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 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 sctp_hdr *)((char *)l3_hdr + info->l3_len);
 439                 sctp_hdr->cksum = 0;
 440                 /* sctp payload must be a multiple of 4 to be
 441                  * offloaded */
 442                 if ((tx_offloads & DEV_TX_OFFLOAD_SCTP_CKSUM) &&
 443                         ((ipv4_hdr->total_length & 0x3) == 0)) {
 444                         ol_flags |= PKT_TX_SCTP_CKSUM;
 445                 } else {
 446                         /* XXX implement CRC32c, example available in
 447                          * RFC3309 */
 448                 }
 449         }
 450
 451         return ol_flags;
 452 }
 453
 454 /* Calculate the checksum of outer header */
 455 static uint64_t
 456 process_outer_cksums(void *outer_l3_hdr, struct testpmd_offload_info *info,
 457         uint64_t tx_offloads, int tso_enabled)
 458 {
 459         struct rte_ipv4_hdr *ipv4_hdr = outer_l3_hdr;
 460         struct rte_ipv6_hdr *ipv6_hdr = outer_l3_hdr;
 461         struct udp_hdr *udp_hdr;
 462         uint64_t ol_flags = 0;
 463
 464         if (info->outer_ethertype == _htons(RTE_ETHER_TYPE_IPv4)) {
 465                 ipv4_hdr->hdr_checksum = 0;
 466                 ol_flags |= PKT_TX_OUTER_IPV4;
 467
 468                 if (tx_offloads & DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM)
 469                         ol_flags |= PKT_TX_OUTER_IP_CKSUM;
 470                 else
 471                         ipv4_hdr->hdr_checksum = rte_ipv4_cksum(ipv4_hdr);
 472         } else
 473                 ol_flags |= PKT_TX_OUTER_IPV6;
 474
 475         if (info->outer_l4_proto != IPPROTO_UDP)
 476                 return ol_flags;
 477
 478         /* Skip SW outer UDP checksum generation if HW supports it */
 479         if (tx_offloads & DEV_TX_OFFLOAD_OUTER_UDP_CKSUM) {
 480                 ol_flags |= PKT_TX_OUTER_UDP_CKSUM;
 481                 return ol_flags;
 482         }
 483
 484         udp_hdr = (struct udp_hdr *)((char *)outer_l3_hdr + info->outer_l3_len);
 485
 486         /* outer UDP checksum is done in software. In the other side, for
 487          * UDP tunneling, like VXLAN or Geneve, outer UDP checksum can be
 488          * set to zero.
 489          *
 490          * If a packet will be TSOed into small packets by NIC, we cannot
 491          * set/calculate a non-zero checksum, because it will be a wrong
 492          * value after the packet be split into several small packets.
 493          */
 494         if (tso_enabled)
 495                 udp_hdr->dgram_cksum = 0;
 496
 497         /* do not recalculate udp cksum if it was 0 */
 498         if (udp_hdr->dgram_cksum != 0) {
 499                 udp_hdr->dgram_cksum = 0;
 500                 if (info->outer_ethertype == _htons(RTE_ETHER_TYPE_IPv4))
 501                         udp_hdr->dgram_cksum =
 502                                 rte_ipv4_udptcp_cksum(ipv4_hdr, udp_hdr);
 503                 else
 504                         udp_hdr->dgram_cksum =
 505                                 rte_ipv6_udptcp_cksum(ipv6_hdr, udp_hdr);
 506         }
 507
 508         return ol_flags;
 509 }
 510
 511 /*
 512  * Helper function.
 513  * Performs actual copying.
 514  * Returns number of segments in the destination mbuf on success,
 515  * or negative error code on failure.
