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 <string.h> 11 #include <sys/queue.h> 12 #include <stdarg.h> 13 #include <errno.h> 14 #include <getopt.h> 15 16 #include <rte_common.h> 17 #include <rte_byteorder.h> 18 #include <rte_log.h> 19 #include <rte_memory.h> 20 #include <rte_memcpy.h> 21 #include <rte_eal.h> 22 #include <rte_launch.h> 23 #include <rte_atomic.h> 24 #include <rte_cycles.h> 25 #include <rte_prefetch.h> 26 #include <rte_lcore.h> 27 #include <rte_per_lcore.h> 28 #include <rte_branch_prediction.h> 29 #include <rte_interrupts.h> 30 #include <rte_random.h> 31 #include <rte_debug.h> 32 #include <rte_ether.h> 33 #include <rte_ethdev.h> 34 #include <rte_mempool.h> 35 #include <rte_mbuf.h> 36 #include <rte_malloc.h> 37 #include <rte_fbk_hash.h> 38 #include <rte_ip.h> 39 40 #define RTE_LOGTYPE_IPv4_MULTICAST RTE_LOGTYPE_USER1 41 42 #define MAX_PORTS 16 43 44 #define MCAST_CLONE_PORTS 2 45 #define MCAST_CLONE_SEGS 2 46 47 #define PKT_MBUF_DATA_SIZE RTE_MBUF_DEFAULT_BUF_SIZE 48 #define NB_PKT_MBUF 8192 49 50 #define HDR_MBUF_DATA_SIZE (2 * RTE_PKTMBUF_HEADROOM) 51 #define NB_HDR_MBUF (NB_PKT_MBUF * MAX_PORTS) 52 53 #define NB_CLONE_MBUF (NB_PKT_MBUF * MCAST_CLONE_PORTS * MCAST_CLONE_SEGS * 2) 54 55 /* allow max jumbo frame 9.5 KB */ 56 #define JUMBO_FRAME_MAX_SIZE 0x2600 57 58 #define MAX_PKT_BURST 32 59 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */ 60 61 /* Configure how many packets ahead to prefetch, when reading packets */ 62 #define PREFETCH_OFFSET 3 63 64 /* 65 * Construct Ethernet multicast address from IPv4 multicast address. 66 * Citing RFC 1112, section 6.4: 67 * "An IP host group address is mapped to an Ethernet multicast address 68 * by placing the low-order 23-bits of the IP address into the low-order 69 * 23 bits of the Ethernet multicast address 01-00-5E-00-00-00 (hex)." 70 */ 71 #define ETHER_ADDR_FOR_IPV4_MCAST(x) \ 72 (rte_cpu_to_be_64(0x01005e000000ULL | ((x) & 0x7fffff)) >> 16) 73 74 /* 75 * Configurable number of RX/TX ring descriptors 76 */ 77 #define RTE_TEST_RX_DESC_DEFAULT 1024 78 #define RTE_TEST_TX_DESC_DEFAULT 1024 79 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT; 80 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT; 81 82 /* ethernet addresses of ports */ 83 static struct ether_addr ports_eth_addr[MAX_PORTS]; 84 85 /* mask of enabled ports */ 86 static uint32_t enabled_port_mask = 0; 87 88 static uint16_t nb_ports; 89 90 static int rx_queue_per_lcore = 1; 91 92 struct mbuf_table { 93 uint16_t len; 94 struct rte_mbuf *m_table[MAX_PKT_BURST]; 95 }; 96 97 #define MAX_RX_QUEUE_PER_LCORE 16 98 #define MAX_TX_QUEUE_PER_PORT 16 99 struct lcore_queue_conf { 100 uint64_t tx_tsc; 101 uint16_t n_rx_queue; 102 uint8_t rx_queue_list[MAX_RX_QUEUE_PER_LCORE]; 103 uint16_t tx_queue_id[MAX_PORTS]; 104 struct mbuf_table tx_mbufs[MAX_PORTS]; 105 } __rte_cache_aligned; 106 static struct lcore_queue_conf lcore_queue_conf[RTE_MAX_LCORE]; 107 108 static struct rte_eth_conf port_conf = { 109 .rxmode = { 110 .max_rx_pkt_len = JUMBO_FRAME_MAX_SIZE, 111 .split_hdr_size = 0, 112 .offloads = DEV_RX_OFFLOAD_JUMBO_FRAME, 113 }, 114 .txmode = { 115 .mq_mode = ETH_MQ_TX_NONE, 116 .offloads = DEV_TX_OFFLOAD_MULTI_SEGS, 117 }, 118 }; 119 120 static struct rte_mempool *packet_pool, *header_pool, *clone_pool; 121 122 123 /* Multicast */ 124 static struct rte_fbk_hash_params mcast_hash_params = { 125 .name = "MCAST_HASH", 126 .entries = 1024, 127 .entries_per_bucket = 4, 128 .socket_id = 0, 129 .hash_func = NULL, 130 .init_val = 0, 131 }; 132 133 struct rte_fbk_hash_table *mcast_hash = NULL; 134 135 struct mcast_group_params { 136 uint32_t ip; 137 uint16_t port_mask; 138 }; 139 140 static struct mcast_group_params mcast_group_table[] = { 141 {IPv4(224,0,0,101), 0x1}, 142 {IPv4(224,0,0,102), 0x2}, 143 {IPv4(224,0,0,103), 0x3}, 144 {IPv4(224,0,0,104), 0x4}, 145 {IPv4(224,0,0,105), 0x5}, 146 {IPv4(224,0,0,106), 0x6}, 147 {IPv4(224,0,0,107), 0x7}, 148 {IPv4(224,0,0,108), 0x8}, 149 {IPv4(224,0,0,109), 0x9}, 150 {IPv4(224,0,0,110), 0xA}, 151 {IPv4(224,0,0,111), 0xB}, 152 {IPv4(224,0,0,112), 0xC}, 153 {IPv4(224,0,0,113), 0xD}, 154 {IPv4(224,0,0,114), 0xE}, 155 {IPv4(224,0,0,115), 0xF}, 156 }; 157 158 #define N_MCAST_GROUPS \ 159 (sizeof (mcast_group_table) / sizeof (mcast_group_table[0])) 160 161 162 /* Send burst of packets on an output interface */ 163 static void 164 send_burst(struct lcore_queue_conf *qconf, uint16_t port) 165 { 166 struct rte_mbuf **m_table; 167 uint16_t n, queueid; 168 int ret; 169 170 queueid = qconf->tx_queue_id[port]; 171 m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table; 172 n = qconf->tx_mbufs[port].len; 173 174 ret = rte_eth_tx_burst(port, queueid, m_table, n); 175 while (unlikely (ret < n)) { 176 rte_pktmbuf_free(m_table[ret]); 177 ret++; 178 } 179 180 qconf->tx_mbufs[port].len = 0; 181 } 182 183 /* Get number of bits set. */ 184 static inline uint32_t 185 bitcnt(uint32_t v) 186 { 187 uint32_t n; 188 189 for (n = 0; v != 0; v &= v - 1, n++) 190 ; 191 192 return n; 193 } 194 195 /** 196 * Create the output multicast packet based on the given input packet. 197 * There are two approaches for creating outgoing packet, though both 198 * are based on data zero-copy idea, they differ in few details: 199 * First one creates a clone of the input packet, e.g - walk though all 200 * segments of the input packet, and for each of them create a new packet 201 * mbuf and attach that new mbuf to the segment (refer to rte_pktmbuf_clone() 202 * for more details). Then new mbuf is allocated for the packet header 203 * and is prepended to the 'clone' mbuf. 204 * Second approach doesn't make a clone, it just increment refcnt for all 205 * input packet segments. Then it allocates new mbuf for the packet header 206 * and prepends it to the input packet. 207 * Basically first approach reuses only input packet's data, but creates 208 * it's own copy of packet's metadata. Second approach reuses both input's 209 * packet data and metadata. 210 * The advantage of first approach - is that each outgoing packet has it's 211 * own copy of metadata, so we can safely modify data pointer of the 212 * input packet. That allows us to skip creation if the output packet for 213 * the last destination port, but instead modify input packet's header inplace, 214 * e.g: for N destination ports we need to invoke mcast_out_pkt (N-1) times. 