1 /* SPDX-License-Identifier: BSD-3-Clause 2 * Copyright(c) 2010-2014 Intel Corporation 3 */ 4 5 #include <stdio.h> 6 #include <stdlib.h> 7 #include <stdint.h> 8 #include <inttypes.h> 9 #include <sys/types.h> 10 #include <sys/param.h> 11 #include <string.h> 12 #include <sys/queue.h> 13 #include <stdarg.h> 14 #include <errno.h> 15 #include <getopt.h> 16 17 #include <rte_common.h> 18 #include <rte_byteorder.h> 19 #include <rte_log.h> 20 #include <rte_memory.h> 21 #include <rte_memcpy.h> 22 #include <rte_eal.h> 23 #include <rte_launch.h> 24 #include <rte_atomic.h> 25 #include <rte_cycles.h> 26 #include <rte_prefetch.h> 27 #include <rte_lcore.h> 28 #include <rte_per_lcore.h> 29 #include <rte_branch_prediction.h> 30 #include <rte_interrupts.h> 31 #include <rte_random.h> 32 #include <rte_debug.h> 33 #include <rte_ether.h> 34 #include <rte_ethdev.h> 35 #include <rte_mempool.h> 36 #include <rte_mbuf.h> 37 #include <rte_lpm.h> 38 #include <rte_lpm6.h> 39 #include <rte_ip.h> 40 #include <rte_string_fns.h> 41 42 #include <rte_ip_frag.h> 43 44 #define RTE_LOGTYPE_IP_FRAG RTE_LOGTYPE_USER1 45 46 /* allow max jumbo frame 9.5 KB */ 47 #define JUMBO_FRAME_MAX_SIZE 0x2600 48 49 #define ROUNDUP_DIV(a, b) (((a) + (b) - 1) / (b)) 50 51 /* 52 * Default byte size for the IPv6 Maximum Transfer Unit (MTU). 53 * This value includes the size of IPv6 header. 54 */ 55 #define IPV4_MTU_DEFAULT RTE_ETHER_MTU 56 #define IPV6_MTU_DEFAULT RTE_ETHER_MTU 57 58 /* 59 * The overhead from max frame size to MTU. 60 * We have to consider the max possible overhead. 61 */ 62 #define MTU_OVERHEAD \ 63 (RTE_ETHER_HDR_LEN + RTE_ETHER_CRC_LEN + \ 64 2 * sizeof(struct rte_vlan_hdr)) 65 66 /* 67 * Default payload in bytes for the IPv6 packet. 68 */ 69 #define IPV4_DEFAULT_PAYLOAD (IPV4_MTU_DEFAULT - sizeof(struct rte_ipv4_hdr)) 70 #define IPV6_DEFAULT_PAYLOAD (IPV6_MTU_DEFAULT - sizeof(struct rte_ipv6_hdr)) 71 72 /* 73 * Max number of fragments per packet expected - defined by config file. 74 */ 75 #define MAX_PACKET_FRAG RTE_LIBRTE_IP_FRAG_MAX_FRAG 76 77 #define NB_MBUF 8192 78 79 #define MAX_PKT_BURST 32 80 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */ 81 82 /* Configure how many packets ahead to prefetch, when reading packets */ 83 #define PREFETCH_OFFSET 3 84 85 /* 86 * Configurable number of RX/TX ring descriptors 87 */ 88 #define RTE_TEST_RX_DESC_DEFAULT 1024 89 #define RTE_TEST_TX_DESC_DEFAULT 1024 90 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT; 91 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT; 92 93 /* ethernet addresses of ports */ 94 static struct rte_ether_addr ports_eth_addr[RTE_MAX_ETHPORTS]; 95 96 #ifndef IPv4_BYTES 97 #define IPv4_BYTES_FMT "%" PRIu8 ".%" PRIu8 ".%" PRIu8 ".%" PRIu8 98 #define IPv4_BYTES(addr) \ 99 (uint8_t) (((addr) >> 24) & 0xFF),\ 100 (uint8_t) (((addr) >> 16) & 0xFF),\ 101 (uint8_t) (((addr) >> 8) & 0xFF),\ 102 (uint8_t) ((addr) & 0xFF) 103 #endif 104 105 #ifndef IPv6_BYTES 106 #define IPv6_BYTES_FMT "%02x%02x:%02x%02x:%02x%02x:%02x%02x:"\ 107 "%02x%02x:%02x%02x:%02x%02x:%02x%02x" 108 #define IPv6_BYTES(addr) \ 109 addr[0], addr[1], addr[2], addr[3], \ 110 addr[4], addr[5], addr[6], addr[7], \ 111 addr[8], addr[9], addr[10], addr[11],\ 112 addr[12], addr[13],addr[14], addr[15] 113 #endif 114 115 #define IPV6_ADDR_LEN 16 116 117 /* mask of enabled ports */ 118 static int enabled_port_mask = 0; 119 120 static int rx_queue_per_lcore = 1; 121 122 #define MBUF_TABLE_SIZE (2 * MAX(MAX_PKT_BURST, MAX_PACKET_FRAG)) 123 124 struct mbuf_table { 125 uint16_t len; 126 struct rte_mbuf *m_table[MBUF_TABLE_SIZE]; 127 }; 128 129 struct rx_queue { 130 struct rte_mempool *direct_pool; 131 