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_KEY_PRESENT 0x2000 53 #define GRE_KEY_LEN 4 54 #define GRE_SUPPORTED_FIELDS GRE_KEY_PRESENT 55 56 /* We cannot use rte_cpu_to_be_16() on a constant in a switch/case */ 57 #if RTE_BYTE_ORDER == RTE_LITTLE_ENDIAN 58 #define _htons(x) ((uint16_t)((((x) & 0x00ffU) << 8) | (((x) & 0xff00U) >> 8))) 59 #else 60 #define _htons(x) (x) 61 #endif 62 63 /* structure that caches offload info for the current packet */ 64 struct testpmd_offload_info { 65 uint16_t ethertype; 66 uint8_t gso_enable; 67 uint16_t l2_len; 68 uint16_t l3_len; 69 uint16_t l4_len; 70 uint8_t l4_proto; 71 uint8_t is_tunnel; 72 uint16_t outer_ethertype; 73 uint16_t outer_l2_len; 74 uint16_t outer_l3_len; 75 uint8_t outer_l4_proto; 76 uint16_t tso_segsz; 77 uint16_t tunnel_tso_segsz; 78 uint32_t pkt_len; 79 }; 80 81 /* simplified GRE header */ 82 struct simple_gre_hdr { 83 uint16_t flags; 84 uint16_t proto; 85 } __attribute__((__packed__)); 86 87 static uint16_t 88 get_udptcp_checksum(void *l3_hdr, void *l4_hdr, uint16_t ethertype) 89 { 90 if (ethertype == _htons(ETHER_TYPE_IPv4)) 91 return rte_ipv4_udptcp_cksum(l3_hdr, l4_hdr); 92 else /* assume ethertype == ETHER_TYPE_IPv6 */ 93 return rte_ipv6_udptcp_cksum(l3_hdr, l4_hdr); 94 } 95 96 /* Parse an IPv4 header to fill l3_len, l4_len, and l4_proto */ 97 static void 98 parse_ipv4(struct ipv4_hdr *ipv4_hdr, struct testpmd_offload_info *info) 99 { 100 struct tcp_hdr *tcp_hdr; 101 102 info->l3_len = (ipv4_hdr->version_ihl & 0x0f) * 4; 103 info->l4_proto = ipv4_hdr->next_proto_id; 104 105 /* only fill l4_len for TCP, it's useful for TSO */ 106 if (info->l4_proto == IPPROTO_TCP) { 107 tcp_hdr = (struct tcp_hdr *)((char *)ipv4_hdr + info->l3_len); 108 info->l4_len = (tcp_hdr->data_off & 0xf0) >> 2; 109 } else 110 info->l4_len = 0; 111 } 112 113 /* Parse an IPv6 header to fill l3_len, l4_len, and l4_proto */ 114 static void 115 parse_ipv6(struct ipv6_hdr *ipv6_hdr, struct testpmd_offload_info *info) 116 { 117 struct tcp_hdr *tcp_hdr; 118 119 info->l3_len = sizeof(struct ipv6_hdr); 120 info->l4_proto = ipv6_hdr->proto; 121 122 /* only fill l4_len for TCP, it's useful for TSO */ 123 if (info->l4_proto == IPPROTO_TCP) { 124 tcp_hdr = (struct tcp_hdr *)((char *)ipv6_hdr + info->l3_len); 125 info->l4_len = (tcp_hdr->data_off & 0xf0) >> 2; 126 } else 127 info->l4_len = 0; 128 } 129 130 /* 131 * Parse an ethernet header to fill the ethertype, l2_len, l3_len and 132 * ipproto. This function is able to recognize IPv4/IPv6 with one optional vlan 133 * header. The l4_len argument is only set in case of TCP (useful for TSO). 134 */ 135 static void 136 parse_ethernet(struct ether_hdr *eth_hdr, struct testpmd_offload_info *info) 137 { 138 struct ipv4_hdr *ipv4_hdr; 139 struct ipv6_hdr *ipv6_hdr; 140 141 info->l2_len = sizeof(struct ether_hdr); 142 info->ethertype = eth_hdr->ether_type; 143 144 if (info->ethertype == _htons(ETHER_TYPE_VLAN)) { 145 struct vlan_hdr *vlan_hdr = (struct vlan_hdr *)(eth_hdr + 1); 146 147 info->l2_len += sizeof(struct vlan_hdr); 148 info->ethertype = vlan_hdr->eth_proto; 149 } 150 151 switch (info->ethertype) { 152 case _htons(ETHER_TYPE_IPv4): 153 ipv4_hdr = (struct ipv4_hdr *) ((char *)eth_hdr + info->l2_len); 154 parse_ipv4(ipv4_hdr, info); 155 