215 * The advantage of second approach - less work for each outgoing packet, 216 * e.g: we skip "clone" operation completely. Though it comes with a price - 217 * input packet's metadata has to be intact. So for N destination ports we 218 * need to invoke mcast_out_pkt N times. 219 * So for small number of outgoing ports (and segments in the input packet) 220 * first approach will be faster. 221 * As number of outgoing ports (and/or input segments) will grow, 222 * second way will become more preferable. 223 * 224 * @param pkt 225 * Input packet mbuf. 226 * @param use_clone 227 * Control which of the two approaches described above should be used: 228 * - 0 - use second approach: 229 * Don't "clone" input packet. 230 * Prepend new header directly to the input packet 231 * - 1 - use first approach: 232 * Make a "clone" of input packet first. 233 * Prepend new header to the clone of the input packet 234 * @return 235 * - The pointer to the new outgoing packet. 236 * - NULL if operation failed. 237 */ 238 static inline struct rte_mbuf * 239 mcast_out_pkt(struct rte_mbuf *pkt, int use_clone) 240 { 241 struct rte_mbuf *hdr; 242 243 /* Create new mbuf for the header. */ 244 if (unlikely ((hdr = rte_pktmbuf_alloc(header_pool)) == NULL)) 245 return NULL; 246 247 /* If requested, then make a new clone packet. */ 248 if (use_clone != 0 && 249 unlikely ((pkt = rte_pktmbuf_clone(pkt, clone_pool)) == NULL)) { 250 rte_pktmbuf_free(hdr); 251 return NULL; 252 } 253 254 /* prepend new header */ 255 hdr->next = pkt; 256 257 258 /* update header's fields */ 259 hdr->pkt_len = (uint16_t)(hdr->data_len + pkt->pkt_len); 260 hdr->nb_segs = pkt->nb_segs + 1; 261 262 /* copy metadata from source packet*/ 263 hdr->port = pkt->port; 264 hdr->vlan_tci = pkt->vlan_tci; 265 hdr->vlan_tci_outer = pkt->vlan_tci_outer; 266 hdr->tx_offload = pkt->tx_offload; 267 hdr->hash = pkt->hash; 268 269 hdr->ol_flags = pkt->ol_flags; 270 271 __rte_mbuf_sanity_check(hdr, 1); 272 return hdr; 273 } 274 275 /* 276 * Write new Ethernet header to the outgoing packet, 277 * and put it into the outgoing queue for the given port. 278 */ 279 static inline void 280 mcast_send_pkt(struct rte_mbuf *pkt, struct ether_addr *dest_addr, 281 struct lcore_queue_conf *qconf, uint16_t port) 282 { 283 struct ether_hdr *ethdr; 284 uint16_t len; 285 286 /* Construct Ethernet header. */ 287 ethdr = (struct ether_hdr *)rte_pktmbuf_prepend(pkt, (uint16_t)sizeof(*ethdr)); 288 RTE_ASSERT(ethdr != NULL); 289 290 ether_addr_copy(dest_addr, ðdr->d_addr); 291 ether_addr_copy(&ports_eth_addr[port], ðdr->s_addr); 292 ethdr->ether_type = rte_be_to_cpu_16(ETHER_TYPE_IPv4); 293 294 /* Put new packet into the output queue */ 295 len = qconf->tx_mbufs[port].len; 296 qconf->tx_mbufs[port].m_table[len] = pkt; 297 qconf->tx_mbufs[port].len = ++len; 298 299 /* Transmit packets */ 300 if (unlikely(MAX_PKT_BURST == len)) 301 send_burst(qconf, port); 302 } 303 304 /* Multicast forward of the input packet */ 305 static inline void 306 mcast_forward(struct rte_mbuf *m, struct lcore_queue_conf *qconf) 307 { 308 struct rte_mbuf *mc; 309 struct ipv4_hdr *iphdr; 310 uint32_t dest_addr, port_mask, port_num, use_clone; 311 int32_t hash; 312 uint16_t port; 313 union { 314 uint64_t as_int; 315 struct ether_addr as_addr; 316 } dst_eth_addr; 317 318 /* Remove the Ethernet header from the input packet */ 319 iphdr = (struct