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_sctp.h>
  39 #include <rte_prefetch.h>
  40 #include <rte_string_fns.h>
  41 #include <rte_flow.h>
  42 #include <rte_gro.h>
  43 #include <rte_gso.h>
  44
  45 #include "testpmd.h"
  46
  47 #define IP_DEFTTL  64   /* from RFC 1340. */
  48 #define IP_VERSION 0x40
  49 #define IP_HDRLEN  0x05 /* default IP header length == five 32-bits words. */
  50 #define IP_VHL_DEF (IP_VERSION | IP_HDRLEN)
  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 = 4790;
  67
  68 /* structure that caches offload info for the current packet */
  69 struct testpmd_offload_info {
  70         uint16_t ethertype;
  71         uint8_t gso_enable;
  72         uint16_t l2_len;
  73         uint16_t l3_len;
  74         uint16_t l4_len;
  75         uint8_t l4_proto;
  76         uint8_t is_tunnel;
  77         uint16_t outer_ethertype;
  78         uint16_t outer_l2_len;
  79         uint16_t outer_l3_len;
  80         uint8_t outer_l4_proto;
  81         uint16_t tso_segsz;
  82         uint16_t tunnel_tso_segsz;
  83         uint32_t pkt_len;
  84 };
  85
  86 /* simplified GRE header */
  87 struct simple_gre_hdr {
  88         uint16_t flags;
  89         uint16_t proto;
  90 } __attribute__((__packed__));
  91
  92 static uint16_t
  93 get_udptcp_checksum(void *l3_hdr, void *l4_hdr, uint16_t ethertype)
  94 {
  95         if (ethertype == _htons(RTE_ETHER_TYPE_IPv4))
  96                 return rte_ipv4_udptcp_cksum(l3_hdr, l4_hdr);
  97         else /* assume ethertype == RTE_ETHER_TYPE_IPv6 */
  98                 return rte_ipv6_udptcp_cksum(l3_hdr, l4_hdr);
  99 }
 100
 101 /* Parse an IPv4 header to fill l3_len, l4_len, and l4_proto */
 102 static void
 103 parse_ipv4(struct rte_ipv4_hdr *ipv4_hdr, struct testpmd_offload_info *info)
 104 {
 105         struct tcp_hdr *tcp_hdr;
 106
 107         info->l3_len = (ipv4_hdr->version_ihl & 0x0f) * 4;
 108         info->l4_proto = ipv4_hdr->next_proto_id;
 109
 110         /* only fill l4_len for TCP, it's useful for TSO */
 111         if (info->l4_proto == IPPROTO_TCP) {
 112                 tcp_hdr = (struct tcp_hdr *)((char *)ipv4_hdr + info->l3_len);
 113                 info->l4_len = (tcp_hdr->data_off & 0xf0) >> 2;
 114         } else if (info->l4_proto == IPPROTO_UDP)
 115                 info->l4_len = sizeof(struct udp_hdr);
 116         else
 117                 info->l4_len = 0;
 118 }
 119
 120 /* Parse an IPv6 header to fill l3_len, l4_len, and l4_proto */
 121 static void
 122 parse_ipv6(struct rte_ipv6_hdr *ipv6_hdr, struct testpmd_offload_info *info)
 123 {
 124         struct tcp_hdr *tcp_hdr;
 125
 126         info->l3_len = sizeof(struct rte_ipv6_hdr);
 127         info->l4_proto = ipv6_hdr->proto;
 128
 129         /* only fill l4_len for TCP, it's useful for TSO */
 130         if (info->l4_proto == IPPROTO_TCP) {
 131                 tcp_hdr = (struct tcp_hdr *)((char *)ipv6_hdr + info->l3_len);
 132                 info->l4_len = (tcp_hdr->data_off & 0xf0) >> 2;
 133         } else if (info->l4_proto == IPPROTO_UDP)
 134                 info->l4_len = sizeof(struct udp_hdr);
 135         else
 136                 info->l4_len = 0;
 137 }
 138
 139 /*
 140  * Parse an ethernet header to fill the ethertype, l2_len, l3_len and
 141  * ipproto. This function is able to recognize IPv4/IPv6 with one optional vlan
 142  * header. The l4_len argument is only set in case of TCP (useful for TSO).