break; 156 case _htons(ETHER_TYPE_IPv6): 157 ipv6_hdr = (struct ipv6_hdr *) ((char *)eth_hdr + info->l2_len); 158 parse_ipv6(ipv6_hdr, info); 159 break; 160 default: 161 info->l4_len = 0; 162 info->l3_len = 0; 163 info->l4_proto = 0; 164 break; 165 } 166 } 167 168 /* Parse a vxlan header */ 169 static void 170 parse_vxlan(struct udp_hdr *udp_hdr, 171 struct testpmd_offload_info *info, 172 uint32_t pkt_type) 173 { 174 struct ether_hdr *eth_hdr; 175 176 /* check udp destination port, 4789 is the default vxlan port 177 * (rfc7348) or that the rx offload flag is set (i40e only 178 * currently) */ 179 if (udp_hdr->dst_port != _htons(4789) && 180 RTE_ETH_IS_TUNNEL_PKT(pkt_type) == 0) 181 return; 182 183 info->is_tunnel = 1; 184 info->outer_ethertype = info->ethertype; 185 info->outer_l2_len = info->l2_len; 186 info->outer_l3_len = info->l3_len; 187 info->outer_l4_proto = info->l4_proto; 188 189 eth_hdr = (struct ether_hdr *)((char *)udp_hdr + 190 sizeof(struct udp_hdr) + 191 sizeof(struct vxlan_hdr)); 192 193 parse_ethernet(eth_hdr, info); 194 info->l2_len += ETHER_VXLAN_HLEN; /* add udp + vxlan */ 195 } 196 197 /* Parse a gre header */ 198 static void 199 parse_gre(struct simple_gre_hdr *gre_hdr, struct testpmd_offload_info *info) 200 { 201 struct ether_hdr *eth_hdr; 202 struct ipv4_hdr *ipv4_hdr; 203 struct ipv6_hdr *ipv6_hdr; 204 uint8_t gre_len = 0; 205 206 /* check which fields are supported */ 207 if ((gre_hdr->flags & _htons(~GRE_SUPPORTED_FIELDS)) != 0) 208 return; 209 210 gre_len += sizeof(struct simple_gre_hdr); 211 212 if (gre_hdr->flags & _htons(GRE_KEY_PRESENT)) 213 gre_len += GRE_KEY_LEN; 214 215 if (gre_hdr->proto == _htons(ETHER_TYPE_IPv4)) { 216 info->is_tunnel = 1; 217 info->outer_ethertype = info->ethertype; 218 info->outer_l2_len = info->l2_len; 219 info->outer_l3_len = info->l3_len; 220 info->outer_l4_proto = info->l4_proto; 221 222 ipv4_hdr = (struct ipv4_hdr *)((char *)gre_hdr + gre_len); 223 224 parse_ipv4(ipv4_hdr, info); 225 info->ethertype = _htons(ETHER_TYPE_IPv4); 226 info->l2_len = 0; 227 228 } else if (gre_hdr->proto == _htons(ETHER_TYPE_IPv6)) { 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 ipv6_hdr = (struct ipv6_hdr *)((char *)gre_hdr + gre_len); 236 237 info->ethertype = _htons(ETHER_TYPE_IPv6); 238 parse_ipv6(ipv6_hdr, info); 239 info->l2_len = 0; 240 241 } else if (gre_hdr->proto == _htons(ETHER_TYPE_TEB)) { 242 info->is_tunnel = 1; 243 info->outer_ethertype = info->ethertype; 244 info->outer_l2_len = info->l2_len; 245 info->outer_l3_len = info->l3_len; 246 info->outer_l4_proto = info->l4_proto; 247 248 eth_hdr = (struct ether_hdr *)((char *)gre_hdr + gre_len); 249 250 parse_ethernet(eth_hdr, info); 251 } else 252 return; 253 254 info->l2_len += gre_len; 255 } 256 257 258 /* Parse an encapsulated ip or ipv6 header */ 259 static void 260 parse_encap_ip(void *encap_ip, struct testpmd_offload_info *info) 261 { 262 struct ipv4_hdr *ipv4_hdr = encap_ip; 263 struct ipv6_hdr *ipv6_hdr = encap_ip; 264 uint8_t ip_version; 265 266 ip_version = (ipv4_hdr->version_ihl & 0xf0) >> 4; 267 268 if (ip_version != 4 && ip_version != 6) 269 return; 270 271 info->is_tunnel = 1; 272 info->outer_ethertype = info->ethertype; 273 info->outer_l2_len = info->l2_len; 274 info->outer_l3_len = info->l3_len; 