 516  */
 517 static int
 518 mbuf_copy_split(const struct rte_mbuf *ms, struct rte_mbuf *md[],
 519         uint16_t seglen[], uint8_t nb_seg)
 520 {
 521         uint32_t dlen, slen, tlen;
 522         uint32_t i, len;
 523         const struct rte_mbuf *m;
 524         const uint8_t *src;
 525         uint8_t *dst;
 526
 527         dlen = 0;
 528         slen = 0;
 529         tlen = 0;
 530
 531         dst = NULL;
 532         src = NULL;
 533
 534         m = ms;
 535         i = 0;
 536         while (ms != NULL && i != nb_seg) {
 537
 538                 if (slen == 0) {
 539                         slen = rte_pktmbuf_data_len(ms);
 540                         src = rte_pktmbuf_mtod(ms, const uint8_t *);
 541                 }
 542
 543                 if (dlen == 0) {
 544                         dlen = RTE_MIN(seglen[i], slen);
 545                         md[i]->data_len = dlen;
 546                         md[i]->next = (i + 1 == nb_seg) ? NULL : md[i + 1];
 547                         dst = rte_pktmbuf_mtod(md[i], uint8_t *);
 548                 }
 549
 550                 len = RTE_MIN(slen, dlen);
 551                 memcpy(dst, src, len);
 552                 tlen += len;
 553                 slen -= len;
 554                 dlen -= len;
 555                 src += len;
 556                 dst += len;
 557
 558                 if (slen == 0)
 559                         ms = ms->next;
 560                 if (dlen == 0)
 561                         i++;
 562         }
 563
 564         if (ms != NULL)
 565                 return -ENOBUFS;
 566         else if (tlen != m->pkt_len)
 567                 return -EINVAL;
 568
 569         md[0]->nb_segs = nb_seg;
 570         md[0]->pkt_len = tlen;
 571         md[0]->vlan_tci = m->vlan_tci;
 572         md[0]->vlan_tci_outer = m->vlan_tci_outer;
 573         md[0]->ol_flags = m->ol_flags;
 574         md[0]->tx_offload = m->tx_offload;
 575
 576         return nb_seg;
 577 }
 578
 579 /*
 580  * Allocate a new mbuf with up to tx_pkt_nb_segs segments.
 581  * Copy packet contents and offload information into the new segmented mbuf.
 582  */
 583 static struct rte_mbuf *
 584 pkt_copy_split(const struct rte_mbuf *pkt)
 585 {
 586         int32_t n, rc;
 587         uint32_t i, len, nb_seg;
 588         struct rte_mempool *mp;
 589         uint16_t seglen[RTE_MAX_SEGS_PER_PKT];
 590         struct rte_mbuf *p, *md[RTE_MAX_SEGS_PER_PKT];
 591
 592         mp = current_fwd_lcore()->mbp;
 593
 594         if (tx_pkt_split == TX_PKT_SPLIT_RND)
 595                 nb_seg = random() % tx_pkt_nb_segs + 1;
 596         else
 597                 nb_seg = tx_pkt_nb_segs;
 598
 599         memcpy(seglen, tx_pkt_seg_lengths, nb_seg * sizeof(seglen[0]));
 600
 601         /* calculate number of segments to use and their length. */
 602         len = 0;
 603         for (i = 0; i != nb_seg && len < pkt->pkt_len; i++) {
 604                 len += seglen[i];
 605                 md[i] = NULL;
 606         }
 607
 608         n = pkt->pkt_len - len;
 609
 610         /* update size of the last segment to fit rest of the packet */
 611         if (n >= 0) {
 612                 seglen[i - 1] += n;
 613                 len += n;
 614         }
 615
 616         nb_seg = i;
 617         while (i != 0) {