struct rte_mempool *indirect_pool; 132 struct rte_lpm *lpm; 133 struct rte_lpm6 *lpm6; 134 uint16_t portid; 135 }; 136 137 #define MAX_RX_QUEUE_PER_LCORE 16 138 #define MAX_TX_QUEUE_PER_PORT 16 139 struct lcore_queue_conf { 140 uint16_t n_rx_queue; 141 uint16_t tx_queue_id[RTE_MAX_ETHPORTS]; 142 struct rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE]; 143 struct mbuf_table tx_mbufs[RTE_MAX_ETHPORTS]; 144 } __rte_cache_aligned; 145 struct lcore_queue_conf lcore_queue_conf[RTE_MAX_LCORE]; 146 147 static struct rte_eth_conf port_conf = { 148 .rxmode = { 149 .max_rx_pkt_len = JUMBO_FRAME_MAX_SIZE, 150 .split_hdr_size = 0, 151 .offloads = (DEV_RX_OFFLOAD_CHECKSUM | 152 DEV_RX_OFFLOAD_SCATTER | 153 DEV_RX_OFFLOAD_JUMBO_FRAME), 154 }, 155 .txmode = { 156 .mq_mode = ETH_MQ_TX_NONE, 157 .offloads = (DEV_TX_OFFLOAD_IPV4_CKSUM | 158 DEV_TX_OFFLOAD_MULTI_SEGS), 159 }, 160 }; 161 162 /* 163 * IPv4 forwarding table 164 */ 165 struct l3fwd_ipv4_route { 166 uint32_t ip; 167 uint8_t depth; 168 uint8_t if_out; 169 }; 170 171 struct l3fwd_ipv4_route l3fwd_ipv4_route_array[] = { 172 {RTE_IPV4(100,10,0,0), 16, 0}, 173 {RTE_IPV4(100,20,0,0), 16, 1}, 174 {RTE_IPV4(100,30,0,0), 16, 2}, 175 {RTE_IPV4(100,40,0,0), 16, 3}, 176 {RTE_IPV4(100,50,0,0), 16, 4}, 177 {RTE_IPV4(100,60,0,0), 16, 5}, 178 {RTE_IPV4(100,70,0,0), 16, 6}, 179 {RTE_IPV4(100,80,0,0), 16, 7}, 180 }; 181 182 /* 183 * IPv6 forwarding table 184 */ 185 186 struct l3fwd_ipv6_route { 187 uint8_t ip[IPV6_ADDR_LEN]; 188 uint8_t depth; 189 uint8_t if_out; 190 }; 191 192 static struct l3fwd_ipv6_route l3fwd_ipv6_route_array[] = { 193 {{1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 0}, 194 {{2,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 1}, 195 {{3,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 2}, 196 {{4,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 3}, 197 {{5,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 4}, 198 {{6,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 5}, 199 {{7,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 6}, 200 {{8,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 7}, 201 }; 202 203 #define LPM_MAX_RULES 1024 204 #define LPM6_MAX_RULES 1024 205 #define LPM6_NUMBER_TBL8S (1 << 16) 206 207 struct rte_lpm6_config lpm6_config = { 208 .max_rules = LPM6_MAX_RULES, 209 .number_tbl8s = LPM6_NUMBER_TBL8S, 210 .flags = 0 211 }; 212 213 static struct rte_mempool *socket_direct_pool[RTE_MAX_NUMA_NODES]; 214 static struct rte_mempool *socket_indirect_pool[RTE_MAX_NUMA_NODES]; 215 static struct rte_lpm *socket_lpm[RTE_MAX_NUMA_NODES]; 216 static struct rte_lpm6 *socket_lpm6[RTE_MAX_NUMA_NODES]; 217 218 /* Send burst of packets on an output interface */ 219 static inline int 220 send_burst(struct lcore_queue_conf *qconf, uint16_t n, uint16_t port) 221 { 222 struct rte_mbuf **m_table; 223 int ret; 224 uint16_t queueid; 225 226 queueid = qconf->tx_queue_id[port]; 227 m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table; 228 229 ret = rte_eth_tx_burst(port, queueid, m_table, n); 230 if (unlikely(ret < n)) { 231 do { 232 rte_pktmbuf_free(m_table[ret]); 233 } while (++ret < n); 234 } 235 236 return 0; 237 } 238 239 static inline void 240 l3fwd_simple_forward(struct rte_mbuf *m, struct lcore_queue_conf *qconf, 241 uint8_t queueid, uint16_t port_in) 242 { 243 struct rx_queue *rxq; 244 uint32_t i, len, next_hop; 245 uint16_t port_out, ether_type; 246 int32_t len2; 247 uint64_t ol_flags; 248 const struct rte_ether_hdr *eth; 249 250 ol_flags = 0; 251 rxq = &qconf->rx_queue_list[queueid]; 252 253 /* by default, send everything back to the source port */ 254 port_out = port_in; 255 256 /* save