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 #include <signal.h> 16 #include <sys/param.h> 17 18 #include <rte_common.h> 19 #include <rte_byteorder.h> 20 #include <rte_log.h> 21 #include <rte_memory.h> 22 #include <rte_memcpy.h> 23 #include <rte_eal.h> 24 #include <rte_launch.h> 25 #include <rte_atomic.h> 26 #include <rte_cycles.h> 27 #include <rte_prefetch.h> 28 #include <rte_lcore.h> 29 #include <rte_per_lcore.h> 30 #include <rte_branch_prediction.h> 31 #include <rte_interrupts.h> 32 #include <rte_random.h> 33 #include <rte_debug.h> 34 #include <rte_ether.h> 35 #include <rte_ethdev.h> 36 #include <rte_mempool.h> 37 #include <rte_mbuf.h> 38 #include <rte_malloc.h> 39 #include <rte_ip.h> 40 #include <rte_tcp.h> 41 #include <rte_udp.h> 42 #include <rte_string_fns.h> 43 #include <rte_lpm.h> 44 #include <rte_lpm6.h> 45 46 #include <rte_ip_frag.h> 47 48 #define MAX_PKT_BURST 32 49 50 51 #define RTE_LOGTYPE_IP_RSMBL RTE_LOGTYPE_USER1 52 53 #define MAX_JUMBO_PKT_LEN 9600 54 55 #define BUF_SIZE RTE_MBUF_DEFAULT_DATAROOM 56 #define MBUF_DATA_SIZE RTE_MBUF_DEFAULT_BUF_SIZE 57 58 #define NB_MBUF 8192 59 #define MEMPOOL_CACHE_SIZE 256 60 61 /* allow max jumbo frame 9.5 KB */ 62 #define JUMBO_FRAME_MAX_SIZE 0x2600 63 64 #define MAX_FLOW_NUM UINT16_MAX 65 #define MIN_FLOW_NUM 1 66 #define DEF_FLOW_NUM 0x1000 67 68 /* TTL numbers are in ms. */ 69 #define MAX_FLOW_TTL (3600 * MS_PER_S) 70 #define MIN_FLOW_TTL 1 71 #define DEF_FLOW_TTL MS_PER_S 72 73 #define MAX_FRAG_NUM RTE_LIBRTE_IP_FRAG_MAX_FRAG 74 75 /* Should be power of two. */ 76 #define IP_FRAG_TBL_BUCKET_ENTRIES 16 77 78 static uint32_t max_flow_num = DEF_FLOW_NUM; 79 static uint32_t max_flow_ttl = DEF_FLOW_TTL; 80 81 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */ 82 83 #define NB_SOCKETS 8 84 85 /* Configure how many packets ahead to prefetch, when reading packets */ 86 #define PREFETCH_OFFSET 3 87 88 /* 89 * Configurable number of RX/TX ring descriptors 90 */ 91 #define RTE_TEST_RX_DESC_DEFAULT 1024 92 #define RTE_TEST_TX_DESC_DEFAULT 1024 93 94 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT; 95 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT; 96 97 /* ethernet addresses of ports */ 98 static struct rte_ether_addr ports_eth_addr[RTE_MAX_ETHPORTS]; 99 100 #ifndef IPv4_BYTES 101 #define IPv4_BYTES_FMT "%" PRIu8 ".%" PRIu8 ".%" PRIu8 ".%" PRIu8 102 #define IPv4_BYTES(addr) \ 103 (uint8_t) (((addr) >> 24) & 0xFF),\ 104 (uint8_t) (((addr) >> 16) & 0xFF),\ 105 (uint8_t) (((addr) >> 8) & 0xFF),\ 106 (uint8_t) ((addr) & 0xFF) 107 #endif 108 109 #ifndef IPv6_BYTES 110 #define IPv6_BYTES_FMT "%02x%02x:%02x%02x:%02x%02x:%02x%02x:"\ 111 "%02x%02x:%02x%02x:%02x%02x:%02x%02x" 112 #define IPv6_BYTES(addr) \ 113 addr[0], addr[1], addr[2], addr[3], \ 114 addr[4], addr[5], addr[6], addr[7], \ 115 addr[8], addr[9], addr[10], addr[11],\ 116 addr[12], addr[13],addr[14], addr[15] 117 #endif 118 119 #define IPV6_ADDR_LEN 16 120 121 /* mask of enabled ports */ 122 static uint32_t enabled_port_mask = 0; 123 124 static int rx_queue_per_lcore = 1; 125 126 struct mbuf_table { 127 uint32_t len; 128 uint32_t head; 129 uint32_t tail; 130 struct rte_mbuf *m_table[0]; 131 }; 132 133 struct rx_queue { 134 struct rte_ip_frag_tbl *frag_tbl; 135 struct rte_mempool *pool; 136 struct rte_lpm *lpm; 137 struct rte_lpm6 *lpm6; 138 uint16_t portid; 139 }; 140 141 struct tx_lcore_stat { 142 uint64_t call; 143 uint64_t drop; 144 uint64_t queue; 145 uint64_t send; 146 }; 147 148 #define MAX_RX_QUEUE_PER_LCORE 16 149 #define MAX_TX_QUEUE_PER_PORT 16 150 #define MAX_RX_QUEUE_PER_PORT 128 151 152 struct lcore_queue_conf { 153 uint16_t n_rx_queue; 154 struct rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE]; 155 uint16_t tx_queue_id[RTE_MAX_ETHPORTS]; 156 struct rte_ip_frag_death_row death_row; 157 struct mbuf_table *tx_mbufs[RTE_MAX_ETHPORTS]; 158 struct tx_lcore_stat tx_stat; 159 } __rte_cache_aligned; 160 static struct lcore_queue_conf