275 276 if (ip_version == 4) { 277 parse_ipv4(ipv4_hdr, info); 278 info->ethertype = _htons(ETHER_TYPE_IPv4); 279 } else { 280 parse_ipv6(ipv6_hdr, info); 281 info->ethertype = _htons(ETHER_TYPE_IPv6); 282 } 283 info->l2_len = 0; 284 } 285 286 /* if possible, calculate the checksum of a packet in hw or sw, 287 * depending on the testpmd command line configuration */ 288 static uint64_t 289 process_inner_cksums(void *l3_hdr, const struct testpmd_offload_info *info, 290 uint64_t tx_offloads) 291 { 292 struct ipv4_hdr *ipv4_hdr = l3_hdr; 293 struct udp_hdr *udp_hdr; 294 struct tcp_hdr *tcp_hdr; 295 struct sctp_hdr *sctp_hdr; 296 uint64_t ol_flags = 0; 297 uint32_t max_pkt_len, tso_segsz = 0; 298 299 /* ensure packet is large enough to require tso */ 300 if (!info->is_tunnel) { 301 max_pkt_len = info->l2_len + info->l3_len + info->l4_len + 302 info->tso_segsz; 303 if (info->tso_segsz != 0 && info->pkt_len > max_pkt_len) 304 tso_segsz = info->tso_segsz; 305 } else { 306 max_pkt_len = info->outer_l2_len + info->outer_l3_len + 307 info->l2_len + info->l3_len + info->l4_len + 308 info->tunnel_tso_segsz; 309 if (info->tunnel_tso_segsz != 0 && info->pkt_len > max_pkt_len) 310 tso_segsz = info->tunnel_tso_segsz; 311 } 312 313 if (info->ethertype == _htons(ETHER_TYPE_IPv4)) { 314 ipv4_hdr = l3_hdr; 315 ipv4_hdr->hdr_checksum = 0; 316 317 ol_flags |= PKT_TX_IPV4; 318 if (info->l4_proto == IPPROTO_TCP && tso_segsz) { 319 ol_flags |= PKT_TX_IP_CKSUM; 320 } else { 321 if (tx_offloads & DEV_TX_OFFLOAD_IPV4_CKSUM) 322 ol_flags |= PKT_TX_IP_CKSUM; 323 else 324 ipv4_hdr->hdr_checksum = 325 rte_ipv4_cksum(ipv4_hdr); 326 } 327 } else if (info->ethertype == _htons(ETHER_TYPE_IPv6)) 328 ol_flags |= PKT_TX_IPV6; 329 else 330 return 0; /* packet type not supported, nothing to do */ 331 332 if (info->l4_proto == IPPROTO_UDP) { 333 udp_hdr = (struct udp_hdr *)((char *)l3_hdr + info->l3_len); 334 /* do not recalculate udp cksum if it was 0 */ 335 if (udp_hdr->dgram_cksum != 0) { 336 udp_hdr->dgram_cksum = 0; 337 if (tx_offloads & DEV_TX_OFFLOAD_UDP_CKSUM) 338 ol_flags |= PKT_TX_UDP_CKSUM; 339 else { 340 udp_hdr->dgram_cksum = 341 get_udptcp_checksum(l3_hdr, udp_hdr, 342 info->ethertype); 343 } 344 } 345 } else if (info->l4_proto == IPPROTO_TCP) { 346 tcp_hdr = (struct tcp_hdr *)((char *)l3_hdr + info->l3_len); 347 tcp_hdr->cksum = 0; 348 if (tso_segsz) 349 ol_flags |= PKT_TX_TCP_SEG; 350 else if (tx_offloads & DEV_TX_OFFLOAD_TCP_CKSUM) 351 ol_flags |= PKT_TX_TCP_CKSUM; 352 else { 353 tcp_hdr->cksum = 354 get_udptcp_checksum(l3_hdr, tcp_hdr, 355 info->ethertype); 356 } 357 if (info->gso_enable) 358 ol_flags |= PKT_TX_TCP_SEG; 359 } else if (info->l4_proto == IPPROTO_SCTP) { 360 sctp_hdr = (struct sctp_hdr *)((char *)l3_hdr + info->l3_len); 361 sctp_hdr->cksum = 0; 362 /* sctp payload must be a multiple of 4 to be 363 * offloaded */ 364 if ((tx_offloads & DEV_TX_OFFLOAD_SCTP_CKSUM) && 365 ((ipv4_hdr->total_length & 0x3) == 0)) { 366 ol_flags |= PKT_TX_SCTP_CKSUM; 367 } else { 368 /* XXX implement CRC32c, example available in 369 * RFC3309 */ 370 } 371 } 372 373 return ol_flags; 374 } 375 376 /* Calculate the checksum of outer header */ 377 static uint64_t 378 process_outer_cksums(void *outer_l3_hdr, struct testpmd_offload_info *info, 379 uint64_t