ether type of the incoming packet */ 257 eth = rte_pktmbuf_mtod(m, const struct rte_ether_hdr *); 258 ether_type = eth->ether_type; 259 260 /* Remove the Ethernet header and trailer from the input packet */ 261 rte_pktmbuf_adj(m, (uint16_t)sizeof(struct rte_ether_hdr)); 262 263 /* Build transmission burst */ 264 len = qconf->tx_mbufs[port_out].len; 265 266 /* if this is an IPv4 packet */ 267 if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) { 268 struct rte_ipv4_hdr *ip_hdr; 269 uint32_t ip_dst; 270 /* Read the lookup key (i.e. ip_dst) from the input packet */ 271 ip_hdr = rte_pktmbuf_mtod(m, struct rte_ipv4_hdr *); 272 ip_dst = rte_be_to_cpu_32(ip_hdr->dst_addr); 273 274 /* Find destination port */ 275 if (rte_lpm_lookup(rxq->lpm, ip_dst, &next_hop) == 0 && 276 (enabled_port_mask & 1 << next_hop) != 0) { 277 port_out = next_hop; 278 279 /* Build transmission burst for new port */ 280 len = qconf->tx_mbufs[port_out].len; 281 } 282 283 /* if we don't need to do any fragmentation */ 284 if (likely (IPV4_MTU_DEFAULT >= m->pkt_len)) { 285 qconf->tx_mbufs[port_out].m_table[len] = m; 286 len2 = 1; 287 } else { 288 len2 = rte_ipv4_fragment_packet(m, 289 &qconf->tx_mbufs[port_out].m_table[len], 290 (uint16_t)(MBUF_TABLE_SIZE - len), 291 IPV4_MTU_DEFAULT, 292 rxq->direct_pool, rxq->indirect_pool); 293 294 /* Free input packet */ 295 rte_pktmbuf_free(m); 296 297 /* request HW to regenerate IPv4 cksum */ 298 ol_flags |= (PKT_TX_IPV4 | PKT_TX_IP_CKSUM); 299 300 /* If we fail to fragment the packet */ 301 if (unlikely (len2 < 0)) 302 return; 303 } 304 } else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) { 305 /* if this is an IPv6 packet */ 306 struct rte_ipv6_hdr *ip_hdr; 307 308 /* Read the lookup key (i.e. ip_dst) from the input packet */ 309 ip_hdr = rte_pktmbuf_mtod(m, struct rte_ipv6_hdr *); 310 311 /* Find destination port */ 312 if (rte_lpm6_lookup(rxq->lpm6, ip_hdr->dst_addr, 313 &next_hop) == 0 && 314 (enabled_port_mask & 1 << next_hop) != 0) { 315 port_out = next_hop; 316 317 /* Build transmission burst for new port */ 318 len = qconf->tx_mbufs[port_out].len; 319 } 320 321 /* if we don't need to do any fragmentation */ 322 if (likely (IPV6_MTU_DEFAULT >= m->pkt_len)) { 323 qconf->tx_mbufs[port_out].m_table[len] = m; 324 len2 = 1; 325 } else { 326 len2 = rte_ipv6_fragment_packet(m, 327 &qconf->tx_mbufs[port_out].m_table[len], 328 (uint16_t)(MBUF_TABLE_SIZE - len), 329 IPV6_MTU_DEFAULT, 330 rxq->direct_pool, rxq->indirect_pool); 331 332 /* Free input packet */ 333 rte_pktmbuf_free(m); 334 335 /* If we fail to fragment the packet */ 336 if (unlikely (len2 < 0)) 337 return; 338 } 339 } 340 /* else, just forward the packet */ 341 else { 342 qconf->tx_mbufs[port_out].m_table[len] = m; 343 len2 = 1; 344 } 345 346 for (i = len; i < len + len2; i ++) { 347 void *d_addr_bytes; 348 349 m = qconf->tx_mbufs[port_out].m_table[i]; 350 struct rte_ether_hdr *eth_hdr = (struct rte_ether_hdr *) 351 rte_pktmbuf_prepend(m, 352 (uint16_t)sizeof(struct rte_ether_hdr)); 353 if (eth_hdr == NULL) { 354 rte_panic("No headroom in mbuf.\n"); 355 } 356 357 m->ol_flags |= ol_flags; 358 m->l2_len = sizeof(struct rte_ether_hdr); 359 360 /* 02:00:00:00:00:xx */ 361 d_addr_bytes = ð_hdr->d_addr.addr_bytes[0]; 362 *((uint64_t *)d_addr_bytes) = 0x000000000002 + 363 ((uint64_t)port_out << 40); 364 365 /* src addr */ 366 rte_ether_addr_copy(&ports_eth_addr[port_out], 367 ð_hdr->s_addr); 368 eth_hdr->ether_type = ether_type; 369 } 370 371 len += len2; 372 373 if (likely(len < MAX_PKT_BURST)) { 374 qconf->tx_mbufs[port_out].len = (uint16_t)len; 375 return; 376 } 377 378 /* Transmit packets */ 379 