 143  */
 144 static void
 145 parse_ethernet(struct rte_ether_hdr *eth_hdr, struct testpmd_offload_info *info)
 146 {
 147         struct rte_ipv4_hdr *ipv4_hdr;
 148         struct rte_ipv6_hdr *ipv6_hdr;
 149
 150         info->l2_len = sizeof(struct rte_ether_hdr);
 151         info->ethertype = eth_hdr->ether_type;
 152
 153         if (info->ethertype == _htons(RTE_ETHER_TYPE_VLAN)) {
 154                 struct rte_vlan_hdr *vlan_hdr = (
 155                         struct rte_vlan_hdr *)(eth_hdr + 1);
 156
 157                 info->l2_len  += sizeof(struct rte_vlan_hdr);
 158                 info->ethertype = vlan_hdr->eth_proto;
 159         }
 160
 161         switch (info->ethertype) {
 162         case _htons(RTE_ETHER_TYPE_IPv4):
 163                 ipv4_hdr = (struct rte_ipv4_hdr *)
 164                         ((char *)eth_hdr + info->l2_len);
 165                 parse_ipv4(ipv4_hdr, info);
 166                 break;
 167         case _htons(RTE_ETHER_TYPE_IPv6):
 168                 ipv6_hdr = (struct rte_ipv6_hdr *)
 169                         ((char *)eth_hdr + info->l2_len);
 170                 parse_ipv6(ipv6_hdr, info);
 171                 break;
 172         default:
 173                 info->l4_len = 0;
 174                 info->l3_len = 0;
 175                 info->l4_proto = 0;
 176                 break;
 177         }
 178 }
 179
 180 /* Parse a vxlan header */
 181 static void
 182 parse_vxlan(struct udp_hdr *udp_hdr,
 183             struct testpmd_offload_info *info,
 184             uint32_t pkt_type)
 185 {
 186         struct rte_ether_hdr *eth_hdr;
 187
 188         /* check udp destination port, 4789 is the default vxlan port
 189          * (rfc7348) or that the rx offload flag is set (i40e only
 190          * currently) */
 191         if (udp_hdr->dst_port != _htons(4789) &&
 192                 RTE_ETH_IS_TUNNEL_PKT(pkt_type) == 0)
 193                 return;
 194
 195         info->is_tunnel = 1;
 196         info->outer_ethertype = info->ethertype;
 197         info->outer_l2_len = info->l2_len;
 198         info->outer_l3_len = info->l3_len;
 199         info->outer_l4_proto = info->l4_proto;
 200
 201         eth_hdr = (struct rte_ether_hdr *)((char *)udp_hdr +
 202                 sizeof(struct udp_hdr) +
 203                 sizeof(struct rte_vxlan_hdr));
 204
 205         parse_ethernet(eth_hdr, info);
 206         info->l2_len += RTE_ETHER_VXLAN_HLEN; /* add udp + vxlan */
 207 }
 208
 209 /* Parse a vxlan-gpe header */
 210 static void
 211 parse_vxlan_gpe(struct udp_hdr *udp_hdr,
 212             struct testpmd_offload_info *info)
 213 {
 214         struct rte_ether_hdr *eth_hdr;
 215         struct rte_ipv4_hdr *ipv4_hdr;
 216         struct rte_ipv6_hdr *ipv6_hdr;
 217         struct rte_vxlan_gpe_hdr *vxlan_gpe_hdr;
 218         uint8_t vxlan_gpe_len = sizeof(*vxlan_gpe_hdr);
 219
 220         /* Check udp destination port. */
 221         if (udp_hdr->dst_port != _htons(vxlan_gpe_udp_port))
 222                 return;
 223
 224         vxlan_gpe_hdr = (struct rte_vxlan_gpe_hdr *)((char *)udp_hdr +
 225                                 sizeof(struct udp_hdr));
 226
 227         if (!vxlan_gpe_hdr->proto || vxlan_gpe_hdr->proto ==
 228             RTE_VXLAN_GPE_TYPE_IPV4) {
 229                 info->is_tunnel = 1;
 230                 info->outer_ethertype = info->ethertype;
 231                 info->outer_l2_len = info->l2_len;
 232                 info->outer_l3_len = info->l3_len;
 233                 info->outer_l4_proto = info->l4_proto;
 234
 235                 ipv4_hdr = (struct rte_ipv4_hdr *)((char *)vxlan_gpe_hdr +
 236                            vxlan_gpe_len);
 237
 238                 parse_ipv4(ipv4_hdr, info);
 239                 info->ethertype = _htons(RTE_ETHER_TYPE_IPv4);
 240                 info->l2_len = 0;
 241
 242         } else if (vxlan_gpe_hdr->proto == RTE_VXLAN_GPE_TYPE_IPV6) {
 243                 info->is_tunnel = 1;
 244                 info->outer_ethertype = info->ethertype;
 245                 info->outer_l2_len = info->l2_len;
 246                 info->outer_l3_len = info->l3_len;
 247                 info->outer_l4_proto = info->l4_proto;
 248
 249                 ipv6_hdr = (struct rte_ipv6_hdr *)((char *)vxlan_gpe_hdr +
 250                            vxlan_gpe_len);
 251
 252                 info->ethertype = _htons(RTE_ETHER_TYPE_IPv6);
 253                 parse_ipv6(ipv6_hdr, info);
 254                 info->l2_len = 0;
 255
 256         } else if (vxlan_gpe_hdr->proto == RTE_VXLAN_GPE_TYPE_ETH) {
 257                 info->is_tunnel = 1;
 258                 info->outer_ethertype = info->ethertype;
 259                 info->outer_l2_len = info->l2_len;
 260                 info->outer_l3_len = info->l3_len;
 261                 info->outer_l4_proto = info->l4_proto;
 262
 263                 eth_hdr = (struct rte_ether_hdr *)((char *)vxlan_gpe_hdr +
 264                           vxlan_gpe_len);
 265
 266                 parse_ethernet(eth_hdr, info);
 267         } else
 268                 return;
 269
 270         info->l2_len += RTE_ETHER_VXLAN_GPE_HLEN;
 271 }
 272
 273 /* Parse a gre header */
 274 static void