tx_offloads, int tso_enabled) 380 { 381 struct ipv4_hdr *ipv4_hdr = outer_l3_hdr; 382 struct ipv6_hdr *ipv6_hdr = outer_l3_hdr; 383 struct udp_hdr *udp_hdr; 384 uint64_t ol_flags = 0; 385 386 if (info->outer_ethertype == _htons(ETHER_TYPE_IPv4)) { 387 ipv4_hdr->hdr_checksum = 0; 388 ol_flags |= PKT_TX_OUTER_IPV4; 389 390 if (tx_offloads & DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM) 391 ol_flags |= PKT_TX_OUTER_IP_CKSUM; 392 else 393 ipv4_hdr->hdr_checksum = rte_ipv4_cksum(ipv4_hdr); 394 } else 395 ol_flags |= PKT_TX_OUTER_IPV6; 396 397 if (info->outer_l4_proto != IPPROTO_UDP) 398 return ol_flags; 399 400 udp_hdr = (struct udp_hdr *)((char *)outer_l3_hdr + info->outer_l3_len); 401 402 /* outer UDP checksum is done in software as we have no hardware 403 * supporting it today, and no API for it. In the other side, for 404 * UDP tunneling, like VXLAN or Geneve, outer UDP checksum can be 405 * set to zero. 406 * 407 * If a packet will be TSOed into small packets by NIC, we cannot 408 * set/calculate a non-zero checksum, because it will be a wrong 409 * value after the packet be split into several small packets. 410 */ 411 if (tso_enabled) 412 udp_hdr->dgram_cksum = 0; 413 414 /* do not recalculate udp cksum if it was 0 */ 415 if (udp_hdr->dgram_cksum != 0) { 416 udp_hdr->dgram_cksum = 0; 417 if (info->outer_ethertype == _htons(ETHER_TYPE_IPv4)) 418 udp_hdr->dgram_cksum = 419 rte_ipv4_udptcp_cksum(ipv4_hdr, udp_hdr); 420 else 421 udp_hdr->dgram_cksum = 422 rte_ipv6_udptcp_cksum(ipv6_hdr, udp_hdr); 423 } 424 425 return ol_flags; 426 } 427 428 /* 429 * Helper function. 430 * Performs actual copying. 431 * Returns number of segments in the destination mbuf on success, 432 * or negative error code on failure. 433 */ 434 static int 435 mbuf_copy_split(const struct rte_mbuf *ms, struct rte_mbuf *md[], 436 uint16_t seglen[], uint8_t nb_seg) 437 { 438 uint32_t dlen, slen, tlen; 439 uint32_t i, len; 440 const struct rte_mbuf *m; 441 const uint8_t *src; 442 uint8_t *dst; 443 444 dlen = 0; 445 slen = 0; 446 tlen = 0; 447 448 dst = NULL; 449 src = NULL; 450 451 m = ms; 452 i = 0; 453 while (ms != NULL && i != nb_seg) { 454 455 if (slen == 0) { 456 slen = rte_pktmbuf_data_len(ms); 457 src = rte_pktmbuf_mtod(ms, const uint8_t *); 458 } 459 460 if (dlen == 0) { 461 dlen = RTE_MIN(seglen[i], slen); 462 md[i]->data_len = dlen; 463 md[i]->next = (i + 1 == nb_seg) ? NULL : md[i + 1]; 464 dst = rte_pktmbuf_mtod(md[i], uint8_t *); 465 } 466 467 len = RTE_MIN(slen, dlen); 468 memcpy(dst, src, len); 469 tlen += len; 470 slen -= len; 471 dlen -= len; 472 src += len; 473 dst += len; 474 475 if (slen == 0) 476 ms = ms->next; 477 if (dlen == 0) 478 i++; 479 } 480 481 if (ms != NULL) 482 return -ENOBUFS; 483 else if (tlen != m->pkt_len) 484 return -EINVAL; 485 486 md[0]->nb_segs = nb_seg; 487 md[0]->pkt_len = tlen; 488 md[0]->vlan_tci = m->vlan_tci; 489 md[0]->vlan_tci_outer = m->vlan_tci_outer; 490 md[0]->ol_flags = m->ol_flags; 491 md[0]->tx_offload = m->tx_offload; 492 493 return nb_seg; 494 } 495 496 /* 497 * Allocate a new mbuf with up to tx_pkt_nb_segs segments. 