1 /* SPDX-License-Identifier: BSD-3-Clause 2 * Copyright(c) 2010-2016 Intel Corporation 3 */ 4 5 #include <locale.h> 6 #include <stdlib.h> 7 #include <string.h> 8 #include <stdint.h> 9 #include <ctype.h> 10 #include <getopt.h> 11 12 #include <rte_common.h> 13 #include <rte_log.h> 14 #include <rte_malloc.h> 15 #include <rte_memory.h> 16 #include <rte_memcpy.h> 17 #include <rte_eal.h> 18 #include <rte_launch.h> 19 #include <rte_atomic.h> 20 #include <rte_cycles.h> 21 #include <rte_prefetch.h> 22 #include <rte_lcore.h> 23 #include <rte_per_lcore.h> 24 #include <rte_branch_prediction.h> 25 #include <rte_interrupts.h> 26 #include <rte_debug.h> 27 #include <rte_ether.h> 28 #include <rte_ethdev.h> 29 #include <rte_mempool.h> 30 #include <rte_mbuf.h> 31 #include <rte_spinlock.h> 32 33 #include <rte_errno.h> 34 #include <rte_jobstats.h> 35 #include <rte_timer.h> 36 #include <rte_alarm.h> 37 #include <rte_pause.h> 38 39 #define RTE_LOGTYPE_L2FWD RTE_LOGTYPE_USER1 40 41 #define NB_MBUF 8192 42 43 #define MAX_PKT_BURST 32 44 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */ 45 46 /* 47 * Configurable number of RX/TX ring descriptors 48 */ 49 #define RTE_TEST_RX_DESC_DEFAULT 1024 50 #define RTE_TEST_TX_DESC_DEFAULT 1024 51 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT; 52 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT; 53 54 /* ethernet addresses of ports */ 55 static struct rte_ether_addr l2fwd_ports_eth_addr[RTE_MAX_ETHPORTS]; 56 57 /* mask of enabled ports */ 58 static uint32_t l2fwd_enabled_port_mask; 59 60 /* list of enabled ports */ 61 static uint32_t l2fwd_dst_ports[RTE_MAX_ETHPORTS]; 62 63 #define UPDATE_STEP_UP 1 64 #define UPDATE_STEP_DOWN 32 65 66 static unsigned int l2fwd_rx_queue_per_lcore = 1; 67 68 #define MAX_RX_QUEUE_PER_LCORE 16 69 #define MAX_TX_QUEUE_PER_PORT 16 70 /* List of queues to be polled for given lcore. 8< */ 71 struct lcore_queue_conf { 72 unsigned n_rx_port; 73 unsigned rx_port_list[MAX_RX_QUEUE_PER_LCORE]; 74 uint64_t next_flush_time[RTE_MAX_ETHPORTS]; 75 76 struct rte_timer rx_timers[MAX_RX_QUEUE_PER_LCORE]; 77 struct rte_jobstats port_fwd_jobs[MAX_RX_QUEUE_PER_LCORE]; 78 79 struct rte_timer flush_timer; 80 struct rte_jobstats flush_job; 81 struct rte_jobstats idle_job; 82 struct rte_jobstats_context jobs_context; 83 84 rte_atomic16_t stats_read_pending; 85 rte_spinlock_t lock; 86 } __rte_cache_aligned; 87 /* >8 End of list of queues to be polled for given lcore. */ 88 struct lcore_queue_conf lcore_queue_conf[RTE_MAX_LCORE]; 89 90 struct rte_eth_dev_tx_buffer *tx_buffer[RTE_MAX_ETHPORTS]; 91 92 static struct rte_eth_conf port_conf = { 93 .rxmode = { 94 .split_hdr_size = 0, 95 }, 96 .txmode = { 97 .mq_mode = ETH_MQ_TX_NONE, 98 }, 99 }; 100 101 struct rte_mempool *l2fwd_pktmbuf_pool = NULL; 102 103 /* Per-port statistics struct */ 104 struct l2fwd_port_statistics { 105 uint64_t tx; 106 uint64_t rx; 107 uint64_t dropped; 108 } __rte_cache_aligned; 109 struct l2fwd_port_statistics port_statistics[RTE_MAX_ETHPORTS]; 110 111 /* 1 day max */ 112 #define MAX_TIMER_PERIOD 86400 113 /* default period is 10 seconds */ 114 static int64_t timer_period = 10; 115 /* default timer frequency */ 116 static double hz; 117 /* BURST_TX_DRAIN_US converted to cycles */ 118 uint64_t drain_tsc; 119 /* Convert cycles to ns */ 120 static inline double 121 cycles_to_ns(uint64_t cycles) 122 { 123 double t = cycles; 124 125 t *= (double)NS_PER_S; 126 t /= hz; 127 return t; 128 } 129 130 static void 131 show_lcore_stats(unsigned lcore_id) 132 { 133 struct lcore_queue_conf *qconf = &lcore_queue_conf[lcore_id]; 134 struct rte_jobstats_context *ctx = &qconf->jobs_context; 135 struct rte_jobstats *job; 136 uint8_t i; 137 138 /* LCore statistics. */ 139 uint64_t stats_period, loop_count; 140 uint64_t exec, exec_min, exec_max; 141 