1 /* SPDX-License-Identifier: BSD-3-Clause 2 * Copyright(c) 2010-2016 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 <unistd.h> 16 #include <signal.h> 17 18 #include <rte_common.h> 19 #include <rte_byteorder.h> 20 #include <rte_log.h> 21 #include <rte_malloc.h> 22 #include <rte_memory.h> 23 #include <rte_memcpy.h> 24 #include <rte_eal.h> 25 #include <rte_launch.h> 26 #include <rte_atomic.h> 27 #include <rte_cycles.h> 28 #include <rte_prefetch.h> 29 #include <rte_lcore.h> 30 #include <rte_per_lcore.h> 31 #include <rte_branch_prediction.h> 32 #include <rte_interrupts.h> 33 #include <rte_random.h> 34 #include <rte_debug.h> 35 #include <rte_ether.h> 36 #include <rte_ethdev.h> 37 #include <rte_mempool.h> 38 #include <rte_mbuf.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_timer.h> 44 #include <rte_power.h> 45 #include <rte_spinlock.h> 46 47 #define RTE_LOGTYPE_L3FWD_POWER RTE_LOGTYPE_USER1 48 49 #define MAX_PKT_BURST 32 50 51 #define MIN_ZERO_POLL_COUNT 10 52 53 /* 100 ms interval */ 54 #define TIMER_NUMBER_PER_SECOND 10 55 /* 100000 us */ 56 #define SCALING_PERIOD (1000000/TIMER_NUMBER_PER_SECOND) 57 #define SCALING_DOWN_TIME_RATIO_THRESHOLD 0.25 58 59 #define APP_LOOKUP_EXACT_MATCH 0 60 #define APP_LOOKUP_LPM 1 61 #define DO_RFC_1812_CHECKS 62 63 #ifndef APP_LOOKUP_METHOD 64 #define APP_LOOKUP_METHOD APP_LOOKUP_LPM 65 #endif 66 67 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH) 68 #include <rte_hash.h> 69 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM) 70 #include <rte_lpm.h> 71 #else 72 #error "APP_LOOKUP_METHOD set to incorrect value" 73 #endif 74 75 #ifndef IPv6_BYTES 76 #define IPv6_BYTES_FMT "%02x%02x:%02x%02x:%02x%02x:%02x%02x:"\ 77 "%02x%02x:%02x%02x:%02x%02x:%02x%02x" 78 #define IPv6_BYTES(addr) \ 79 addr[0], addr[1], addr[2], addr[3], \ 80 addr[4], addr[5], addr[6], addr[7], \ 81 addr[8], addr[9], addr[10], addr[11],\ 82 addr[12], addr[13],addr[14], addr[15] 83 #endif 84 85 #define MAX_JUMBO_PKT_LEN 9600 86 87 #define IPV6_ADDR_LEN 16 88 89 #define MEMPOOL_CACHE_SIZE 256 90 91 /* 92 * This expression is used to calculate the number of mbufs needed depending on 93 * user input, taking into account memory for rx and tx hardware rings, cache 94 * per lcore and mtable per port per lcore. RTE_MAX is used to ensure that 95 * NB_MBUF never goes below a minimum value of 8192. 96 */ 97 98 #define NB_MBUF RTE_MAX ( \ 99 (nb_ports*nb_rx_queue*nb_rxd + \ 100 nb_ports*nb_lcores*MAX_PKT_BURST + \ 101 nb_ports*n_tx_queue*nb_txd + \ 102 nb_lcores*MEMPOOL_CACHE_SIZE), \ 103 (unsigned)8192) 104 105 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */ 106 107 #define NB_SOCKETS 8 108 109 /* Configure how many packets ahead to prefetch, when reading packets */ 110 #define PREFETCH_OFFSET 3 111 112 /* 113 * Configurable number of RX/TX ring descriptors 114 */ 115 #define RTE_TEST_RX_DESC_DEFAULT 1024 116 #define RTE_TEST_TX_DESC_DEFAULT 1024 117 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT; 118 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT; 119 120 /* ethernet addresses of ports */ 121 static struct ether_addr ports_eth_addr[RTE_MAX_ETHPORTS]; 122 123 /* ethernet addresses of ports */ 124 static rte_spinlock_t locks[RTE_MAX_ETHPORTS]; 125 126 /* mask of enabled ports */ 127 static uint32_t enabled_port_mask = 0; 128 /* Ports set in promiscuous mode off by default. */ 129 static int promiscuous_on = 0; 130 /* NUMA is enabled by default. */ 131 static int numa_on = 1; 132 static int parse_ptype; /**< Parse packet type using rx callback, and */ 133 /**< disabled by default */ 134 135 enum freq_scale_hint_t 136 { 137 FREQ_LOWER = -1, 138 FREQ_CURRENT = 0, 139 FREQ_HIGHER = 1, 140 FREQ_HIGHEST = 2 141 }; 142 143 struct lcore_rx_queue { 144 uint16_t port_id; 145 uint8_t queue_id; 146 enum freq_scale_hint_t freq_up_hint; 147 uint32_t zero_rx_packet_count; 148 uint32_t idle_hint; 149 } __rte_cache_aligned; 150 151 #define MAX_RX_QUEUE_PER_LCORE 16 152 #define MAX_TX_QUEUE_PER_PORT RTE_MAX_ETHPORTS 153 #define MAX_RX_QUEUE_PER_PORT 128 154 155 #define MAX_RX_QUEUE_INTERRUPT_PER_PORT 16 156 157 158 #define MAX_LCORE_PARAMS 1024 159 struct lcore_params { 160 uint16_t port_id; 161 uint8_t queue_id; 162 uint8_t lcore_id; 163 } __rte_cache_aligned; 164 165 static struct lcore_params lcore_params_array[MAX_LCORE_PARAMS]; 166 static struct lcore_params lcore_params_array_default[] = { 167 {0, 0, 2}, 168 {0, 1, 2}, 169 {0, 2, 2}, 170 {1, 0, 2}, 171 {1, 1, 2}, 172 {1, 2, 2}, 173 {2, 0, 2}, 174 {3, 0, 3}, 175 {3, 1, 3}, 176 }; 177 178 static struct lcore_params * lcore_params = lcore_params_array_default; 179 static uint16_t nb_lcore_params = sizeof(lcore_params_array_default) / 180 sizeof(lcore_params_array_default[0]); 181 182 static struct rte_eth_conf port_conf = { 183 .rxmode = { 184 .mq_mode = ETH_MQ_RX_RSS, 185 .max_rx_pkt_len = ETHER_MAX_LEN, 186 .split_hdr_size = 0, 187 .ignore_offload_bitfield = 1, 188 .offloads = (DEV_RX_OFFLOAD_CRC_STRIP | 189 DEV_RX_OFFLOAD_CHECKSUM), 190 }, 191 .rx_adv_conf = { 192 .rss_conf = { 193 .rss_key = NULL, 194 .rss_hf = ETH_RSS_UDP, 195 }, 196 }, 197 .txmode = { 198 .mq_mode = ETH_MQ_TX_NONE, 199 }, 200 .intr_conf = { 201 .rxq = 1, 202 }, 203 }; 204 205 static struct rte_mempool * pktmbuf_pool[NB_SOCKETS]; 206 207 208 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH) 209 210 #ifdef RTE_ARCH_X86 211 #include <rte_hash_crc.h> 212 #define DEFAULT_HASH_FUNC rte_hash_crc 213 #else 214 #include <rte_jhash.h> 215 #define DEFAULT_HASH_FUNC rte_jhash 216 #endif 217 218 struct ipv4_5tuple { 219 uint32_t ip_dst; 220 uint32_t ip_src; 221 uint16_t port_dst; 222 uint16_t port_src; 223 uint8_t proto; 224 } __attribute__((__packed__)); 225 226 struct ipv6_5tuple { 227 uint8_t ip_dst[IPV6_ADDR_LEN]; 228 uint8_t ip_src[IPV6_ADDR_LEN]; 229 uint16_t port_dst; 230 uint16_t port_src; 231 uint8_t proto; 232 } __attribute__((__packed__)); 233 234 struct ipv4_l3fwd_route { 235 struct ipv4_5tuple key; 236 uint8_t if_out; 237 }; 238 239 struct ipv6_l3fwd_route { 240 struct ipv6_5tuple key; 241 uint8_t if_out; 242 }; 243 244 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = { 245 {{IPv4(100,10,0,1), IPv4(200,10,0,1), 101, 11, IPPROTO_TCP}, 0}, 246 {{IPv4(100,20,0,2), IPv4(200,20,0,2), 102, 12, IPPROTO_TCP}, 1}, 247 {{IPv4(100,30,0,3), IPv4(200,30,0,3), 103, 13, IPPROTO_TCP}, 2}, 248 {{IPv4(100,40,0,4), IPv4(200,40,0,4), 104, 14, IPPROTO_TCP}, 3}, 249 }; 250 251 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = { 252 { 253 { 254 {0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 255 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05}, 