send_burst(qconf, (uint16_t)len, port_out); 380 qconf->tx_mbufs[port_out].len = 0; 381 } 382 383 /* main processing loop */ 384 static int 385 main_loop(__attribute__((unused)) void *dummy) 386 { 387 struct rte_mbuf *pkts_burst[MAX_PKT_BURST]; 388 unsigned lcore_id; 389 uint64_t prev_tsc, diff_tsc, cur_tsc; 390 int i, j, nb_rx; 391 uint16_t portid; 392 struct lcore_queue_conf *qconf; 393 const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US; 394 395 prev_tsc = 0; 396 397 lcore_id = rte_lcore_id(); 398 qconf = &lcore_queue_conf[lcore_id]; 399 400 if (qconf->n_rx_queue == 0) { 401 RTE_LOG(INFO, IP_FRAG, "lcore %u has nothing to do\n", lcore_id); 402 return 0; 403 } 404 405 RTE_LOG(INFO, IP_FRAG, "entering main loop on lcore %u\n", lcore_id); 406 407 for (i = 0; i < qconf->n_rx_queue; i++) { 408 409 portid = qconf->rx_queue_list[i].portid; 410 RTE_LOG(INFO, IP_FRAG, " -- lcoreid=%u portid=%d\n", lcore_id, 411 portid); 412 } 413 414 while (1) { 415 416 cur_tsc = rte_rdtsc(); 417 418 /* 419 * TX burst queue drain 420 */ 421 diff_tsc = cur_tsc - prev_tsc; 422 if (unlikely(diff_tsc > drain_tsc)) { 423 424 /* 425 * This could be optimized (use queueid instead of 426 * portid), but it is not called so often 427 */ 428 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) { 429 if (qconf->tx_mbufs[portid].len == 0) 430 continue; 431 send_burst(&lcore_queue_conf[lcore_id], 432 qconf->tx_mbufs[portid].len, 433 portid); 434 qconf->tx_mbufs[portid].len = 0; 435 } 436 437 prev_tsc = cur_tsc; 438 } 439 440 /* 441 * Read packet from RX queues 442 */ 443 for (i = 0; i < qconf->n_rx_queue; i++) { 444 445 portid = qconf->rx_queue_list[i].portid; 446 nb_rx = rte_eth_rx_burst(portid, 0, pkts_burst, 447 MAX_PKT_BURST); 448 449 /* Prefetch first packets */ 450 for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) { 451 rte_prefetch0(rte_pktmbuf_mtod( 452 pkts_burst[j], void *)); 453 } 454 455 /* Prefetch and forward already prefetched packets */ 456 for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) { 457 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[ 458 j + PREFETCH_OFFSET], void *)); 459 l3fwd_simple_forward(pkts_burst[j], qconf, i, portid); 460 } 461 462 /* Forward remaining prefetched packets */ 463 for (; j < nb_rx; j++) { 464 l3fwd_simple_forward(pkts_burst[j], qconf, i, portid); 465 } 466 } 467 } 468 } 469 470 /* display usage */ 471 static void 472 print_usage(const char *prgname) 473 { 474 printf("%s [EAL options] -- -p PORTMASK [-q NQ]\n" 475 " -p PORTMASK: hexadecimal bitmask of ports to configure\n" 476 " -q NQ: number of queue (=ports) per lcore (default is 1)\n", 477 prgname); 478 } 479 480 static int 481 parse_portmask(const char *portmask) 482 { 483 char *end = NULL; 484 unsigned long pm; 485 486 /* parse hexadecimal string */ 487 pm = strtoul(portmask, &end, 16); 488 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0')) 489 return -1; 490 491 if (pm == 0) 492 return -1; 493 494 return pm; 495 } 496 497 static int 498 parse_nqueue(const char *q_arg) 499 { 500 char *end = NULL; 501 unsigned long n; 502 503 /* parse hexadecimal string */ 504 n = strtoul(q_arg, &end, 10); 505 if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0')) 506 return -1; 507 if (n == 0) 508 return -1; 509 if (n >= MAX_RX_QUEUE_PER_LCORE) 510 return -1; 511 512 return n; 513 } 514 515 /* Parse the argument given in the command line of the application */ 516 static int 517 parse_args(int argc, char **argv) 518 { 519 int opt, ret; 520 char **argvopt; 521 int option_index; 522 char *prgname = argv[0]; 523 static struct option lgopts[] = { 524 {NULL, 0, 0, 0} 525 }; 526 527 argvopt = argv; 528 529 while ((opt = getopt_long(argc, argvopt, "p:q:", 530 lgopts, &option_index)) != EOF) { 531 532 switch (opt) { 533 /* portmask */ 534 case 'p': 535 enabled_port_mask = parse_portmask(optarg); 536 if (enabled_port_mask < 0) { 537 printf("invalid portmask\n"); 538 print_usage(prgname); 539 return -1; 540 } 541 break; 542 543 /* nqueue */ 544 case 'q': 545 rx_queue_per_lcore = parse_nqueue(optarg); 546 if (rx_queue_per_lcore < 0) { 547 printf("invalid queue number\n"); 548 print_usage(prgname); 549 return -1; 550 } 551 break; 552 553 /* long options */ 554 case 0: 555 print_usage(prgname); 556 return -1; 557 558 default: 559 print_usage(prgname); 560 return -1; 561 } 562 } 563 564 if (enabled_port_mask == 0) { 565 printf("portmask not specified\n"); 566 print_usage(prgname); 567 return -1; 568 } 569 570 if (optind >= 0) 571 argv[optind-1] = prgname; 572 573 ret = optind-1; 574 optind = 1; /* reset getopt lib */ 575 return ret; 576 } 577 578 static void 579 print_ethaddr(const char *name, struct rte_ether_addr *eth_addr) 580 { 581 char buf[RTE_ETHER_ADDR_FMT_SIZE]; 582 rte_ether_format_addr(buf, RTE_ETHER_ADDR_FMT_SIZE, eth_addr); 583 printf("%s%s", name, buf); 584 } 585 586 /* Check the link status of all ports in up to 9s, and print them finally */ 587 static void 588 check_all_ports_link_status(uint32_t port_mask) 589 { 590 #define CHECK_INTERVAL 100 /* 100ms */ 591 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */ 592 uint16_t portid; 593 uint8_t count, all_ports_up, print_flag = 0; 594 struct rte_eth_link link; 595 596 printf("\nChecking link status"); 597 fflush(stdout); 598 for (count = 0; count <= MAX_CHECK_TIME; count++) { 599 all_ports_up = 1; 600 RTE_ETH_FOREACH_DEV(portid) { 601 if ((port_mask & (1 << portid)) == 0) 602 continue; 603 memset(&link, 0, sizeof(link)); 604 rte_eth_link_get_nowait(portid, &link); 605 /* print link status if flag set */ 606 if (print_flag == 1) { 607 if (link.link_status) 608 printf( 609 "Port%d Link Up .Speed %u Mbps - %s\n", 610 portid, link.link_speed, 611 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ? 612 ("full-duplex") : ("half-duplex\n")); 613 else 614 printf("Port %d Link Down\n", portid); 615 continue; 616 } 617 /* clear all_ports_up flag if any link down */ 618 if (link.link_status == ETH_LINK_DOWN) { 619 all_ports_up = 0; 620 break; 621 } 622 } 623 /* after finally printing all link status, get out */ 624 if (print_flag == 1) 625 break; 626 627 if (all_ports_up == 0) { 628 printf("."); 629 fflush(stdout); 630 rte_delay_ms(CHECK_INTERVAL); 631 } 632 633 /* set the print_flag if all ports up or timeout */ 634 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) { 635 print_flag = 1; 636 printf("\ndone\n"); 637 } 638 } 639 } 640 641 /* Check L3 packet type detection capablity of the NIC port */ 642 static int 643 check_ptype(int portid) 644 { 645 int i, ret; 646 int ptype_l3_ipv4 = 0, ptype_l3_ipv6 = 0; 647 uint32_t ptype_mask = RTE_PTYPE_L3_MASK; 648 649 ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, NULL, 0); 650 if (ret <= 0) 651 return 0; 652 653 uint32_t ptypes[ret]; 654 655 ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, ptypes, ret); 656 for (i = 0; i < ret; ++i) { 657 if (ptypes[i] & RTE_PTYPE_L3_IPV4) 658 ptype_l3_ipv4 = 1; 659 if (ptypes[i] & RTE_PTYPE_L3_IPV6) 660 ptype_l3_ipv6 = 1; 661 } 662 663 if (ptype_l3_ipv4 == 0) 664 printf("port %d cannot parse RTE_PTYPE_L3_IPV4\n", portid); 665 666 if (ptype_l3_ipv6 == 0) 667 printf("port %d cannot parse RTE_PTYPE_L3_IPV6\n", portid); 668 669 if (ptype_l3_ipv4 && ptype_l3_ipv6) 670 return 1; 671 672 return 0; 673 674 } 675 676 /* Parse packet type of a packet by SW */ 677 static