uint64_t management, management_min, management_max; 142 uint64_t busy, busy_min, busy_max; 143 144 /* Jobs statistics. */ 145 const uint16_t port_cnt = qconf->n_rx_port; 146 uint64_t jobs_exec_cnt[port_cnt], jobs_period[port_cnt]; 147 uint64_t jobs_exec[port_cnt], jobs_exec_min[port_cnt], 148 jobs_exec_max[port_cnt]; 149 150 uint64_t flush_exec_cnt, flush_period; 151 uint64_t flush_exec, flush_exec_min, flush_exec_max; 152 153 uint64_t idle_exec_cnt; 154 uint64_t idle_exec, idle_exec_min, idle_exec_max; 155 uint64_t collection_time = rte_get_timer_cycles(); 156 157 /* Ask forwarding thread to give us stats. */ 158 rte_atomic16_set(&qconf->stats_read_pending, 1); 159 rte_spinlock_lock(&qconf->lock); 160 rte_atomic16_set(&qconf->stats_read_pending, 0); 161 162 /* Collect context statistics. */ 163 stats_period = ctx->state_time - ctx->start_time; 164 loop_count = ctx->loop_cnt; 165 166 exec = ctx->exec_time; 167 exec_min = ctx->min_exec_time; 168 exec_max = ctx->max_exec_time; 169 170 management = ctx->management_time; 171 management_min = ctx->min_management_time; 172 management_max = ctx->max_management_time; 173 174 rte_jobstats_context_reset(ctx); 175 176 for (i = 0; i < port_cnt; i++) { 177 job = &qconf->port_fwd_jobs[i]; 178 179 jobs_exec_cnt[i] = job->exec_cnt; 180 jobs_period[i] = job->period; 181 182 jobs_exec[i] = job->exec_time; 183 jobs_exec_min[i] = job->min_exec_time; 184 jobs_exec_max[i] = job->max_exec_time; 185 186 rte_jobstats_reset(job); 187 } 188 189 flush_exec_cnt = qconf->flush_job.exec_cnt; 190 flush_period = qconf->flush_job.period; 191 flush_exec = qconf->flush_job.exec_time; 192 flush_exec_min = qconf->flush_job.min_exec_time; 193 flush_exec_max = qconf->flush_job.max_exec_time; 194 rte_jobstats_reset(&qconf->flush_job); 195 196 idle_exec_cnt = qconf->idle_job.exec_cnt; 197 idle_exec = qconf->idle_job.exec_time; 198 idle_exec_min = qconf->idle_job.min_exec_time; 199 idle_exec_max = qconf->idle_job.max_exec_time; 200 rte_jobstats_reset(&qconf->idle_job); 201 202 rte_spinlock_unlock(&qconf->lock); 203 204 exec -= idle_exec; 205 busy = exec + management; 206 busy_min = exec_min + management_min; 207 busy_max = exec_max + management_max; 208 209 210 collection_time = rte_get_timer_cycles() - collection_time; 211 212 #define STAT_FMT "\n%-18s %'14.0f %6.1f%% %'10.0f %'10.0f %'10.0f" 213 214 printf("\n----------------" 215 "\nLCore %3u: statistics (time in ns, collected in %'9.0f)" 216 "\n%-18s %14s %7s %10s %10s %10s " 217 "\n%-18s %'14.0f" 218 "\n%-18s %'14" PRIu64 219 STAT_FMT /* Exec */ 220 STAT_FMT /* Management */ 221 STAT_FMT /* Busy */ 222 STAT_FMT, /* Idle */ 223 lcore_id, cycles_to_ns(collection_time), 224 "Stat type", "total", "%total", "avg", "min", "max", 225 "Stats duration:", cycles_to_ns(stats_period), 226 "Loop count:", loop_count, 227 "Exec time", 228 cycles_to_ns(exec), exec * 100.0 / stats_period, 229 cycles_to_ns(loop_count ? exec / loop_count : 0), 230 cycles_to_ns(exec_min), 231 cycles_to_ns(exec_max), 232 "Management time", 233 cycles_to_ns(management), management * 100.0 / stats_period, 234 cycles_to_ns(loop_count ? management / loop_count : 0), 235 cycles_to_ns(management_min), 236 cycles_to_ns(management_max), 237 "Exec + management", 238 cycles_to_ns(busy), busy * 100.0 / stats_period, 239 cycles_to_ns(loop_count ? busy / loop_count : 0), 240 cycles_to_ns(busy_min), 241 cycles_to_ns(busy_max), 242 "Idle (job)", 243 cycles_to_ns(idle_exec), idle_exec * 100.0 / stats_period, 244 cycles_to_ns(idle_exec_cnt ? idle_exec / idle_exec_cnt : 0), 245 cycles_to_ns(idle_exec_min), 246 cycles_to_ns(idle_exec_max)); 247 248 for (i = 0; i < qconf->n_rx_port; i++) { 249 job = &qconf->port_fwd_jobs[i]; 250 printf("\n\nJob %" PRIu32 ": %-20s " 251 "\n%-18s %'14" PRIu64 252 "\n%-18s %'14.0f" 253 STAT_FMT, 254 