lcore_queue_conf[RTE_MAX_LCORE]; 161 162 static struct rte_eth_conf port_conf = { 163 .rxmode = { 164 .mq_mode = ETH_MQ_RX_RSS, 165 .max_rx_pkt_len = JUMBO_FRAME_MAX_SIZE, 166 .split_hdr_size = 0, 167 .offloads = (DEV_RX_OFFLOAD_CHECKSUM | 168 DEV_RX_OFFLOAD_JUMBO_FRAME), 169 }, 170 .rx_adv_conf = { 171 .rss_conf = { 172 .rss_key = NULL, 173 .rss_hf = ETH_RSS_IP, 174 }, 175 }, 176 .txmode = { 177 .mq_mode = ETH_MQ_TX_NONE, 178 .offloads = (DEV_TX_OFFLOAD_IPV4_CKSUM | 179 DEV_TX_OFFLOAD_MULTI_SEGS), 180 }, 181 }; 182 183 /* 184 * IPv4 forwarding table 185 */ 186 struct l3fwd_ipv4_route { 187 uint32_t ip; 188 uint8_t depth; 189 uint8_t if_out; 190 }; 191 192 struct l3fwd_ipv4_route l3fwd_ipv4_route_array[] = { 193 {RTE_IPV4(100,10,0,0), 16, 0}, 194 {RTE_IPV4(100,20,0,0), 16, 1}, 195 {RTE_IPV4(100,30,0,0), 16, 2}, 196 {RTE_IPV4(100,40,0,0), 16, 3}, 197 {RTE_IPV4(100,50,0,0), 16, 4}, 198 {RTE_IPV4(100,60,0,0), 16, 5}, 199 {RTE_IPV4(100,70,0,0), 16, 6}, 200 {RTE_IPV4(100,80,0,0), 16, 7}, 201 }; 202 203 /* 204 * IPv6 forwarding table 205 */ 206 207 struct l3fwd_ipv6_route { 208 uint8_t ip[IPV6_ADDR_LEN]; 209 uint8_t depth; 210 uint8_t if_out; 211 }; 212 213 static struct l3fwd_ipv6_route l3fwd_ipv6_route_array[] = { 214 {{1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 0}, 215 {{2,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 1}, 216 {{3,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 2}, 217 {{4,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 3}, 218 {{5,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 4}, 219 {{6,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 5}, 220 {{7,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 6}, 221 {{8,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 7}, 222 }; 223 224 #define LPM_MAX_RULES 1024 225 #define LPM6_MAX_RULES 1024 226 #define LPM6_NUMBER_TBL8S (1 << 16) 227 228 struct rte_lpm6_config lpm6_config = { 229 .max_rules = LPM6_MAX_RULES, 230 .number_tbl8s = LPM6_NUMBER_TBL8S, 231 .flags = 0 232 }; 233 234 static struct rte_lpm *socket_lpm[RTE_MAX_NUMA_NODES]; 235 static struct rte_lpm6 *socket_lpm6[RTE_MAX_NUMA_NODES]; 236 237 #ifdef RTE_LIBRTE_IP_FRAG_TBL_STAT 238 #define TX_LCORE_STAT_UPDATE(s, f, v) ((s)->f += (v)) 239 #else 240 #define TX_LCORE_STAT_UPDATE(s, f, v) do {} while (0) 241 #endif /* RTE_LIBRTE_IP_FRAG_TBL_STAT */ 242 243 /* 244 * If number of queued packets reached given threahold, then 245 * send burst of packets on an output interface. 246 */ 247 static inline uint32_t 248 send_burst(struct lcore_queue_conf *qconf, uint32_t thresh, uint16_t port) 249 { 250 uint32_t fill, len, k, n; 251 struct mbuf_table *txmb; 252 253 txmb = qconf->tx_mbufs[port]; 254 len = txmb->len; 255 256 if ((int32_t)(fill = txmb->head - txmb->tail) < 0) 257 fill += len; 258 259 if (fill >= thresh) { 260 n = RTE_MIN(len - txmb->tail, fill); 261 262 k = rte_eth_tx_burst(port, qconf->tx_queue_id[port], 263 txmb->m_table + txmb->tail, (uint16_t)n); 264 265 TX_LCORE_STAT_UPDATE(&qconf->tx_stat, call, 1); 266 TX_LCORE_STAT_UPDATE(&qconf->tx_stat, send, k); 267 268 fill -= k; 269 if ((txmb->tail += k) == len) 270 txmb->tail = 0; 271 } 272 273 return fill; 274 } 275 276 /* Enqueue a single packet, and send burst if queue is filled */ 277 static inline int 278 send_single_packet(struct rte_mbuf *m, uint16_t port) 279 { 280 uint32_t fill, lcore_id, len; 281 struct lcore_queue_conf *qconf; 282 struct mbuf_table *txmb; 283 284 lcore_id = rte_lcore_id(); 285 qconf = &lcore_queue_conf[lcore_id]; 286 287 txmb = qconf->tx_mbufs[port]; 288 len = txmb->len; 289 290 fill = send_burst(qconf, MAX_PKT_BURST, port); 291 292 if (fill == len - 1) { 293 TX_LCORE_STAT_UPDATE(&qconf->tx_stat, drop, 1); 294 rte_pktmbuf_free(txmb->m_table[txmb->tail]); 295 if (++txmb->tail == len) 296 txmb->tail = 0; 297 } 298 299 TX_LCORE_STAT_UPDATE(&qconf->tx_stat, queue, 1); 300 txmb->m_table[txmb->head] = m; 301 if(++txmb->head == len) 302 txmb->head = 0; 303 