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 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 ipv6_hdr *ipv6_hdr, struct testpmd_offload_info *info)
 123 {
 124         struct tcp_hdr *tcp_hdr;
 125
 126         info->l3_len = sizeof(struct 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 ipv4_hdr *ipv4_hdr;
 148         struct 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 ipv4_hdr *) ((char *)eth_hdr + info->l2_len);
 164                 parse_ipv4(ipv4_hdr, info);
 165                 break;
 166         case _htons(RTE_ETHER_TYPE_IPv6):
 167                 ipv6_hdr = (struct ipv6_hdr *) ((char *)eth_hdr + info->l2_len);
 168                 parse_ipv6(ipv6_hdr, info);
 169                 break;
 170         default:
 171                 info->l4_len = 0;
 172                 info->l3_len = 0;
 173                 info->l4_proto = 0;
 174                 break;
 175         }
 176 }
 177
 178 /* Parse a vxlan header */
 179 static void
 180 parse_vxlan(struct udp_hdr *udp_hdr,
 181             struct testpmd_offload_info *info,
 182             uint32_t pkt_type)
 183 {
 184         struct rte_ether_hdr *eth_hdr;
 185
 186         /* check udp destination port, 4789 is the default vxlan port
 187          * (rfc7348) or that the rx offload flag is set (i40e only
 188          * currently) */
 189         if (udp_hdr->dst_port != _htons(4789) &&
 190                 RTE_ETH_IS_TUNNEL_PKT(pkt_type) == 0)
 191                 return;
 192
 193         info->is_tunnel = 1;
 194         info->outer_ethertype = info->ethertype;
 195         info->outer_l2_len = info->l2_len;
 196         info->outer_l3_len = info->l3_len;
 197         info->outer_l4_proto = info->l4_proto;
 198
 199         eth_hdr = (struct rte_ether_hdr *)((char *)udp_hdr +
 200                 sizeof(struct udp_hdr) +
 201                 sizeof(struct rte_vxlan_hdr));
 202
 203         parse_ethernet(eth_hdr, info);
 204         info->l2_len += RTE_ETHER_VXLAN_HLEN; /* add udp + vxlan */
 205 }
 206
 207 /* Parse a vxlan-gpe header */
 208 static void
 209 parse_vxlan_gpe(struct udp_hdr *udp_hdr,
 210             struct testpmd_offload_info *info)
 211 {
 212         struct rte_ether_hdr *eth_hdr;
 213         struct ipv4_hdr *ipv4_hdr;
 214         struct ipv6_hdr *ipv6_hdr;
 215         struct rte_vxlan_gpe_hdr *vxlan_gpe_hdr;
 216         uint8_t vxlan_gpe_len = sizeof(*vxlan_gpe_hdr);
 217
 218         /* Check udp destination port. */
 219         if (udp_hdr->dst_port != _htons(vxlan_gpe_udp_port))
 220                 return;
 221
 222         vxlan_gpe_hdr = (struct rte_vxlan_gpe_hdr *)((char *)udp_hdr +
 223                                 sizeof(struct udp_hdr));
 224
 225         if (!vxlan_gpe_hdr->proto || vxlan_gpe_hdr->proto ==
 226             RTE_VXLAN_GPE_TYPE_IPV4) {
 227                 info->is_tunnel = 1;
 228                 info->outer_ethertype = info->ethertype;
 229                 info->outer_l2_len = info->l2_len;
 230                 info->outer_l3_len = info->l3_len;
 231                 info->outer_l4_proto = info->l4_proto;
 232
 233                 ipv4_hdr = (struct ipv4_hdr *)((char *)vxlan_gpe_hdr +
 234                            vxlan_gpe_len);
 235
 236                 parse_ipv4(ipv4_hdr, info);
 237                 info->ethertype = _htons(RTE_ETHER_TYPE_IPv4);
 238                 info->l2_len = 0;
 239
 240         } else if (vxlan_gpe_hdr->proto == RTE_VXLAN_GPE_TYPE_IPV6) {
 241                 info->is_tunnel = 1;
 242                 info->outer_ethertype = info->ethertype;
 243                 info->outer_l2_len = info->l2_len;
 244                 info->outer_l3_len = info->l3_len;
 245                 info->outer_l4_proto = info->l4_proto;
 246
 247                 ipv6_hdr = (struct ipv6_hdr *)((char *)vxlan_gpe_hdr +
 248                            vxlan_gpe_len);
 249
 250                 info->ethertype = _htons(RTE_ETHER_TYPE_IPv6);
 251                 parse_ipv6(ipv6_hdr, info);
 252                 info->l2_len = 0;
 253
 254         } else if (vxlan_gpe_hdr->proto == RTE_VXLAN_GPE_TYPE_ETH) {
 255                 info->is_tunnel = 1;
 256                 info->outer_ethertype = info->ethertype;
 257                 info->outer_l2_len = info->l2_len;
 258                 info->outer_l3_len = info->l3_len;
 259                 info->outer_l4_proto = info->l4_proto;
 260
 261                 eth_hdr = (struct rte_ether_hdr *)((char *)vxlan_gpe_hdr +
 262                           vxlan_gpe_len);
 263
 264                 parse_ethernet(eth_hdr, info);
 265         } else
 266                 return;
 267
 268         info->l2_len += RTE_ETHER_VXLAN_GPE_HLEN;
 269 }
 270
 271 /* Parse a gre header */
