122                         rte_delay_us(burst_tx_delay_time);
1123                         nb_tx += rte_eth_tx_burst(fs->tx_port, fs->tx_queue,
1124                                         &tx_pkts_burst[nb_tx], nb_rx - nb_tx);
1125                 }
1126         }
1127         fs->tx_packets += nb_tx;
1128         fs->rx_bad_ip_csum += rx_bad_ip_csum;
1129         fs->rx_bad_l4_csum += rx_bad_l4_csum;
1130         fs->rx_bad_outer_l4_csum += rx_bad_outer_l4_csum;
1131         fs->rx_bad_outer_ip_csum += rx_bad_outer_ip_csum;
1132
1133         inc_tx_burst_stats(fs, nb_tx);
1134         if (unlikely(nb_tx < nb_rx)) {
1135                 fs->fwd_dropped += (nb_rx - nb_tx);
1136                 do {
1137                         rte_pktmbuf_free(tx_pkts_burst[nb_tx]);
1138                 } while (++nb_tx < nb_rx);
1139         }
1140
1141         get_end_cycles(fs, start_tsc);
1142 }
1143
1144 struct fwd_engine csum_fwd_engine = {
1145         .fwd_mode_name  = "csum",
1146         .port_fwd_begin = NULL,
1147         .port_fwd_end   = NULL,
1148         .packet_fwd     = pkt_burst_checksum_forward,
1149 };