256 {0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 257 0x02, 0x1e, 0x67, 0xff, 0xfe, 0x0d, 0xb6, 0x0a}, 258 1, 10, IPPROTO_UDP 259 }, 4 260 }, 261 }; 262 263 typedef struct rte_hash lookup_struct_t; 264 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS]; 265 static lookup_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS]; 266 267 #define L3FWD_HASH_ENTRIES 1024 268 269 #define IPV4_L3FWD_NUM_ROUTES \ 270 (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0])) 271 272 #define IPV6_L3FWD_NUM_ROUTES \ 273 (sizeof(ipv6_l3fwd_route_array) / sizeof(ipv6_l3fwd_route_array[0])) 274 275 static uint16_t ipv4_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned; 276 static uint16_t ipv6_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned; 277 #endif 278 279 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM) 280 struct ipv4_l3fwd_route { 281 uint32_t ip; 282 uint8_t depth; 283 uint8_t if_out; 284 }; 285 286 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = { 287 {IPv4(1,1,1,0), 24, 0}, 288 {IPv4(2,1,1,0), 24, 1}, 289 {IPv4(3,1,1,0), 24, 2}, 290 {IPv4(4,1,1,0), 24, 3}, 291 {IPv4(5,1,1,0), 24, 4}, 292 {IPv4(6,1,1,0), 24, 5}, 293 {IPv4(7,1,1,0), 24, 6}, 294 {IPv4(8,1,1,0), 24, 7}, 295 }; 296 297 #define IPV4_L3FWD_NUM_ROUTES \ 298 (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0])) 299 300 #define IPV4_L3FWD_LPM_MAX_RULES 1024 301 302 typedef struct rte_lpm lookup_struct_t; 303 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS]; 304 #endif 305 306 struct lcore_conf { 307 uint16_t n_rx_queue; 308 struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE]; 309 uint16_t n_tx_port; 310 uint16_t tx_port_id[RTE_MAX_ETHPORTS]; 311 uint16_t tx_queue_id[RTE_MAX_ETHPORTS]; 312 struct rte_eth_dev_tx_buffer *tx_buffer[RTE_MAX_ETHPORTS]; 313 lookup_struct_t * ipv4_lookup_struct; 314 lookup_struct_t * ipv6_lookup_struct; 315 } __rte_cache_aligned; 316 317 struct lcore_stats { 318 /* total sleep time in ms since last frequency scaling down */ 319 uint32_t sleep_time; 320 /* number of long sleep recently */ 321 uint32_t nb_long_sleep; 322 /* freq. scaling up trend */ 323 uint32_t trend; 324 /* total packet processed recently */ 325 uint64_t nb_rx_processed; 326 /* total iterations looped recently */ 327 uint64_t nb_iteration_looped; 328 uint32_t padding[9]; 329 } __rte_cache_aligned; 330 331 static struct lcore_conf lcore_conf[RTE_MAX_LCORE] __rte_cache_aligned; 332 static struct lcore_stats stats[RTE_MAX_LCORE] __rte_cache_aligned; 333 static struct rte_timer power_timers[RTE_MAX_LCORE]; 334 335 static inline uint32_t power_idle_heuristic(uint32_t zero_rx_packet_count); 336 static inline enum freq_scale_hint_t power_freq_scaleup_heuristic( \ 337 unsigned int lcore_id, uint16_t port_id, uint16_t queue_id); 338 339 /* exit signal handler */ 340 static void 341 signal_exit_now(int sigtype) 342 { 343 unsigned lcore_id; 344 unsigned int portid; 345 int ret; 346 347 if (sigtype == SIGINT) { 348 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) { 349 if (rte_lcore_is_enabled(lcore_id) == 0) 350 continue; 351 352 /* init power management library */ 353 ret = rte_power_exit(lcore_id); 354 if (ret) 355 rte_exit(EXIT_FAILURE, "Power management " 356 "library de-initialization failed on " 357 "core%u\n", lcore_id); 358 } 359 360 RTE_ETH_FOREACH_DEV(portid) { 361 if ((enabled_port_mask & (1 << portid)) == 0) 362 continue; 363 364 rte_eth_dev_stop(portid); 365 rte_eth_dev_close(portid); 366 } 367 } 368 369 rte_exit(EXIT_SUCCESS, "User forced exit\n"); 370 } 371 372 /* Freqency scale down timer callback */ 373 static void 374 power_timer_cb(__attribute__((unused)) struct rte_timer *tim, 375 __attribute__((unused)) void *arg) 376 { 377 uint64_t hz; 378 float sleep_time_ratio; 379 unsigned lcore_id = rte_lcore_id(); 380 381 /* accumulate total execution time in us when callback is invoked */ 382 sleep_time_ratio = (float)(stats[lcore_id].sleep_time) / 383 (float)SCALING_PERIOD; 384 /** 385 * check whether need to scale down frequency a step if it sleep a lot. 386 */ 387 if (sleep_time_ratio >= SCALING_DOWN_TIME_RATIO_THRESHOLD) { 388 if (rte_power_freq_down) 389 rte_power_freq_down(lcore_id); 390 } 391 else if ( (unsigned)(stats[lcore_id].nb_rx_processed / 392 stats[lcore_id].nb_iteration_looped) < MAX_PKT_BURST) { 393 /** 394 * scale down a step if average packet per iteration less 395 * than expectation. 396 */ 397 if (rte_power_freq_down) 398 rte_power_freq_down(lcore_id); 399 } 400 401 /** 402 * initialize another timer according to current frequency to ensure 403 * timer interval is relatively fixed. 404 */ 405 hz = rte_get_timer_hz(); 406 rte_timer_reset(&power_timers[lcore_id], hz/TIMER_NUMBER_PER_SECOND, 407 SINGLE, lcore_id, power_timer_cb, NULL); 408 409 stats[lcore_id].nb_rx_processed = 0; 410 stats[lcore_id].nb_iteration_looped = 0; 411 412 stats[lcore_id].sleep_time = 0; 413 } 414 415 /* Enqueue a single packet, and send burst if queue is filled */ 416 static inline int 417 send_single_packet(struct rte_mbuf *m, uint16_t port) 418 { 419 uint32_t lcore_id; 420 struct lcore_conf *qconf; 421 422 lcore_id = rte_lcore_id(); 423 qconf = &lcore_conf[lcore_id]; 424 425 rte_eth_tx_buffer(port, qconf->tx_queue_id[port], 426 qconf->tx_buffer[port], m); 427 428 return 0; 429 } 430 431 #ifdef DO_RFC_1812_CHECKS 432 static inline int 433 is_valid_ipv4_pkt(struct ipv4_hdr *pkt, uint32_t link_len) 434 { 435 /* From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2 */ 436 /* 437 * 1. The packet length reported by the Link Layer must be large 438 * enough to hold the minimum length legal IP datagram (20 bytes). 439 */ 440 if (link_len < sizeof(struct ipv4_hdr)) 441 return -1; 442 443 /* 2. The IP checksum must be correct. */ 444 /* this is checked in H/W */ 445 446 /* 447 * 3. The IP version number must be 4. If the version number is not 4 448 * then the packet may be another version of IP, such as IPng or 449 * ST-II. 450 */ 451 if (((pkt->version_ihl) >> 4) != 4) 452 return -3; 453 /* 454 * 4. The IP header length field must be large enough to hold the 455 * minimum length legal IP datagram (20 bytes = 5 words). 456 */ 457 if ((pkt->version_ihl & 0xf) < 5) 458 return -4; 459 460 /* 461 * 5. The IP total length field must be large enough to hold the IP 462 * datagram header, whose length is specified in the IP header length 463 * field. 