ipv4_hdr *)rte_pktmbuf_adj(m, (uint16_t)sizeof(struct ether_hdr)); 320 RTE_ASSERT(iphdr != NULL); 321 322 dest_addr = rte_be_to_cpu_32(iphdr->dst_addr); 323 324 /* 325 * Check that it is a valid multicast address and 326 * we have some active ports assigned to it. 327 */ 328 if(!IS_IPV4_MCAST(dest_addr) || 329 (hash = rte_fbk_hash_lookup(mcast_hash, dest_addr)) <= 0 || 330 (port_mask = hash & enabled_port_mask) == 0) { 331 rte_pktmbuf_free(m); 332 return; 333 } 334 335 /* Calculate number of destination ports. */ 336 port_num = bitcnt(port_mask); 337 338 /* Should we use rte_pktmbuf_clone() or not. */ 339 use_clone = (port_num <= MCAST_CLONE_PORTS && 340 m->nb_segs <= MCAST_CLONE_SEGS); 341 342 /* Mark all packet's segments as referenced port_num times */ 343 if (use_clone == 0) 344 rte_pktmbuf_refcnt_update(m, (uint16_t)port_num); 345 346 /* construct destination ethernet address */ 347 dst_eth_addr.as_int = ETHER_ADDR_FOR_IPV4_MCAST(dest_addr); 348 349 for (port = 0; use_clone != port_mask; port_mask >>= 1, port++) { 350 351 /* Prepare output packet and send it out. */ 352 if ((port_mask & 1) != 0) { 353 if (likely ((mc = mcast_out_pkt(m, use_clone)) != NULL)) 354 mcast_send_pkt(mc, &dst_eth_addr.as_addr, 355 qconf, port); 356 else if (use_clone == 0) 357 rte_pktmbuf_free(m); 358 } 359 } 360 361 /* 362 * If we making clone packets, then, for the last destination port, 363 * we can overwrite input packet's metadata. 364 */ 365 if (use_clone != 0) 366 mcast_send_pkt(m, &dst_eth_addr.as_addr, qconf, port); 367 else 368 rte_pktmbuf_free(m); 369 } 370 371 /* Send burst of outgoing packet, if timeout expires. */ 372 static inline void 373 send_timeout_burst(struct lcore_queue_conf *qconf) 374 { 375 uint64_t cur_tsc; 376 uint16_t portid; 377 const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US; 378 379 cur_tsc = rte_rdtsc(); 380 if (likely (cur_tsc < qconf->tx_tsc + drain_tsc)) 381 return; 382 383 for (portid = 0; portid < MAX_PORTS; portid++) { 384 if (qconf->tx_mbufs[portid].len != 0) 385 send_burst(qconf, portid); 386 } 387 qconf->tx_tsc = cur_tsc; 388 } 389 390 /* main processing loop */ 391 static int 392 main_loop(__rte_unused void *dummy) 393 { 394 struct rte_mbuf *pkts_burst[MAX_PKT_BURST]; 395 unsigned lcore_id; 396 int i, j, nb_rx; 397 uint16_t portid; 398 struct lcore_queue_conf *qconf; 399 400 lcore_id = rte_lcore_id(); 401 qconf = &lcore_queue_conf[lcore_id]; 402 403 404 if (qconf->n_rx_queue == 0) { 405 RTE_LOG(INFO, IPv4_MULTICAST, "lcore %u has nothing to do\n", 406 lcore_id); 407 return 0; 408 } 409 410 RTE_LOG(INFO, IPv4_MULTICAST, "entering main loop on lcore %u\n", 411 lcore_id); 412 413 for (i = 0; i < qconf->n_rx_queue; i++) { 414 415 portid = qconf->rx_queue_list[i]; 416 RTE_LOG(INFO, IPv4_MULTICAST, " -- lcoreid=%u portid=%d\n", 417 lcore_id, portid); 418 } 419 420 while (1) { 421 422 /* 423 * Read packet from RX queues 424 */ 425 for (i = 0; i < qconf->n_rx_queue; i++) { 426 427 portid = qconf->rx_queue_list[i]; 428 nb_rx = rte_eth_rx_burst(portid, 0, pkts_burst, 429 MAX_PKT_BURST); 430 431 /* Prefetch first packets */ 432 for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) { 433 rte_prefetch0(rte_pktmbuf_mtod( 434 pkts_burst[j], void *)); 435 } 436 437 /* Prefetch