 275 parse_gre(struct simple_gre_hdr *gre_hdr, struct testpmd_offload_info *info)
 276 {
 277         struct rte_ether_hdr *eth_hdr;
 278         struct rte_ipv4_hdr *ipv4_hdr;
 279         struct rte_ipv6_hdr *ipv6_hdr;
 280         uint8_t gre_len = 0;
 281
 282         gre_len += sizeof(struct simple_gre_hdr);
 283
 284         if (gre_hdr->flags & _htons(GRE_KEY_PRESENT))
 285                 gre_len += GRE_EXT_LEN;
 286         if (gre_hdr->flags & _htons(GRE_SEQUENCE_PRESENT))
 287                 gre_len += GRE_EXT_LEN;
 288         if (gre_hdr->flags & _htons(GRE_CHECKSUM_PRESENT))
 289                 gre_len += GRE_EXT_LEN;
 290
 291         if (gre_hdr->proto == _htons(RTE_ETHER_TYPE_IPv4)) {
 292                 info->is_tunnel = 1;
 293                 info->outer_ethertype = info->ethertype;
 294                 info->outer_l2_len = info->l2_len;
 295                 info->outer_l3_len = info->l3_len;
 296                 info->outer_l4_proto = info->l4_proto;
 297
 298                 ipv4_hdr = (struct rte_ipv4_hdr *)((char *)gre_hdr + gre_len);
 299
 300                 parse_ipv4(ipv4_hdr, info);
 301                 info->ethertype = _htons(RTE_ETHER_TYPE_IPv4);
 302                 info->l2_len = 0;
 303
 304         } else if (gre_hdr->proto == _htons(RTE_ETHER_TYPE_IPv6)) {
 305                 info->is_tunnel = 1;
 306                 info->outer_ethertype = info->ethertype;
 307                 info->outer_l2_len = info->l2_len;
 308                 info->outer_l3_len = info->l3_len;
 309                 info->outer_l4_proto = info->l4_proto;
 310
 311                 ipv6_hdr = (struct rte_ipv6_hdr *)((char *)gre_hdr + gre_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 (gre_hdr->proto == _htons(RTE_ETHER_TYPE_TEB)) {
 318                 info->is_tunnel = 1;
 319                 info->outer_ethertype = info->ethertype;
 320                 info->outer_l2_len = info->l2_len;
 321                 info->outer_l3_len = info->l3_len;
 322                 info->outer_l4_proto = info->l4_proto;
 323
 324                 eth_hdr = (struct rte_ether_hdr *)((char *)gre_hdr + gre_len);
 325
 326                 parse_ethernet(eth_hdr, info);
 327         } else
 328                 return;
 329
 330         info->l2_len += gre_len;
 331 }
 332
 333
 334 /* Parse an encapsulated ip or ipv6 header */
 335 static void
 336 parse_encap_ip(void *encap_ip, struct testpmd_offload_info *info)
 337 {
 338         struct rte_ipv4_hdr *ipv4_hdr = encap_ip;
 339         struct rte_ipv6_hdr *ipv6_hdr = encap_ip;
 340         uint8_t ip_version;
 341
 342         ip_version = (ipv4_hdr->version_ihl & 0xf0) >> 4;
 343
 344         if (ip_version != 4 && ip_version != 6)
 345                 return;
 346
 347         info->is_tunnel = 1;
 348         info->outer_ethertype = info->ethertype;
 349         info->outer_l2_len = info->l2_len;
 350         info->outer_l3_len = info->l3_len;
 351
 352         if (ip_version == 4) {
 353                 parse_ipv4(ipv4_hdr, info);
 354                 info->ethertype = _htons(RTE_ETHER_TYPE_IPv4);
 355         } else {
 356                 parse_ipv6(ipv6_hdr, info);
 357                 info->ethertype = _htons(RTE_ETHER_TYPE_IPv6);
 358         }
 359         info->l2_len = 0;
 360 }
 361
 362 /* if possible, calculate the checksum of a packet in hw or sw,
 363  * depending on the testpmd command line configuration */
 364 static uint64_t
 365 process_inner_cksums(void *l3_hdr, const struct testpmd_offload_info *info,
 366         uint64_t tx_offloads)
 367 {
 368         struct rte_ipv4_hdr *ipv4_hdr = l3_hdr;
 369         struct udp_hdr *udp_hdr;
 370         struct tcp_hdr *tcp_hdr;
 371         struct rte_sctp_hdr *sctp_hdr;
 372         uint64_t ol_flags = 0;
 373         uint32_t max_pkt_len, tso_segsz = 0;
 374
 375         /* ensure packet is large enough to require tso */
 376         if (!info->is_tunnel) {
 377                 max_pkt_len = info->l2_len + info->l3_len + info->l4_len +
 378                         info->tso_segsz;
 379                 if (info->tso_segsz != 0 && info->pkt_len > max_pkt_len)
 380                         tso_segsz = info->tso_segsz;
 381         } else {
 382                 max_pkt_len = info->outer_l2_len + info->outer_l3_len +
 383                         info->l2_len + info->l3_len + info->l4_len +
 384                         info->tunnel_tso_segsz;
 385                 if (info->tunnel_tso_segsz != 0 && info->pkt_len > max_pkt_len)
 386                         tso_segsz = info->tunnel_tso_segsz;
 387         }
 388
 389         if (info->ethertype == _htons(RTE_ETHER_TYPE_IPv4)) {
 390                 ipv4_hdr = l3_hdr;
 391                 ipv4_hdr->hdr_checksum = 0;
 392
 393                 ol_flags |= PKT_TX_IPV4;
 394                 if (info->l4_proto == IPPROTO_TCP && tso_segsz) {
 395                         ol_flags |= PKT_TX_IP_CKSUM;
 396                 } else {
 397                         if (tx_offloads & DEV_TX_OFFLOAD_IPV4_CKSUM)
 398                                 ol_flags |= PKT_TX_IP_CKSUM;
 399                         else
 400                                 ipv4_hdr->hdr_checksum =
 401                                         rte_ipv4_cksum(ipv4_hdr);