 618                 p = rte_pktmbuf_alloc(mp);
 619                 if (p == NULL) {
 620                         TESTPMD_LOG(ERR,
 621                                 "failed to allocate %u-th of %u mbuf "
 622                                 "from mempool: %s\n",
 623                                 nb_seg - i, nb_seg, mp->name);
 624                         break;
 625                 }
 626
 627                 md[--i] = p;
 628                 if (rte_pktmbuf_tailroom(md[i]) < seglen[i]) {
 629                         TESTPMD_LOG(ERR, "mempool %s, %u-th segment: "
 630                                 "expected seglen: %u, "
 631                                 "actual mbuf tailroom: %u\n",
 632                                 mp->name, i, seglen[i],
 633                                 rte_pktmbuf_tailroom(md[i]));
 634                         break;
 635                 }
 636         }
 637
 638         /* all mbufs successfully allocated, do copy */
 639         if (i == 0) {
 640                 rc = mbuf_copy_split(pkt, md, seglen, nb_seg);
 641                 if (rc < 0)
 642                         TESTPMD_LOG(ERR,
 643                                 "mbuf_copy_split for %p(len=%u, nb_seg=%u) "
 644                                 "into %u segments failed with error code: %d\n",
 645                                 pkt, pkt->pkt_len, pkt->nb_segs, nb_seg, rc);
 646
 647                 /* figure out how many mbufs to free. */
 648                 i = RTE_MAX(rc, 0);
 649         }
 650
 651         /* free unused mbufs */
 652         for (; i != nb_seg; i++) {
 653                 rte_pktmbuf_free_seg(md[i]);
 654                 md[i] = NULL;
 655         }
 656
 657         return md[0];
 658 }
 659
 660 /*
 661  * Receive a burst of packets, and for each packet:
 662  *  - parse packet, and try to recognize a supported packet type (1)
 663  *  - if it's not a supported packet type, don't touch the packet, else:
 664  *  - reprocess the checksum of all supported layers. This is done in SW
 665  *    or HW, depending on testpmd command line configuration
 666  *  - if TSO is enabled in testpmd command line, also flag the mbuf for TCP
 667  *    segmentation offload (this implies HW TCP checksum)
 668  * Then transmit packets on the output port.
 669  *
 670  * (1) Supported packets are:
 671  *   Ether / (vlan) / IP|IP6 / UDP|TCP|SCTP .
 672  *   Ether / (vlan) / outer IP|IP6 / outer UDP / VxLAN / Ether / IP|IP6 /
 673  *           UDP|TCP|SCTP
 674  *   Ether / (vlan) / outer IP|IP6 / outer UDP / VXLAN-GPE / Ether / IP|IP6 /
 675  *           UDP|TCP|SCTP
 676  *   Ether / (vlan) / outer IP|IP6 / outer UDP / VXLAN-GPE / IP|IP6 /
 677  *           UDP|TCP|SCTP
 678  *   Ether / (vlan) / outer IP|IP6 / GRE / Ether / IP|IP6 / UDP|TCP|SCTP
 679  *   Ether / (vlan) / outer IP|IP6 / GRE / IP|IP6 / UDP|TCP|SCTP
 680  *   Ether / (vlan) / outer IP|IP6 / IP|IP6 / UDP|TCP|SCTP
 681  *
 682  * The testpmd command line for this forward engine sets the flags
 683  * TESTPMD_TX_OFFLOAD_* in ports[tx_port].tx_ol_flags. They control
 684  * wether a checksum must be calculated in software or in hardware. The
 685  * IP, UDP, TCP and SCTP flags always concern the inner layer. The
 686  * OUTER_IP is only useful for tunnel packets.