 272 static void
 273 parse_gre(struct simple_gre_hdr *gre_hdr, struct testpmd_offload_info *info)
 274 {
 275         struct rte_ether_hdr *eth_hdr;
 276         struct ipv4_hdr *ipv4_hdr;
 277         struct ipv6_hdr *ipv6_hdr;
 278         uint8_t gre_len = 0;
 279
 280         gre_len += sizeof(struct simple_gre_hdr);
 281
 282         if (gre_hdr->flags & _htons(GRE_KEY_PRESENT))
 283                 gre_len += GRE_EXT_LEN;
 284         if (gre_hdr->flags & _htons(GRE_SEQUENCE_PRESENT))
 285                 gre_len += GRE_EXT_LEN;
 286         if (gre_hdr->flags & _htons(GRE_CHECKSUM_PRESENT))
 287                 gre_len += GRE_EXT_LEN;
 288
 289         if (gre_hdr->proto == _htons(RTE_ETHER_TYPE_IPv4)) {
 290                 info->is_tunnel = 1;
 291                 info->outer_ethertype = info->ethertype;
 292                 info->outer_l2_len = info->l2_len;
 293                 info->outer_l3_len = info->l3_len;
 294                 info->outer_l4_proto = info->l4_proto;
 295
 296                 ipv4_hdr = (struct ipv4_hdr *)((char *)gre_hdr + gre_len);
 297
 298                 parse_ipv4(ipv4_hdr, info);
 299                 info->ethertype = _htons(RTE_ETHER_TYPE_IPv4);
 300                 info->l2_len = 0;
 301
 302         } else if (gre_hdr->proto == _htons(RTE_ETHER_TYPE_IPv6)) {
 303                 info->is_tunnel = 1;
 304                 info->outer_ethertype = info->ethertype;
 305                 info->outer_l2_len = info->l2_len;
 306                 info->outer_l3_len = info->l3_len;
 307                 info->outer_l4_proto = info->l4_proto;
 308
 309                 ipv6_hdr = (struct ipv6_hdr *)((char *)gre_hdr + gre_len);
 310
 311                 info->ethertype = _htons(RTE_ETHER_TYPE_IPv6);
 312                 parse_ipv6(ipv6_hdr, info);
 313                 info->l2_len = 0;
 314
 315         } else if (gre_hdr->proto == _htons(RTE_ETHER_TYPE_TEB)) {
 316                 info->is_tunnel = 1;
 317                 info->outer_ethertype = info->ethertype;
 318                 info->outer_l2_len = info->l2_len;
 319                 info->outer_l3_len = info->l3_len;
 320                 info->outer_l4_proto = info->l4_proto;
 321
 322                 eth_hdr = (struct rte_ether_hdr *)((char *)gre_hdr + gre_len);
 323
 324                 parse_ethernet(eth_hdr, info);
 325         } else
 326                 return;
 327
 328         info->l2_len += gre_len;
 329 }
 330
 331
 332 /* Parse an encapsulated ip or ipv6 header */
 333 static void
 334 parse_encap_ip(void *encap_ip, struct testpmd_offload_info *info)
 335 {
 336         struct ipv4_hdr *ipv4_hdr = encap_ip;
 337         struct ipv6_hdr *ipv6_hdr = encap_ip;
 338         uint8_t ip_version;
 339
 340         ip_version = (ipv4_hdr->version_ihl & 0xf0) >> 4;
 341
 342         if (ip_version != 4 && ip_version != 6)
 343                 return;
 344
 345         info->is_tunnel = 1;
 346         info->outer_ethertype = info->ethertype;
 347         info->outer_l2_len = info->l2_len;
 348         info->outer_l3_len = info->l3_len;
 349
 350         if (ip_version == 4) {
 351                 parse_ipv4(ipv4_hdr, info);
 352                 info->ethertype = _htons(RTE_ETHER_TYPE_IPv4);
 353         } else {
 354                 parse_ipv6(ipv6_hdr, info);
 355                 info->ethertype = _htons(RTE_ETHER_TYPE_IPv6);
 356         }
 357         info->l2_len = 0;
 358 }
 359
 360 /* if possible, calculate the checksum of a packet in hw or sw,
 361  * depending on the testpmd command line configuration */
 362 static uint64_t
 363 process_inner_cksums(void *l3_hdr, const struct testpmd_offload_info *info,
 364         uint64_t tx_offloads)
 365 {
 366         struct ipv4_hdr *ipv4_hdr = l3_hdr;
 367         struct udp_hdr *udp_hdr;
 368         struct tcp_hdr *tcp_hdr;
 369         struct sctp_hdr *sctp_hdr;
 370         uint64_t ol_flags = 0;
 371         uint32_t max_pkt_len, tso_segsz = 0;
 372
 373         /* ensure packet is large enough to require tso */
 374         if (!info->is_tunnel) {
 375                 max_pkt_len = info->l2_len + info->l3_len + info->l4_len +
 376                         info->tso_segsz;
 377                 if (info->tso_segsz != 0 && info->pkt_len > max_pkt_len)
 378                         tso_segsz = info->tso_segsz;
 379         } else {
 380                 max_pkt_len = info->outer_l2_len + info->outer_l3_len +
 381                         info->l2_len + info->l3_len + info->l4_len +
 382                         info->tunnel_tso_segsz;
 383                 if (info->tunnel_tso_segsz != 0 && info->pkt_len > max_pkt_len)
 384                         tso_segsz = info->tunnel_tso_segsz;
 385         }
 386