 402                 }
 403         } else if (info->ethertype == _htons(RTE_ETHER_TYPE_IPv6))
 404                 ol_flags |= PKT_TX_IPV6;
 405         else
 406                 return 0; /* packet type not supported, nothing to do */
 407
 408         if (info->l4_proto == IPPROTO_UDP) {
 409                 udp_hdr = (struct udp_hdr *)((char *)l3_hdr + info->l3_len);
 410                 /* do not recalculate udp cksum if it was 0 */
 411                 if (udp_hdr->dgram_cksum != 0) {
 412                         udp_hdr->dgram_cksum = 0;
 413                         if (tx_offloads & DEV_TX_OFFLOAD_UDP_CKSUM)
 414                                 ol_flags |= PKT_TX_UDP_CKSUM;
 415                         else {
 416                                 udp_hdr->dgram_cksum =
 417                                         get_udptcp_checksum(l3_hdr, udp_hdr,
 418                                                 info->ethertype);
 419                         }
 420                 }
 421                 if (info->gso_enable)
 422                         ol_flags |= PKT_TX_UDP_SEG;
 423         } else if (info->l4_proto == IPPROTO_TCP) {
 424                 tcp_hdr = (struct tcp_hdr *)((char *)l3_hdr + info->l3_len);
 425                 tcp_hdr->cksum = 0;
 426                 if (tso_segsz)
 427                         ol_flags |= PKT_TX_TCP_SEG;
 428                 else if (tx_offloads & DEV_TX_OFFLOAD_TCP_CKSUM)
 429                         ol_flags |= PKT_TX_TCP_CKSUM;
 430                 else {
 431                         tcp_hdr->cksum =
 432                                 get_udptcp_checksum(l3_hdr, tcp_hdr,
 433                                         info->ethertype);
 434                 }
 435                 if (info->gso_enable)
 436                         ol_flags |= PKT_TX_TCP_SEG;
 437         } else if (info->l4_proto == IPPROTO_SCTP) {
 438                 sctp_hdr = (struct rte_sctp_hdr *)
 439                         ((char *)l3_hdr + info->l3_len);
 440                 sctp_hdr->cksum = 0;
 441                 /* sctp payload must be a multiple of 4 to be
 442                  * offloaded */
 443                 if ((tx_offloads & DEV_TX_OFFLOAD_SCTP_CKSUM) &&
 444                         ((ipv4_hdr->total_length & 0x3) == 0)) {
 445                         ol_flags |= PKT_TX_SCTP_CKSUM;
 446                 } else {
 447                         /* XXX implement CRC32c, example available in
 448                          * RFC3309 */
 449                 }
 450         }
 451
 452         return ol_flags;
 453 }
 454
 455 /* Calculate the checksum of outer header */
 456 static uint64_t
 457 process_outer_cksums(void *outer_l3_hdr, struct testpmd_offload_info *info,
 458         uint64_t tx_offloads, int tso_enabled)
 459 {
 460         struct rte_ipv4_hdr *ipv4_hdr = outer_l3_hdr;
 461         struct rte_ipv6_hdr *ipv6_hdr = outer_l3_hdr;
 462         struct udp_hdr *udp_hdr;
 463         uint64_t ol_flags = 0;
 464
 465         if (info->outer_ethertype == _htons(RTE_ETHER_TYPE_IPv4)) {
 466                 ipv4_hdr->hdr_checksum = 0;
 467                 ol_flags |= PKT_TX_OUTER_IPV4;
 468
 469                 if (tx_offloads & DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM)
 470                         ol_flags |= PKT_TX_OUTER_IP_CKSUM;
 471                 else
 472                         ipv4_hdr->hdr_checksum = rte_ipv4_cksum(ipv4_hdr);
 473         } else
 474                 ol_flags |= PKT_TX_OUTER_IPV6;
 475
 476         if (info->outer_l4_proto != IPPROTO_UDP)
 477                 return ol_flags;
 478
 479         /* Skip SW outer UDP checksum generation if HW supports it */
 480         if (tx_offloads & DEV_TX_OFFLOAD_OUTER_UDP_CKSUM) {
 481                 ol_flags |= PKT_TX_OUTER_UDP_CKSUM;
 482                 return ol_flags;
 483         }
 484
 485         udp_hdr = (struct udp_hdr *)((char *)outer_l3_hdr + info->outer_l3_len);
 486
 487         /* outer UDP checksum is done in software. In the other side, for
 488          * UDP tunneling, like VXLAN or Geneve, outer UDP checksum can be
 489          * set to zero.
 490          *
 491          * If a packet will be TSOed into small packets by NIC, we cannot
 492          * set/calculate a non-zero checksum, because it will be a wrong
 493          * value after the packet be split into several small packets.
 494          */
 495         if (tso_enabled)
 496                 udp_hdr->dgram_cksum = 0;
 497
 498         /* do not recalculate udp cksum if it was 0 */
 499         if (udp_hdr->dgram_cksum != 0) {
 500                 udp_hdr->dgram_cksum = 0;
 501                 if (info->outer_ethertype == _htons(RTE_ETHER_TYPE_IPv4))
 502                         udp_hdr->dgram_cksum =
 503                                 rte_ipv4_udptcp_cksum(ipv4_hdr, udp_hdr);
 504                 else
 505                         udp_hdr->dgram_cksum =
 506                                 rte_ipv6_udptcp_cksum(ipv6_hdr, udp_hdr);
 507         }
 508
 509         return ol_flags;
 510 }
 511
 512 /*
 513  * Helper function.
 514  * Performs actual copying.
 515  * Returns number of segments in the destination mbuf on success,
 516  * or negative error code on failure.