464 */ 465 if (rte_cpu_to_be_16(pkt->total_length) < sizeof(struct ipv4_hdr)) 466 return -5; 467 468 return 0; 469 } 470 #endif 471 472 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH) 473 static void 474 print_ipv4_key(struct ipv4_5tuple key) 475 { 476 printf("IP dst = %08x, IP src = %08x, port dst = %d, port src = %d, " 477 "proto = %d\n", (unsigned)key.ip_dst, (unsigned)key.ip_src, 478 key.port_dst, key.port_src, key.proto); 479 } 480 static void 481 print_ipv6_key(struct ipv6_5tuple key) 482 { 483 printf( "IP dst = " IPv6_BYTES_FMT ", IP src = " IPv6_BYTES_FMT ", " 484 "port dst = %d, port src = %d, proto = %d\n", 485 IPv6_BYTES(key.ip_dst), IPv6_BYTES(key.ip_src), 486 key.port_dst, key.port_src, key.proto); 487 } 488 489 static inline uint16_t 490 get_ipv4_dst_port(struct ipv4_hdr *ipv4_hdr, uint16_t portid, 491 lookup_struct_t * ipv4_l3fwd_lookup_struct) 492 { 493 struct ipv4_5tuple key; 494 struct tcp_hdr *tcp; 495 struct udp_hdr *udp; 496 int ret = 0; 497 498 key.ip_dst = rte_be_to_cpu_32(ipv4_hdr->dst_addr); 499 key.ip_src = rte_be_to_cpu_32(ipv4_hdr->src_addr); 500 key.proto = ipv4_hdr->next_proto_id; 501 502 switch (ipv4_hdr->next_proto_id) { 503 case IPPROTO_TCP: 504 tcp = (struct tcp_hdr *)((unsigned char *)ipv4_hdr + 505 sizeof(struct ipv4_hdr)); 506 key.port_dst = rte_be_to_cpu_16(tcp->dst_port); 507 key.port_src = rte_be_to_cpu_16(tcp->src_port); 508 break; 509 510 case IPPROTO_UDP: 511 udp = (struct udp_hdr *)((unsigned char *)ipv4_hdr + 512 sizeof(struct ipv4_hdr)); 513 key.port_dst = rte_be_to_cpu_16(udp->dst_port); 514 key.port_src = rte_be_to_cpu_16(udp->src_port); 515 break; 516 517 default: 518 key.port_dst = 0; 519 key.port_src = 0; 520 break; 521 } 522 523 /* Find destination port */ 524 ret = rte_hash_lookup(ipv4_l3fwd_lookup_struct, (const void *)&key); 525 return ((ret < 0) ? portid : ipv4_l3fwd_out_if[ret]); 526 } 527 528 static inline uint16_t 529 get_ipv6_dst_port(struct ipv6_hdr *ipv6_hdr, uint16_t portid, 530 lookup_struct_t *ipv6_l3fwd_lookup_struct) 531 { 532 struct ipv6_5tuple key; 533 struct tcp_hdr *tcp; 534 struct udp_hdr *udp; 535 int ret = 0; 536 537 memcpy(key.ip_dst, ipv6_hdr->dst_addr, IPV6_ADDR_LEN); 538 memcpy(key.ip_src, ipv6_hdr->src_addr, IPV6_ADDR_LEN); 539 540 key.proto = ipv6_hdr->proto; 541 542 switch (ipv6_hdr->proto) { 543 case IPPROTO_TCP: 544 tcp = (struct tcp_hdr *)((unsigned char *) ipv6_hdr + 545 sizeof(struct ipv6_hdr)); 546 key.port_dst = rte_be_to_cpu_16(tcp->dst_port); 547 key.port_src = rte_be_to_cpu_16(tcp->src_port); 548 break; 549 550 case IPPROTO_UDP: 551 udp = (struct udp_hdr *)((unsigned char *) ipv6_hdr + 552 sizeof(struct ipv6_hdr)); 553 key.port_dst = rte_be_to_cpu_16(udp->dst_port); 554 key.port_src = rte_be_to_cpu_16(udp->src_port); 555 break; 556 557 default: 558 key.port_dst = 0; 559 key.port_src = 0; 560 break; 561 } 562 563 /* Find destination port */ 564 ret = rte_hash_lookup(ipv6_l3fwd_lookup_struct, (const void *)&key); 565 return ((ret < 0) ? portid : ipv6_l3fwd_out_if[ret]); 566 } 567 #endif 568 569 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM) 570 static inline uint16_t 571 get_ipv4_dst_port(struct ipv4_hdr *ipv4_hdr, uint16_t portid, 572 lookup_struct_t *ipv4_l3fwd_lookup_struct) 573 { 574 uint32_t next_hop; 575 576 return ((rte_lpm_lookup(ipv4_l3fwd_lookup_struct, 577 rte_be_to_cpu_32(ipv4_hdr->dst_addr), &next_hop) == 0)? 578 next_hop : portid); 579 } 580 #endif 581 582 static inline void 583 parse_ptype_one(struct rte_mbuf *m) 584 { 585 struct ether_hdr *eth_hdr; 586 uint32_t packet_type = RTE_PTYPE_UNKNOWN; 587 uint16_t ether_type; 588 589 eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *); 590 ether_type = eth_hdr->ether_type; 591 if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv4)) 592 packet_type |= RTE_PTYPE_L3_IPV4_EXT_UNKNOWN; 593 else if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv6)) 594 packet_type |= RTE_PTYPE_L3_IPV6_EXT_UNKNOWN; 595 596 m->packet_type = packet_type; 597 } 598 599 static uint16_t 600 cb_parse_ptype(uint16_t port __rte_unused, uint16_t queue __rte_unused, 601 struct rte_mbuf *pkts[], uint16_t nb_pkts, 602 uint16_t max_pkts __rte_unused, 603 void *user_param __rte_unused) 604 { 605 unsigned int i; 606 607 for (i = 0; i < nb_pkts; ++i) 608 parse_ptype_one(pkts[i]); 609 610 return nb_pkts; 611 } 612 613 static int 614 add_cb_parse_ptype(uint16_t portid, uint16_t queueid) 615 { 616 printf("Port %d: softly parse packet type info\n", portid); 617 if (rte_eth_add_rx_callback(portid, queueid, cb_parse_ptype, NULL)) 618 return 0; 619 620 printf("Failed to add rx callback: port=%d\n", portid); 621 return -1; 622 } 623 624 static inline void 625 l3fwd_simple_forward(struct rte_mbuf *m, uint16_t portid, 626 struct lcore_conf *qconf) 627 { 628 struct ether_hdr *eth_hdr; 629 struct ipv4_hdr *ipv4_hdr; 630 void *d_addr_bytes; 631 uint16_t dst_port; 632 633 eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *); 634 635 if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) { 636 /* Handle IPv4 headers.*/ 637 ipv4_hdr = 638 rte_pktmbuf_mtod_offset(m, struct ipv4_hdr *, 639 sizeof(struct ether_hdr)); 640 641 #ifdef DO_RFC_1812_CHECKS 642 /* Check to make sure the packet is valid (RFC1812) */ 643 if (is_valid_ipv4_pkt(ipv4_hdr, m->pkt_len) < 0) { 644 rte_pktmbuf_free(m); 645 return; 646 } 647 #endif 648 649 dst_port = get_ipv4_dst_port(ipv4_hdr, portid, 650 qconf->ipv4_lookup_struct); 651 if (dst_port >= RTE_MAX_ETHPORTS || 652 (enabled_port_mask & 1 << dst_port) == 0) 653 dst_port = portid; 654 655 /* 02:00:00:00:00:xx */ 656 d_addr_bytes = ð_hdr->d_addr.addr_bytes[0]; 657 *((uint64_t *)d_addr_bytes) = 658 0x000000000002 + ((uint64_t)dst_port << 40); 659 660 #ifdef DO_RFC_1812_CHECKS 661 /* Update time to live and header checksum */ 662 --(ipv4_hdr->time_to_live); 663 ++(ipv4_hdr->hdr_checksum); 664 #endif 665 666 /* src addr */ 667 ether_addr_copy(&ports_eth_addr[dst_port], ð_hdr->s_addr); 668 669 send_single_packet(m, dst_port); 670 } else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) { 671 /* Handle IPv6 headers.