and forward already prefetched packets */ 438 for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) { 439 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[ 440 j + PREFETCH_OFFSET], void *)); 441 mcast_forward(pkts_burst[j], qconf); 442 } 443 444 /* Forward remaining prefetched packets */ 445 for (; j < nb_rx; j++) { 446 mcast_forward(pkts_burst[j], qconf); 447 } 448 } 449 450 /* Send out packets from TX queues */ 451 send_timeout_burst(qconf); 452 } 453 } 454 455 /* display usage */ 456 static void 457 print_usage(const char *prgname) 458 { 459 printf("%s [EAL options] -- -p PORTMASK [-q NQ]\n" 460 " -p PORTMASK: hexadecimal bitmask of ports to configure\n" 461 " -q NQ: number of queue (=ports) per lcore (default is 1)\n", 462 prgname); 463 } 464 465 static uint32_t 466 parse_portmask(const char *portmask) 467 { 468 char *end = NULL; 469 unsigned long pm; 470 471 /* parse hexadecimal string */ 472 pm = strtoul(portmask, &end, 16); 473 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0')) 474 return 0; 475 476 return (uint32_t)pm; 477 } 478 479 static int 480 parse_nqueue(const char *q_arg) 481 { 482 char *end = NULL; 483 unsigned long n; 484 485 /* parse numerical string */ 486 errno = 0; 487 n = strtoul(q_arg, &end, 0); 488 if (errno != 0 || end == NULL || *end != '\0' || 489 n == 0 || n >= MAX_RX_QUEUE_PER_LCORE) 490 return -1; 491 492 return n; 493 } 494 495 /* Parse the argument given in the command line of the application */ 496 static int 497 parse_args(int argc, char **argv) 498 { 499 int opt, ret; 500 char **argvopt; 501 int option_index; 502 char *prgname = argv[0]; 503 static struct option lgopts[] = { 504 {NULL, 0, 0, 0} 505 }; 506 507 argvopt = argv; 508 509 while ((opt = getopt_long(argc, argvopt, "p:q:", 510 lgopts, &option_index)) != EOF) { 511 512 switch (opt) { 513 /* portmask */ 514 case 'p': 515 enabled_port_mask = parse_portmask(optarg); 516 if (enabled_port_mask == 0) { 517 printf("invalid portmask\n"); 518 print_usage(prgname); 519 return -1; 520 } 521 break; 522 523 /* nqueue */ 524 case 'q': 525 rx_queue_per_lcore = parse_nqueue(optarg); 526 if (rx_queue_per_lcore < 0) { 527 printf("invalid queue number\n"); 528 print_usage(prgname); 529 return -1; 530 } 531 break; 532 533 default: 534 print_usage(prgname); 535 return -1; 536 } 537 } 538 539 if (optind >= 0) 540 argv[optind-1] = prgname; 541 542 ret = optind-1; 543 optind = 1; /* reset getopt lib */ 544 return ret; 545 } 546 547 static void 548 print_ethaddr(const char *name, struct ether_addr *eth_addr) 549 { 550 char buf[ETHER_ADDR_FMT_SIZE]; 551 ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr); 552 printf("%s%s", name, buf); 553 } 554 555 static int 556 init_mcast_hash(void) 557 { 558 uint32_t i; 559 560 mcast_hash_params.socket_id = rte_socket_id(); 561 mcast_hash = rte_fbk_hash_create(&mcast_hash_params); 562 if (mcast_hash == NULL){ 563 return -1; 564 } 565 566 for (i = 0; i < N_MCAST_GROUPS; i ++){ 567 if (rte_fbk_hash_add_key(mcast_hash, 568 mcast_group_table[i].ip, 569 mcast_group_table[i].port_mask) < 0) { 570 return -1; 571 } 572 } 573 574 return 0; 575 } 576 577 /* Check the link status of all ports in up to 9s, and print them finally */ 578 static void 579 check_all_ports_link_status(uint32_t