i, job->name, 255 "Exec count:", jobs_exec_cnt[i], 256 "Exec period: ", cycles_to_ns(jobs_period[i]), 257 "Exec time", 258 cycles_to_ns(jobs_exec[i]), jobs_exec[i] * 100.0 / stats_period, 259 cycles_to_ns(jobs_exec_cnt[i] ? jobs_exec[i] / jobs_exec_cnt[i] 260 : 0), 261 cycles_to_ns(jobs_exec_min[i]), 262 cycles_to_ns(jobs_exec_max[i])); 263 } 264 265 if (qconf->n_rx_port > 0) { 266 job = &qconf->flush_job; 267 printf("\n\nJob %" PRIu32 ": %-20s " 268 "\n%-18s %'14" PRIu64 269 "\n%-18s %'14.0f" 270 STAT_FMT, 271 i, job->name, 272 "Exec count:", flush_exec_cnt, 273 "Exec period: ", cycles_to_ns(flush_period), 274 "Exec time", 275 cycles_to_ns(flush_exec), flush_exec * 100.0 / stats_period, 276 cycles_to_ns(flush_exec_cnt ? flush_exec / flush_exec_cnt : 0), 277 cycles_to_ns(flush_exec_min), 278 cycles_to_ns(flush_exec_max)); 279 } 280 } 281 282 /* Print out statistics on packets dropped */ 283 static void 284 show_stats_cb(__rte_unused void *param) 285 { 286 uint64_t total_packets_dropped, total_packets_tx, total_packets_rx; 287 unsigned portid, lcore_id; 288 289 total_packets_dropped = 0; 290 total_packets_tx = 0; 291 total_packets_rx = 0; 292 293 const char clr[] = { 27, '[', '2', 'J', '\0' }; 294 const char topLeft[] = { 27, '[', '1', ';', '1', 'H', '\0' }; 295 296 /* Clear screen and move to top left */ 297 printf("%s%s" 298 "\nPort statistics ===================================", 299 clr, topLeft); 300 301 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) { 302 /* skip disabled ports */ 303 if ((l2fwd_enabled_port_mask & (1 << portid)) == 0) 304 continue; 305 printf("\nStatistics for port %u ------------------------------" 306 "\nPackets sent: %24"PRIu64 307 "\nPackets received: %20"PRIu64 308 "\nPackets dropped: %21"PRIu64, 309 portid, 310 port_statistics[portid].tx, 311 port_statistics[portid].rx, 312 port_statistics[portid].dropped); 313 314 total_packets_dropped += port_statistics[portid].dropped; 315 total_packets_tx += port_statistics[portid].tx; 316 total_packets_rx += port_statistics[portid].rx; 317 } 318 319 printf("\nAggregate statistics ===============================" 320 "\nTotal packets sent: %18"PRIu64 321 "\nTotal packets received: %14"PRIu64 322 "\nTotal packets dropped: %15"PRIu64 323 "\n====================================================", 324 total_packets_tx, 325 total_packets_rx, 326 total_packets_dropped); 327 328 RTE_LCORE_FOREACH(lcore_id) { 329 if (lcore_queue_conf[lcore_id].n_rx_port > 0) 330 show_lcore_stats(lcore_id); 331 } 332 333 printf("\n====================================================\n"); 334 335 fflush(stdout); 336 337 rte_eal_alarm_set(timer_period * US_PER_S, show_stats_cb, NULL); 338 } 339 340 /* Start of l2fwd_simple_forward. 8< */ 341 static void 342 l2fwd_simple_forward(struct rte_mbuf *m, unsigned portid) 343 { 344 struct rte_ether_hdr *eth; 345 void *tmp; 346 int sent; 347 unsigned dst_port; 348 struct rte_eth_dev_tx_buffer *buffer; 349 350 dst_port = l2fwd_dst_ports[portid]; 351 eth = rte_pktmbuf_mtod(m, struct rte_ether_hdr *); 352 353 /* 02:00:00:00:00:xx */ 354 tmp = ð->d_addr.addr_bytes[0]; 355 *((uint64_t *)tmp) = 0x000000000002 + ((uint64_t)dst_port << 40); 356 357 /* src addr */ 358 rte_ether_addr_copy(&l2fwd_ports_eth_addr[dst_port], ð->s_addr); 359 360 buffer = tx_buffer[dst_port]; 361 sent = rte_eth_tx_buffer(dst_port, 0, buffer, m); 362 if (sent) 363 port_statistics[dst_port].tx += sent; 364 } 365 /* >8 End of l2fwd_simple_forward. */ 366 367 static void 368 l2fwd_job_update_cb(struct rte_jobstats *job, int64_t result) 369 { 370 int64_t err = job->target - result; 371 int64_t histeresis = job->target / 8; 372 373 if (err < -histeresis) { 374 if (job->min_period + UPDATE_STEP_DOWN < job->period) 375 job->period -= UPDATE_STEP_DOWN; 376 } else if (err > histeresis) { 377 if (job->period + UPDATE_STEP_UP < job->max_period) 378 job->period += UPDATE_STEP_UP; 379 } 380 } 381 382 static void 383 l2fwd_fwd_job(__rte_unused struct rte_timer *timer, void *arg) 384 { 385 struct rte_mbuf *pkts_burst[MAX_PKT_BURST]; 386 struct rte_mbuf *m; 387 388 const uint16_t port_idx = (uintptr_t) arg; 389 const unsigned lcore_id = rte_lcore_id(); 390 struct lcore_queue_conf *qconf = &lcore_queue_conf[lcore_id]; 391 struct rte_jobstats *job = &qconf->port_fwd_jobs[port_idx]; 392 const uint16_t portid = qconf->rx_port_list[port_idx]; 393 394 uint8_t j; 395 uint16_t total_nb_rx; 396 397 rte_jobstats_start(&qconf->jobs_context, job); 398 399 /* Call rx burst 2 times. This allow rte_jobstats logic to see if this 400 * function must be called more frequently. */ 401 402 /* Call rx burst 2 times. 8< */ 403 total_nb_rx = rte_eth_rx_burst(portid, 0, pkts_burst, 404 MAX_PKT_BURST); 405 406 for (j = 0; j < total_nb_rx; j++) { 407 m = pkts_burst[j]; 408 rte_prefetch0(rte_pktmbuf_mtod(m, void *)); 409 l2fwd_simple_forward(m, portid); 410 } 411 /* >8 End of call rx burst 2 times. */ 412 413 /* Read second try. 8< */ 414 if (total_nb_rx == MAX_PKT_BURST) { 415 const uint16_t nb_rx = rte_eth_rx_burst(portid, 0, pkts_burst, 416 MAX_PKT_BURST); 417 418 total_nb_rx += nb_rx; 419 for (j = 0; j < nb_rx; j++) { 420 m = pkts_burst[j]; 421 rte_prefetch0(rte_pktmbuf_mtod(m, void *)); 422 l2fwd_simple_forward(m, portid); 423 } 424 } 425 /* >8 End of read second try. */ 426 427 port_statistics[portid].rx += total_nb_rx; 428 429 /* Adjust period time in which we are running here. 8< */ 430 if (rte_jobstats_finish(job, total_nb_rx) != 0) { 431 rte_timer_reset(&qconf->rx_timers[port_idx], job->period, PERIODICAL, 432 lcore_id, l2fwd_fwd_job, arg); 433 } 434 /* >8 End of adjust period time in which we are running. */ 435 } 436 437 /* Draining TX queue of each port. 8< */ 438 static void 439 l2fwd_flush_job(__rte_unused struct rte_timer *timer, __rte_unused void *arg) 440 { 441 uint64_t now; 442 unsigned lcore_id; 443 struct lcore_queue_conf *qconf; 444 uint16_t portid; 445 unsigned i; 446 uint32_t sent; 447 struct rte_eth_dev_tx_buffer *buffer; 448 449 lcore_id = rte_lcore_id(); 450 qconf = &lcore_queue_conf[lcore_id]; 451 452 rte_jobstats_start(&qconf->jobs_context, &qconf->flush_job); 453 454 now = rte_get_timer_cycles(); 455 lcore_id = rte_lcore_id(); 456 qconf = &lcore_queue_conf[lcore_id]; 457 458 for (i = 0; i < qconf->n_rx_port; i++) { 459 portid = l2fwd_dst_ports[qconf->rx_port_list[i]]; 460 461 if (qconf->next_flush_time[portid] <= now) 462 continue; 463 464 buffer = tx_buffer[portid]; 465 sent = rte_eth_tx_buffer_flush(portid, 0, buffer); 466 if (sent) 467 port_statistics[portid].tx += sent; 468 469 qconf->next_flush_time[portid] = rte_get_timer_cycles() + drain_tsc; 470 } 471 472 /* Pass target to indicate that this job is happy of time interwal 473 * in which it was called. */ 474 rte_jobstats_finish(&qconf->flush_job, qconf->flush_job.target); 475 } 476 /* >8 End of draining TX queue of each port. */ 477 478 /* main processing loop */ 479 static void 480 l2fwd_main_loop(void) 481 { 482 unsigned lcore_id; 483 unsigned i, portid; 484 struct lcore_queue_conf *qconf; 485 uint8_t stats_read_pending = 0; 486 uint8_t need_manage; 487 488 lcore_id = rte_lcore_id(); 489 qconf = &lcore_queue_conf[lcore_id]; 490 491 if (qconf->n_rx_port == 0) { 492 RTE_LOG(INFO, L2FWD, "lcore %u has nothing to do\n", lcore_id); 493 return; 494 } 495 496 RTE_LOG(INFO, L2FWD, "entering main loop on lcore %u\n", lcore_id); 497 498 for (i = 0; i < qconf->n_rx_port; i++) { 499 500 portid = qconf->rx_port_list[i]; 501 RTE_LOG(INFO, L2FWD, " -- lcoreid=%u portid=%u\n", lcore_id, 502 portid); 503 } 504 505 rte_jobstats_init(&qconf->idle_job, "idle", 0, 0, 0, 0); 506 507 /* Minimize impact of stats reading. 