304 return 0; 305 } 306 307 static inline void 308 reassemble(struct rte_mbuf *m, uint16_t portid, uint32_t queue, 309 struct lcore_queue_conf *qconf, uint64_t tms) 310 { 311 struct rte_ether_hdr *eth_hdr; 312 struct rte_ip_frag_tbl *tbl; 313 struct rte_ip_frag_death_row *dr; 314 struct rx_queue *rxq; 315 void *d_addr_bytes; 316 uint32_t next_hop; 317 uint16_t dst_port; 318 319 rxq = &qconf->rx_queue_list[queue]; 320 321 eth_hdr = rte_pktmbuf_mtod(m, struct rte_ether_hdr *); 322 323 dst_port = portid; 324 325 /* if packet is IPv4 */ 326 if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) { 327 struct rte_ipv4_hdr *ip_hdr; 328 uint32_t ip_dst; 329 330 ip_hdr = (struct rte_ipv4_hdr *)(eth_hdr + 1); 331 332 /* if it is a fragmented packet, then try to reassemble. */ 333 if (rte_ipv4_frag_pkt_is_fragmented(ip_hdr)) { 334 struct rte_mbuf *mo; 335 336 tbl = rxq->frag_tbl; 337 dr = &qconf->death_row; 338 339 /* prepare mbuf: setup l2_len/l3_len. */ 340 m->l2_len = sizeof(*eth_hdr); 341 m->l3_len = sizeof(*ip_hdr); 342 343 /* process this fragment. */ 344 mo = rte_ipv4_frag_reassemble_packet(tbl, dr, m, tms, ip_hdr); 345 if (mo == NULL) 346 /* no packet to send out. */ 347 return; 348 349 /* we have our packet reassembled. */ 350 if (mo != m) { 351 m = mo; 352 eth_hdr = rte_pktmbuf_mtod(m, 353 struct rte_ether_hdr *); 354 ip_hdr = (struct rte_ipv4_hdr *)(eth_hdr + 1); 355 } 356 357 /* update offloading flags */ 358 m->ol_flags |= (PKT_TX_IPV4 | PKT_TX_IP_CKSUM); 359 } 360 ip_dst = rte_be_to_cpu_32(ip_hdr->dst_addr); 361 362 /* Find destination port */ 363 if (rte_lpm_lookup(rxq->lpm, ip_dst, &next_hop) == 0 && 364 (enabled_port_mask & 1 << next_hop) != 0) { 365 dst_port = next_hop; 366 } 367 368 eth_hdr->ether_type = rte_be_to_cpu_16(RTE_ETHER_TYPE_IPV4); 369 } else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) { 370 /* if packet is IPv6 */ 371 struct ipv6_extension_fragment *frag_hdr; 372 struct rte_ipv6_hdr *ip_hdr; 373 374 ip_hdr = (struct rte_ipv6_hdr *)(eth_hdr + 1); 375 376 frag_hdr = rte_ipv6_frag_get_ipv6_fragment_header(ip_hdr); 377 378 if (frag_hdr != NULL) { 379 struct rte_mbuf *mo; 380 381 tbl = rxq->frag_tbl; 382 dr = &qconf->death_row; 383 384 /* prepare mbuf: setup l2_len/l3_len. */ 385 m->l2_len = sizeof(*eth_hdr); 386 m->l3_len = sizeof(*ip_hdr) + sizeof(*frag_hdr); 387 388 mo = rte_ipv6_frag_reassemble_packet(tbl, dr, m, tms, ip_hdr, frag_hdr); 389 if (mo == NULL) 390 return; 391 392 if (mo != m) { 393 m = mo; 394 eth_hdr = rte_pktmbuf_mtod(m, 395 struct rte_ether_hdr *); 396 ip_hdr = (struct rte_ipv6_hdr *)(eth_hdr + 1); 397 } 398 } 399 400 /* Find destination port */ 401 if (rte_lpm6_lookup(rxq->lpm6, ip_hdr->dst_addr, 402 &next_hop) == 0 && 403 (enabled_port_mask & 1 << next_hop) != 0) { 404 dst_port = next_hop; 405 } 406 407 eth_hdr->ether_type = rte_be_to_cpu_16(RTE_ETHER_TYPE_IPV6); 408 } 409 /* if packet wasn't IPv4 or IPv6, it's forwarded to the port it came from */ 410 411 /* 02:00:00:00:00:xx */ 412 d_addr_bytes = ð_hdr->d_addr.addr_bytes[0]; 413 *((uint64_t *)d_addr_bytes) = 0x000000000002 + ((uint64_t)dst_port << 40); 414 415 /* src addr */ 416 rte_ether_addr_copy(&ports_eth_addr[dst_port], ð_hdr->s_addr); 417 418 send_single_packet(m, dst_port); 419 } 420 421 /* main processing loop */ 422 static int 423 main_loop(__attribute__((unused)) void *dummy) 424 { 425 struct rte_mbuf *pkts_burst[MAX_PKT_BURST]; 426 unsigned lcore_id; 427 uint64_t diff_tsc, cur_tsc, prev_tsc; 428 int i, j, nb_rx; 429 uint16_t portid; 430 struct lcore_queue_conf *qconf; 431 const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US; 432 433 prev_tsc = 0; 434 435 lcore_id = rte_lcore_id(); 436 qconf = &lcore_queue_conf[lcore_id]; 437 438 if (qconf->n_rx_queue == 0) { 439 RTE_LOG(INFO, IP_RSMBL, "lcore %u has nothing to do\n", lcore_id); 440 return 0; 441 } 442 443 RTE_LOG(INFO, IP_RSMBL, "entering main loop on lcore %u\n", lcore_id); 444 445 for (i = 0; i < qconf->n_rx_queue; i++) { 