 387         if (info->ethertype == _htons(RTE_ETHER_TYPE_IPv4)) {
 388                 ipv4_hdr = l3_hdr;
 389                 ipv4_hdr->hdr_checksum = 0;
 390
 391                 ol_flags |= PKT_TX_IPV4;
 392                 if (info->l4_proto == IPPROTO_TCP && tso_segsz) {
 393                         ol_flags |= PKT_TX_IP_CKSUM;
 394                 } else {
 395                         if (tx_offloads & DEV_TX_OFFLOAD_IPV4_CKSUM)
 396                                 ol_flags |= PKT_TX_IP_CKSUM;
 397                         else
 398                                 ipv4_hdr->hdr_checksum =
 399                                         rte_ipv4_cksum(ipv4_hdr);
 400                 }
 401         } else if (info->ethertype == _htons(RTE_ETHER_TYPE_IPv6))
 402                 ol_flags |= PKT_TX_IPV6;
 403         else
 404                 return 0; /* packet type not supported, nothing to do */
 405
 406         if (info->l4_proto == IPPROTO_UDP) {
 407                 udp_hdr = (struct udp_hdr *)((char *)l3_hdr + info->l3_len);
 408                 /* do not recalculate udp cksum if it was 0 */
 409                 if (udp_hdr->dgram_cksum != 0) {
 410                         udp_hdr->dgram_cksum = 0;
 411                         if (tx_offloads & DEV_TX_OFFLOAD_UDP_CKSUM)
 412                                 ol_flags |= PKT_TX_UDP_CKSUM;
 413                         else {
 414                                 udp_hdr->dgram_cksum =
 415                                         get_udptcp_checksum(l3_hdr, udp_hdr,
 416                                                 info->ethertype);
 417                         }
 418                 }
 419                 if (info->gso_enable)
 420                         ol_flags |= PKT_TX_UDP_SEG;
 421         } else if (info->l4_proto == IPPROTO_TCP) {
 422                 tcp_hdr = (struct tcp_hdr *)((char *)l3_hdr + info->l3_len);
 423                 tcp_hdr->cksum = 0;
 424                 if (tso_segsz)
 425                         ol_flags |= PKT_TX_TCP_SEG;
 426                 else if (tx_offloads & DEV_TX_OFFLOAD_TCP_CKSUM)
 427                         ol_flags |= PKT_TX_TCP_CKSUM;
 428                 else {
 429                         tcp_hdr->cksum =
 430                                 get_udptcp_checksum(l3_hdr, tcp_hdr,
 431                                         info->ethertype);
 432                 }
 433                 if (info->gso_enable)
 434                         ol_flags |= PKT_TX_TCP_SEG;
 435         } else if (info->l4_proto == IPPROTO_SCTP) {
 436                 sctp_hdr = (struct sctp_hdr *)((char *)l3_hdr + info->l3_len);
 437                 sctp_hdr->cksum = 0;
 438                 /* sctp payload must be a multiple of 4 to be
 439                  * offloaded */
 440                 if ((tx_offloads & DEV_TX_OFFLOAD_SCTP_CKSUM) &&
 441                         ((ipv4_hdr->total_length & 0x3) == 0)) {
 442                         ol_flags |= PKT_TX_SCTP_CKSUM;
 443                 } else {
 444                         /* XXX implement CRC32c, example available in
 445                          * RFC3309 */
 446                 }
 447         }
 448
 449         return ol_flags;
 450 }
 451
 452 /* Calculate the checksum of outer header */
 453 static uint64_t
 454 process_outer_cksums(void *outer_l3_hdr, struct testpmd_offload_info *info,
 455         uint64_t tx_offloads, int tso_enabled)
 456 {
 457         struct ipv4_hdr *ipv4_hdr = outer_l3_hdr;
 458         struct ipv6_hdr *ipv6_hdr = outer_l3_hdr;
 459         struct udp_hdr *udp_hdr;
 460         uint64_t ol_flags = 0;
 461
 462         if (info->outer_ethertype == _htons(RTE_ETHER_TYPE_IPv4)) {
 463                 ipv4_hdr->hdr_checksum = 0;
 464                 ol_flags |= PKT_TX_OUTER_IPV4;
 465
 466                 if (tx_offloads & DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM)
 467                         ol_flags |= PKT_TX_OUTER_IP_CKSUM;
 468                 else
 469                         ipv4_hdr->hdr_checksum = rte_ipv4_cksum(ipv4_hdr);
 470         } else
 471                 ol_flags |= PKT_TX_OUTER_IPV6;
 472
 473         if (info->outer_l4_proto != IPPROTO_UDP)
 474                 return ol_flags;
 475
 476         /* Skip SW outer UDP checksum generation if HW supports it */
 477         if (tx_offloads & DEV_TX_OFFLOAD_OUTER_UDP_CKSUM) {
 478                 ol_flags |= PKT_TX_OUTER_UDP_CKSUM;
 479                 return ol_flags;
 480         }
 481
 482         udp_hdr = (struct udp_hdr *)((char *)outer_l3_hdr + info->outer_l3_len);
 483
 484         /* outer UDP checksum is done in software. In the other side, for
 485          * UDP tunneling, like VXLAN or Geneve, outer UDP checksum can be
 486          * set to zero.