498 * Copy packet contents and offload information into then new segmented mbuf. 499 */ 500 static struct rte_mbuf * 501 pkt_copy_split(const struct rte_mbuf *pkt) 502 { 503 int32_t n, rc; 504 uint32_t i, len, nb_seg; 505 struct rte_mempool *mp; 506 uint16_t seglen[RTE_MAX_SEGS_PER_PKT]; 507 struct rte_mbuf *p, *md[RTE_MAX_SEGS_PER_PKT]; 508 509 mp = current_fwd_lcore()->mbp; 510 511 if (tx_pkt_split == TX_PKT_SPLIT_RND) 512 nb_seg = random() % tx_pkt_nb_segs + 1; 513 else 514 nb_seg = tx_pkt_nb_segs; 515 516 memcpy(seglen, tx_pkt_seg_lengths, nb_seg * sizeof(seglen[0])); 517 518 /* calculate number of segments to use and their length. */ 519 len = 0; 520 for (i = 0; i != nb_seg && len < pkt->pkt_len; i++) { 521 len += seglen[i]; 522 md[i] = NULL; 523 } 524 525 n = pkt->pkt_len - len; 526 527 /* update size of the last segment to fit rest of the packet */ 528 if (n >= 0) { 529 seglen[i - 1] += n; 530 len += n; 531 } 532 533 nb_seg = i; 534 while (i != 0) { 535 p = rte_pktmbuf_alloc(mp); 536 if (p == NULL) { 537 TESTPMD_LOG(ERR, 538 "failed to allocate %u-th of %u mbuf " 539 "from mempool: %s\n", 540 nb_seg - i, nb_seg, mp->name); 541 break; 542 } 543 544 md[--i] = p; 545 if (rte_pktmbuf_tailroom(md[i]) < seglen[i]) { 546 TESTPMD_LOG(ERR, "mempool %s, %u-th segment: " 547 "expected seglen: %u, " 548 "actual mbuf tailroom: %u\n", 549 mp->name, i, seglen[i], 550 rte_pktmbuf_tailroom(md[i])); 551 break; 552 } 553 } 554 555 /* all mbufs successfully allocated, do copy */ 556 if (i == 0) { 557 rc = mbuf_copy_split(pkt, md, seglen, nb_seg); 558 if (rc < 0) 559 TESTPMD_LOG(ERR, 560 "mbuf_copy_split for %p(len=%u, nb_seg=%u) " 561 "into %u segments failed with error code: %d\n", 562 pkt, pkt->pkt_len, pkt->nb_segs, nb_seg, rc); 563 564 /* figure out how many mbufs to free. */ 565 i = RTE_MAX(rc, 0); 566 } 567 568 /* free unused mbufs */ 569 for (; i != nb_seg; i++) { 570 rte_pktmbuf_free_seg(md[i]); 571 md[i] = NULL; 572 } 573 574 return md[0]; 575 } 576 577 /* 578 * Receive a burst of packets, and for each packet: 579 * - parse packet, and try to recognize a supported packet type (1) 580 * - if it's not a supported packet type, don't touch the packet, else: 581 * - reprocess the checksum of all supported layers. This is done in SW 582 * or HW, depending on testpmd command line configuration 583 * - if TSO is enabled in testpmd command line, also flag the mbuf for TCP 584 * segmentation offload (this implies HW TCP checksum) 585 * Then transmit packets on the output port. 586 * 587 * (1) Supported packets are: 588 * Ether / (vlan) / IP|IP6 / UDP|TCP|SCTP . 589 * Ether / (vlan) / outer IP|IP6 / outer UDP / VxLAN / Ether / IP|IP6 / 590 * UDP|TCP|SCTP 591 * Ether / (vlan) / outer IP|IP6 / GRE / Ether / IP|IP6 / UDP|TCP|SCTP 592 * Ether / (vlan) / outer IP|IP6 / GRE / IP|IP6 / UDP|TCP|SCTP 593 * Ether / (vlan) / outer IP|IP6 / IP|IP6 / UDP|TCP|SCTP 594 * 595 * The testpmd command line for this forward engine sets the flags 596 * TESTPMD_TX_OFFLOAD_* in ports[tx_port].tx_ol_flags. They control 597 * wether a checksum must be calculated in software or in hardware. The 598 * IP, UDP, TCP and SCTP flags always concern the inner layer. The 599 * OUTER_IP is only useful for tunnel packets. 