8< */ 508 for (;;) { 509 rte_spinlock_lock(&qconf->lock); 510 511 do { 512 rte_jobstats_context_start(&qconf->jobs_context); 513 514 /* Do the Idle job: 515 * - Read stats_read_pending flag 516 * - check if some real job need to be executed 517 */ 518 rte_jobstats_start(&qconf->jobs_context, &qconf->idle_job); 519 520 uint64_t repeats = 0; 521 522 do { 523 uint8_t i; 524 uint64_t now = rte_get_timer_cycles(); 525 526 repeats++; 527 need_manage = qconf->flush_timer.expire < now; 528 /* Check if we was esked to give a stats. */ 529 stats_read_pending = 530 rte_atomic16_read(&qconf->stats_read_pending); 531 need_manage |= stats_read_pending; 532 533 for (i = 0; i < qconf->n_rx_port && !need_manage; i++) 534 need_manage = qconf->rx_timers[i].expire < now; 535 536 } while (!need_manage); 537 538 if (likely(repeats != 1)) 539 rte_jobstats_finish(&qconf->idle_job, qconf->idle_job.target); 540 else 541 rte_jobstats_abort(&qconf->idle_job); 542 543 rte_timer_manage(); 544 rte_jobstats_context_finish(&qconf->jobs_context); 545 } while (likely(stats_read_pending == 0)); 546 547 rte_spinlock_unlock(&qconf->lock); 548 rte_pause(); 549 } 550 /* >8 End of minimize impact of stats reading. */ 551 } 552 553 static int 554 l2fwd_launch_one_lcore(__rte_unused void *dummy) 555 { 556 l2fwd_main_loop(); 557 return 0; 558 } 559 560 /* display usage */ 561 static void 562 l2fwd_usage(const char *prgname) 563 { 564 printf("%s [EAL options] -- -p PORTMASK [-q NQ]\n" 565 " -p PORTMASK: hexadecimal bitmask of ports to configure\n" 566 " -q NQ: number of queue (=ports) per lcore (default is 1)\n" 567 " -T PERIOD: statistics will be refreshed each PERIOD seconds (0 to disable, 10 default, 86400 maximum)\n" 568 " -l set system default locale instead of default (\"C\" locale) for thousands separator in stats.", 569 prgname); 570 } 571 572 static int 573 l2fwd_parse_portmask(const char *portmask) 574 { 575 char *end = NULL; 576 unsigned long pm; 577 578 /* parse hexadecimal string */ 579 pm = strtoul(portmask, &end, 16); 580 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0')) 581 return 0; 582 583 return pm; 584 } 585 586 static unsigned int 587 l2fwd_parse_nqueue(const char *q_arg) 588 { 589 char *end = NULL; 590 unsigned long n; 591 592 /* parse hexadecimal string */ 593 n = strtoul(q_arg, &end, 10); 594 if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0')) 595 return 0; 596 if (n == 0) 597 return 0; 598 if (n >= MAX_RX_QUEUE_PER_LCORE) 599 return 0; 600 601 return n; 602 } 603 604 static int 605 l2fwd_parse_timer_period(const char *q_arg) 606 { 607 char *end = NULL; 608 int n; 609 610 /* parse number string */ 611 n = strtol(q_arg, &end, 10); 612 if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0')) 613 return -1; 614 if (n >= MAX_TIMER_PERIOD) 615 return -1; 616 617 return n; 618 } 619 620 /* Parse the argument given in the command line of the application */ 621 static int 622 l2fwd_parse_args(int argc, char **argv) 623 { 624 int opt, ret; 625 char **argvopt; 626 int option_index; 627 char *prgname = argv[0]; 628 static struct option lgopts[] = { 629 {NULL, 0, 0, 0} 630 }; 631 632 argvopt = argv; 633 634 while ((opt = getopt_long(argc, argvopt, "p:q:T:l", 635 lgopts, &option_index)) != EOF) { 636 637 switch (opt) { 638 /* portmask */ 639 case 'p': 640 l2fwd_enabled_port_mask = l2fwd_parse_portmask(optarg); 641 if (l2fwd_enabled_port_mask == 0) { 642 printf("invalid portmask\n"); 643 l2fwd_usage(prgname); 644 return -1; 645 } 646 break; 647 648 /* nqueue */ 649 case 'q': 650 l2fwd_rx_queue_per_lcore = l2fwd_parse_nqueue(optarg); 651 if (l2fwd_rx_queue_per_lcore == 0) { 652 printf("invalid queue number\n"); 653 l2fwd_usage(prgname); 654 return -1; 655 } 656 break; 657 658 /* timer period */ 659 case 'T': 660 timer_period = l2fwd_parse_timer_period(optarg); 661 if (timer_period < 0) { 662 printf("invalid timer period\n"); 663 l2fwd_usage(prgname); 664 return -1; 665 } 666 break; 667 668 /* For thousands separator in printf. */ 669 case 'l': 670 setlocale(LC_ALL, ""); 671 break; 672 673 /* long options */ 674 case 0: 675 l2fwd_usage(prgname); 676 return -1; 677 678 default: 679 l2fwd_usage(prgname); 680 return -1; 681 } 682 } 683 684 if (optind >= 0) 685 argv[optind-1] = prgname; 686 687 ret = optind-1; 688 optind = 1; /* reset getopt lib */ 689 return ret; 690 } 691 692 /* Check the link status of all ports in up to 9s, and print them finally */ 693 static void 694 check_all_ports_link_status(uint32_t port_mask) 695 { 696 #define CHECK_INTERVAL 100 /* 100ms */ 697 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */ 698 uint16_t portid; 699 uint8_t count, all_ports_up, print_flag = 0; 700 struct rte_eth_link link; 701 int ret; 702 char link_status_text[RTE_ETH_LINK_MAX_STR_LEN]; 703 704 printf("\nChecking link status"); 705 fflush(stdout); 706 for (count = 0; count <= MAX_CHECK_TIME; count++) { 707 all_ports_up = 1; 708 RTE_ETH_FOREACH_DEV(portid) { 709 if ((port_mask & (1 << portid)) == 0) 710 continue; 711 memset(&link, 0, sizeof(link)); 712 ret = rte_eth_link_get_nowait(portid, &link); 713 if (ret < 0) { 714 all_ports_up = 0; 715 if (print_flag == 1) 716 printf("Port %u link get failed: %s\n", 717 portid, rte_strerror(-ret)); 718 continue; 719 } 720 /* print link status if flag set */ 721 if (print_flag == 1) { 722 rte_eth_link_to_str(link_status_text, 723 sizeof(link_status_text), &link); 724 printf("Port %d %s\n", portid, 725 link_status_text); 726 continue; 727 } 728 /* clear all_ports_up flag if any link down */ 729 if (link.link_status == ETH_LINK_DOWN) { 730 all_ports_up = 0; 731 break; 732 } 733 } 734 /* after finally printing all link status, get out */ 735 if (print_flag == 1) 736 break; 737 738 if (all_ports_up == 0) { 739 printf("."); 740 fflush(stdout); 741 rte_delay_ms(CHECK_INTERVAL); 742 } 743 744 /* set the print_flag if all ports up or timeout */ 745 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) { 746 print_flag = 1; 747 printf("done\n"); 748 } 749 } 750 } 751 752 int 753 main(int argc, char **argv) 754 { 755 struct lcore_queue_conf *qconf; 756 unsigned lcore_id, rx_lcore_id; 757 unsigned nb_ports_in_mask = 0; 758 int ret; 759 char name[RTE_JOBSTATS_NAMESIZE]; 760 uint16_t nb_ports; 761 uint16_t nb_ports_available = 0; 762 uint16_t portid, last_port; 763 uint8_t i; 764 765 /* Init EAL. 8< */ 766 ret = rte_eal_init(argc, argv); 767 if (ret < 0) 768 rte_exit(EXIT_FAILURE, "Invalid EAL arguments\n"); 769 argc -= ret; 770 argv += ret; 771 772 /* parse application arguments (after the EAL ones) */ 773 ret = l2fwd_parse_args(argc, argv); 774 if (ret < 0) 775 rte_exit(EXIT_FAILURE, "Invalid L2FWD arguments\n"); 776 /* >8 End of init EAL. */ 777 778 rte_timer_subsystem_init(); 779 780 /* fetch default timer frequency. */ 781 hz = rte_get_timer_hz(); 782 783 /* Create the mbuf pool. 8< */ 784 l2fwd_pktmbuf_pool = 785 rte_pktmbuf_pool_create("mbuf_pool", NB_MBUF, 32, 786 0, RTE_MBUF_DEFAULT_BUF_SIZE, rte_socket_id()); 787 if (l2fwd_pktmbuf_pool == NULL) 788 rte_exit(EXIT_FAILURE, "Cannot init mbuf pool\n"); 789 /* >8 End of creation of mbuf pool. */ 790 nb_ports = rte_eth_dev_count_avail(); 791 if (nb_ports == 0) 792 rte_exit(EXIT_FAILURE, "No Ethernet ports - bye\n"); 793 794 /* Reset l2fwd_dst_ports. 8< */ 795 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) 796 l2fwd_dst_ports[portid] = 0; 797 last_port = 0; 798 799 /* 800 * Each logical core is assigned a dedicated TX queue on each port. 801 */ 802 RTE_ETH_FOREACH_DEV(portid) { 803 /* skip ports that are not enabled */ 804 if ((l2fwd_enabled_port_mask & (1 << portid)) == 0) 805 continue; 806 807 if (nb_ports_in_mask % 2) { 808 l2fwd_dst_ports[portid] = last_port; 809 l2fwd_dst_ports[last_port] = portid; 810 } else 811 last_port = portid; 812 813 nb_ports_in_mask++; 814 } 815 /* >8 End of reset l2fwd_dst_ports. */ 816 if (nb_ports_in_mask % 2) { 817 printf("Notice: odd number of ports in portmask.