 517  */
 518 static int
 519 mbuf_copy_split(const struct rte_mbuf *ms, struct rte_mbuf *md[],
 520         uint16_t seglen[], uint8_t nb_seg)
 521 {
 522         uint32_t dlen, slen, tlen;
 523         uint32_t i, len;
 524         const struct rte_mbuf *m;
 525         const uint8_t *src;
 526         uint8_t *dst;
 527
 528         dlen = 0;
 529         slen = 0;
 530         tlen = 0;
 531
 532         dst = NULL;
 533         src = NULL;
 534
 535         m = ms;
 536         i = 0;
 537         while (ms != NULL && i != nb_seg) {
 538
 539                 if (slen == 0) {
 540                         slen = rte_pktmbuf_data_len(ms);
 541                         src = rte_pktmbuf_mtod(ms, const uint8_t *);
 542                 }
 543
 544                 if (dlen == 0) {
 545                         dlen = RTE_MIN(seglen[i], slen);
 546                         md[i]->data_len = dlen;
 547                         md[i]->next = (i + 1 == nb_seg) ? NULL : md[i + 1];
 548                         dst = rte_pktmbuf_mtod(md[i], uint8_t *);
 549                 }
 550
 551                 len = RTE_MIN(slen, dlen);
 552                 memcpy(dst, src, len);
 553                 tlen += len;
 554                 slen -= len;
 555                 dlen -= len;
 556                 src += len;
 557                 dst += len;
 558
 559                 if (slen == 0)
 560                         ms = ms->next;
 561                 if (dlen == 0)
 562                         i++;
 563         }
 564
 565         if (ms != NULL)
 566                 return -ENOBUFS;
 567         else if (tlen != m->pkt_len)
 568                 return -EINVAL;
 569
 570         md[0]->nb_segs = nb_seg;
 571         md[0]->pkt_len = tlen;
 572         md[0]->vlan_tci = m->vlan_tci;
 573         md[0]->vlan_tci_outer = m->vlan_tci_outer;
 574         md[0]->ol_flags = m->ol_flags;
 575         md[0]->tx_offload = m->tx_offload;
 576
 577         return nb_seg;
 578 }
 579
 580 /*
 581  * Allocate a new mbuf with up to tx_pkt_nb_segs segments.
 582  * Copy packet contents and offload information into the new segmented mbuf.
 583  */
 584 static struct rte_mbuf *
 585 pkt_copy_split(const struct rte_mbuf *pkt)
 586 {
 587         int32_t n, rc;
 588         uint32_t i, len, nb_seg;
 589         struct rte_mempool *mp;
 590         uint16_t seglen[RTE_MAX_SEGS_PER_PKT];
 591         struct rte_mbuf *p, *md[RTE_MAX_SEGS_PER_PKT];
 592
 593         mp = current_fwd_lcore()->mbp;
 594
 595         if (tx_pkt_split == TX_PKT_SPLIT_RND)
 596                 nb_seg = random() % tx_pkt_nb_segs + 1;
 597         else
 598                 nb_seg = tx_pkt_nb_segs;
 599
 600         memcpy(seglen, tx_pkt_seg_lengths, nb_seg * sizeof(seglen[0]));
 601
 602         /* calculate number of segments to use and their length. */
 603         len = 0;
 604         for (i = 0; i != nb_seg && len < pkt->pkt_len; i++) {
 605                 len += seglen[i];
 606                 md[i] = NULL;
 607         }
 608
 609         n = pkt->pkt_len - len;
 610
 611         /* update size of the last segment to fit rest of the packet */
 612         if (n >= 0) {
 613                 seglen[i - 1] += n;
 614                 len += n;
 615         }
 616
 617         nb_seg = i;
 618         while (i != 0) {
 619                 p = rte_pktmbuf_alloc(mp);
 620                 if (p == NULL) {
 621                         TESTPMD_LOG(ERR,
 622                                 "failed to allocate %u-th of %u mbuf "
 623                                 "from mempool: %s\n",
 624                                 nb_seg - i, nb_seg, mp->name);
 625                         break;
 626                 }
 627
 628                 md[--i] = p;
 629                 if (rte_pktmbuf_tailroom(md[i]) < seglen[i]) {
 630                         TESTPMD_LOG(ERR, "mempool %s, %u-th segment: "
 631                                 "expected seglen: %u, "
 632                                 "actual mbuf tailroom: %u\n",
 633                                 mp->name, i, seglen[i],
 634                                 rte_pktmbuf_tailroom(md[i]));
 635                         break;
 636                 }
 637         }
 638
 639         /* all mbufs successfully allocated, do copy */
 640         if (i == 0) {
 641                 rc = mbuf_copy_split(pkt, md, seglen, nb_seg);
 642                 if (rc < 0)
 643                         TESTPMD_LOG(ERR,
 644                                 "mbuf_copy_split for %p(len=%u, nb_seg=%u) "
 645                                 "into %u segments failed with error code: %d\n",
 646                                 pkt, pkt->pkt_len, pkt->nb_segs, nb_seg, rc);
 647
 648                 /* figure out how many mbufs to free. */
 649                 i = RTE_MAX(rc, 0);
 650         }
 651
 652         /* free unused mbufs */
 653         for (; i != nb_seg; i++) {
 654                 rte_pktmbuf_free_seg(md[i]);
 655                 md[i] = NULL;
 656         }
 657
 658         return md[0];
 659 }
 660
 661 /*
 662  * Receive a burst of packets, and for each packet:
 663  *  - parse packet, and try to recognize a supported packet type (1)
 664  *  - if it's not a supported packet type, don't touch the packet, else:
 665  *  - reprocess the checksum of all supported layers. This is done in SW
 666  *    or HW, depending on testpmd command line configuration
 667  *  - if TSO is enabled in testpmd command line, also flag the mbuf for TCP
 668  *    segmentation offload (this implies HW TCP checksum)
 669  * Then transmit packets on the output port.
 670  *
 671  * (1) Supported packets are:
 672  *   Ether / (vlan) / IP|IP6 / UDP|TCP|SCTP .
 673  *   Ether / (vlan) / outer IP|IP6 / outer UDP / VxLAN / Ether / IP|IP6 /
 674  *           UDP|TCP|SCTP
 675  *   Ether / (vlan) / outer IP|IP6 / outer UDP / VXLAN-GPE / Ether / IP|IP6 /
 676  *           UDP|TCP|SCTP
 677  *   Ether / (vlan) / outer IP|IP6 / outer UDP / VXLAN-GPE / IP|IP6 /
 678  *           UDP|TCP|SCTP
 679  *   Ether / (vlan) / outer IP|IP6 / GRE / Ether / IP|IP6 / UDP|TCP|SCTP
 680  *   Ether / (vlan) / outer IP|IP6 / GRE / IP|IP6 / UDP|TCP|SCTP
 681  *   Ether / (vlan) / outer IP|IP6 / IP|IP6 / UDP|TCP|SCTP
 682  *
 683  * The testpmd command line for this forward engine sets the flags
 684  * TESTPMD_TX_OFFLOAD_* in ports[tx_port].tx_ol_flags. They control
 685  * wether a checksum must be calculated in software or in hardware. The
 686  * IP, UDP, TCP and SCTP flags always concern the inner layer. The
 687  * OUTER_IP is only useful for tunnel packets.