 487          *
 488          * If a packet will be TSOed into small packets by NIC, we cannot
 489          * set/calculate a non-zero checksum, because it will be a wrong
 490          * value after the packet be split into several small packets.
 491          */
 492         if (tso_enabled)
 493                 udp_hdr->dgram_cksum = 0;
 494
 495         /* do not recalculate udp cksum if it was 0 */
 496         if (udp_hdr->dgram_cksum != 0) {
 497                 udp_hdr->dgram_cksum = 0;
 498                 if (info->outer_ethertype == _htons(RTE_ETHER_TYPE_IPv4))
 499                         udp_hdr->dgram_cksum =
 500                                 rte_ipv4_udptcp_cksum(ipv4_hdr, udp_hdr);
 501                 else
 502                         udp_hdr->dgram_cksum =
 503                                 rte_ipv6_udptcp_cksum(ipv6_hdr, udp_hdr);
 504         }
 505
 506         return ol_flags;
 507 }
 508
 509 /*
 510  * Helper function.
 511  * Performs actual copying.
 512  * Returns number of segments in the destination mbuf on success,
 513  * or negative error code on failure.
 514  */
 515 static int
 516 mbuf_copy_split(const struct rte_mbuf *ms, struct rte_mbuf *md[],
 517         uint16_t seglen[], uint8_t nb_seg)
 518 {
 519         uint32_t dlen, slen, tlen;
 520         uint32_t i, len;
 521         const struct rte_mbuf *m;
 522         const uint8_t *src;
 523         uint8_t *dst;
 524
 525         dlen = 0;
 526         slen = 0;
 527         tlen = 0;
 528
 529         dst = NULL;
 530         src = NULL;
 531
 532         m = ms;
 533         i = 0;
 534         while (ms != NULL && i != nb_seg) {
 535
 536                 if (slen == 0) {
 537                         slen = rte_pktmbuf_data_len(ms);
 538                         src = rte_pktmbuf_mtod(ms, const uint8_t *);
 539                 }
 540
 541                 if (dlen == 0) {
 542                         dlen = RTE_MIN(seglen[i], slen);
 543                         md[i]->data_len = dlen;
 544                         md[i]->next = (i + 1 == nb_seg) ? NULL : md[i + 1];
 545                         dst = rte_pktmbuf_mtod(md[i], uint8_t *);
 546                 }
 547
 548                 len = RTE_MIN(slen, dlen);
 549                 memcpy(dst, src, len);
 550                 tlen += len;
 551                 slen -= len;
 552                 dlen -= len;
 553                 src += len;
 554                 dst += len;
 555
 556                 if (slen == 0)
 557                         ms = ms->next;
 558                 if (dlen == 0)
 559                         i++;
 560         }
 561
 562         if (ms != NULL)
 563                 return -ENOBUFS;
 564         else if (tlen != m->pkt_len)
 565                 return -EINVAL;
 566
 567         md[0]->nb_segs = nb_seg;
 568         md[0]->pkt_len = tlen;
 569         md[0]->vlan_tci = m->vlan_tci;
 570         md[0]->vlan_tci_outer = m->vlan_tci_outer;
 571         md[0]->ol_flags = m->ol_flags;
 572         md[0]->tx_offload = m->tx_offload;
 573
 574         return nb_seg;
 575 }
 576
 577 /*
 578  * Allocate a new mbuf with up to tx_pkt_nb_segs segments.
 579  * Copy packet contents and offload information into the new segmented mbuf.