inline void 678 parse_ptype(struct rte_mbuf *m) 679 { 680 struct rte_ether_hdr *eth_hdr; 681 uint32_t packet_type = RTE_PTYPE_UNKNOWN; 682 uint16_t ether_type; 683 684 eth_hdr = rte_pktmbuf_mtod(m, struct rte_ether_hdr *); 685 ether_type = eth_hdr->ether_type; 686 if (ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4)) 687 packet_type |= RTE_PTYPE_L3_IPV4_EXT_UNKNOWN; 688 else if (ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6)) 689 packet_type |= RTE_PTYPE_L3_IPV6_EXT_UNKNOWN; 690 691 m->packet_type = packet_type; 692 } 693 694 /* callback function to detect packet type for a queue of a port */ 695 static uint16_t 696 cb_parse_ptype(uint16_t port __rte_unused, uint16_t queue __rte_unused, 697 struct rte_mbuf *pkts[], uint16_t nb_pkts, 698 uint16_t max_pkts __rte_unused, 699 void *user_param __rte_unused) 700 { 701 uint16_t i; 702 703 for (i = 0; i < nb_pkts; ++i) 704 parse_ptype(pkts[i]); 705 706 return nb_pkts; 707 } 708 709 static int 710 init_routing_table(void) 711 { 712 struct rte_lpm *lpm; 713 struct rte_lpm6 *lpm6; 714 int socket, ret; 715 unsigned i; 716 717 for (socket = 0; socket < RTE_MAX_NUMA_NODES; socket++) { 718 if (socket_lpm[socket]) { 719 lpm = socket_lpm[socket]; 720 /* populate the LPM table */ 721 for (i = 0; i < RTE_DIM(l3fwd_ipv4_route_array); i++) { 722 ret = rte_lpm_add(lpm, 723 l3fwd_ipv4_route_array[i].ip, 724 l3fwd_ipv4_route_array[i].depth, 725 l3fwd_ipv4_route_array[i].if_out); 726 727 if (ret < 0) { 728 RTE_LOG(ERR, IP_FRAG, "Unable to add entry %i to the l3fwd " 729 "LPM table\n", i); 730 return -1; 731 } 732 733 RTE_LOG(INFO, IP_FRAG, "Socket %i: adding route " IPv4_BYTES_FMT 734 "/%d (port %d)\n", 735 socket, 736 IPv4_BYTES(l3fwd_ipv4_route_array[i].ip), 737 l3fwd_ipv4_route_array[i].depth, 738 l3fwd_ipv4_route_array[i].if_out); 739 } 740 } 741 742 if (socket_lpm6[socket]) { 743 lpm6 = socket_lpm6[socket]; 744 /* populate the LPM6 table */ 745 for (i = 0; i < RTE_DIM(l3fwd_ipv6_route_array); i++) { 746 ret = rte_lpm6_add(lpm6, 747 l3fwd_ipv6_route_array[i].ip, 748 l3fwd_ipv6_route_array[i].depth, 749 l3fwd_ipv6_route_array[i].if_out); 750 751 if (ret < 0) { 752 RTE_LOG(ERR, IP_FRAG, "Unable to add entry %i to the l3fwd " 753 "LPM6 table\n", i); 754 return -1; 755 } 756 757 RTE_LOG(INFO, IP_FRAG, "Socket %i: adding route " IPv6_BYTES_FMT 758 "/%d (port %d)\n", 759 socket, 760 IPv6_BYTES(l3fwd_ipv6_route_array[i].ip), 761 l3fwd_ipv6_route_array[i].depth, 762 l3fwd_ipv6_route_array[i].if_out); 763 } 764 } 765 } 766 return 0; 767 } 768 769 static int 770 init_mem(void) 771 { 772 char buf[PATH_MAX]; 773 struct rte_mempool *mp; 774 struct rte_lpm *lpm; 775 struct rte_lpm6 *lpm6; 776 struct rte_lpm_config lpm_config; 777 int socket; 778 unsigned lcore_id; 779 780 /* traverse through lcores and initialize structures on each socket */ 781 782 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) { 783 784 if (rte_lcore_is_enabled(lcore_id) == 0) 785 continue; 786 787 socket = rte_lcore_to_socket_id(lcore_id); 788 789 if (socket == SOCKET_ID_ANY) 790 socket = 0; 791 792 if (socket_direct_pool[socket] == NULL) { 793 RTE_LOG(INFO, IP_FRAG, "Creating direct mempool on socket %i\n", 794 socket); 795 snprintf(buf, sizeof(buf), "pool_direct_%i", socket); 796 797 mp = rte_pktmbuf_pool_create(buf, NB_MBUF, 32, 798 0, RTE_MBUF_DEFAULT_BUF_SIZE, socket); 799 if (mp == NULL) { 800 RTE_LOG(ERR, IP_FRAG, "Cannot create direct mempool\n"); 801 return -1; 802 } 803 socket_direct_pool[socket] = mp; 804 } 805 806 if (socket_indirect_pool[socket] == NULL) { 807 RTE_LOG(INFO, IP_FRAG, "Creating indirect mempool on socket %i\n", 808 socket); 809 