*/ 672 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH) 673 struct ipv6_hdr *ipv6_hdr; 674 675 ipv6_hdr = 676 rte_pktmbuf_mtod_offset(m, struct ipv6_hdr *, 677 sizeof(struct ether_hdr)); 678 679 dst_port = get_ipv6_dst_port(ipv6_hdr, portid, 680 qconf->ipv6_lookup_struct); 681 682 if (dst_port >= RTE_MAX_ETHPORTS || 683 (enabled_port_mask & 1 << dst_port) == 0) 684 dst_port = portid; 685 686 /* 02:00:00:00:00:xx */ 687 d_addr_bytes = ð_hdr->d_addr.addr_bytes[0]; 688 *((uint64_t *)d_addr_bytes) = 689 0x000000000002 + ((uint64_t)dst_port << 40); 690 691 /* src addr */ 692 ether_addr_copy(&ports_eth_addr[dst_port], ð_hdr->s_addr); 693 694 send_single_packet(m, dst_port); 695 #else 696 /* We don't currently handle IPv6 packets in LPM mode. */ 697 rte_pktmbuf_free(m); 698 #endif 699 } else 700 rte_pktmbuf_free(m); 701 702 } 703 704 #define MINIMUM_SLEEP_TIME 1 705 #define SUSPEND_THRESHOLD 300 706 707 static inline uint32_t 708 power_idle_heuristic(uint32_t zero_rx_packet_count) 709 { 710 /* If zero count is less than 100, sleep 1us */ 711 if (zero_rx_packet_count < SUSPEND_THRESHOLD) 712 return MINIMUM_SLEEP_TIME; 713 /* If zero count is less than 1000, sleep 100 us which is the 714 minimum latency switching from C3/C6 to C0 715 */ 716 else 717 return SUSPEND_THRESHOLD; 718 } 719 720 static inline enum freq_scale_hint_t 721 power_freq_scaleup_heuristic(unsigned lcore_id, 722 uint16_t port_id, 723 uint16_t queue_id) 724 { 725 uint32_t rxq_count = rte_eth_rx_queue_count(port_id, queue_id); 726 /** 727 * HW Rx queue size is 128 by default, Rx burst read at maximum 32 entries 728 * per iteration 729 */ 730 #define FREQ_GEAR1_RX_PACKET_THRESHOLD MAX_PKT_BURST 731 #define FREQ_GEAR2_RX_PACKET_THRESHOLD (MAX_PKT_BURST*2) 732 #define FREQ_GEAR3_RX_PACKET_THRESHOLD (MAX_PKT_BURST*3) 733 #define FREQ_UP_TREND1_ACC 1 734 #define FREQ_UP_TREND2_ACC 100 735 #define FREQ_UP_THRESHOLD 10000 736 737 if (likely(rxq_count > FREQ_GEAR3_RX_PACKET_THRESHOLD)) { 738 stats[lcore_id].trend = 0; 739 return FREQ_HIGHEST; 740 } else if (likely(rxq_count > FREQ_GEAR2_RX_PACKET_THRESHOLD)) 741 stats[lcore_id].trend += FREQ_UP_TREND2_ACC; 742 else if (likely(rxq_count > FREQ_GEAR1_RX_PACKET_THRESHOLD)) 743 stats[lcore_id].trend += FREQ_UP_TREND1_ACC; 744 745 if (likely(stats[lcore_id].trend > FREQ_UP_THRESHOLD)) { 746 stats[lcore_id].trend = 0; 747 return FREQ_HIGHER; 748 } 749 750 return FREQ_CURRENT; 751 } 752 753 /** 754 * force polling thread sleep until one-shot rx interrupt triggers 755 * @param port_id 756 * Port id. 757 * @param queue_id 758 * Rx queue id. 759 * @return 760 * 0 on success 761 */ 762 static int 763 sleep_until_rx_interrupt(int num) 764 { 765 struct rte_epoll_event event[num]; 766 int n, i; 767 uint16_t port_id; 768 uint8_t queue_id; 769 void *data; 770 771 RTE_LOG(INFO, L3FWD_POWER, 772 "lcore %u sleeps until interrupt triggers\n", 773 rte_lcore_id()); 774 775 n = rte_epoll_wait(RTE_EPOLL_PER_THREAD, event, num, -1); 776 for (i = 0; i < n; i++) { 777 data = event[i].epdata.data; 778 port_id = ((uintptr_t)data) >> CHAR_BIT; 779 queue_id = ((uintptr_t)data) & 780 RTE_LEN2MASK(CHAR_BIT, uint8_t); 781 rte_eth_dev_rx_intr_disable(port_id, queue_id); 782 RTE_LOG(INFO, L3FWD_POWER, 783 "lcore %u is waked up from rx interrupt on" 784 " port %d queue %d\n", 785 rte_lcore_id(), port_id, queue_id); 786 } 787 788 return 0; 789 } 790 791 static void turn_on_intr(struct lcore_conf *qconf) 792 { 793 int i; 794 struct lcore_rx_queue *rx_queue; 795 uint8_t queue_id; 796 uint16_t port_id; 797 798 for (i = 0; i < qconf->n_rx_queue; ++i) { 799 rx_queue = &(qconf->rx_queue_list[i]); 800 port_id = rx_queue->port_id; 801 queue_id = rx_queue->queue_id; 802 803 rte_spinlock_lock(&(locks[port_id])); 804 rte_eth_dev_rx_intr_enable(port_id, queue_id); 805 rte_spinlock_unlock(&(locks[port_id])); 806 } 807 } 808 809 static int event_register(struct lcore_conf *qconf) 810 { 811 struct lcore_rx_queue *rx_queue; 812 uint8_t queueid; 813 uint16_t portid; 814 uint32_t data; 815 int ret; 816 int i; 817 818 for (i = 0; i < qconf->n_rx_queue; ++i) { 819 rx_queue = &(qconf->rx_queue_list[i]); 820 portid = rx_queue->port_id; 821 queueid = rx_queue->queue_id; 822 data = portid << CHAR_BIT | queueid; 823 824 ret = rte_eth_dev_rx_intr_ctl_q(portid, queueid, 825 RTE_EPOLL_PER_THREAD, 826 RTE_INTR_EVENT_ADD, 827 (void *)((uintptr_t)data)); 828 if (ret) 829 return ret; 830 } 831 832 return 0; 833 } 834 835 /* main processing loop */ 836 static int 837 main_loop(__attribute__((unused)) void *dummy) 838 { 839 struct rte_mbuf *pkts_burst[MAX_PKT_BURST]; 840 unsigned lcore_id; 841 uint64_t prev_tsc, diff_tsc, cur_tsc, tim_res_tsc, hz; 842 uint64_t prev_tsc_power = 0, cur_tsc_power, diff_tsc_power; 843 int i, j, nb_rx; 844 uint8_t queueid; 845 uint16_t portid; 846 struct lcore_conf *qconf; 847 struct lcore_rx_queue *rx_queue; 848 enum freq_scale_hint_t lcore_scaleup_hint; 849 uint32_t lcore_rx_idle_count = 0; 850 uint32_t lcore_idle_hint = 0; 851 int intr_en = 0; 852 853 const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US; 854 855 prev_tsc = 0; 856 hz = rte_get_timer_hz(); 857 tim_res_tsc = hz/TIMER_NUMBER_PER_SECOND; 858 859 lcore_id = rte_lcore_id(); 860 qconf = &lcore_conf[lcore_id]; 861 862 if (qconf->n_rx_queue == 0) { 863 RTE_LOG(INFO, L3FWD_POWER, "lcore %u has nothing to do\n", lcore_id); 864 return 0; 865 } 866 867 RTE_LOG(INFO, L3FWD_POWER, "entering main loop on lcore %u\n", lcore_id); 868 869 for (i = 0; i < qconf->n_rx_queue; i++) { 870 portid = qconf->rx_queue_list[i].port_id; 871 queueid = qconf->rx_queue_list[i].queue_id; 872 RTE_LOG(INFO, L3FWD_POWER, " -- lcoreid=%u portid=%u " 873 "rxqueueid=%hhu\n", lcore_id, portid, queueid); 874 } 875 876 /* add into event wait list */ 877 if (event_register(qconf) == 0) 878 intr_en = 1; 879 else 880 RTE_LOG(INFO, L3FWD_POWER, "RX interrupt won't enable.\n"); 881 882 while (1) { 883 stats[lcore_id].nb_iteration_looped++; 884 885 cur_tsc = rte_rdtsc(); 886 cur_tsc_power = cur_tsc; 887 888 /* 889 * TX burst queue drain 890 */ 891 diff_tsc = cur_tsc - prev_tsc; 892 if (unlikely(diff_tsc > drain_tsc)) { 893 for (i = 0; i < qconf->n_tx_port; ++i) { 894 portid = qconf->tx_port_id[i]; 895 rte_eth_tx_buffer_flush(portid, 896 qconf->tx_queue_id[portid], 897 qconf->tx_buffer[portid]); 898 } 899 prev_tsc = cur_tsc; 900 } 901 902 diff_tsc_power = cur_tsc_power - prev_tsc_power; 903 if (diff_tsc_power > tim_res_tsc) { 904 rte_timer_manage(); 905 prev_tsc_power = cur_tsc_power; 906 } 907 908 start_rx: 909 /* 910 * Read packet from RX queues 911 */ 912 lcore_scaleup_hint = FREQ_CURRENT; 913 lcore_rx_idle_count = 0; 914 for (i = 0; i < qconf->n_rx_queue; ++i) { 915 rx_queue = &(qconf->rx_queue_list[i]); 916 rx_queue->idle_hint = 0; 917 portid = rx_queue->port_id; 918 queueid = rx_queue->queue_id; 919 920 nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst, 921 MAX_PKT_BURST); 922 923 stats[lcore_id].nb_rx_processed += nb_rx; 924 if (unlikely(nb_rx == 0)) { 925 /** 926 * no packet received from rx queue, try to 927 * sleep for a while forcing CPU enter deeper 928 * C states. 