446 447 portid = qconf->rx_queue_list[i].portid; 448 RTE_LOG(INFO, IP_RSMBL, " -- lcoreid=%u portid=%u\n", lcore_id, 449 portid); 450 } 451 452 while (1) { 453 454 cur_tsc = rte_rdtsc(); 455 456 /* 457 * TX burst queue drain 458 */ 459 diff_tsc = cur_tsc - prev_tsc; 460 if (unlikely(diff_tsc > drain_tsc)) { 461 462 /* 463 * This could be optimized (use queueid instead of 464 * portid), but it is not called so often 465 */ 466 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) { 467 if ((enabled_port_mask & (1 << portid)) != 0) 468 send_burst(qconf, 1, portid); 469 } 470 471 prev_tsc = cur_tsc; 472 } 473 474 /* 475 * Read packet from RX queues 476 */ 477 for (i = 0; i < qconf->n_rx_queue; ++i) { 478 479 portid = qconf->rx_queue_list[i].portid; 480 481 nb_rx = rte_eth_rx_burst(portid, 0, pkts_burst, 482 MAX_PKT_BURST); 483 484 /* Prefetch first packets */ 485 for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) { 486 rte_prefetch0(rte_pktmbuf_mtod( 487 pkts_burst[j], void *)); 488 } 489 490 /* Prefetch and forward already prefetched packets */ 491 for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) { 492 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[ 493 j + PREFETCH_OFFSET], void *)); 494 reassemble(pkts_burst[j], portid, 495 i, qconf, cur_tsc); 496 } 497 498 /* Forward remaining prefetched packets */ 499 for (; j < nb_rx; j++) { 500 reassemble(pkts_burst[j], portid, 501 i, qconf, cur_tsc); 502 } 503 504 rte_ip_frag_free_death_row(&qconf->death_row, 505 PREFETCH_OFFSET); 506 } 507 } 508 } 509 510 /* display usage */ 511 static void 512 print_usage(const char *prgname) 513 { 514 printf("%s [EAL options] -- -p PORTMASK [-q NQ]" 515 " [--max-pkt-len PKTLEN]" 516 " [--maxflows=<flows>] [--flowttl=<ttl>[(s|ms)]]\n" 517 " -p PORTMASK: hexadecimal bitmask of ports to configure\n" 518 " -q NQ: number of RX queues per lcore\n" 519 " --maxflows=<flows>: optional, maximum number of flows " 520 "supported\n" 521 " --flowttl=<ttl>[(s|ms)]: optional, maximum TTL for each " 522 "flow\n", 523 prgname); 524 } 525 526 static uint32_t 527 parse_flow_num(const char *str, uint32_t min, uint32_t max, uint32_t *val) 528 { 529 char *end; 530 uint64_t v; 531 532 /* parse decimal string */ 533 errno = 0; 534 v = strtoul(str, &end, 10); 535 if (errno != 0 || *end != '\0') 536 return -EINVAL; 537 538 if (v < min || v > max) 539 return -EINVAL; 540 541 *val = (uint32_t)v; 542 return 0; 543 } 544 545 static int 546 parse_flow_ttl(const char *str, uint32_t min, uint32_t max, uint32_t *val) 547 { 548 char *end; 549 uint64_t v; 550 551 static const char frmt_sec[] = "s"; 552 static const char frmt_msec[] = "ms"; 553 554 /* parse decimal string */ 555 errno = 0; 556 v = strtoul(str, &end, 10); 557 if (errno != 0) 558 return -EINVAL; 559 560 if (*end != '\0') { 561 if (strncmp(frmt_sec, end, sizeof(frmt_sec)) == 0) 562 v *= MS_PER_S; 563 else if (strncmp(frmt_msec, end, sizeof (frmt_msec)) != 0) 564 return -EINVAL; 565 } 566 567 if (v < min || v > max) 568 return -EINVAL; 569 570 *val = (uint32_t)v; 571 return 0; 572 } 573 574 static int 575 parse_portmask(const char *portmask) 576 { 577 char *end = NULL; 578 unsigned long pm; 579 580 /* parse hexadecimal string */ 581 pm = strtoul(portmask, &end, 16); 582 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0')) 583 return -1; 584 585 if (pm == 0) 586 return -1; 587 588 return pm; 589 } 590 591 static int 592 parse_nqueue(const char *q_arg) 593 { 594 char *end = NULL; 595 unsigned long n; 596 597 printf("%p\n", q_arg); 598 599 /* parse hexadecimal string */ 600 n = strtoul(q_arg, &end, 10); 601 if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0')) 602 return -1; 603 if (n == 0) 604 return -1; 605 if (n >= MAX_RX_QUEUE_PER_LCORE) 606 return -1; 607 608 return n; 609 } 610 611 /* Parse the argument given in the command line of the application */ 612 static int 613 parse_args(int argc, char **argv) 614 { 615 int opt, ret; 616 char **argvopt; 617 int option_index; 618 char *prgname = argv[0]; 619 static struct option lgopts[] = { 620 {"max-pkt-len", 1, 0, 0}, 621 {"maxflows", 1, 0, 0}, 622 {"flowttl", 1, 0, 0}, 623 {NULL, 0, 0, 0} 624 }; 625 626 argvopt = argv; 627 628 while ((opt = getopt_long(argc, argvopt, "p:q:", 629 lgopts, &option_index)) != EOF) { 630 631 switch (opt) { 632 /* portmask */ 633 case 'p': 634 enabled_port_mask = parse_portmask(optarg); 635 if (enabled_port_mask == 0) { 636 printf("invalid portmask\n"); 637 print_usage(prgname); 638 return -1; 639 } 640 break; 641 642 /* nqueue */ 643 case 'q': 644 rx_queue_per_lcore = parse_nqueue(optarg); 645 if (rx_queue_per_lcore < 0) { 646 printf("invalid queue number\n"); 647 print_usage(prgname); 648 return -1; 649 } 650 break; 651 652 /* long options */ 653 case 0: 654 if (!strncmp(lgopts[option_index].name, 655 "maxflows", 8)) { 656 if ((ret = parse_flow_num(optarg, MIN_FLOW_NUM, 657 MAX_FLOW_NUM, 658 &max_flow_num)) != 0) { 659 printf("invalid value: \"%s\" for " 660 "parameter %s\n", 661 optarg, 662 lgopts[option_index].name); 663 print_usage(prgname); 664 return ret; 665 } 666 } 667 668 if (!strncmp(lgopts[option_index].name, "flowttl", 7)) { 669 if ((ret = parse_flow_ttl(optarg, MIN_FLOW_TTL, 670 MAX_FLOW_TTL, 671 &max_flow_ttl)) != 0) { 672 printf("invalid value: \"%s\" for " 673 "parameter %s\n", 674 optarg, 675 lgopts[option_index].name); 676 print_usage(prgname); 677 return ret; 678 } 679 } 680 681 break; 682 683 default: 684 print_usage(prgname); 685 return -1; 686 } 687 } 688 689 if (optind >= 0) 690 argv[optind-1] = prgname; 691 692 ret = optind-1; 693 optind = 1; /* reset getopt lib */ 694 return ret; 695 } 696 697 static void 698 print_ethaddr(const char *name, const struct rte_ether_addr *eth_addr) 699 { 700 char buf[RTE_ETHER_ADDR_FMT_SIZE]; 701 rte_ether_format_addr(buf, RTE_ETHER_ADDR_FMT_SIZE, eth_addr); 702 printf("%s%s", name, buf); 703 } 704 705 /* Check the link status of all ports in up to 9s, and print them finally */ 706 static void 707 check_all_ports_link_status(uint32_t port_mask) 708 { 709 #define CHECK_INTERVAL 100 /* 100ms */ 710 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */ 711 uint16_t portid; 712 uint8_t count, all_ports_up, print_flag = 0; 713 struct rte_eth_link link; 714 715 printf("\nChecking link status"); 716 fflush(stdout); 717 for (count = 0; count <= MAX_CHECK_TIME; count++) { 718 all_ports_up = 1; 719 RTE_ETH_FOREACH_DEV(portid) { 720 if ((port_mask & (1 << portid)) == 0) 721 continue; 722 memset(&link, 0, sizeof(link)); 723 rte_eth_link_get_nowait(portid, &link); 724 /* print link status if flag set */ 725 if (print_flag == 1) { 726 if (link.link_status) 727 printf( 728 "Port%d Link Up. Speed %u Mbps - %s\n", 729 portid, link.link_speed, 730 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ? 731 ("full-duplex") : ("half-duplex\n")); 732 else 733 printf("Port %d Link Down\n", portid); 734 continue; 735 } 736 /* clear all_ports_up flag if any link down */ 737 if (link.link_status == ETH_LINK_DOWN) { 738 all_ports_up = 0; 739 break; 740 } 741 } 742 /* after finally printing all link status, get out */ 743 if (print_flag == 1) 744 break; 745 746 if (all_ports_up == 0) { 747 printf("."); 748 fflush(stdout); 749 rte_delay_ms(CHECK_INTERVAL); 750 } 751 752 /* set the print_flag if all ports up or timeout */ 753 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) { 754 print_flag = 1; 755 printf("\ndone\n"); 756 } 757 } 758 } 759 760 static int 761 init_routing_table(void) 762 { 763 struct rte_lpm *lpm; 764 struct rte_lpm6 *lpm6; 765 int socket, ret; 766 unsigned i; 767 768 for (socket = 0; socket < RTE_MAX_NUMA_NODES; socket++) { 769 if (socket_lpm[socket]) { 770 lpm = socket_lpm[socket]; 771 /* populate the LPM table */ 772 for (i = 0; i < RTE_DIM(l3fwd_ipv4_route_array); i++) { 773 ret = rte_lpm_add(lpm, 774 l3fwd_ipv4_route_array[i].ip, 775 l3fwd_ipv4_route_array[i].depth, 776 