 580  */
 581 static struct rte_mbuf *
 582 pkt_copy_split(const struct rte_mbuf *pkt)
 583 {
 584         int32_t n, rc;
 585         uint32_t i, len, nb_seg;
 586         struct rte_mempool *mp;
 587         uint16_t seglen[RTE_MAX_SEGS_PER_PKT];
 588         struct rte_mbuf *p, *md[RTE_MAX_SEGS_PER_PKT];
 589
 590         mp = current_fwd_lcore()->mbp;
 591
 592         if (tx_pkt_split == TX_PKT_SPLIT_RND)
 593                 nb_seg = random() % tx_pkt_nb_segs + 1;
 594         else
 595                 nb_seg = tx_pkt_nb_segs;
 596
 597         memcpy(seglen, tx_pkt_seg_lengths, nb_seg * sizeof(seglen[0]));
 598
 599         /* calculate number of segments to use and their length. */
 600         len = 0;
 601         for (i = 0; i != nb_seg && len < pkt->pkt_len; i++) {
 602                 len += seglen[i];
 603                 md[i] = NULL;
 604         }
 605
 606         n = pkt->pkt_len - len;
 607
 608         /* update size of the last segment to fit rest of the packet */
 609         if (n >= 0) {
 610                 seglen[i - 1] += n;
 611                 len += n;
 612         }
 613
 614         nb_seg = i;
 615         while (i != 0) {
 616                 p = rte_pktmbuf_alloc(mp);
 617                 if (p == NULL) {
 618                         TESTPMD_LOG(ERR,
 619                                 "failed to allocate %u-th of %u mbuf "
 620                                 "from mempool: %s\n",
 621                                 nb_seg - i, nb_seg, mp->name);
 622                         break;
 623                 }
 624
 625                 md[--i] = p;
 626                 if (rte_pktmbuf_tailroom(md[i]) < seglen[i]) {
 627                         TESTPMD_LOG(ERR, "mempool %s, %u-th segment: "
 628                                 "expected seglen: %u, "
 629                                 "actual mbuf tailroom: %u\n",
 630                                 mp->name, i, seglen[i],
 631                                 rte_pktmbuf_tailroom(md[i]));
 632                         break;
 633                 }
 634         }
 635
 636         /* all mbufs successfully allocated, do copy */
 637         if (i == 0) {
 638                 rc = mbuf_copy_split(pkt, md, seglen, nb_seg);
 639                 if (rc < 0)
 640                         TESTPMD_LOG(ERR,
 641                                 "mbuf_copy_split for %p(len=%u, nb_seg=%u) "
 642                                 "into %u segments failed with error code: %d\n",
 643                                 pkt, pkt->pkt_len, pkt->nb_segs, nb_seg, rc);
 644
 645                 /* figure out how many mbufs to free. */
 646                 i = RTE_MAX(rc, 0);
 647         }
 648
 649         /* free unused mbufs */
 650         for (; i != nb_seg; i++) {
 651                 rte_pktmbuf_free_seg(md[i]);
 652                 md[i] = NULL;
 653         }
 654
 655         return md[0];
 656 }
 657
 658 /*
 659  * Receive a burst of packets, and for each packet:
 660  *  - parse packet, and try to recognize a supported packet type (1)
 661  *  - if it's not a supported packet type, don't touch the packet, else:
 662  *  - reprocess the checksum of all supported layers. This is done in SW
 663  *    or HW, depending on testpmd command line configuration
 664  *  - if TSO is enabled in testpmd command line, also flag the mbuf for TCP
 665  *    segmentation offload (this implies HW TCP checksum)
 666  * Then transmit packets on the output port.
 667  *
 668  * (1) Supported packets are:
 669  *   Ether / (vlan) / IP|IP6 / UDP|TCP|SCTP .
 670  *   Ether / (vlan) / outer IP|IP6 / outer UDP / VxLAN / Ether / IP|IP6 /
 671  *           UDP|TCP|SCTP
 672  *   Ether / (vlan) / outer IP|IP6 / outer UDP / VXLAN-GPE / Ether / IP|IP6 /
 673  *           UDP|TCP|SCTP
 674  *   Ether / (vlan) / outer IP|IP6 / outer UDP / VXLAN-GPE / IP|IP6 /
 675  *           UDP|TCP|SCTP
 676  *   Ether / (vlan) / outer IP|IP6 / GRE / Ether / IP|IP6 / UDP|TCP|SCTP
 677  *   Ether / (vlan) / outer IP|IP6 / GRE / IP|IP6 / UDP|TCP|SCTP
 678  *   Ether / (vlan) / outer IP|IP6 / IP|IP6 / UDP|TCP|SCTP
 679  *
 680  * The testpmd command line for this forward engine sets the flags
 681  * TESTPMD_TX_OFFLOAD_* in ports[tx_port].tx_ol_flags. They control
 682  * wether a checksum must be calculated in software or in hardware. The
 683  * IP, UDP, TCP and SCTP flags always concern the inner layer. The
 684  * OUTER_IP is only useful for tunnel packets.