 688  */
 689 static void
 690 pkt_burst_checksum_forward(struct fwd_stream *fs)
 691 {
 692         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
 693         struct rte_mbuf *gso_segments[GSO_MAX_PKT_BURST];
 694         struct rte_gso_ctx *gso_ctx;
 695         struct rte_mbuf **tx_pkts_burst;
 696         struct rte_port *txp;
 697         struct rte_mbuf *m, *p;
 698         struct rte_ether_hdr *eth_hdr;
 699         void *l3_hdr = NULL, *outer_l3_hdr = NULL; /* can be IPv4 or IPv6 */
 700         void **gro_ctx;
 701         uint16_t gro_pkts_num;
 702         uint8_t gro_enable;
 703         uint16_t nb_rx;
 704         uint16_t nb_tx;
 705         uint16_t nb_prep;
 706         uint16_t i;
 707         uint64_t rx_ol_flags, tx_ol_flags;
 708         uint64_t tx_offloads;
 709         uint32_t retry;
 710         uint32_t rx_bad_ip_csum;
 711         uint32_t rx_bad_l4_csum;
 712         uint32_t rx_bad_outer_l4_csum;
 713         struct testpmd_offload_info info;
 714         uint16_t nb_segments = 0;
 715         int ret;
 716
 717 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
 718         uint64_t start_tsc;
 719         uint64_t end_tsc;
 720         uint64_t core_cycles;
 721 #endif
 722
 723 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
 724         start_tsc = rte_rdtsc();
 725 #endif
 726
 727         /* receive a burst of packet */
 728         nb_rx = rte_eth_rx_burst(fs->rx_port, fs->rx_queue, pkts_burst,
 729                                  nb_pkt_per_burst);
 730         if (unlikely(nb_rx == 0))
 731                 return;
 732 #ifdef RTE_TEST_PMD_RECORD_BURST_STATS
 733         fs->rx_burst_stats.pkt_burst_spread[nb_rx]++;
 734 #endif
 735         fs->rx_packets += nb_rx;
 736         rx_bad_ip_csum = 0;
 737         rx_bad_l4_csum = 0;
 738         rx_bad_outer_l4_csum = 0;
 739         gro_enable = gro_ports[fs->rx_port].enable;
 740
 741         txp = &ports[fs->tx_port];
 742         tx_offloads = txp->dev_conf.txmode.offloads;
 743         memset(&info, 0, sizeof(info));
 744         info.tso_segsz = txp->tso_segsz;
 745         info.tunnel_tso_segsz = txp->tunnel_tso_segsz;
 746         if (gso_ports[fs->tx_port].enable)
 747                 info.gso_enable = 1;
 748
 749         for (i = 0; i < nb_rx; i++) {
 750                 if (likely(i < nb_rx - 1))
 751                         rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[i + 1],
 752                                                        void *));
 753
 754                 m = pkts_burst[i];
 755                 info.is_tunnel = 0;
 756                 info.pkt_len = rte_pktmbuf_pkt_len(m);
 757                 tx_ol_flags = m->ol_flags &
 758                               (IND_ATTACHED_MBUF | EXT_ATTACHED_MBUF);
 759                 rx_ol_flags = m->ol_flags;
 760
 761                 /* Update the L3/L4 checksum error packet statistics */
 762                 if ((rx_ol_flags & PKT_RX_IP_CKSUM_MASK) == PKT_RX_IP_CKSUM_BAD)
 763                         rx_bad_ip_csum += 1;
 764                 if ((rx_ol_flags & PKT_RX_L4_CKSUM_MASK) == PKT_RX_L4_CKSUM_BAD)
 765                         rx_bad_l4_csum += 1;
 766                 if (rx_ol_flags & PKT_RX_OUTER_L4_CKSUM_BAD)
 767                         rx_bad_outer_l4_csum += 1;
 768
 769                 /* step 1: dissect packet, parsing optional vlan, ip4/ip6, vxlan
 770                  * and inner headers */
 771
 772                 eth_hdr = rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
 773                 rte_ether_addr_copy(&peer_eth_addrs[fs->peer_addr],
 774                                 &eth_hdr->d_addr);
 775                 rte_ether_addr_copy(&ports[fs->tx_port].eth_addr,
 776                                 &eth_hdr->s_addr);
 777                 parse_ethernet(eth_hdr, &info);
 778                 l3_hdr = (char *)eth_hdr + info.l2_len;
 779
 780                 /* check if it's a supported tunnel */
 781                 if (txp->parse_tunnel) {
 782                         if (info.l4_proto == IPPROTO_UDP) {
 783                                 struct udp_hdr *udp_hdr;
 784
 785                                 udp_hdr = (struct udp_hdr *)((char *)l3_hdr +
 786                                         info.l3_len);
 787                                 parse_vxlan_gpe(udp_hdr, &info);
 788                                 if (info.is_tunnel) {
 789                                         tx_ol_flags |= PKT_TX_TUNNEL_VXLAN_GPE;
 790                                 } else {
 791                                         parse_vxlan(udp_hdr, &info,
 792                                                     m->packet_type);
 793                                         if (info.is_tunnel)
 794                                                 tx_ol_flags |=
 795                                                         PKT_TX_TUNNEL_VXLAN;
 796                                 }
 797                         } else if (info.l4_proto == IPPROTO_GRE) {
 798                                 struct simple_gre_hdr *gre_hdr;
 799
 800                                 gre_hdr = (struct simple_gre_hdr *)
 801                                         ((char *)l3_hdr + info.l3_len);
 802                                 parse_gre(gre_hdr, &info);
 803                                 if (info.is_tunnel)
 804                                         tx_ol_flags |= PKT_TX_TUNNEL_GRE;
 805                         } else if (info.l4_proto == IPPROTO_IPIP) {
 806                                 void *encap_ip_hdr;
 807
 808                                 encap_ip_hdr = (char *)l3_hdr + info.l3_len;
 809                                 parse_encap_ip(encap_ip_hdr, &info);
 810                                 if (info.is_tunnel)
 811                                         tx_ol_flags |= PKT_TX_TUNNEL_IPIP;