l3fwd_ipv4_route_array[i].if_out); 777 778 if (ret < 0) { 779 RTE_LOG(ERR, IP_RSMBL, "Unable to add entry %i to the l3fwd " 780 "LPM table\n", i); 781 return -1; 782 } 783 784 RTE_LOG(INFO, IP_RSMBL, "Socket %i: adding route " IPv4_BYTES_FMT 785 "/%d (port %d)\n", 786 socket, 787 IPv4_BYTES(l3fwd_ipv4_route_array[i].ip), 788 l3fwd_ipv4_route_array[i].depth, 789 l3fwd_ipv4_route_array[i].if_out); 790 } 791 } 792 793 if (socket_lpm6[socket]) { 794 lpm6 = socket_lpm6[socket]; 795 /* populate the LPM6 table */ 796 for (i = 0; i < RTE_DIM(l3fwd_ipv6_route_array); i++) { 797 ret = rte_lpm6_add(lpm6, 798 l3fwd_ipv6_route_array[i].ip, 799 l3fwd_ipv6_route_array[i].depth, 800 l3fwd_ipv6_route_array[i].if_out); 801 802 if (ret < 0) { 803 RTE_LOG(ERR, IP_RSMBL, "Unable to add entry %i to the l3fwd " 804 "LPM6 table\n", i); 805 return -1; 806 } 807 808 RTE_LOG(INFO, IP_RSMBL, "Socket %i: adding route " IPv6_BYTES_FMT 809 "/%d (port %d)\n", 810 socket, 811 IPv6_BYTES(l3fwd_ipv6_route_array[i].ip), 812 l3fwd_ipv6_route_array[i].depth, 813 l3fwd_ipv6_route_array[i].if_out); 814 } 815 } 816 } 817 return 0; 818 } 819 820 static int 821 setup_port_tbl(struct lcore_queue_conf *qconf, uint32_t lcore, int socket, 822 uint32_t port) 823 { 824 struct mbuf_table *mtb; 825 uint32_t n; 826 size_t sz; 827 828 n = RTE_MAX(max_flow_num, 2UL * MAX_PKT_BURST); 829 sz = sizeof (*mtb) + sizeof (mtb->m_table[0]) * n; 830 831 if ((mtb = rte_zmalloc_socket(__func__, sz, RTE_CACHE_LINE_SIZE, 832 socket)) == NULL) { 833 RTE_LOG(ERR, IP_RSMBL, "%s() for lcore: %u, port: %u " 834 "failed to allocate %zu bytes\n", 835 __func__, lcore, port, sz); 836 return -1; 837 } 838 839 mtb->len = n; 840 qconf->tx_mbufs[port] = mtb; 841 842 return 0; 843 } 844 845 static int 846 setup_queue_tbl(struct rx_queue *rxq, uint32_t lcore, uint32_t queue) 847 { 848 int socket; 849 uint32_t nb_mbuf; 850 uint64_t frag_cycles; 851 char buf[RTE_MEMPOOL_NAMESIZE]; 852 853 socket = rte_lcore_to_socket_id(lcore); 854 if (socket == SOCKET_ID_ANY) 855 socket = 0; 856 857 frag_cycles = (rte_get_tsc_hz() + MS_PER_S - 1) / MS_PER_S * 858 max_flow_ttl; 859 860 if ((rxq->frag_tbl = rte_ip_frag_table_create(max_flow_num, 861 IP_FRAG_TBL_BUCKET_ENTRIES, max_flow_num, frag_cycles, 862 socket)) == NULL) { 863 RTE_LOG(ERR, IP_RSMBL, "ip_frag_tbl_create(%u) on " 864 "lcore: %u for queue: %u failed\n", 865 max_flow_num, lcore, queue); 866 return -1; 867 } 868 869 /* 870 * At any given moment up to <max_flow_num * (MAX_FRAG_NUM)> 871 * mbufs could be stored int the fragment table. 872 * Plus, each TX queue can hold up to <max_flow_num> packets. 873 */ 874 875 nb_mbuf = RTE_MAX(max_flow_num, 2UL * MAX_PKT_BURST) * MAX_FRAG_NUM; 876 nb_mbuf *= (port_conf.rxmode.max_rx_pkt_len + BUF_SIZE - 1) / BUF_SIZE; 877 nb_mbuf *= 2; /* ipv4 and ipv6 */ 878 nb_mbuf += nb_rxd + nb_txd; 879 880 nb_mbuf = RTE_MAX(nb_mbuf, (uint32_t)NB_MBUF); 881 882 snprintf(buf, sizeof(buf), "mbuf_pool_%u_%u", lcore, queue); 883 884 rxq->pool = rte_pktmbuf_pool_create(buf, nb_mbuf, MEMPOOL_CACHE_SIZE, 0, 885 MBUF_DATA_SIZE, socket); 886 if (rxq->pool == NULL) { 887 RTE_LOG(ERR, IP_RSMBL, 888 "rte_pktmbuf_pool_create(%s) failed", buf); 889 return -1; 890 } 891 892 return 0; 893 } 894 895 static int 896 init_mem(void) 897 { 898 char buf[PATH_MAX]; 899 struct rte_lpm *lpm; 900 struct rte_lpm6 *lpm6; 901 struct rte_lpm_config lpm_config; 902 int socket; 903 unsigned lcore_id; 904 905 /* traverse through lcores and initialize structures on each socket */ 906 907 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) { 908 909 if (rte_lcore_is_enabled(lcore_id) == 0) 910 continue; 911 912 socket = rte_lcore_to_socket_id(lcore_id); 913 914 if (socket == SOCKET_ID_ANY) 915 socket = 0; 916 917 if (socket_lpm[socket] == NULL) { 918 RTE_LOG(INFO, IP_RSMBL, "Creating LPM table on socket %i\n", socket); 919 snprintf(buf, sizeof(buf), "IP_RSMBL_LPM_%i", socket); 920 921 lpm_config.max_rules = LPM_MAX_RULES; 922 lpm_config.number_tbl8s = 256; 923 lpm_config.flags = 0; 924 925 lpm = rte_lpm_create(buf, socket, &lpm_config); 926 if (lpm == NULL) { 927 RTE_LOG(ERR, IP_RSMBL, "Cannot create LPM table\n"); 928 return -1; 929 } 930 socket_lpm[socket] = lpm; 931 } 932 933 if (socket_lpm6[socket] == NULL) { 934 RTE_LOG(INFO, IP_RSMBL, "Creating LPM6 table on socket %i\n", socket); 935 snprintf(buf, sizeof(buf), "IP_RSMBL_LPM_%i", socket); 936 937 lpm6 = rte_lpm6_create(buf, socket, &lpm6_config); 938 if (lpm6 == NULL) { 939 RTE_LOG(ERR, IP_RSMBL, "Cannot create LPM table\n"); 940 return -1; 941 } 942 socket_lpm6[socket] = lpm6; 943 } 944 } 945 946 return 0; 947 } 948 949 static void 950 queue_dump_stat(void) 951 { 952 uint32_t i, lcore; 953 const struct lcore_queue_conf *qconf; 954 955 for (lcore = 0; lcore < RTE_MAX_LCORE; lcore++) { 956 if (rte_lcore_is_enabled(lcore) == 0) 957 continue; 958 959 qconf = &lcore_queue_conf[lcore]; 960 for (i = 0; i < qconf->n_rx_queue; i++) { 961 962 fprintf(stdout, " -- lcoreid=%u portid=%u " 963 "frag tbl stat:\n", 964 lcore, qconf->rx_queue_list[i].portid); 965 rte_ip_frag_table_statistics_dump(stdout, 966 qconf->rx_queue_list[i].frag_tbl); 967 fprintf(stdout, "TX bursts:\t%" PRIu64 "\n" 968 "TX packets _queued:\t%" PRIu64 "\n" 969 "TX packets dropped:\t%" PRIu64 "\n" 970 "TX packets send:\t%" PRIu64 "\n", 971 qconf->tx_stat.call, 972 qconf->tx_stat.queue, 973 qconf->tx_stat.drop, 974 qconf->tx_stat.send); 975 } 976 } 977 } 978 979 static void 980 signal_handler(int signum) 981 { 982 queue_dump_stat(); 983 if (signum != SIGUSR1) 984 rte_exit(0, "received signal: %d, exiting\n", signum); 985 } 986 987 int 988 main(int argc, char **argv) 989 { 990 struct lcore_queue_conf *qconf; 991 struct rte_eth_dev_info dev_info; 992 struct rte_eth_txconf *txconf; 993 struct rx_queue *rxq; 994 int ret, socket; 995 unsigned nb_ports; 996 uint16_t queueid; 997 unsigned lcore_id = 0, rx_lcore_id = 0; 998 uint32_t n_tx_queue, nb_lcores; 999 uint16_t portid; 1000 1001 /* init EAL */ 1002 ret = rte_eal_init(argc, argv); 1003 if (ret < 0) 1004 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n"); 1005 argc -= ret; 1006 argv += ret; 1007 1008 /* parse application arguments (after the EAL ones) */ 1009 ret = parse_args(argc, argv); 1010 if (ret < 0) 1011 rte_exit(EXIT_FAILURE, "Invalid IP reassembly parameters\n"); 1012 1013 nb_ports = rte_eth_dev_count_avail(); 1014 if (nb_ports == 0) 1015 rte_exit(EXIT_FAILURE, "No ports found!\n"); 1016 1017 nb_lcores = rte_lcore_count(); 1018 1019 /* initialize structures (mempools, lpm etc.) */ 1020 if (init_mem() < 0) 1021 rte_panic("Cannot initialize memory structures!\n"); 1022 1023 /* check if portmask has non-existent ports */ 1024 if (enabled_port_mask & ~(RTE_LEN2MASK(nb_ports, unsigned))) 1025 rte_exit(EXIT_FAILURE, "Non-existent ports in portmask!\n"); 1026 1027 /* initialize all ports */ 1028 RTE_ETH_FOREACH_DEV(portid) { 1029 struct rte_eth_rxconf rxq_conf; 1030 struct rte_eth_conf local_port_conf = port_conf; 1031 1032 /* skip ports that are not enabled */ 1033 if ((enabled_port_mask & (1 << portid)) == 0) { 1034 printf("\nSkipping disabled port %d\n", portid); 1035 continue; 1036 } 1037 1038 qconf = &lcore_queue_conf[rx_lcore_id]; 1039 1040 /* limit the frame size to the maximum supported by NIC */ 1041 ret = rte_eth_dev_info_get(portid, &dev_info); 1042 if (ret != 0) 1043 rte_exit(EXIT_FAILURE, 1044 "Error during getting device (port %u) info: %s\n", 1045 portid, strerror(-ret)); 1046 1047 local_port_conf.rxmode.max_rx_pkt_len = RTE_MIN( 1048 dev_info.max_rx_pktlen, 1049 local_port_conf.rxmode.max_rx_pkt_len); 1050 1051 /* get the lcore_id for this port */ 1052 while (rte_lcore_is_enabled(rx_lcore_id) == 0 || 1053 qconf->n_rx_queue == (unsigned)rx_queue_per_lcore) { 1054 1055 rx_lcore_id++; 1056 if (rx_lcore_id >= RTE_MAX_LCORE) 1057 rte_exit(EXIT_FAILURE, "Not enough cores\n"); 1058 1059 qconf = &lcore_queue_conf[rx_lcore_id]; 1060 } 1061 1062 