929 */ 930 rx_queue->zero_rx_packet_count++; 931 932 if (rx_queue->zero_rx_packet_count <= 933 MIN_ZERO_POLL_COUNT) 934 continue; 935 936 rx_queue->idle_hint = power_idle_heuristic(\ 937 rx_queue->zero_rx_packet_count); 938 lcore_rx_idle_count++; 939 } else { 940 rx_queue->zero_rx_packet_count = 0; 941 942 /** 943 * do not scale up frequency immediately as 944 * user to kernel space communication is costly 945 * which might impact packet I/O for received 946 * packets. 947 */ 948 rx_queue->freq_up_hint = 949 power_freq_scaleup_heuristic(lcore_id, 950 portid, queueid); 951 } 952 953 /* Prefetch first packets */ 954 for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) { 955 rte_prefetch0(rte_pktmbuf_mtod( 956 pkts_burst[j], void *)); 957 } 958 959 /* Prefetch and forward already prefetched packets */ 960 for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) { 961 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[ 962 j + PREFETCH_OFFSET], void *)); 963 l3fwd_simple_forward(pkts_burst[j], portid, 964 qconf); 965 } 966 967 /* Forward remaining prefetched packets */ 968 for (; j < nb_rx; j++) { 969 l3fwd_simple_forward(pkts_burst[j], portid, 970 qconf); 971 } 972 } 973 974 if (likely(lcore_rx_idle_count != qconf->n_rx_queue)) { 975 for (i = 1, lcore_scaleup_hint = 976 qconf->rx_queue_list[0].freq_up_hint; 977 i < qconf->n_rx_queue; ++i) { 978 rx_queue = &(qconf->rx_queue_list[i]); 979 if (rx_queue->freq_up_hint > 980 lcore_scaleup_hint) 981 lcore_scaleup_hint = 982 rx_queue->freq_up_hint; 983 } 984 985 if (lcore_scaleup_hint == FREQ_HIGHEST) { 986 if (rte_power_freq_max) 987 rte_power_freq_max(lcore_id); 988 } else if (lcore_scaleup_hint == FREQ_HIGHER) { 989 if (rte_power_freq_up) 990 rte_power_freq_up(lcore_id); 991 } 992 } else { 993 /** 994 * All Rx queues empty in recent consecutive polls, 995 * sleep in a conservative manner, meaning sleep as 996 * less as possible. 997 */ 998 for (i = 1, lcore_idle_hint = 999 qconf->rx_queue_list[0].idle_hint; 1000 i < qconf->n_rx_queue; ++i) { 1001 rx_queue = &(qconf->rx_queue_list[i]); 1002 if (rx_queue->idle_hint < lcore_idle_hint) 1003 lcore_idle_hint = rx_queue->idle_hint; 1004 } 1005 1006 if (lcore_idle_hint < SUSPEND_THRESHOLD) 1007 /** 1008 * execute "pause" instruction to avoid context 1009 * switch which generally take hundred of 1010 * microseconds for short sleep. 1011 */ 1012 rte_delay_us(lcore_idle_hint); 1013 else { 1014 /* suspend until rx interrupt trigges */ 1015 if (intr_en) { 1016 turn_on_intr(qconf); 1017 sleep_until_rx_interrupt( 1018 qconf->n_rx_queue); 1019 /** 1020 * start receiving packets immediately 1021 */ 1022 goto start_rx; 1023 } 1024 } 1025 stats[lcore_id].sleep_time += lcore_idle_hint; 1026 } 1027 } 1028 } 1029 1030 static int 1031 check_lcore_params(void) 1032 { 1033 uint8_t queue, lcore; 1034 uint16_t i; 1035 int socketid; 1036 1037 for (i = 0; i < nb_lcore_params; ++i) { 1038 queue = lcore_params[i].queue_id; 1039 if (queue >= MAX_RX_QUEUE_PER_PORT) { 1040 printf("invalid queue number: %hhu\n", queue); 1041 return -1; 1042 } 1043 lcore = lcore_params[i].lcore_id; 1044 if (!rte_lcore_is_enabled(lcore)) { 1045 printf("error: lcore %hhu is not enabled in lcore " 1046 "mask\n", lcore); 1047 return -1; 1048 } 1049 if ((socketid = rte_lcore_to_socket_id(lcore) != 0) && 1050 (numa_on == 0)) { 1051 printf("warning: lcore %hhu is on socket %d with numa " 1052 "off\n", lcore, socketid); 1053 } 1054 } 1055 return 0; 1056 } 1057 1058 static int 1059 check_port_config(void) 1060 { 1061 unsigned portid; 1062 uint16_t i; 1063 1064 for (i = 0; i < nb_lcore_params; ++i) { 1065 portid = lcore_params[i].port_id; 1066 if ((enabled_port_mask & (1 << portid)) == 0) { 1067 printf("port %u is not enabled in port mask\n", 1068 portid); 1069 return -1; 1070 } 1071 if (!rte_eth_dev_is_valid_port(portid)) { 1072 printf("port %u is not present on the board\n", 1073 portid); 1074 return -1; 1075 } 1076 } 1077 return 0; 1078 } 1079 1080 static uint8_t 1081 get_port_n_rx_queues(const uint16_t port) 1082 { 1083 int queue = -1; 1084 uint16_t i; 1085 1086 for (i = 0; i < nb_lcore_params; ++i) { 1087 if (lcore_params[i].port_id == port && 1088 lcore_params[i].queue_id > queue) 1089 queue = lcore_params[i].queue_id; 1090 } 1091 return (uint8_t)(++queue); 1092 } 1093 1094 static int 1095 init_lcore_rx_queues(void) 1096 { 1097 uint16_t i, nb_rx_queue; 1098 uint8_t lcore; 1099 1100 for (i = 0; i < nb_lcore_params; ++i) { 1101 lcore = lcore_params[i].lcore_id; 1102 nb_rx_queue = lcore_conf[lcore].n_rx_queue; 1103 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) { 1104 printf("error: too many queues (%u) for lcore: %u\n", 1105 (unsigned)nb_rx_queue + 1, (unsigned)lcore); 1106 return -1; 1107 } else { 1108 lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id = 1109 lcore_params[i].port_id; 1110 lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id = 1111 lcore_params[i].queue_id; 1112 lcore_conf[lcore].n_rx_queue++; 1113 } 1114 } 1115 return 0; 1116 } 1117 1118 /* display usage */ 1119 static void 1120 print_usage(const char *prgname) 1121 { 1122 printf ("%s [EAL options] -- -p PORTMASK -P" 1123 " [--config (port,queue,lcore)[,(port,queue,lcore]]" 1124 " [--enable-jumbo [--max-pkt-len PKTLEN]]\n" 1125 " -p PORTMASK: hexadecimal bitmask of ports to configure\n" 1126 " -P : enable promiscuous mode\n" 1127 " --config (port,queue,lcore): rx queues configuration\n" 1128 " --no-numa: optional, disable numa awareness\n" 1129 " --enable-jumbo: enable jumbo frame" 1130 " which max packet len is PKTLEN in decimal (64-9600)\n" 1131 " --parse-ptype: parse packet type by software\n", 1132 prgname); 1133 } 1134 1135 static int parse_max_pkt_len(const char *pktlen) 1136 { 1137 char *end = NULL; 1138 unsigned long len; 1139 1140 /* parse decimal string */ 1141 len = strtoul(pktlen, &end, 10); 1142 if ((pktlen[0] == '\0') || (end == NULL) || (*end != '\0')) 1143 return -1; 1144 1145 if (len == 0) 1146 return -1; 1147 1148 return len; 1149 } 1150 1151 static int 1152 parse_portmask(const char *portmask) 1153 { 1154 char *end = NULL; 1155 unsigned long pm; 1156 1157 /* parse hexadecimal string */ 1158 pm = strtoul(portmask, &end, 16); 1159 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0')) 1160 return -1; 1161 1162 if (pm == 0) 1163 return -1; 1164 1165 return pm; 1166 } 1167 1168 static int 1169 parse_config(const char *q_arg) 1170 { 1171 char s[256]; 1172 const char *p, *p0 = q_arg; 1173 char *end; 1174 enum fieldnames { 1175 FLD_PORT = 0, 1176 FLD_QUEUE, 1177 FLD_LCORE, 1178 _NUM_FLD 1179 }; 1180 unsigned long int_fld[_NUM_FLD]; 1181 char *str_fld[_NUM_FLD]; 1182 int i; 1183 unsigned size; 1184 1185 nb_lcore_params = 0; 1186 1187 while ((p = strchr(p0,'(')) != NULL) { 1188 ++p; 1189 if((p0 = strchr(p,')')) == NULL) 1190 return -1; 1191 1192 size = p0 - p; 1193 if(size >= sizeof(s)) 1194 return -1; 1195 1196 snprintf(s, sizeof(s), "%.