port_mask) 580 { 581 #define CHECK_INTERVAL 100 /* 100ms */ 582 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */ 583 uint16_t portid; 584 uint8_t count, all_ports_up, print_flag = 0; 585 struct rte_eth_link link; 586 587 printf("\nChecking link status"); 588 fflush(stdout); 589 for (count = 0; count <= MAX_CHECK_TIME; count++) { 590 all_ports_up = 1; 591 RTE_ETH_FOREACH_DEV(portid) { 592 if ((port_mask & (1 << portid)) == 0) 593 continue; 594 memset(&link, 0, sizeof(link)); 595 rte_eth_link_get_nowait(portid, &link); 596 /* print link status if flag set */ 597 if (print_flag == 1) { 598 if (link.link_status) 599 printf( 600 "Port%d Link Up. Speed %u Mbps - %s\n", 601 portid, link.link_speed, 602 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ? 603 ("full-duplex") : ("half-duplex\n")); 604 else 605 printf("Port %d Link Down\n", portid); 606 continue; 607 } 608 /* clear all_ports_up flag if any link down */ 609 if (link.link_status == ETH_LINK_DOWN) { 610 all_ports_up = 0; 611 break; 612 } 613 } 614 /* after finally printing all link status, get out */ 615 if (print_flag == 1) 616 break; 617 618 if (all_ports_up == 0) { 619 printf("."); 620 fflush(stdout); 621 rte_delay_ms(CHECK_INTERVAL); 622 } 623 624 /* set the print_flag if all ports up or timeout */ 625 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) { 626 print_flag = 1; 627 printf("done\n"); 628 } 629 } 630 } 631 632 int 633 main(int argc, char **argv) 634 { 635 struct lcore_queue_conf *qconf; 636 struct rte_eth_dev_info dev_info; 637 struct rte_eth_txconf *txconf; 638 int ret; 639 uint16_t queueid; 640 unsigned lcore_id = 0, rx_lcore_id = 0; 641 uint32_t n_tx_queue, nb_lcores; 642 uint16_t portid; 643 644 /* init EAL */ 645 ret = rte_eal_init(argc, argv); 646 if (ret < 0) 647 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n"); 648 argc -= ret; 649 argv += ret; 650 651 /* parse application arguments (after the EAL ones) */ 652 ret = parse_args(argc, argv); 653 if (ret < 0) 654 rte_exit(EXIT_FAILURE, "Invalid IPV4_MULTICAST parameters\n"); 655 656 /* create the mbuf pools */ 657 packet_pool = rte_pktmbuf_pool_create("packet_pool", NB_PKT_MBUF, 32, 658 0, PKT_MBUF_DATA_SIZE, rte_socket_id()); 659 660 if (packet_pool == NULL) 661 rte_exit(EXIT_FAILURE, "Cannot init packet mbuf pool\n"); 662 663 header_pool = rte_pktmbuf_pool_create("header_pool", NB_HDR_MBUF, 32, 664 0, HDR_MBUF_DATA_SIZE, rte_socket_id()); 665 666 if (header_pool == NULL) 667 rte_exit(EXIT_FAILURE, "Cannot init header mbuf pool\n"); 668 669 clone_pool = rte_pktmbuf_pool_create("clone_pool", NB_CLONE_MBUF, 32, 670 0, 0, rte_socket_id()); 671 672 if (clone_pool == NULL) 673 rte_exit(EXIT_FAILURE, "Cannot init clone mbuf pool\n"); 674 675 nb_ports = rte_eth_dev_count_avail(); 676 if (nb_ports == 0) 677 rte_exit(EXIT_FAILURE, "No physical ports!