600 */ 601 static void 602 pkt_burst_checksum_forward(struct fwd_stream *fs) 603 { 604 struct rte_mbuf *pkts_burst[MAX_PKT_BURST]; 605 struct rte_mbuf *gso_segments[GSO_MAX_PKT_BURST]; 606 struct rte_gso_ctx *gso_ctx; 607 struct rte_mbuf **tx_pkts_burst; 608 struct rte_port *txp; 609 struct rte_mbuf *m, *p; 610 struct ether_hdr *eth_hdr; 611 void *l3_hdr = NULL, *outer_l3_hdr = NULL; /* can be IPv4 or IPv6 */ 612 void **gro_ctx; 613 uint16_t gro_pkts_num; 614 uint8_t gro_enable; 615 uint16_t nb_rx; 616 uint16_t nb_tx; 617 uint16_t nb_prep; 618 uint16_t i; 619 uint64_t rx_ol_flags, tx_ol_flags; 620 uint64_t tx_offloads; 621 uint32_t retry; 622 uint32_t rx_bad_ip_csum; 623 uint32_t rx_bad_l4_csum; 624 struct testpmd_offload_info info; 625 uint16_t nb_segments = 0; 626 int ret; 627 628 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES 629 uint64_t start_tsc; 630 uint64_t end_tsc; 631 uint64_t core_cycles; 632 #endif 633 634 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES 635 start_tsc = rte_rdtsc(); 636 #endif 637 638 /* receive a burst of packet */ 639 nb_rx = rte_eth_rx_burst(fs->rx_port, fs->rx_queue, pkts_burst, 640 nb_pkt_per_burst); 641 if (unlikely(nb_rx == 0)) 642 return; 643 #ifdef RTE_TEST_PMD_RECORD_BURST_STATS 644 fs->rx_burst_stats.pkt_burst_spread[nb_rx]++; 645 #endif 646 fs->rx_packets += nb_rx; 647 rx_bad_ip_csum = 0; 648 rx_bad_l4_csum = 0; 649 gro_enable = gro_ports[fs->rx_port].enable; 650 651 txp = &ports[fs->tx_port]; 652 tx_offloads = txp->dev_conf.txmode.offloads; 653 memset(&info, 0, sizeof(info)); 654 info.tso_segsz = txp->tso_segsz; 655 info.tunnel_tso_segsz = txp->tunnel_tso_segsz; 656 if (gso_ports[fs->tx_port].enable) 657 info.gso_enable = 1; 658 659 for (i = 0; i < nb_rx; i++) { 660 if (likely(i < nb_rx - 1)) 661 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[i + 1], 662 void *)); 663 664 m = pkts_burst[i]; 665 info.is_tunnel = 0; 666 info.pkt_len = rte_pktmbuf_pkt_len(m); 667 tx_ol_flags = 0; 668 rx_ol_flags = m->ol_flags; 669 670 /* Update the L3/L4 checksum error packet statistics */ 671 if ((rx_ol_flags & PKT_RX_IP_CKSUM_MASK) == PKT_RX_IP_CKSUM_BAD) 672 rx_bad_ip_csum += 1; 673 if ((rx_ol_flags & PKT_RX_L4_CKSUM_MASK) == PKT_RX_L4_CKSUM_BAD) 674 rx_bad_l4_csum += 1; 675 676 /* step 1: dissect packet, parsing optional vlan, ip4/ip6, vxlan 677 * and inner headers */ 678 679 eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *); 680 ether_addr_copy(&peer_eth_addrs[fs->peer_addr], 681 ð_hdr->d_addr); 682 ether_addr_copy(&ports[fs->tx_port].eth_addr, 683 ð_hdr->s_addr); 684 parse_ethernet(eth_hdr, &info); 685 l3_hdr = (char *)eth_hdr + info.l2_len; 686 687 /* check if it's a supported tunnel */ 688 if (txp->parse_tunnel) { 689 if (info.l4_proto == IPPROTO_UDP) { 690 struct udp_hdr *udp_hdr; 691 692 udp_hdr = (struct udp_hdr *)((char *)l3_hdr + 693 info.l3_len); 694 parse_vxlan(udp_hdr, &info, m->packet_type); 695 if (info.is_tunnel) 696 tx_ol_flags |= PKT_TX_TUNNEL_VXLAN; 697 } else if (info.l4_proto == IPPROTO_GRE) { 698 struct simple_gre_hdr *gre_hdr; 699 700 gre_hdr = (struct simple_gre_hdr *) 701 ((char *)l3_hdr + info.l3_len); 702 parse_gre(gre_hdr, &info); 703 if (info.is_tunnel) 704 tx_ol_flags |= PKT_TX_TUNNEL_GRE; 705 } else if (info.l4_proto == IPPROTO_IPIP) { 706 void *encap_ip_hdr; 707 708 encap_ip_hdr = (char *)l3_hdr + info.l3_len; 709 parse_encap_ip(encap_ip_hdr, &info); 710 if (info.is_tunnel) 711 tx_ol_flags |= PKT_TX_TUNNEL_IPIP; 712 } 713 } 714 715 /* update