*s", size, p); 1197 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') != 1198 _NUM_FLD) 1199 return -1; 1200 for (i = 0; i < _NUM_FLD; i++){ 1201 errno = 0; 1202 int_fld[i] = strtoul(str_fld[i], &end, 0); 1203 if (errno != 0 || end == str_fld[i] || int_fld[i] > 1204 255) 1205 return -1; 1206 } 1207 if (nb_lcore_params >= MAX_LCORE_PARAMS) { 1208 printf("exceeded max number of lcore params: %hu\n", 1209 nb_lcore_params); 1210 return -1; 1211 } 1212 lcore_params_array[nb_lcore_params].port_id = 1213 (uint8_t)int_fld[FLD_PORT]; 1214 lcore_params_array[nb_lcore_params].queue_id = 1215 (uint8_t)int_fld[FLD_QUEUE]; 1216 lcore_params_array[nb_lcore_params].lcore_id = 1217 (uint8_t)int_fld[FLD_LCORE]; 1218 ++nb_lcore_params; 1219 } 1220 lcore_params = lcore_params_array; 1221 1222 return 0; 1223 } 1224 1225 #define CMD_LINE_OPT_PARSE_PTYPE "parse-ptype" 1226 1227 /* Parse the argument given in the command line of the application */ 1228 static int 1229 parse_args(int argc, char **argv) 1230 { 1231 int opt, ret; 1232 char **argvopt; 1233 int option_index; 1234 char *prgname = argv[0]; 1235 static struct option lgopts[] = { 1236 {"config", 1, 0, 0}, 1237 {"no-numa", 0, 0, 0}, 1238 {"enable-jumbo", 0, 0, 0}, 1239 {CMD_LINE_OPT_PARSE_PTYPE, 0, 0, 0}, 1240 {NULL, 0, 0, 0} 1241 }; 1242 1243 argvopt = argv; 1244 1245 while ((opt = getopt_long(argc, argvopt, "p:P", 1246 lgopts, &option_index)) != EOF) { 1247 1248 switch (opt) { 1249 /* portmask */ 1250 case 'p': 1251 enabled_port_mask = parse_portmask(optarg); 1252 if (enabled_port_mask == 0) { 1253 printf("invalid portmask\n"); 1254 print_usage(prgname); 1255 return -1; 1256 } 1257 break; 1258 case 'P': 1259 printf("Promiscuous mode selected\n"); 1260 promiscuous_on = 1; 1261 break; 1262 1263 /* long options */ 1264 case 0: 1265 if (!strncmp(lgopts[option_index].name, "config", 6)) { 1266 ret = parse_config(optarg); 1267 if (ret) { 1268 printf("invalid config\n"); 1269 print_usage(prgname); 1270 return -1; 1271 } 1272 } 1273 1274 if (!strncmp(lgopts[option_index].name, 1275 "no-numa", 7)) { 1276 printf("numa is disabled \n"); 1277 numa_on = 0; 1278 } 1279 1280 if (!strncmp(lgopts[option_index].name, 1281 "enable-jumbo", 12)) { 1282 struct option lenopts = 1283 {"max-pkt-len", required_argument, \ 1284 0, 0}; 1285 1286 printf("jumbo frame is enabled \n"); 1287 port_conf.rxmode.offloads |= 1288 DEV_RX_OFFLOAD_JUMBO_FRAME; 1289 port_conf.txmode.offloads |= 1290 DEV_TX_OFFLOAD_MULTI_SEGS; 1291 1292 /** 1293 * if no max-pkt-len set, use the default value 1294 * ETHER_MAX_LEN 1295 */ 1296 if (0 == getopt_long(argc, argvopt, "", 1297 &lenopts, &option_index)) { 1298 ret = parse_max_pkt_len(optarg); 1299 if ((ret < 64) || 1300 (ret > MAX_JUMBO_PKT_LEN)){ 1301 printf("invalid packet " 1302 "length\n"); 1303 print_usage(prgname); 1304 return -1; 1305 } 1306 port_conf.rxmode.max_rx_pkt_len = ret; 1307 } 1308 printf("set jumbo frame " 1309 "max packet length to %u\n", 1310 (unsigned int)port_conf.rxmode.max_rx_pkt_len); 1311 } 1312 1313 if (!strncmp(lgopts[option_index].name, 1314 CMD_LINE_OPT_PARSE_PTYPE, 1315 sizeof(CMD_LINE_OPT_PARSE_PTYPE))) { 1316 printf("soft parse-ptype is enabled\n"); 1317 parse_ptype = 1; 1318 } 1319 1320 break; 1321 1322 default: 1323 print_usage(prgname); 1324 return -1; 1325 } 1326 } 1327 1328 if (optind >= 0) 1329 argv[optind-1] = prgname; 1330 1331 ret = optind-1; 1332 optind = 1; /* reset getopt lib */ 1333 return ret; 1334 } 1335 1336 static void 1337 print_ethaddr(const char *name, const struct ether_addr *eth_addr) 1338 { 1339 char buf[ETHER_ADDR_FMT_SIZE]; 1340 ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr); 1341 printf("%s%s", name, buf); 1342 } 1343 1344 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH) 1345 static void 1346 setup_hash(int socketid) 1347 { 1348 struct rte_hash_parameters ipv4_l3fwd_hash_params = { 1349 .name = NULL, 1350 .entries = L3FWD_HASH_ENTRIES, 1351 .key_len = sizeof(struct ipv4_5tuple), 1352 .hash_func = DEFAULT_HASH_FUNC, 1353 .hash_func_init_val = 0, 1354 }; 1355 1356 struct rte_hash_parameters ipv6_l3fwd_hash_params = { 1357 .name = NULL, 1358 .entries = L3FWD_HASH_ENTRIES, 1359 .key_len = sizeof(struct ipv6_5tuple), 1360 .hash_func = DEFAULT_HASH_FUNC, 1361 .hash_func_init_val = 0, 1362 }; 1363 1364 unsigned i; 1365 int ret; 1366 char s[64]; 1367 1368 /* create ipv4 hash */ 1369 snprintf(s, sizeof(s), "ipv4_l3fwd_hash_%d", socketid); 1370 ipv4_l3fwd_hash_params.name = s; 1371 ipv4_l3fwd_hash_params.socket_id = socketid; 1372 ipv4_l3fwd_lookup_struct[socketid] = 1373 rte_hash_create(&ipv4_l3fwd_hash_params); 1374 if (ipv4_l3fwd_lookup_struct[socketid] == NULL) 1375 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on " 1376 "socket %d\n", socketid); 1377 1378 /* create ipv6 hash */ 1379 snprintf(s, sizeof(s), "ipv6_l3fwd_hash_%d", socketid); 1380 ipv6_l3fwd_hash_params.name = s; 1381 ipv6_l3fwd_hash_params.socket_id = socketid; 1382 ipv6_l3fwd_lookup_struct[socketid] = 1383 rte_hash_create(&ipv6_l3fwd_hash_params); 1384 if (ipv6_l3fwd_lookup_struct[socketid] == NULL) 1385 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on " 1386 "socket %d\n", socketid); 1387 1388 1389 /* populate the ipv4 hash */ 1390 for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) { 1391 ret = rte_hash_add_key (ipv4_l3fwd_lookup_struct[socketid], 1392 (void *) &ipv4_l3fwd_route_array[i].key); 1393 if (ret < 0) { 1394 rte_exit(EXIT_FAILURE, "Unable to add entry %u to the" 1395 "l3fwd hash on socket %d\n", i, socketid); 1396 } 1397 ipv4_l3fwd_out_if[ret] = ipv4_l3fwd_route_array[i].if_out; 1398 printf("Hash: Adding key\n"); 1399 print_ipv4_key(ipv4_l3fwd_route_array[i].key); 1400 } 1401 1402 /* populate the ipv6 hash */ 1403 for (i = 0; i < IPV6_L3FWD_NUM_ROUTES; i++) { 1404 ret = rte_hash_add_key (ipv6_l3fwd_lookup_struct[socketid], 1405 (void *) &ipv6_l3fwd_route_array[i].key); 1406 if (ret < 0) { 1407 rte_exit(EXIT_FAILURE, "Unable to add entry %u to the" 1408 "l3fwd hash on socket %d\n", i, socketid); 1409 } 1410 ipv6_l3fwd_out_if[ret] = ipv6_l3fwd_route_array[i].if_out; 1411 printf("Hash: Adding key\n"); 1412 print_ipv6_key(ipv6_l3fwd_route_array[i].key); 1413 } 1414 } 1415 #endif 1416 1417 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM) 1418 static void 1419 setup_lpm(int socketid) 1420 { 1421 unsigned i; 1422 int ret; 1423 char s[64]; 1424 1425 /* create the LPM table */ 1426 struct rte_lpm_config