 687  */
 688 static void
 689 pkt_burst_checksum_forward(struct fwd_stream *fs)
 690 {
 691         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
 692         struct rte_mbuf *gso_segments[GSO_MAX_PKT_BURST];
 693         struct rte_gso_ctx *gso_ctx;
 694         struct rte_mbuf **tx_pkts_burst;
 695         struct rte_port *txp;
 696         struct rte_mbuf *m, *p;
 697         struct rte_ether_hdr *eth_hdr;
 698         void *l3_hdr = NULL, *outer_l3_hdr = NULL; /* can be IPv4 or IPv6 */
 699         void **gro_ctx;
 700         uint16_t gro_pkts_num;
 701         uint8_t gro_enable;
 702         uint16_t nb_rx;
 703         uint16_t nb_tx;
 704         uint16_t nb_prep;
 705         uint16_t i;
 706         uint64_t rx_ol_flags, tx_ol_flags;
 707         uint64_t tx_offloads;
 708         uint32_t retry;
 709         uint32_t rx_bad_ip_csum;
 710         uint32_t rx_bad_l4_csum;
 711         uint32_t rx_bad_outer_l4_csum;
 712         struct testpmd_offload_info info;
 713         uint16_t nb_segments = 0;
 714         int ret;
 715
 716 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
 717         uint64_t start_tsc;
 718         uint64_t end_tsc;
 719         uint64_t core_cycles;
 720 #endif
 721
 722 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
 723         start_tsc = rte_rdtsc();
 724 #endif
 725
 726         /* receive a burst of packet */
 727         nb_rx = rte_eth_rx_burst(fs->rx_port, fs->rx_queue, pkts_burst,
 728                                  nb_pkt_per_burst);
 729         if (unlikely(nb_rx == 0))
 730                 return;
 731 #ifdef RTE_TEST_PMD_RECORD_BURST_STATS
 732         fs->rx_burst_stats.pkt_burst_spread[nb_rx]++;
 733 #endif
 734         fs->rx_packets += nb_rx;
 735         rx_bad_ip_csum = 0;
 736         rx_bad_l4_csum = 0;
 737         rx_bad_outer_l4_csum = 0;
 738         gro_enable = gro_ports[fs->rx_port].enable;
 739
 740         txp = &ports[fs->tx_port];
 741         tx_offloads = txp->dev_conf.txmode.offloads;
 742         memset(&info, 0, sizeof(info));
 743         info.tso_segsz = txp->tso_segsz;
 744         info.tunnel_tso_segsz = txp->tunnel_tso_segsz;
 745         if (gso_ports[fs->tx_port].enable)
 746                 info.gso_enable = 1;
 747
 748         for (i = 0; i < nb_rx; i++) {
 749                 if (likely(i < nb_rx - 1))
 750                         rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[i + 1],
 751                                                        void *));
 752
 753                 m = pkts_burst[i];
 754                 info.is_tunnel = 0;
 755                 info.pkt_len = rte_pktmbuf_pkt_len(m);
 756                 tx_ol_flags = m->ol_flags &
 757                               (IND_ATTACHED_MBUF | EXT_ATTACHED_MBUF);
 758                 rx_ol_flags = m->ol_flags;
 759
 760                 /* Update the L3/L4 checksum error packet statistics */
 761                 if ((rx_ol_flags & PKT_RX_IP_CKSUM_MASK) == PKT_RX_IP_CKSUM_BAD)
 762                         rx_bad_ip_csum += 1;
 763                 if ((rx_ol_flags & PKT_RX_L4_CKSUM_MASK) == PKT_RX_L4_CKSUM_BAD)
 764                         rx_bad_l4_csum += 1;
 765                 if (rx_ol_flags & PKT_RX_OUTER_L4_CKSUM_BAD)
 766                         rx_bad_outer_l4_csum += 1;
 767
 768                 /* step 1: dissect packet, parsing optional vlan, ip4/ip6, vxlan
 769                  * and inner headers */
 770
 771                 eth_hdr = rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
 772                 rte_ether_addr_copy(&peer_eth_addrs[fs->peer_addr],
 773                                 &eth_hdr->d_addr);
 774                 rte_ether_addr_copy(&ports[fs->tx_port].eth_addr,
 775                                 &eth_hdr->s_addr);
 776                 parse_ethernet(eth_hdr, &info);
 777                 l3_hdr = (char *)eth_hdr + info.l2_len;
 778
 779                 /* check if it's a supported tunnel */
 780                 if (txp->parse_tunnel) {
 781                         if (info.l4_proto == IPPROTO_UDP) {
 782                                 struct udp_hdr *udp_hdr;
 783
 784                                 udp_hdr = (struct udp_hdr *)((char *)l3_hdr +
 785                                         info.l3_len);
 786                                 parse_vxlan_gpe(udp_hdr, &info);