 685  */
 686 static void
 687 pkt_burst_checksum_forward(struct fwd_stream *fs)
 688 {
 689         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
 690         struct rte_mbuf *gso_segments[GSO_MAX_PKT_BURST];
 691         struct rte_gso_ctx *gso_ctx;
 692         struct rte_mbuf **tx_pkts_burst;
 693         struct rte_port *txp;
 694         struct rte_mbuf *m, *p;
 695         struct rte_ether_hdr *eth_hdr;
 696         void *l3_hdr = NULL, *outer_l3_hdr = NULL; /* can be IPv4 or IPv6 */
 697         void **gro_ctx;
 698         uint16_t gro_pkts_num;
 699         uint8_t gro_enable;
 700         uint16_t nb_rx;
 701         uint16_t nb_tx;
 702         uint16_t nb_prep;
 703         uint16_t i;
 704         uint64_t rx_ol_flags, tx_ol_flags;
 705         uint64_t tx_offloads;
 706         uint32_t retry;
 707         uint32_t rx_bad_ip_csum;
 708         uint32_t rx_bad_l4_csum;
 709         uint32_t rx_bad_outer_l4_csum;
 710         struct testpmd_offload_info info;
 711         uint16_t nb_segments = 0;
 712         int ret;
 713
 714 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
 715         uint64_t start_tsc;
 716         uint64_t end_tsc;
 717         uint64_t core_cycles;
 718 #endif
 719
 720 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
 721         start_tsc = rte_rdtsc();
 722 #endif
 723
 724         /* receive a burst of packet */
 725         nb_rx = rte_eth_rx_burst(fs->rx_port, fs->rx_queue, pkts_burst,
 726                                  nb_pkt_per_burst);
 727         if (unlikely(nb_rx == 0))
 728                 return;
 729 #ifdef RTE_TEST_PMD_RECORD_BURST_STATS
 730         fs->rx_burst_stats.pkt_burst_spread[nb_rx]++;
 731 #endif
 732         fs->rx_packets += nb_rx;
 733         rx_bad_ip_csum = 0;
 734         rx_bad_l4_csum = 0;
 735         rx_bad_outer_l4_csum = 0;
 736         gro_enable = gro_ports[fs->rx_port].enable;
 737
 738         txp = &ports[fs->tx_port];
 739         tx_offloads = txp->dev_conf.txmode.offloads;
 740         memset(&info, 0, sizeof(info));
 741         info.tso_segsz = txp->tso_segsz;
 742         info.tunnel_tso_segsz = txp->tunnel_tso_segsz;
 743         if (gso_ports[fs->tx_port].enable)
 744                 info.gso_enable = 1;
 745
 746         for (i = 0; i < nb_rx; i++) {
 747                 if (likely(i < nb_rx - 1))
 748                         rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[i + 1],
 749                                                        void *));
 750
 751                 m = pkts_burst[i];
 752                 info.is_tunnel = 0;
 753                 info.pkt_len = rte_pktmbuf_pkt_len(m);
 754                 tx_ol_flags = m->ol_flags &
 755                               (IND_ATTACHED_MBUF | EXT_ATTACHED_MBUF);
 756                 rx_ol_flags = m->ol_flags;
 757
 758                 /* Update the L3/L4 checksum error packet statistics */
 759                 if ((rx_ol_flags & PKT_RX_IP_CKSUM_MASK) == PKT_RX_IP_CKSUM_BAD)
 760                         rx_bad_ip_csum += 1;
 761                 if ((rx_ol_flags & PKT_RX_L4_CKSUM_MASK) == PKT_RX_L4_CKSUM_BAD)
 762                         rx_bad_l4_csum += 1;
 763                 if (rx_ol_flags & PKT_RX_OUTER_L4_CKSUM_BAD)
 764                         rx_bad_outer_l4_csum += 1;
 765
 766                 /* step 1: dissect packet, parsing optional vlan, ip4/ip6, vxlan
 767                  * and inner headers */
 768
 769                 eth_hdr = rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
 770                 rte_ether_addr_copy(&peer_eth_addrs[fs->peer_addr],
 771                                 &eth_hdr->d_addr);
 772                 rte_ether_addr_copy(&ports[fs->tx_port].eth_addr,
 773                                 &eth_hdr->s_addr);
 774                 parse_ethernet(eth_hdr, &info);
 775                 l3_hdr = (char *)eth_hdr + info.l2_len;
 776
 777                 /* check if it's a supported tunnel */
 778                 if (txp->parse_tunnel) {
 779                         if (info.l4_proto == IPPROTO_UDP) {
 780                                 struct udp_hdr *udp_hdr;
 781
 782                                 udp_hdr = (struct udp_hdr *)((char *)l3_hdr +
 783                                         info.l3_len);
 784                                 parse_vxlan_gpe(udp_hdr, &info);
 785                                 if (info.is_tunnel) {
 786                                         tx_ol_flags |= PKT_TX_TUNNEL_VXLAN_GPE;
 787                                 } else {
 788                                         parse_vxlan(udp_hdr, &info,
 789                                                     m->packet_type);
 790                                         if (info.is_tunnel)
 791                                                 tx_ol_flags |=
 792                                                         PKT_TX_TUNNEL_VXLAN;
 793                                 }
 794                         } else if (info.l4_proto == IPPROTO_GRE) {
 795                                 struct simple_gre_hdr *gre_hdr;
 796
 797                                 gre_hdr = (struct simple_gre_hdr *)
 798                                         ((char *)l3_hdr + info.l3_len);
 799                                 parse_gre(gre_hdr, &info);
 800                                 if (info.is_tunnel)
 801                                         tx_ol_flags |= PKT_TX_TUNNEL_GRE;
 802                         } else if (info.l4_proto == IPPROTO_IPIP) {
 803                                 void *encap_ip_hdr;
 804
 805                                 encap_ip_hdr = (char *)l3_hdr + info.l3_len;
 806                                 parse_encap_ip(encap_ip_hdr, &info);
 807                                 if (info.is_tunnel)
 808                                         tx_ol_flags |= PKT_TX_TUNNEL_IPIP;