lpm_ipv4_config; 1427 1428 lpm_ipv4_config.max_rules = IPV4_L3FWD_LPM_MAX_RULES; 1429 lpm_ipv4_config.number_tbl8s = 256; 1430 lpm_ipv4_config.flags = 0; 1431 1432 snprintf(s, sizeof(s), "IPV4_L3FWD_LPM_%d", socketid); 1433 ipv4_l3fwd_lookup_struct[socketid] = 1434 rte_lpm_create(s, socketid, &lpm_ipv4_config); 1435 if (ipv4_l3fwd_lookup_struct[socketid] == NULL) 1436 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table" 1437 " on socket %d\n", socketid); 1438 1439 /* populate the LPM table */ 1440 for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) { 1441 ret = rte_lpm_add(ipv4_l3fwd_lookup_struct[socketid], 1442 ipv4_l3fwd_route_array[i].ip, 1443 ipv4_l3fwd_route_array[i].depth, 1444 ipv4_l3fwd_route_array[i].if_out); 1445 1446 if (ret < 0) { 1447 rte_exit(EXIT_FAILURE, "Unable to add entry %u to the " 1448 "l3fwd LPM table on socket %d\n", 1449 i, socketid); 1450 } 1451 1452 printf("LPM: Adding route 0x%08x / %d (%d)\n", 1453 (unsigned)ipv4_l3fwd_route_array[i].ip, 1454 ipv4_l3fwd_route_array[i].depth, 1455 ipv4_l3fwd_route_array[i].if_out); 1456 } 1457 } 1458 #endif 1459 1460 static int 1461 init_mem(unsigned nb_mbuf) 1462 { 1463 struct lcore_conf *qconf; 1464 int socketid; 1465 unsigned lcore_id; 1466 char s[64]; 1467 1468 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) { 1469 if (rte_lcore_is_enabled(lcore_id) == 0) 1470 continue; 1471 1472 if (numa_on) 1473 socketid = rte_lcore_to_socket_id(lcore_id); 1474 else 1475 socketid = 0; 1476 1477 if (socketid >= NB_SOCKETS) { 1478 rte_exit(EXIT_FAILURE, "Socket %d of lcore %u is " 1479 "out of range %d\n", socketid, 1480 lcore_id, NB_SOCKETS); 1481 } 1482 if (pktmbuf_pool[socketid] == NULL) { 1483 snprintf(s, sizeof(s), "mbuf_pool_%d", socketid); 1484 pktmbuf_pool[socketid] = 1485 rte_pktmbuf_pool_create(s, nb_mbuf, 1486 MEMPOOL_CACHE_SIZE, 0, 1487 RTE_MBUF_DEFAULT_BUF_SIZE, 1488 socketid); 1489 if (pktmbuf_pool[socketid] == NULL) 1490 rte_exit(EXIT_FAILURE, 1491 "Cannot init mbuf pool on socket %d\n", 1492 socketid); 1493 else 1494 printf("Allocated mbuf pool on socket %d\n", 1495 socketid); 1496 1497 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM) 1498 setup_lpm(socketid); 1499 #else 1500 setup_hash(socketid); 1501 #endif 1502 } 1503 qconf = &lcore_conf[lcore_id]; 1504 qconf->ipv4_lookup_struct = ipv4_l3fwd_lookup_struct[socketid]; 1505 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH) 1506 qconf->ipv6_lookup_struct = ipv6_l3fwd_lookup_struct[socketid]; 1507 #endif 1508 } 1509 return 0; 1510 } 1511 1512 /* Check the link status of all ports in up to 9s, and print them finally */ 1513 static void 1514 check_all_ports_link_status(uint32_t port_mask) 1515 { 1516 #define CHECK_INTERVAL 100 /* 100ms */ 1517 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */ 1518 uint8_t count, all_ports_up, print_flag = 0; 1519 uint16_t portid; 1520 struct rte_eth_link link; 1521 1522 printf("\nChecking link status"); 1523 fflush(stdout); 1524 for (count = 0; count <= MAX_CHECK_TIME; count++) { 1525 all_ports_up = 1; 1526 RTE_ETH_FOREACH_DEV(portid) { 1527 if ((port_mask & (1 << portid)) == 0) 1528 continue; 1529 memset(&link, 0, sizeof(link)); 1530 rte_eth_link_get_nowait(portid, &link); 1531 /* print link status if flag set */ 1532 if (print_flag == 1) { 1533 if (link.link_status) 1534 printf("Port %d Link Up - speed %u " 1535 "Mbps - %s\n", (uint8_t)portid, 1536 (unsigned)link.link_speed, 1537 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ? 1538 ("full-duplex") : ("half-duplex\n")); 1539 else 1540 printf("Port %d Link Down\n", 1541 (uint8_t)portid); 1542 continue; 1543 } 1544 /* clear all_ports_up flag if any link down */ 1545 if (link.link_status == ETH_LINK_DOWN) { 1546 all_ports_up = 0; 1547 break; 1548 } 1549 } 1550 /* after finally printing all link status, get out */ 1551 if (print_flag == 1) 1552 break; 1553 1554 if (all_ports_up == 0) { 1555 printf("."); 1556 fflush(stdout); 1557 rte_delay_ms(CHECK_INTERVAL); 1558 } 1559 1560 /* set the print_flag if all ports up or timeout */ 1561 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) { 1562 print_flag = 1; 1563 printf("done\n"); 1564 } 1565 } 1566 } 1567 1568 static int check_ptype(uint16_t portid) 1569 { 1570 int i, ret; 1571 int ptype_l3_ipv4 = 0; 1572 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH) 1573 int ptype_l3_ipv6 = 0; 1574 #endif 1575 uint32_t ptype_mask = RTE_PTYPE_L3_MASK; 1576 1577 ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, NULL, 0); 1578 if (ret <= 0) 1579 return 0; 1580 1581 uint32_t ptypes[ret]; 1582 1583 ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, ptypes, ret); 1584 for (i = 0; i < ret; ++i) { 1585 if (ptypes[i] & RTE_PTYPE_L3_IPV4) 1586 ptype_l3_ipv4 = 1; 1587 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH) 1588 if (ptypes[i] & RTE_PTYPE_L3_IPV6) 1589 ptype_l3_ipv6 = 1; 1590 #endif 1591 } 1592 1593 if (ptype_l3_ipv4 == 0) 1594 printf("port %d cannot parse RTE_PTYPE_L3_IPV4\n", portid); 1595 1596 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH) 1597 if (ptype_l3_ipv6 == 0) 1598 printf("port %d cannot parse RTE_PTYPE_L3_IPV6\n", portid); 1599 #endif 1600 1601 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM) 1602 if (ptype_l3_ipv4) 1603 #else /* APP_LOOKUP_EXACT_MATCH */ 1604 if (ptype_l3_ipv4 && ptype_l3_ipv6) 1605 #endif 1606 return 1; 1607 1608 return 0; 1609 1610 } 1611 1612 int 1613 main(int argc, char **argv) 1614 { 1615 struct lcore_conf *qconf; 1616 struct rte_eth_dev_info dev_info; 1617 struct rte_eth_txconf *txconf; 1618 int ret; 1619 uint16_t nb_ports; 1620 uint16_t queueid; 1621 unsigned lcore_id; 1622 uint64_t hz; 1623 uint32_t n_tx_queue, nb_lcores; 1624 uint32_t dev_rxq_num, dev_txq_num; 1625 uint8_t nb_rx_queue, queue, socketid; 1626 uint16_t portid; 1627 1628 /* catch SIGINT and restore cpufreq governor to ondemand */ 1629 signal(SIGINT, signal_exit_now); 1630 1631 /* init EAL */ 1632 ret = rte_eal_init(argc, argv); 1633 if (ret < 0) 1634 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n"); 1635 argc -= ret; 1636 argv += ret; 1637 1638 /* init RTE timer library to be used late */ 1639 rte_timer_subsystem_init(); 1640 1641 /* parse application arguments (after the EAL ones) */ 1642 ret = parse_args(argc, argv); 1643 if (ret < 0) 1644 rte_exit(EXIT_FAILURE, "Invalid L3FWD parameters\n"); 1645 1646 if (check_lcore_params() < 0) 1647 rte_exit(EXIT_FAILURE, "check_lcore_params failed\n"); 1648 1649 ret = init_lcore_rx_queues(); 1650 if (ret < 0) 1651 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n"); 1652 1653 nb_ports = rte_eth_dev_count_avail(); 