 812                         }
 813                 }
 814
 815                 /* update l3_hdr and outer_l3_hdr if a tunnel was parsed */
 816                 if (info.is_tunnel) {
 817                         outer_l3_hdr = l3_hdr;
 818                         l3_hdr = (char *)l3_hdr + info.outer_l3_len + info.l2_len;
 819                 }
 820
 821                 /* step 2: depending on user command line configuration,
 822                  * recompute checksum either in software or flag the
 823                  * mbuf to offload the calculation to the NIC. If TSO
 824                  * is configured, prepare the mbuf for TCP segmentation. */
 825
 826                 /* process checksums of inner headers first */
 827                 tx_ol_flags |= process_inner_cksums(l3_hdr, &info,
 828                         tx_offloads);
 829
 830                 /* Then process outer headers if any. Note that the software
 831                  * checksum will be wrong if one of the inner checksums is
 832                  * processed in hardware. */
 833                 if (info.is_tunnel == 1) {
 834                         tx_ol_flags |= process_outer_cksums(outer_l3_hdr, &info,
 835                                         tx_offloads,
 836                                         !!(tx_ol_flags & PKT_TX_TCP_SEG));
 837                 }
 838
 839                 /* step 3: fill the mbuf meta data (flags and header lengths) */
 840
 841                 m->tx_offload = 0;
 842                 if (info.is_tunnel == 1) {
 843                         if (info.tunnel_tso_segsz ||
 844                             (tx_offloads &
 845                              DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM) ||
 846                             (tx_offloads &
 847                              DEV_TX_OFFLOAD_OUTER_UDP_CKSUM) ||
 848                             (tx_ol_flags & PKT_TX_OUTER_IPV6)) {
 849                                 m->outer_l2_len = info.outer_l2_len;
 850                                 m->outer_l3_len = info.outer_l3_len;
 851                                 m->l2_len = info.l2_len;
 852                                 m->l3_len = info.l3_len;
 853                                 m->l4_len = info.l4_len;
 854                                 m->tso_segsz = info.tunnel_tso_segsz;
 855                         }
 856                         else {
 857                                 /* if there is a outer UDP cksum
 858                                    processed in sw and the inner in hw,
 859                                    the outer checksum will be wrong as
 860                                    the payload will be modified by the
 861                                    hardware */
 862                                 m->l2_len = info.outer_l2_len +
 863                                         info.outer_l3_len + info.l2_len;
 864                                 m->l3_len = info.l3_len;
 865                                 m->l4_len = info.l4_len;
 866                         }
 867                 } else {
 868                         /* this is only useful if an offload flag is
 869                          * set, but it does not hurt to fill it in any
 870                          * case */
 871                         m->l2_len = info.l2_len;
 872                         m->l3_len = info.l3_len;
 873                         m->l4_len = info.l4_len;
 874                         m->tso_segsz = info.tso_segsz;
 875                 }
 876                 m->ol_flags = tx_ol_flags;
 877
 878                 /* Do split & copy for the packet. */
 879                 if (tx_pkt_split != TX_PKT_SPLIT_OFF) {
 880                         p = pkt_copy_split(m);
 881                         if (p != NULL) {
 882                                 rte_pktmbuf_free(m);
 883                                 m = p;
 884                                 pkts_burst[i] = m;
 885                         }
 886                 }
 887
 888                 /* if verbose mode is enabled, dump debug info */
 889                 if (verbose_level > 0) {
 890                         char buf[256];
 891
 892                         printf("-----------------\n");
 893                         printf("port=%u, mbuf=%p, pkt_len=%u, nb_segs=%u:\n",
 894                                 fs->rx_port, m, m->pkt_len, m->nb_segs);
 895                         /* dump rx parsed packet info */
 896                         rte_get_rx_ol_flag_list(rx_ol_flags, buf, sizeof(buf));
 897                         printf("rx: l2_len=%d ethertype=%x l3_len=%d "
 898                                 "l4_proto=%d l4_len=%d flags=%s\n",
 899                                 info.l2_len, rte_be_to_cpu_16(info.ethertype),
 900                                 info.l3_len, info.l4_proto, info.l4_len, buf);
 901                         if (rx_ol_flags & PKT_RX_LRO)
 902                                 printf("rx: m->lro_segsz=%u\n", m->tso_segsz);
 903                         if (info.is_tunnel == 1)
 904                                 printf("rx: outer_l2_len=%d outer_ethertype=%x "
 905                                         "outer_l3_len=%d\n", info.outer_l2_len,
 906                                         rte_be_to_cpu_16(info.outer_ethertype),
 907                                         info.outer_l3_len);
 908                         /* dump tx packet info */
 909                         if ((tx_offloads & (DEV_TX_OFFLOAD_IPV4_CKSUM |
 910                                             DEV_TX_OFFLOAD_UDP_CKSUM |
 911                                             DEV_TX_OFFLOAD_TCP_CKSUM |
 912                                             DEV_TX_OFFLOAD_SCTP_CKSUM)) ||
 913                                 info.tso_segsz != 0)