 809                         }
 810                 }
 811
 812                 /* update l3_hdr and outer_l3_hdr if a tunnel was parsed */
 813                 if (info.is_tunnel) {
 814                         outer_l3_hdr = l3_hdr;
 815                         l3_hdr = (char *)l3_hdr + info.outer_l3_len + info.l2_len;
 816                 }
 817
 818                 /* step 2: depending on user command line configuration,
 819                  * recompute checksum either in software or flag the
 820                  * mbuf to offload the calculation to the NIC. If TSO
 821                  * is configured, prepare the mbuf for TCP segmentation. */
 822
 823                 /* process checksums of inner headers first */
 824                 tx_ol_flags |= process_inner_cksums(l3_hdr, &info,
 825                         tx_offloads);
 826
 827                 /* Then process outer headers if any. Note that the software
 828                  * checksum will be wrong if one of the inner checksums is
 829                  * processed in hardware. */
 830                 if (info.is_tunnel == 1) {
 831                         tx_ol_flags |= process_outer_cksums(outer_l3_hdr, &info,
 832                                         tx_offloads,
 833                                         !!(tx_ol_flags & PKT_TX_TCP_SEG));
 834                 }
 835
 836                 /* step 3: fill the mbuf meta data (flags and header lengths) */
 837
 838                 m->tx_offload = 0;
 839                 if (info.is_tunnel == 1) {
 840                         if (info.tunnel_tso_segsz ||
 841                             (tx_offloads &
 842                              DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM) ||
 843                             (tx_offloads &
 844                              DEV_TX_OFFLOAD_OUTER_UDP_CKSUM) ||
 845                             (tx_ol_flags & PKT_TX_OUTER_IPV6)) {
 846                                 m->outer_l2_len = info.outer_l2_len;
 847                                 m->outer_l3_len = info.outer_l3_len;
 848                                 m->l2_len = info.l2_len;
 849                                 m->l3_len = info.l3_len;
 850                                 m->l4_len = info.l4_len;
 851                                 m->tso_segsz = info.tunnel_tso_segsz;
 852                         }
 853                         else {
 854                                 /* if there is a outer UDP cksum
 855                                    processed in sw and the inner in hw,
 856                                    the outer checksum will be wrong as
 857                                    the payload will be modified by the
 858                                    hardware */
 859                                 m->l2_len = info.outer_l2_len +
 860                                         info.outer_l3_len + info.l2_len;
 861                                 m->l3_len = info.l3_len;
 862                                 m->l4_len = info.l4_len;
 863                         }
 864                 } else {
 865                         /* this is only useful if an offload flag is
 866                          * set, but it does not hurt to fill it in any
 867                          * case */
 868                         m->l2_len = info.l2_len;
 869                         m->l3_len = info.l3_len;
 870                         m->l4_len = info.l4_len;
 871                         m->tso_segsz = info.tso_segsz;
 872                 }
 873                 m->ol_flags = tx_ol_flags;
 874
 875                 /* Do split & copy for the packet. */
 876                 if (tx_pkt_split != TX_PKT_SPLIT_OFF) {
 877                         p = pkt_copy_split(m);
 878                         if (p != NULL) {
 879                                 rte_pktmbuf_free(m);
 880                                 m = p;
 881                                 pkts_burst[i] = m;
 882                         }
 883                 }
 884
 885                 /* if verbose mode is enabled, dump debug info */
 886                 if (verbose_level > 0) {
 887                         char buf[256];
 888
 889                         printf("-----------------\n");
 890                         printf("port=%u, mbuf=%p, pkt_len=%u, nb_segs=%u:\n",
 891                                 fs->rx_port, m, m->pkt_len, m->nb_segs);
 892                         /* dump rx parsed packet info */
 893                         rte_get_rx_ol_flag_list(rx_ol_flags, buf, sizeof(buf));
 894                         printf("rx: l2_len=%d ethertype=%x l3_len=%d "
 895                                 "l4_proto=%d l4_len=%d flags=%s\n",
 896                                 info.l2_len, rte_be_to_cpu_16(info.ethertype),
 897                                 info.l3_len, info.l4_proto, info.l4_len, buf);
 898                         if (rx_ol_flags & PKT_RX_LRO)
 899                                 printf("rx: m->lro_segsz=%u\n", m->tso_segsz);
 900                         if (info.is_tunnel == 1)
 901                                 printf("rx: outer_l2_len=%d outer_ethertype=%x "
 902                                         "outer_l3_len=%d\n", info.outer_l2_len,
 903                                         rte_be_to_cpu_16(info.outer_ethertype),
 904                                         info.outer_l3_len);
 905                         /* dump tx packet info */
 906                         if ((tx_offloads & (DEV_TX_OFFLOAD_IPV4_CKSUM |
 907                                             DEV_TX_OFFLOAD_UDP_CKSUM |
 908                                             DEV_TX_OFFLOAD_TCP_CKSUM |
 909                                             DEV_TX_OFFLOAD_SCTP_CKSUM)) ||
 910                                 info.tso_segsz != 0)