 787                                 if (info.is_tunnel) {
 788                                         tx_ol_flags |= PKT_TX_TUNNEL_VXLAN_GPE;
 789                                 } else {
 790                                         parse_vxlan(udp_hdr, &info,
 791                                                     m->packet_type);
 792                                         if (info.is_tunnel)
 793                                                 tx_ol_flags |=
 794                                                         PKT_TX_TUNNEL_VXLAN;
 795                                 }
 796                         } else if (info.l4_proto == IPPROTO_GRE) {
 797                                 struct simple_gre_hdr *gre_hdr;
 798
 799                                 gre_hdr = (struct simple_gre_hdr *)
 800                                         ((char *)l3_hdr + info.l3_len);
 801                                 parse_gre(gre_hdr, &info);
 802                                 if (info.is_tunnel)
 803                                         tx_ol_flags |= PKT_TX_TUNNEL_GRE;
 804                         } else if (info.l4_proto == IPPROTO_IPIP) {
 805                                 void *encap_ip_hdr;
 806
 807                                 encap_ip_hdr = (char *)l3_hdr + info.l3_len;
 808                                 parse_encap_ip(encap_ip_hdr, &info);
 809                                 if (info.is_tunnel)
 810                                         tx_ol_flags |= PKT_TX_TUNNEL_IPIP;
 811                         }
 812                 }
 813
 814                 /* update l3_hdr and outer_l3_hdr if a tunnel was parsed */
 815                 if (info.is_tunnel) {
 816                         outer_l3_hdr = l3_hdr;
 817                         l3_hdr = (char *)l3_hdr + info.outer_l3_len + info.l2_len;
 818                 }
 819
 820                 /* step 2: depending on user command line configuration,
 821                  * recompute checksum either in software or flag the
 822                  * mbuf to offload the calculation to the NIC. If TSO
 823                  * is configured, prepare the mbuf for TCP segmentation. */
 824
 825                 /* process checksums of inner headers first */
 826                 tx_ol_flags |= process_inner_cksums(l3_hdr, &info,
 827                         tx_offloads);
 828
 829                 /* Then process outer headers if any. Note that the software
 830                  * checksum will be wrong if one of the inner checksums is
 831                  * processed in hardware. */
 832                 if (info.is_tunnel == 1) {
 833                         tx_ol_flags |= process_outer_cksums(outer_l3_hdr, &info,
 834                                         tx_offloads,
 835                                         !!(tx_ol_flags & PKT_TX_TCP_SEG));
 836                 }
 837
 838                 /* step 3: fill the mbuf meta data (flags and header lengths) */
 839
 840                 m->tx_offload = 0;
 841                 if (info.is_tunnel == 1) {
 842                         if (info.tunnel_tso_segsz ||
 843                             (tx_offloads &
 844                              DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM) ||
 845                             (tx_offloads &
 846                              DEV_TX_OFFLOAD_OUTER_UDP_CKSUM) ||
 847                             (tx_ol_flags & PKT_TX_OUTER_IPV6)) {
 848                                 m->outer_l2_len = info.outer_l2_len;
 849                                 m->outer_l3_len = info.outer_l3_len;
 850                                 m->l2_len = info.l2_len;
 851                                 m->l3_len = info.l3_len;
 852                                 m->l4_len = info.l4_len;
 853                                 m->tso_segsz = info.tunnel_tso_segsz;
 854                         }
 855                         else {
 856                                 /* if there is a outer UDP cksum
 857                                    processed in sw and the inner in hw,
 858                                    the outer checksum will be wrong as
 859                                    the payload will be modified by the
 860                                    hardware */
 861                                 m->l2_len = info.outer_l2_len +
 862                                         info.outer_l3_len + info.l2_len;