\n"); 678 if (nb_ports > MAX_PORTS) 679 nb_ports = MAX_PORTS; 680 681 nb_lcores = rte_lcore_count(); 682 683 /* initialize all ports */ 684 RTE_ETH_FOREACH_DEV(portid) { 685 struct rte_eth_rxconf rxq_conf; 686 struct rte_eth_conf local_port_conf = port_conf; 687 688 /* skip ports that are not enabled */ 689 if ((enabled_port_mask & (1 << portid)) == 0) { 690 printf("Skipping disabled port %d\n", portid); 691 continue; 692 } 693 694 qconf = &lcore_queue_conf[rx_lcore_id]; 695 696 /* limit the frame size to the maximum supported by NIC */ 697 rte_eth_dev_info_get(portid, &dev_info); 698 local_port_conf.rxmode.max_rx_pkt_len = RTE_MIN( 699 dev_info.max_rx_pktlen, 700 local_port_conf.rxmode.max_rx_pkt_len); 701 702 /* get the lcore_id for this port */ 703 while (rte_lcore_is_enabled(rx_lcore_id) == 0 || 704 qconf->n_rx_queue == (unsigned)rx_queue_per_lcore) { 705 706 rx_lcore_id ++; 707 qconf = &lcore_queue_conf[rx_lcore_id]; 708 709 if (rx_lcore_id >= RTE_MAX_LCORE) 710 rte_exit(EXIT_FAILURE, "Not enough cores\n"); 711 } 712 qconf->rx_queue_list[qconf->n_rx_queue] = portid; 713 qconf->n_rx_queue++; 714 715 /* init port */ 716 printf("Initializing port %d on lcore %u... ", portid, 717 rx_lcore_id); 718 fflush(stdout); 719 720 n_tx_queue = nb_lcores; 721 if (n_tx_queue > MAX_TX_QUEUE_PER_PORT) 722 n_tx_queue = MAX_TX_QUEUE_PER_PORT; 723 724 ret = rte_eth_dev_configure(portid, 1, (uint16_t)n_tx_queue, 725 &local_port_conf); 726 if (ret < 0) 727 rte_exit(EXIT_FAILURE, "Cannot configure device: err=%d, port=%d\n", 728 ret, portid); 729 730 ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd, 731 &nb_txd); 732 if (ret < 0) 733 rte_exit(EXIT_FAILURE, 734 "Cannot adjust number of descriptors: err=%d, port=%d\n", 735 ret, portid); 736 737 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]); 738 print_ethaddr(" Address:", &ports_eth_addr[portid]); 739 printf(", "); 740 741 /* init one RX queue */ 742 queueid = 0; 743 printf("rxq=%hu ", queueid); 744 fflush(stdout); 745 rxq_conf = dev_info.default_rxconf; 746 rxq_conf.offloads = local_port_conf.rxmode.offloads; 747 ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd, 748 rte_eth_dev_socket_id(portid), 749 &rxq_conf, 750 packet_pool); 751 if (ret < 0) 752 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: err=%d, port=%d\n", 753 ret, portid); 754 755 /* init one TX queue per couple (lcore,port) */ 756 queueid = 0; 757 758 RTE_LCORE_FOREACH(lcore_id) { 759 if (rte_lcore_is_enabled(lcore_id) == 0) 760 continue; 761 printf("txq=%u,%hu ", lcore_id, queueid); 762 fflush(stdout); 763 764 txconf = &dev_info.default_txconf; 765 txconf->offloads = local_port_conf.txmode.offloads; 766 ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd, 767 rte_lcore_to_socket_id(lcore_id), txconf); 768 if (ret < 0) 769 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: err=%d, " 770 "port=%d\n", ret, portid); 771 772 qconf = &lcore_queue_conf[lcore_id]; 773 qconf->tx_queue_id[portid] = queueid; 774 queueid++; 775 } 776 777 /* Start device */ 778 ret = rte_eth_dev_start(portid); 779 if (ret < 0) 780 rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%d\n", 781 ret, portid); 782 783 printf("done:\n"); 784 } 785 786 check_all_ports_link_status(enabled_port_mask); 787 788 /* initialize the multicast hash */ 789 int retval = init_mcast_hash(); 790 if (retval != 0) 791 rte_exit(EXIT_FAILURE, "Cannot build the multicast hash\n"); 792 793 /* launch per-lcore init on every lcore */ 794 rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER); 795 RTE_LCORE_FOREACH_SLAVE(lcore_id) { 796 if (rte_eal_wait_lcore(lcore_id) < 0) 797 return -1; 798 } 799 800 return 0; 801 } 802