l3_hdr and outer_l3_hdr if a tunnel was parsed */ 716 if (info.is_tunnel) { 717 outer_l3_hdr = l3_hdr; 718 l3_hdr = (char *)l3_hdr + info.outer_l3_len + info.l2_len; 719 } 720 721 /* step 2: depending on user command line configuration, 722 * recompute checksum either in software or flag the 723 * mbuf to offload the calculation to the NIC. If TSO 724 * is configured, prepare the mbuf for TCP segmentation. */ 725 726 /* process checksums of inner headers first */ 727 tx_ol_flags |= process_inner_cksums(l3_hdr, &info, 728 tx_offloads); 729 730 /* Then process outer headers if any. Note that the software 731 * checksum will be wrong if one of the inner checksums is 732 * processed in hardware. */ 733 if (info.is_tunnel == 1) { 734 tx_ol_flags |= process_outer_cksums(outer_l3_hdr, &info, 735 tx_offloads, 736 !!(tx_ol_flags & PKT_TX_TCP_SEG)); 737 } 738 739 /* step 3: fill the mbuf meta data (flags and header lengths) */ 740 741 if (info.is_tunnel == 1) { 742 if (info.tunnel_tso_segsz || 743 (tx_offloads & 744 DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM) || 745 (tx_ol_flags & PKT_TX_OUTER_IPV6)) { 746 m->outer_l2_len = info.outer_l2_len; 747 m->outer_l3_len = info.outer_l3_len; 748 m->l2_len = info.l2_len; 749 m->l3_len = info.l3_len; 750 m->l4_len = info.l4_len; 751 m->tso_segsz = info.tunnel_tso_segsz; 752 } 753 else { 754 /* if there is a outer UDP cksum 755 processed in sw and the inner in hw, 756 the outer checksum will be wrong as 757 the payload will be modified by the 758 hardware */ 759 m->l2_len = info.outer_l2_len + 760 info.outer_l3_len + info.l2_len; 761 m->l3_len = info.l3_len; 762 m->l4_len = info.l4_len; 763 } 764 } else { 765 /* this is only useful if an offload flag is 766 * set, but it does not hurt to fill it in any 767 * case */ 768 m->l2_len = info.l2_len; 769 m->l3_len = info.l3_len; 770 m->l4_len = info.l4_len; 771 m->tso_segsz = info.tso_segsz; 772 } 773 m->ol_flags = tx_ol_flags; 774 775 /* Do split & copy for the packet. */ 776 if (tx_pkt_split != TX_PKT_SPLIT_OFF) { 777 p = pkt_copy_split(m); 778 if (p != NULL) { 779 rte_pktmbuf_free(m); 780 m = p; 781 pkts_burst[i] = m; 782 } 783 } 784 785 /* if verbose mode is enabled, dump debug info */ 786 if (verbose_level > 0) { 787 char buf[256]; 788 789 printf("-----------------\n"); 790 printf("port=%u, mbuf=%p, pkt_len=%u, nb_segs=%u:\n", 791 fs->rx_port, m, m->pkt_len, m->nb_segs); 792 /* dump rx parsed packet info */ 793 rte_get_rx_ol_flag_list(rx_ol_flags, buf, sizeof(buf)); 794 printf("rx: l2_len=%d ethertype=%x l3_len=%d " 795 "l4_proto=%d l4_len=%d flags=%s\n", 796 info.l2_len, rte_be_to_cpu_16(info.ethertype), 797 info.l3_len, info.l4_proto, info.l4_len, buf); 798 if (rx_ol_flags & PKT_RX_LRO) 799 printf("rx: m->lro_segsz=%u\n", m->tso_segsz); 800 if (info.is_tunnel == 1) 801 printf("rx: outer_l2_len=%d outer_ethertype=%x " 802 "outer_l3_len=%d\n", info.outer_l2_len, 803 rte_be_to_cpu_16(info.outer_ethertype), 804 info.outer_l3_len); 805 /* dump tx packet info */ 806 if ((tx_offloads & (DEV_TX_OFFLOAD_IPV4_CKSUM | 807 DEV_TX_OFFLOAD_UDP_CKSUM | 808 DEV_TX_OFFLOAD_TCP_CKSUM | 809 DEV_TX_OFFLOAD_SCTP_CKSUM)) || 810 info.tso_segsz != 0) 811 printf("tx: m->l2_len=%d m->l3_len=%d " 812 "m->l4_len=%d\n", 813 m->l2_len, m->l3_len, m->l4_len); 