socket = rte_lcore_to_socket_id(portid); 1063 if (socket == SOCKET_ID_ANY) 1064 socket = 0; 1065 1066 queueid = qconf->n_rx_queue; 1067 rxq = &qconf->rx_queue_list[queueid]; 1068 rxq->portid = portid; 1069 rxq->lpm = socket_lpm[socket]; 1070 rxq->lpm6 = socket_lpm6[socket]; 1071 1072 ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd, 1073 &nb_txd); 1074 if (ret < 0) 1075 rte_exit(EXIT_FAILURE, 1076 "Cannot adjust number of descriptors: err=%d, port=%d\n", 1077 ret, portid); 1078 1079 if (setup_queue_tbl(rxq, rx_lcore_id, queueid) < 0) 1080 rte_exit(EXIT_FAILURE, "Failed to set up queue table\n"); 1081 qconf->n_rx_queue++; 1082 1083 /* init port */ 1084 printf("Initializing port %d ... ", portid ); 1085 fflush(stdout); 1086 1087 n_tx_queue = nb_lcores; 1088 if (n_tx_queue > MAX_TX_QUEUE_PER_PORT) 1089 n_tx_queue = MAX_TX_QUEUE_PER_PORT; 1090 if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE) 1091 local_port_conf.txmode.offloads |= 1092 DEV_TX_OFFLOAD_MBUF_FAST_FREE; 1093 1094 local_port_conf.rx_adv_conf.rss_conf.rss_hf &= 1095 dev_info.flow_type_rss_offloads; 1096 if (local_port_conf.rx_adv_conf.rss_conf.rss_hf != 1097 port_conf.rx_adv_conf.rss_conf.rss_hf) { 1098 printf("Port %u modified RSS hash function based on hardware support," 1099 "requested:%#"PRIx64" configured:%#"PRIx64"\n", 1100 portid, 1101 port_conf.rx_adv_conf.rss_conf.rss_hf, 1102 local_port_conf.rx_adv_conf.rss_conf.rss_hf); 1103 } 1104 1105 ret = rte_eth_dev_configure(portid, 1, (uint16_t)n_tx_queue, 1106 &local_port_conf); 1107 if (ret < 0) { 1108 printf("\n"); 1109 rte_exit(EXIT_FAILURE, "Cannot configure device: " 1110 "err=%d, port=%d\n", 1111 ret, portid); 1112 } 1113 1114 /* init one RX queue */ 1115 rxq_conf = dev_info.default_rxconf; 1116 rxq_conf.offloads = local_port_conf.rxmode.offloads; 1117 ret = rte_eth_rx_queue_setup(portid, 0, nb_rxd, 1118 socket, &rxq_conf, 1119 rxq->pool); 1120 if (ret < 0) { 1121 printf("\n"); 1122 rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: " 1123 "err=%d, port=%d\n", 1124 ret, portid); 1125 } 1126 1127 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]); 1128 print_ethaddr(" Address:", &ports_eth_addr[portid]); 1129 printf("\n"); 1130 1131 /* init one TX queue per couple (lcore,port) */ 1132 queueid = 0; 1133 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) { 1134 if (rte_lcore_is_enabled(lcore_id) == 0) 1135 continue; 1136 1137 socket = (int) rte_lcore_to_socket_id(lcore_id); 1138 1139 printf("txq=%u,%d,%d ", lcore_id, queueid, socket); 1140 fflush(stdout); 1141 1142 txconf = &dev_info.default_txconf; 1143 txconf->offloads = local_port_conf.txmode.offloads; 1144 1145 ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd, 1146 socket, txconf); 1147 if (ret < 0) 1148 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: err=%d, " 1149 "port=%d\n", ret, portid); 1150 1151 qconf = &lcore_queue_conf[lcore_id]; 1152 qconf->tx_queue_id[portid] = queueid; 1153 setup_port_tbl(qconf, lcore_id, socket, portid); 1154 queueid++; 1155 } 1156 printf("\n"); 1157 } 1158 1159 printf("\n"); 1160 1161 /* start ports */ 1162 RTE_ETH_FOREACH_DEV(portid) { 1163 if ((enabled_port_mask & (1 << portid)) == 0) { 1164 continue; 1165 } 1166 /* Start device */ 1167 ret = rte_eth_dev_start(portid); 1168 if (ret < 0) 1169 rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%d\n", 1170 ret, portid); 1171 1172 rte_eth_promiscuous_enable(portid); 1173 } 1174 1175 if (init_routing_table() < 0) 1176 rte_exit(EXIT_FAILURE, "Cannot init routing table\n"); 1177 1178 check_all_ports_link_status(enabled_port_mask); 1179 1180 signal(SIGUSR1, signal_handler); 1181 signal(SIGTERM, signal_handler); 1182 signal(SIGINT, signal_handler); 1183 1184 /* launch per-lcore init on every lcore */ 1185 rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER); 1186 RTE_LCORE_FOREACH_SLAVE(lcore_id) { 1187 if (rte_eal_wait_lcore(lcore_id) < 0) 1188 return -1; 1189 } 1190 1191 return 0; 1192 } 1193