\n"); 818 l2fwd_dst_ports[last_port] = last_port; 819 } 820 821 rx_lcore_id = 0; 822 qconf = NULL; 823 824 /* Initialize the port/queue configuration of each logical core */ 825 RTE_ETH_FOREACH_DEV(portid) { 826 /* skip ports that are not enabled */ 827 if ((l2fwd_enabled_port_mask & (1 << portid)) == 0) 828 continue; 829 830 /* get the lcore_id for this port */ 831 while (rte_lcore_is_enabled(rx_lcore_id) == 0 || 832 lcore_queue_conf[rx_lcore_id].n_rx_port == 833 l2fwd_rx_queue_per_lcore) { 834 rx_lcore_id++; 835 if (rx_lcore_id >= RTE_MAX_LCORE) 836 rte_exit(EXIT_FAILURE, "Not enough cores\n"); 837 } 838 839 if (qconf != &lcore_queue_conf[rx_lcore_id]) 840 /* Assigned a new logical core in the loop above. */ 841 qconf = &lcore_queue_conf[rx_lcore_id]; 842 843 qconf->rx_port_list[qconf->n_rx_port] = portid; 844 qconf->n_rx_port++; 845 printf("Lcore %u: RX port %u\n", rx_lcore_id, portid); 846 } 847 848 /* Initialise each port */ 849 RTE_ETH_FOREACH_DEV(portid) { 850 struct rte_eth_dev_info dev_info; 851 struct rte_eth_rxconf rxq_conf; 852 struct rte_eth_txconf txq_conf; 853 struct rte_eth_conf local_port_conf = port_conf; 854 855 /* skip ports that are not enabled */ 856 if ((l2fwd_enabled_port_mask & (1 << portid)) == 0) { 857 printf("Skipping disabled port %u\n", portid); 858 continue; 859 } 860 nb_ports_available++; 861 862 /* init port */ 863 printf("Initializing port %u... ", portid); 864 fflush(stdout); 865 866 ret = rte_eth_dev_info_get(portid, &dev_info); 867 if (ret != 0) 868 rte_exit(EXIT_FAILURE, 869 "Error during getting device (port %u) info: %s\n", 870 portid, strerror(-ret)); 871 872 if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE) 873 local_port_conf.txmode.offloads |= 874 DEV_TX_OFFLOAD_MBUF_FAST_FREE; 875 /* Configure the RX and TX queues. 8< */ 876 ret = rte_eth_dev_configure(portid, 1, 1, &local_port_conf); 877 if (ret < 0) 878 rte_exit(EXIT_FAILURE, "Cannot configure device: err=%d, port=%u\n", 879 ret, portid); 880 /* >8 End of configuring the RX and TX queues. */ 881 882 ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd, 883 &nb_txd); 884 if (ret < 0) 885 rte_exit(EXIT_FAILURE, 886 "Cannot adjust number of descriptors: err=%d, port=%u\n", 887 ret, portid); 888 889 ret = rte_eth_macaddr_get(portid, 890 &l2fwd_ports_eth_addr[portid]); 891 if (ret < 0) 892 rte_exit(EXIT_FAILURE, 893 "Cannot get MAC address: err=%d, port=%u\n", 894 ret, portid); 895 896 /* init one RX queue */ 897 fflush(stdout); 898 rxq_conf = dev_info.default_rxconf; 899 rxq_conf.offloads = local_port_conf.rxmode.offloads; 900 /* RX queue initialization. 8< */ 901 ret = rte_eth_rx_queue_setup(portid, 0, nb_rxd, 902 rte_eth_dev_socket_id(portid), 903 &rxq_conf, 904 l2fwd_pktmbuf_pool); 905 if (ret < 0) 906 rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup:err=%d, port=%u\n", 907 ret, portid); 908 /* >8 End of RX queue initialization. */ 909 910 /* Init one TX queue on each port. 8< */ 911 txq_conf = dev_info.default_txconf; 912 txq_conf.offloads = local_port_conf.txmode.offloads; 913 fflush(stdout); 914 ret = rte_eth_tx_queue_setup(portid, 0, nb_txd, 915 rte_eth_dev_socket_id(portid), 916 &txq_conf); 917 if (ret < 0) 918 rte_exit(EXIT_FAILURE, 919 "rte_eth_tx_queue_setup:err=%d, port=%u\n", 920 ret, portid); 921 /* >8 End of init one TX queue on each port. */ 922 923 /* Initialize TX buffers */ 924 tx_buffer[portid] = rte_zmalloc_socket("tx_buffer", 925 RTE_ETH_TX_BUFFER_SIZE(MAX_PKT_BURST), 0, 926 rte_eth_dev_socket_id(portid)); 