 914                                 printf("tx: m->l2_len=%d m->l3_len=%d "
 915                                         "m->l4_len=%d\n",
 916                                         m->l2_len, m->l3_len, m->l4_len);
 917                         if (info.is_tunnel == 1) {
 918                                 if ((tx_offloads &
 919                                     DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM) ||
 920                                     (tx_offloads &
 921                                     DEV_TX_OFFLOAD_OUTER_UDP_CKSUM) ||
 922                                     (tx_ol_flags & PKT_TX_OUTER_IPV6))
 923                                         printf("tx: m->outer_l2_len=%d "
 924                                                 "m->outer_l3_len=%d\n",
 925                                                 m->outer_l2_len,
 926                                                 m->outer_l3_len);
 927                                 if (info.tunnel_tso_segsz != 0 &&
 928                                                 (m->ol_flags & PKT_TX_TCP_SEG))
 929                                         printf("tx: m->tso_segsz=%d\n",
 930                                                 m->tso_segsz);
 931                         } else if (info.tso_segsz != 0 &&
 932                                         (m->ol_flags & PKT_TX_TCP_SEG))
 933                                 printf("tx: m->tso_segsz=%d\n", m->tso_segsz);
 934                         rte_get_tx_ol_flag_list(m->ol_flags, buf, sizeof(buf));
 935                         printf("tx: flags=%s", buf);
 936                         printf("\n");
 937                 }
 938         }
 939
 940         if (unlikely(gro_enable)) {
 941                 if (gro_flush_cycles == GRO_DEFAULT_FLUSH_CYCLES) {
 942                         nb_rx = rte_gro_reassemble_burst(pkts_burst, nb_rx,
 943                                         &(gro_ports[fs->rx_port].param));
 944                 } else {
 945                         gro_ctx = current_fwd_lcore()->gro_ctx;
 946                         nb_rx = rte_gro_reassemble(pkts_burst, nb_rx, gro_ctx);
 947
 948                         if (++fs->gro_times >= gro_flush_cycles) {
 949                                 gro_pkts_num = rte_gro_get_pkt_count(gro_ctx);
 950                                 if (gro_pkts_num > MAX_PKT_BURST - nb_rx)
 951                                         gro_pkts_num = MAX_PKT_BURST - nb_rx;
 952
 953                                 nb_rx += rte_gro_timeout_flush(gro_ctx, 0,
 954                                                 RTE_GRO_TCP_IPV4,
 955                                                 &pkts_burst[nb_rx],
 956                                                 gro_pkts_num);
 957                                 fs->gro_times = 0;
 958                         }
 959                 }
 960         }
 961
 962         if (gso_ports[fs->tx_port].enable == 0)
 963                 tx_pkts_burst = pkts_burst;
 964         else {
 965                 gso_ctx = &(current_fwd_lcore()->gso_ctx);
 966                 gso_ctx->gso_size = gso_max_segment_size;
 967                 for (i = 0; i < nb_rx; i++) {
 968                         ret = rte_gso_segment(pkts_burst[i], gso_ctx,
 969                                         &gso_segments[nb_segments],
 970                                         GSO_MAX_PKT_BURST - nb_segments);
 971                         if (ret >= 0)
 972                                 nb_segments += ret;
 973                         else {
 974                                 TESTPMD_LOG(DEBUG, "Unable to segment packet");
 975                                 rte_pktmbuf_free(pkts_burst[i]);
 976                         }
 977                 }
 978
 979                 tx_pkts_burst = gso_segments;
 980                 nb_rx = nb_segments;
 981         }
 982
 983         nb_prep = rte_eth_tx_prepare(fs->tx_port, fs->tx_queue,
 984                         tx_pkts_burst, nb_rx);
 985         if (nb_prep != nb_rx)
 986                 printf("Preparing packet burst to transmit failed: %s\n",
 987                                 rte_strerror(rte_errno));
 988
 989         nb_tx = rte_eth_tx_burst(fs->tx_port, fs->tx_queue, tx_pkts_burst,
 990                         nb_prep);
 991
 992         /*
 993          * Retry if necessary
 994          */
 995         if (unlikely(nb_tx < nb_rx) && fs->retry_enabled) {
 996                 retry = 0;
 997                 while (nb_tx < nb_rx && retry++ < burst_tx_retry_num) {
 998                         rte_delay_us(burst_tx_delay_time);
 999                         nb_tx += rte_eth_tx_burst(fs->tx_port, fs->tx_queue,
1000                                         &tx_pkts_burst[nb_tx], nb_rx - nb_tx);
1001                 }
1002         }
1003         fs->tx_packets += nb_tx;
1004         fs->rx_bad_ip_csum += rx_bad_ip_csum;
1005         fs->rx_bad_l4_csum += rx_bad_l4_csum;
1006         fs->rx_bad_outer_l4_csum += rx_bad_outer_l4_csum;
1007
1008 #ifdef RTE_TEST_PMD_RECORD_BURST_STATS
1009         fs->tx_burst_stats.pkt_burst_spread[nb_tx]++;
1010 #endif
1011         if (unlikely(nb_tx < nb_rx)) {
1012                 fs->fwd_dropped += (nb_rx - nb_tx);
1013                 do {
1014                         rte_pktmbuf_free(tx_pkts_burst[nb_tx]);
1015                 } while (++nb_tx < nb_rx);
1016         }
1017
1018 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
1019         end_tsc = rte_rdtsc();
1020         core_cycles = (end_tsc - start_tsc);
1021         fs->core_cycles = (uint64_t) (fs->core_cycles + core_cycles);
1022 #endif
1023 }
1024
1025 struct fwd_engine csum_fwd_engine = {
1026         .fwd_mode_name  = "csum",
1027         .port_fwd_begin = NULL,
1028         .port_fwd_end   = NULL,
1029         .packet_fwd     = pkt_burst_checksum_forward,
1030 };