 863                                 m->l3_len = info.l3_len;
 864                                 m->l4_len = info.l4_len;
 865                         }
 866                 } else {
 867                         /* this is only useful if an offload flag is
 868                          * set, but it does not hurt to fill it in any
 869                          * case */
 870                         m->l2_len = info.l2_len;
 871                         m->l3_len = info.l3_len;
 872                         m->l4_len = info.l4_len;
 873                         m->tso_segsz = info.tso_segsz;
 874                 }
 875                 m->ol_flags = tx_ol_flags;
 876
 877                 /* Do split & copy for the packet. */
 878                 if (tx_pkt_split != TX_PKT_SPLIT_OFF) {
 879                         p = pkt_copy_split(m);
 880                         if (p != NULL) {
 881                                 rte_pktmbuf_free(m);
 882                                 m = p;
 883                                 pkts_burst[i] = m;
 884                         }
 885                 }
 886
 887                 /* if verbose mode is enabled, dump debug info */
 888                 if (verbose_level > 0) {
 889                         char buf[256];
 890
 891                         printf("-----------------\n");
 892                         printf("port=%u, mbuf=%p, pkt_len=%u, nb_segs=%u:\n",
 893                                 fs->rx_port, m, m->pkt_len, m->nb_segs);
 894                         /* dump rx parsed packet info */
 895                         rte_get_rx_ol_flag_list(rx_ol_flags, buf, sizeof(buf));
 896                         printf("rx: l2_len=%d ethertype=%x l3_len=%d "
 897                                 "l4_proto=%d l4_len=%d flags=%s\n",
 898                                 info.l2_len, rte_be_to_cpu_16(info.ethertype),
 899                                 info.l3_len, info.l4_proto, info.l4_len, buf);
 900                         if (rx_ol_flags & PKT_RX_LRO)
 901                                 printf("rx: m->lro_segsz=%u\n", m->tso_segsz);
 902                         if (info.is_tunnel == 1)
 903                                 printf("rx: outer_l2_len=%d outer_ethertype=%x "
 904                                         "outer_l3_len=%d\n", info.outer_l2_len,
 905                                         rte_be_to_cpu_16(info.outer_ethertype),
 906                                         info.outer_l3_len);
 907                         /* dump tx packet info */
 908                         if ((tx_offloads & (DEV_TX_OFFLOAD_IPV4_CKSUM |
 909                                             DEV_TX_OFFLOAD_UDP_CKSUM |
 910                                             DEV_TX_OFFLOAD_TCP_CKSUM |
 911                                             DEV_TX_OFFLOAD_SCTP_CKSUM)) ||
 912                                 info.tso_segsz != 0)
 913                                 printf("tx: m->l2_len=%d m->l3_len=%d "
 914                                         "m->l4_len=%d\n",
 915                                         m->l2_len, m->l3_len, m->l4_len);
 916                         if (info.is_tunnel == 1) {
 917                                 if ((tx_offloads &
 918                                     DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM) ||
 919                                     (tx_offloads &
 920                                     DEV_TX_OFFLOAD_OUTER_UDP_CKSUM) ||
 921                                     (tx_ol_flags & PKT_TX_OUTER_IPV6))
 922                                         printf("tx: m->outer_l2_len=%d "
 923                                                 "m->outer_l3_len=%d\n",
 924                                                 m->outer_l2_len,
 925                                                 m->outer_l3_len);
 926                                 if (info.tunnel_tso_segsz != 0 &&
 927                                                 (m->ol_flags & PKT_TX_TCP_SEG))
 928                                         printf("tx: m->tso_segsz=%d\n",
 929                                                 m->tso_segsz);
 930                         } else if (info.tso_segsz != 0 &&
 931                                         (m->ol_flags & PKT_TX_TCP_SEG))
 932                                 printf("tx: m->tso_segsz=%d\n", m->tso_segsz);