927 if (tx_buffer[portid] == NULL) 928 rte_exit(EXIT_FAILURE, "Cannot allocate buffer for tx on port %u\n", 929 portid); 930 931 rte_eth_tx_buffer_init(tx_buffer[portid], MAX_PKT_BURST); 932 933 ret = rte_eth_tx_buffer_set_err_callback(tx_buffer[portid], 934 rte_eth_tx_buffer_count_callback, 935 &port_statistics[portid].dropped); 936 if (ret < 0) 937 rte_exit(EXIT_FAILURE, 938 "Cannot set error callback for tx buffer on port %u\n", 939 portid); 940 941 /* Start device */ 942 ret = rte_eth_dev_start(portid); 943 if (ret < 0) 944 rte_exit(EXIT_FAILURE, "rte_eth_dev_start:err=%d, port=%u\n", 945 ret, portid); 946 947 printf("done:\n"); 948 949 ret = rte_eth_promiscuous_enable(portid); 950 if (ret != 0) { 951 rte_exit(EXIT_FAILURE, 952 "rte_eth_promiscuous_enable:err=%s, port=%u\n", 953 rte_strerror(-ret), portid); 954 return ret; 955 956 } 957 958 printf("Port %u, MAC address: " RTE_ETHER_ADDR_PRT_FMT "\n\n", 959 portid, 960 RTE_ETHER_ADDR_BYTES(&l2fwd_ports_eth_addr[portid])); 961 962 /* initialize port stats */ 963 memset(&port_statistics, 0, sizeof(port_statistics)); 964 } 965 966 if (!nb_ports_available) { 967 rte_exit(EXIT_FAILURE, 968 "All available ports are disabled. Please set portmask.\n"); 969 } 970 971 check_all_ports_link_status(l2fwd_enabled_port_mask); 972 973 drain_tsc = (hz + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US; 974 975 RTE_LCORE_FOREACH(lcore_id) { 976 qconf = &lcore_queue_conf[lcore_id]; 977 978 rte_spinlock_init(&qconf->lock); 979 980 if (rte_jobstats_context_init(&qconf->jobs_context) != 0) 981 rte_panic("Jobs stats context for core %u init failed\n", lcore_id); 982 983 if (qconf->n_rx_port == 0) { 984 RTE_LOG(INFO, L2FWD, 985 "lcore %u: no ports so no jobs stats context initialization\n", 986 lcore_id); 987 continue; 988 } 989 /* Add flush job. 8< */ 990 991 /* Set fixed period by setting min = max = initial period. Set target to 992 * zero as it is irrelevant for this job. 993 */ 994 rte_jobstats_init(&qconf->flush_job, "flush", drain_tsc, drain_tsc, 995 drain_tsc, 0); 996 997 rte_timer_init(&qconf->flush_timer); 998 ret = rte_timer_reset(&qconf->flush_timer, drain_tsc, PERIODICAL, 999 lcore_id, &l2fwd_flush_job, NULL); 1000 1001 if (ret < 0) { 1002 rte_exit(1, "Failed to reset flush job timer for lcore %u: %s", 1003 lcore_id, rte_strerror(-ret)); 1004 } 1005 /* >8 End of add flush job. */ 1006 1007 for (i = 0; i < qconf->n_rx_port; i++) { 1008 struct rte_jobstats *job = &qconf->port_fwd_jobs[i]; 1009 1010 portid = qconf->rx_port_list[i]; 1011 printf("Setting forward job for port %u\n", portid); 1012 1013 snprintf(name, RTE_DIM(name), "port %u fwd", portid); 1014 /* Setup forward job. 8< */ 1015 1016 /* Set min, max and initial period. Set target to MAX_PKT_BURST as 1017 * this is desired optimal RX/TX burst size. 1018 */ 1019 rte_jobstats_init(job, name, 0, drain_tsc, 0, MAX_PKT_BURST); 1020 rte_jobstats_set_update_period_function(job, l2fwd_job_update_cb); 1021 1022 rte_timer_init(&qconf->rx_timers[i]); 1023 ret = rte_timer_reset(&qconf->rx_timers[i], 0, PERIODICAL, lcore_id, 1024 &l2fwd_fwd_job, (void *)(uintptr_t)i); 1025 1026 if (ret < 0) { 1027 rte_exit(1, "Failed to reset lcore %u port %u job timer: %s", 1028 lcore_id, qconf->rx_port_list[i], rte_strerror(-ret)); 1029 } 1030 /* >8 End of forward job. */ 1031 } 1032 } 1033 1034 if (timer_period) 1035 rte_eal_alarm_set(timer_period * MS_PER_S, show_stats_cb, NULL); 1036 else 1037 RTE_LOG(INFO, L2FWD, "Stats display disabled\n"); 1038 1039 /* launch per-lcore init on every lcore */ 1040 rte_eal_mp_remote_launch(l2fwd_launch_one_lcore, NULL, CALL_MAIN); 1041 RTE_LCORE_FOREACH_WORKER(lcore_id) { 1042 if (rte_eal_wait_lcore(lcore_id) < 0) 1043 return -1; 1044 } 1045 1046 /* clean up the EAL */ 1047 rte_eal_cleanup(); 1048 1049 return 0; 1050 } 1051