814 if (info.is_tunnel == 1) { 815 if ((tx_offloads & 816 DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM) || 817 (tx_ol_flags & PKT_TX_OUTER_IPV6)) 818 printf("tx: m->outer_l2_len=%d " 819 "m->outer_l3_len=%d\n", 820 m->outer_l2_len, 821 m->outer_l3_len); 822 if (info.tunnel_tso_segsz != 0 && 823 (m->ol_flags & PKT_TX_TCP_SEG)) 824 printf("tx: m->tso_segsz=%d\n", 825 m->tso_segsz); 826 } else if (info.tso_segsz != 0 && 827 (m->ol_flags & PKT_TX_TCP_SEG)) 828 printf("tx: m->tso_segsz=%d\n", m->tso_segsz); 829 rte_get_tx_ol_flag_list(m->ol_flags, buf, sizeof(buf)); 830 printf("tx: flags=%s", buf); 831 printf("\n"); 832 } 833 } 834 835 if (unlikely(gro_enable)) { 836 if (gro_flush_cycles == GRO_DEFAULT_FLUSH_CYCLES) { 837 nb_rx = rte_gro_reassemble_burst(pkts_burst, nb_rx, 838 &(gro_ports[fs->rx_port].param)); 839 } else { 840 gro_ctx = current_fwd_lcore()->gro_ctx; 841 nb_rx = rte_gro_reassemble(pkts_burst, nb_rx, gro_ctx); 842 843 if (++fs->gro_times >= gro_flush_cycles) { 844 gro_pkts_num = rte_gro_get_pkt_count(gro_ctx); 845 if (gro_pkts_num > MAX_PKT_BURST - nb_rx) 846 gro_pkts_num = MAX_PKT_BURST - nb_rx; 847 848 nb_rx += rte_gro_timeout_flush(gro_ctx, 0, 849 RTE_GRO_TCP_IPV4, 850 &pkts_burst[nb_rx], 851 gro_pkts_num); 852 fs->gro_times = 0; 853 } 854 } 855 } 856 857 if (gso_ports[fs->tx_port].enable == 0) 858 tx_pkts_burst = pkts_burst; 859 else { 860 gso_ctx = &(current_fwd_lcore()->gso_ctx); 861 gso_ctx->gso_size = gso_max_segment_size; 862 for (i = 0; i < nb_rx; i++) { 863 ret = rte_gso_segment(pkts_burst[i], gso_ctx, 864 &gso_segments[nb_segments], 865 GSO_MAX_PKT_BURST - nb_segments); 866 if (ret >= 0) 867 nb_segments += ret; 868 else { 869 TESTPMD_LOG(DEBUG, "Unable to segment packet"); 870 rte_pktmbuf_free(pkts_burst[i]); 871 } 872 } 873 874 tx_pkts_burst = gso_segments; 875 nb_rx = nb_segments; 876 } 877 878 nb_prep = rte_eth_tx_prepare(fs->tx_port, fs->tx_queue, 879 tx_pkts_burst, nb_rx); 880 if (nb_prep != nb_rx) 881 printf("Preparing packet burst to transmit failed: %s\n", 882 rte_strerror(rte_errno)); 883 884 nb_tx = rte_eth_tx_burst(fs->tx_port, fs->tx_queue, tx_pkts_burst, 885 nb_prep); 886 887 /* 888 * Retry if necessary 889 */ 890 if (unlikely(nb_tx < nb_rx) && fs->retry_enabled) { 891 retry = 0; 892 while (nb_tx < nb_rx && retry++ < burst_tx_retry_num) { 893 rte_delay_us(burst_tx_delay_time); 894 nb_tx += rte_eth_tx_burst(fs->tx_port, fs->tx_queue, 895 &tx_pkts_burst[nb_tx], nb_rx - nb_tx); 896 } 897 } 898 fs->tx_packets += nb_tx; 899 fs->rx_bad_ip_csum += rx_bad_ip_csum; 900 fs->rx_bad_l4_csum += rx_bad_l4_csum; 901 902 #ifdef RTE_TEST_PMD_RECORD_BURST_STATS 903 fs->tx_burst_stats.pkt_burst_spread[nb_tx]++; 904 #endif 905 if (unlikely(nb_tx < nb_rx)) { 906 fs->fwd_dropped += (nb_rx - nb_tx); 907 do { 908 rte_pktmbuf_free(tx_pkts_burst[nb_tx]); 909 } while (++nb_tx < nb_rx); 910 } 911 912 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES 913 end_tsc = rte_rdtsc(); 914 core_cycles = (end_tsc - start_tsc); 915 fs->core_cycles = (uint64_t) (fs->core_cycles + core_cycles); 916 #endif 917 } 918 919 struct fwd_engine csum_fwd_engine = { 920 .fwd_mode_name = "csum", 921 .port_fwd_begin = NULL, 922 .port_fwd_end = NULL, 923 .packet_fwd = pkt_burst_checksum_forward, 924 }; 925