 911                                 printf("tx: m->l2_len=%d m->l3_len=%d "
 912                                         "m->l4_len=%d\n",
 913                                         m->l2_len, m->l3_len, m->l4_len);
 914                         if (info.is_tunnel == 1) {
 915                                 if ((tx_offloads &
 916                                     DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM) ||
 917                                     (tx_offloads &
 918                                     DEV_TX_OFFLOAD_OUTER_UDP_CKSUM) ||
 919                                     (tx_ol_flags & PKT_TX_OUTER_IPV6))
 920                                         printf("tx: m->outer_l2_len=%d "
 921                                                 "m->outer_l3_len=%d\n",
 922                                                 m->outer_l2_len,
 923                                                 m->outer_l3_len);
 924                                 if (info.tunnel_tso_segsz != 0 &&
 925                                                 (m->ol_flags & PKT_TX_TCP_SEG))
 926                                         printf("tx: m->tso_segsz=%d\n",
 927                                                 m->tso_segsz);
 928                         } else if (info.tso_segsz != 0 &&
 929                                         (m->ol_flags & PKT_TX_TCP_SEG))
 930                                 printf("tx: m->tso_segsz=%d\n", m->tso_segsz);
 931                         rte_get_tx_ol_flag_list(m->ol_flags, buf, sizeof(buf));
 932                         printf("tx: flags=%s", buf);
 933                         printf("\n");
 934                 }
 935         }
 936
 937         if (unlikely(gro_enable)) {
 938                 if (gro_flush_cycles == GRO_DEFAULT_FLUSH_CYCLES) {
 939                         nb_rx = rte_gro_reassemble_burst(pkts_burst, nb_rx,
 940                                         &(gro_ports[fs->rx_port].param));
 941                 } else {
 942                         gro_ctx = current_fwd_lcore()->gro_ctx;
 943                         nb_rx = rte_gro_reassemble(pkts_burst, nb_rx, gro_ctx);
 944
 945                         if (++fs->gro_times >= gro_flush_cycles) {
 946                                 gro_pkts_num = rte_gro_get_pkt_count(gro_ctx);
 947                                 if (gro_pkts_num > MAX_PKT_BURST - nb_rx)
 948                                         gro_pkts_num = MAX_PKT_BURST - nb_rx;
 949
 950                                 nb_rx += rte_gro_timeout_flush(gro_ctx, 0,
 951                                                 RTE_GRO_TCP_IPV4,
 952                                                 &pkts_burst[nb_rx],
 953                                                 gro_pkts_num);
 954                                 fs->gro_times = 0;
 955                         }
 956                 }
 957         }
 958
 959         if (gso_ports[fs->tx_port].enable == 0)
 960                 tx_pkts_burst = pkts_burst;
 961         else {
 962                 gso_ctx = &(current_fwd_lcore()->gso_ctx);
 963                 gso_ctx->gso_size = gso_max_segment_size;
 964                 for (i = 0; i < nb_rx; i++) {
 965                         ret = rte_gso_segment(pkts_burst[i], gso_ctx,
 966                                         &gso_segments[nb_segments],
 967                                         GSO_MAX_PKT_BURST - nb_segments);
 968                         if (ret >= 0)
 969                                 nb_segments += ret;
 970                         else {
 971                                 TESTPMD_LOG(DEBUG, "Unable to segment packet");
 972                                 rte_pktmbuf_free(pkts_burst[i]);
 973                         }
 974                 }
 975
 976                 tx_pkts_burst = gso_segments;
 977                 nb_rx = nb_segments;
 978         }
 979
 980         nb_prep = rte_eth_tx_prepare(fs->tx_port, fs->tx_queue,
 981                         tx_pkts_burst, nb_rx);
 982         if (nb_prep != nb_rx)
 983                 printf("Preparing packet burst to transmit failed: %s\n",
 984                                 rte_strerror(rte_errno));
 985
 986         nb_tx = rte_eth_tx_burst(fs->tx_port, fs->tx_queue, tx_pkts_burst,
 987                         nb_prep);
 988
 989         /*
 990          * Retry if necessary
 991          */
 992         if (unlikely(nb_tx < nb_rx) && fs->retry_enabled) {
 993                 retry = 0;
 994                 while (nb_tx < nb_rx && retry++ < burst_tx_retry_num) {
 995                         rte_delay_us(burst_tx_delay_time);
 996                         nb_tx += rte_eth_tx_burst(fs->tx_port, fs->tx_queue,
 997                                         &tx_pkts_burst[nb_tx], nb_rx - nb_tx);
 998                 }
 999         }
1000         fs->tx_packets += nb_tx;
1001         fs->rx_bad_ip_csum += rx_bad_ip_csum;
1002         fs->rx_bad_l4_csum += rx_bad_l4_csum;
1003         fs->rx_bad_outer_l4_csum += rx_bad_outer_l4_csum;
1004
1005 #ifdef RTE_TEST_PMD_RECORD_BURST_STATS
1006         fs->tx_burst_stats.pkt_burst_spread[nb_tx]++;
1007 #endif
1008         if (unlikely(nb_tx < nb_rx)) {
1009                 fs->fwd_dropped += (nb_rx - nb_tx);
1010                 do {
1011                         rte_pktmbuf_free(tx_pkts_burst[nb_tx]);
1012                 } while (++nb_tx < nb_rx);
1013         }
1014
1015 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
1016         end_tsc = rte_rdtsc();
1017         core_cycles = (end_tsc - start_tsc);
1018         fs->core_cycles = (uint64_t) (fs->core_cycles + core_cycles);
1019 #endif
1020 }
1021
1022 struct fwd_engine csum_fwd_engine = {
1023         .fwd_mode_name  = "csum",
1024         .port_fwd_begin = NULL,
1025         .port_fwd_end   = NULL,
1026         .packet_fwd     = pkt_burst_checksum_forward,
1027 };