1654 1655 if (check_port_config() < 0) 1656 rte_exit(EXIT_FAILURE, "check_port_config failed\n"); 1657 1658 nb_lcores = rte_lcore_count(); 1659 1660 /* initialize all ports */ 1661 RTE_ETH_FOREACH_DEV(portid) { 1662 struct rte_eth_conf local_port_conf = port_conf; 1663 1664 /* skip ports that are not enabled */ 1665 if ((enabled_port_mask & (1 << portid)) == 0) { 1666 printf("\nSkipping disabled port %d\n", portid); 1667 continue; 1668 } 1669 1670 /* init port */ 1671 printf("Initializing port %d ... ", portid ); 1672 fflush(stdout); 1673 1674 rte_eth_dev_info_get(portid, &dev_info); 1675 dev_rxq_num = dev_info.max_rx_queues; 1676 dev_txq_num = dev_info.max_tx_queues; 1677 1678 nb_rx_queue = get_port_n_rx_queues(portid); 1679 if (nb_rx_queue > dev_rxq_num) 1680 rte_exit(EXIT_FAILURE, 1681 "Cannot configure not existed rxq: " 1682 "port=%d\n", portid); 1683 1684 n_tx_queue = nb_lcores; 1685 if (n_tx_queue > dev_txq_num) 1686 n_tx_queue = dev_txq_num; 1687 printf("Creating queues: nb_rxq=%d nb_txq=%u... ", 1688 nb_rx_queue, (unsigned)n_tx_queue ); 1689 /* If number of Rx queue is 0, no need to enable Rx interrupt */ 1690 if (nb_rx_queue == 0) 1691 local_port_conf.intr_conf.rxq = 0; 1692 rte_eth_dev_info_get(portid, &dev_info); 1693 if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE) 1694 local_port_conf.txmode.offloads |= 1695 DEV_TX_OFFLOAD_MBUF_FAST_FREE; 1696 ret = rte_eth_dev_configure(portid, nb_rx_queue, 1697 (uint16_t)n_tx_queue, &local_port_conf); 1698 if (ret < 0) 1699 rte_exit(EXIT_FAILURE, "Cannot configure device: " 1700 "err=%d, port=%d\n", ret, portid); 1701 1702 ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd, 1703 &nb_txd); 1704 if (ret < 0) 1705 rte_exit(EXIT_FAILURE, 1706 "Cannot adjust number of descriptors: err=%d, port=%d\n", 1707 ret, portid); 1708 1709 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]); 1710 print_ethaddr(" Address:", &ports_eth_addr[portid]); 1711 printf(", "); 1712 1713 /* init memory */ 1714 ret = init_mem(NB_MBUF); 1715 if (ret < 0) 1716 rte_exit(EXIT_FAILURE, "init_mem failed\n"); 1717 1718 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) { 1719 if (rte_lcore_is_enabled(lcore_id) == 0) 1720 continue; 1721 1722 /* Initialize TX buffers */ 1723 qconf = &lcore_conf[lcore_id]; 1724 qconf->tx_buffer[portid] = rte_zmalloc_socket("tx_buffer", 1725 RTE_ETH_TX_BUFFER_SIZE(MAX_PKT_BURST), 0, 1726 rte_eth_dev_socket_id(portid)); 1727 if (qconf->tx_buffer[portid] == NULL) 1728 rte_exit(EXIT_FAILURE, "Can't allocate tx buffer for port %u\n", 1729 portid); 1730 1731 rte_eth_tx_buffer_init(qconf->tx_buffer[portid], MAX_PKT_BURST); 1732 } 1733 1734 /* init one TX queue per couple (lcore,port) */ 1735 queueid = 0; 1736 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) { 1737 if (rte_lcore_is_enabled(lcore_id) == 0) 1738 continue; 1739 1740 if (queueid >= dev_txq_num) 1741 continue; 1742 1743 if (numa_on) 1744 socketid = \ 1745 (uint8_t)rte_lcore_to_socket_id(lcore_id); 1746 else 1747 socketid = 0; 1748 1749 printf("txq=%u,%d,%d ", lcore_id, queueid, socketid); 1750 fflush(stdout); 1751 1752 txconf = &dev_info.default_txconf; 1753 txconf->txq_flags = ETH_TXQ_FLAGS_IGNORE; 1754 txconf->offloads = local_port_conf.txmode.offloads; 1755 ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd, 1756 socketid, txconf); 1757 if (ret < 0) 1758 rte_exit(EXIT_FAILURE, 1759 "rte_eth_tx_queue_setup: err=%d, " 1760 "port=%d\n", ret, portid); 1761 1762 qconf = &lcore_conf[lcore_id]; 1763 qconf->tx_queue_id[portid] = queueid; 1764 queueid++; 1765 1766 qconf->tx_port_id[qconf->n_tx_port] = portid; 1767 qconf->n_tx_port++; 1768 } 1769 printf("\n"); 1770 } 1771 1772 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) { 1773 if (rte_lcore_is_enabled(lcore_id) == 0) 1774 continue; 1775 1776 /* init power management library */ 1777 ret = rte_power_init(lcore_id); 1778 if (ret) 1779 RTE_LOG(ERR, POWER, 1780 "Library initialization failed on core %u\n", lcore_id); 1781 1782 /* init timer structures for each enabled lcore */ 1783 rte_timer_init(&power_timers[lcore_id]); 1784 hz = rte_get_timer_hz(); 1785 rte_timer_reset(&power_timers[lcore_id], 1786 hz/TIMER_NUMBER_PER_SECOND, SINGLE, lcore_id, 1787 power_timer_cb, NULL); 1788 1789 qconf = &lcore_conf[lcore_id]; 1790 printf("\nInitializing rx queues on lcore %u ... ", lcore_id ); 1791 fflush(stdout); 1792 /* init RX queues */ 1793 for(queue = 0; queue < qconf->n_rx_queue; ++queue) { 1794 struct rte_eth_rxconf rxq_conf; 1795 struct rte_eth_dev *dev; 1796 struct rte_eth_conf *conf; 1797 1798 portid = qconf->rx_queue_list[queue].port_id; 1799 queueid = qconf->rx_queue_list[queue].queue_id; 1800 dev = &rte_eth_devices[portid]; 1801 conf = &dev->data->dev_conf; 1802 1803 if (numa_on) 1804 socketid = \ 1805 (uint8_t)rte_lcore_to_socket_id(lcore_id); 1806 else 1807 socketid = 0; 1808 1809 printf("rxq=%d,%d,%d ", portid, queueid, socketid); 1810 fflush(stdout); 1811 1812 rte_eth_dev_info_get(portid, &dev_info); 1813 rxq_conf = dev_info.default_rxconf; 1814 rxq_conf.offloads = conf->rxmode.offloads; 1815 ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd, 1816 socketid, &rxq_conf, 1817 pktmbuf_pool[socketid]); 1818 if (ret < 0) 1819 rte_exit(EXIT_FAILURE, 1820 "rte_eth_rx_queue_setup: err=%d, " 1821 "port=%d\n", ret, portid); 1822 1823 if (parse_ptype) { 1824 if (add_cb_parse_ptype(portid, queueid) < 0) 1825 rte_exit(EXIT_FAILURE, 1826 "Fail to add ptype cb\n"); 1827 } else if (!check_ptype(portid)) 1828 rte_exit(EXIT_FAILURE, 1829 "PMD can not provide needed ptypes\n"); 1830 } 1831 } 1832 1833 printf("\n"); 1834 1835 /* start ports */ 1836 RTE_ETH_FOREACH_DEV(portid) { 1837 if ((enabled_port_mask & (1 << portid)) == 0) { 1838 continue; 1839 } 1840 /* Start device */ 1841 ret = rte_eth_dev_start(portid); 1842 if (ret < 0) 1843 rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, " 1844 "port=%d\n", ret, portid); 1845 /* 1846 * If enabled, put device in promiscuous mode. 1847 * This allows IO forwarding mode to forward packets 1848 * to itself through 2 cross-connected ports of the 1849 * target machine. 1850 */ 1851 if (promiscuous_on) 1852 rte_eth_promiscuous_enable(portid); 1853 /* initialize spinlock for each port */ 1854 rte_spinlock_init(&(locks[portid])); 1855 } 1856 1857 check_all_ports_link_status(enabled_port_mask); 1858 1859 /* launch per-lcore init on every lcore */ 1860 rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER); 1861 RTE_LCORE_FOREACH_SLAVE(lcore_id) { 1862 if (rte_eal_wait_lcore(lcore_id) < 0) 1863 return -1; 1864 } 1865 1866 return 0; 1867 } 1868