snprintf(buf, sizeof(buf), "pool_indirect_%i", socket); 810 811 mp = rte_pktmbuf_pool_create(buf, NB_MBUF, 32, 0, 0, 812 socket); 813 if (mp == NULL) { 814 RTE_LOG(ERR, IP_FRAG, "Cannot create indirect mempool\n"); 815 return -1; 816 } 817 socket_indirect_pool[socket] = mp; 818 } 819 820 if (socket_lpm[socket] == NULL) { 821 RTE_LOG(INFO, IP_FRAG, "Creating LPM table on socket %i\n", socket); 822 snprintf(buf, sizeof(buf), "IP_FRAG_LPM_%i", socket); 823 824 lpm_config.max_rules = LPM_MAX_RULES; 825 lpm_config.number_tbl8s = 256; 826 lpm_config.flags = 0; 827 828 lpm = rte_lpm_create(buf, socket, &lpm_config); 829 if (lpm == NULL) { 830 RTE_LOG(ERR, IP_FRAG, "Cannot create LPM table\n"); 831 return -1; 832 } 833 socket_lpm[socket] = lpm; 834 } 835 836 if (socket_lpm6[socket] == NULL) { 837 RTE_LOG(INFO, IP_FRAG, "Creating LPM6 table on socket %i\n", socket); 838 snprintf(buf, sizeof(buf), "IP_FRAG_LPM_%i", socket); 839 840 lpm6 = rte_lpm6_create(buf, socket, &lpm6_config); 841 if (lpm6 == NULL) { 842 RTE_LOG(ERR, IP_FRAG, "Cannot create LPM table\n"); 843 return -1; 844 } 845 socket_lpm6[socket] = lpm6; 846 } 847 } 848 849 return 0; 850 } 851 852 int 853 main(int argc, char **argv) 854 { 855 struct lcore_queue_conf *qconf; 856 struct rte_eth_dev_info dev_info; 857 struct rte_eth_txconf *txconf; 858 struct rx_queue *rxq; 859 int socket, ret; 860 uint16_t nb_ports; 861 uint16_t queueid = 0; 862 unsigned lcore_id = 0, rx_lcore_id = 0; 863 uint32_t n_tx_queue, nb_lcores; 864 uint16_t portid; 865 866 /* init EAL */ 867 ret = rte_eal_init(argc, argv); 868 if (ret < 0) 869 rte_exit(EXIT_FAILURE, "rte_eal_init failed"); 870 argc -= ret; 871 argv += ret; 872 873 /* parse application arguments (after the EAL ones) */ 874 ret = parse_args(argc, argv); 875 if (ret < 0) 876 rte_exit(EXIT_FAILURE, "Invalid arguments"); 877 878 nb_ports = rte_eth_dev_count_avail(); 879 if (nb_ports == 0) 880 rte_exit(EXIT_FAILURE, "No ports found!\n"); 881 882 nb_lcores = rte_lcore_count(); 883 884 /* initialize structures (mempools, lpm etc.) */ 885 if (init_mem() < 0) 886 rte_panic("Cannot initialize memory structures!\n"); 887 888 /* check if portmask has non-existent ports */ 889 if (enabled_port_mask & ~(RTE_LEN2MASK(nb_ports, unsigned))) 890 rte_exit(EXIT_FAILURE, "Non-existent ports in portmask!\n"); 891 892 /* initialize all ports */ 893 RTE_ETH_FOREACH_DEV(portid) { 894 struct rte_eth_conf local_port_conf = port_conf; 895 struct rte_eth_rxconf rxq_conf; 896 897 /* skip ports that are not enabled */ 898 if ((enabled_port_mask & (1 << portid)) == 0) { 899 printf("Skipping disabled port %d\n", portid); 900 continue; 901 } 902 903 qconf = &lcore_queue_conf[rx_lcore_id]; 904 905 /* limit the frame size to the maximum supported by NIC */ 906 ret = rte_eth_dev_info_get(portid, &dev_info); 907 if (ret != 0) 908 rte_exit(EXIT_FAILURE, 909 "Error during getting device (port %u) info: %s\n", 910 portid, strerror(-ret)); 911 912 local_port_conf.rxmode.max_rx_pkt_len = RTE_MIN( 913 dev_info.max_rx_pktlen, 914 local_port_conf.rxmode.max_rx_pkt_len); 915 916 /* get the lcore_id for this port */ 917 while (rte_lcore_is_enabled(rx_lcore_id) == 0 || 918 qconf->n_rx_queue == (unsigned)rx_queue_per_lcore) { 919 920 rx_lcore_id ++; 921 if (rx_lcore_id >= RTE_MAX_LCORE) 922 rte_exit(EXIT_FAILURE, "Not enough cores\n"); 923 924 qconf = &lcore_queue_conf[rx_lcore_id]; 925 } 926 927 socket = (int) rte_lcore_to_socket_id(rx_lcore_id); 928 if (socket == SOCKET_ID_ANY) 929 socket = 0; 930 931 rxq = &qconf->rx_queue_list[qconf->n_rx_queue]; 932 rxq->portid = portid; 933 rxq->direct_pool = socket_direct_pool[socket]; 934 rxq->indirect_pool = socket_indirect_pool[socket]; 