 933                         rte_get_tx_ol_flag_list(m->ol_flags, buf, sizeof(buf));
 934                         printf("tx: flags=%s", buf);
 935                         printf("\n");
 936                 }
 937         }
 938
 939         if (unlikely(gro_enable)) {
 940                 if (gro_flush_cycles == GRO_DEFAULT_FLUSH_CYCLES) {
 941                         nb_rx = rte_gro_reassemble_burst(pkts_burst, nb_rx,
 942                                         &(gro_ports[fs->rx_port].param));
 943                 } else {
 944                         gro_ctx = current_fwd_lcore()->gro_ctx;
 945                         nb_rx = rte_gro_reassemble(pkts_burst, nb_rx, gro_ctx);
 946
 947                         if (++fs->gro_times >= gro_flush_cycles) {
 948                                 gro_pkts_num = rte_gro_get_pkt_count(gro_ctx);
 949                                 if (gro_pkts_num > MAX_PKT_BURST - nb_rx)
 950                                         gro_pkts_num = MAX_PKT_BURST - nb_rx;
 951
 952                                 nb_rx += rte_gro_timeout_flush(gro_ctx, 0,
 953                                                 RTE_GRO_TCP_IPV4,
 954                                                 &pkts_burst[nb_rx],
 955                                                 gro_pkts_num);
 956                                 fs->gro_times = 0;
 957                         }
 958                 }
 959         }
 960
 961         if (gso_ports[fs->tx_port].enable == 0)
 962                 tx_pkts_burst = pkts_burst;
 963         else {
 964                 gso_ctx = &(current_fwd_lcore()->gso_ctx);
 965                 gso_ctx->gso_size = gso_max_segment_size;
 966                 for (i = 0; i < nb_rx; i++) {
 967                         ret = rte_gso_segment(pkts_burst[i], gso_ctx,
 968                                         &gso_segments[nb_segments],
 969                                         GSO_MAX_PKT_BURST - nb_segments);
 970                         if (ret >= 0)
 971                                 nb_segments += ret;
 972                         else {
 973                                 TESTPMD_LOG(DEBUG, "Unable to segment packet");
 974                                 rte_pktmbuf_free(pkts_burst[i]);
 975                         }
 976                 }
 977
 978                 tx_pkts_burst = gso_segments;
 979                 nb_rx = nb_segments;
 980         }
 981
 982         nb_prep = rte_eth_tx_prepare(fs->tx_port, fs->tx_queue,
 983                         tx_pkts_burst, nb_rx);
 984         if (nb_prep != nb_rx)
 985                 printf("Preparing packet burst to transmit failed: %s\n",
 986                                 rte_strerror(rte_errno));
 987
 988         nb_tx = rte_eth_tx_burst(fs->tx_port, fs->tx_queue, tx_pkts_burst,
 989                         nb_prep);
 990
 991         /*
 992          * Retry if necessary
 993          */
 994         if (unlikely(nb_tx < nb_rx) && fs->retry_enabled) {
 995                 retry = 0;
 996                 while (nb_tx < nb_rx && retry++ < burst_tx_retry_num) {
 997                         rte_delay_us(burst_tx_delay_time);
 998                         nb_tx += rte_eth_tx_burst(fs->tx_port, fs->tx_queue,
 999                                         &tx_pkts_burst[nb_tx], nb_rx - nb_tx);
1000                 }
1001         }
1002         fs->tx_packets += nb_tx;
1003         fs->rx_bad_ip_csum += rx_bad_ip_csum;
1004         fs->rx_bad_l4_csum += rx_bad_l4_csum;
1005         fs->rx_bad_outer_l4_csum += rx_bad_outer_l4_csum;
1006
1007 #ifdef RTE_TEST_PMD_RECORD_BURST_STATS
1008         fs->tx_burst_stats.pkt_burst_spread[nb_tx]++;
1009 #endif
1010         if (unlikely(nb_tx < nb_rx)) {
1011                 fs->fwd_dropped += (nb_rx - nb_tx);
1012                 do {
1013                         rte_pktmbuf_free(tx_pkts_burst[nb_tx]);
1014                 } while (++nb_tx < nb_rx);
1015         }
1016
1017 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
1018         end_tsc = rte_rdtsc();
1019         core_cycles = (end_tsc - start_tsc);
1020         fs->core_cycles = (uint64_t) (fs->core_cycles + core_cycles);
1021 #endif
1022 }
1023
1024 struct fwd_engine csum_fwd_engine = {
1025         .fwd_mode_name  = "csum",
1026         .port_fwd_begin = NULL,
1027         .port_fwd_end   = NULL,
1028         .packet_fwd     = pkt_burst_checksum_forward,
1029 };