935 rxq->lpm = socket_lpm[socket]; 936 rxq->lpm6 = socket_lpm6[socket]; 937 qconf->n_rx_queue++; 938 939 /* init port */ 940 printf("Initializing port %d on lcore %u...", portid, 941 rx_lcore_id); 942 fflush(stdout); 943 944 n_tx_queue = nb_lcores; 945 if (n_tx_queue > MAX_TX_QUEUE_PER_PORT) 946 n_tx_queue = MAX_TX_QUEUE_PER_PORT; 947 ret = rte_eth_dev_configure(portid, 1, (uint16_t)n_tx_queue, 948 &local_port_conf); 949 if (ret < 0) { 950 printf("\n"); 951 rte_exit(EXIT_FAILURE, "Cannot configure device: " 952 "err=%d, port=%d\n", 953 ret, portid); 954 } 955 956 /* set the mtu to the maximum received packet size */ 957 ret = rte_eth_dev_set_mtu(portid, 958 local_port_conf.rxmode.max_rx_pkt_len - MTU_OVERHEAD); 959 if (ret < 0) { 960 printf("\n"); 961 rte_exit(EXIT_FAILURE, "Set MTU failed: " 962 "err=%d, port=%d\n", 963 ret, portid); 964 } 965 966 ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd, 967 &nb_txd); 968 if (ret < 0) { 969 printf("\n"); 970 rte_exit(EXIT_FAILURE, "Cannot adjust number of " 971 "descriptors: err=%d, port=%d\n", ret, portid); 972 } 973 974 /* init one RX queue */ 975 rxq_conf = dev_info.default_rxconf; 976 rxq_conf.offloads = local_port_conf.rxmode.offloads; 977 ret = rte_eth_rx_queue_setup(portid, 0, nb_rxd, 978 socket, &rxq_conf, 979 socket_direct_pool[socket]); 980 if (ret < 0) { 981 printf("\n"); 982 rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: " 983 "err=%d, port=%d\n", 984 ret, portid); 985 } 986 987 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]); 988 print_ethaddr(" Address:", &ports_eth_addr[portid]); 989 printf("\n"); 990 991 /* init one TX queue per couple (lcore,port) */ 992 ret = rte_eth_dev_info_get(portid, &dev_info); 993 if (ret != 0) 994 rte_exit(EXIT_FAILURE, 995 "Error during getting device (port %u) info: %s\n", 996 portid, strerror(-ret)); 997 998 queueid = 0; 999 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) { 1000 if (rte_lcore_is_enabled(lcore_id) == 0) 1001 continue; 1002 1003 if (queueid >= dev_info.nb_tx_queues) 1004 break; 1005 1006 socket = (int) rte_lcore_to_socket_id(lcore_id); 1007 printf("txq=%u,%d ", lcore_id, queueid); 1008 fflush(stdout); 1009 1010 txconf = &dev_info.default_txconf; 1011 txconf->offloads = local_port_conf.txmode.offloads; 1012 ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd, 1013 socket, txconf); 1014 if (ret < 0) { 1015 printf("\n"); 1016 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: " 1017 "err=%d, port=%d\n", ret, portid); 1018 } 1019 1020 qconf = &lcore_queue_conf[lcore_id]; 1021 qconf->tx_queue_id[portid] = queueid; 1022 queueid++; 1023 } 1024 1025 printf("\n"); 1026 } 1027 1028 printf("\n"); 1029 1030 /* start ports */ 1031 RTE_ETH_FOREACH_DEV(portid) { 1032 if ((enabled_port_mask & (1 << portid)) == 0) { 1033 continue; 1034 } 1035 /* Start device */ 1036 ret = rte_eth_dev_start(portid); 1037 if (ret < 0) 1038 rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%d\n", 1039 ret, portid); 1040 1041 rte_eth_promiscuous_enable(portid); 1042 1043 if (check_ptype(portid) == 0) { 1044 rte_eth_add_rx_callback(portid, 0, cb_parse_ptype, NULL); 1045 printf("Add Rx callback function to detect L3 packet type by SW :" 1046 " port = %d\n", portid); 1047 } 1048 } 1049 1050 if (init_routing_table() < 0) 1051 rte_exit(EXIT_FAILURE, "Cannot init routing table\n"); 1052 1053 check_all_ports_link_status(enabled_port_mask); 1054 1055 /* launch per-lcore init on every lcore */ 1056 rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER); 1057 RTE_LCORE_FOREACH_SLAVE(lcore_id) { 1058 if (rte_eal_wait_lcore(lcore_id) < 0) 1059 return -1; 1060 } 1061 1062 return 0; 1063 } 1064