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 512 116 #define RTE_TEST_TX_DESC_DEFAULT 512 117 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, nb_ports; 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 nb_ports = rte_eth_dev_count(); 361 for (portid = 0; portid < nb_ports; portid++) { 362 if ((enabled_port_mask & (1 << portid)) == 0) 363 continue; 364 365 rte_eth_dev_stop(portid); 366 rte_eth_dev_close(portid); 367 } 368 } 369 370 rte_exit(EXIT_SUCCESS, "User forced exit\n"); 371 } 372 373 /* Freqency scale down timer callback */ 374 static void 375 power_timer_cb(__attribute__((unused)) struct rte_timer *tim, 376 __attribute__((unused)) void *arg) 377 { 378 uint64_t hz; 379 float sleep_time_ratio; 380 unsigned lcore_id = rte_lcore_id(); 381 382 /* accumulate total execution time in us when callback is invoked */ 383 sleep_time_ratio = (float)(stats[lcore_id].sleep_time) / 384 (float)SCALING_PERIOD; 385 /** 386 * check whether need to scale down frequency a step if it sleep a lot. 387 */ 388 if (sleep_time_ratio >= SCALING_DOWN_TIME_RATIO_THRESHOLD) { 389 if (rte_power_freq_down) 390 rte_power_freq_down(lcore_id); 391 } 392 else if ( (unsigned)(stats[lcore_id].nb_rx_processed / 393 stats[lcore_id].nb_iteration_looped) < MAX_PKT_BURST) { 394 /** 395 * scale down a step if average packet per iteration less 396 * than expectation. 397 */ 398 if (rte_power_freq_down) 399 rte_power_freq_down(lcore_id); 400 } 401 402 /** 403 * initialize another timer according to current frequency to ensure 404 * timer interval is relatively fixed. 405 */ 406 hz = rte_get_timer_hz(); 407 rte_timer_reset(&power_timers[lcore_id], hz/TIMER_NUMBER_PER_SECOND, 408 SINGLE, lcore_id, power_timer_cb, NULL); 409 410 stats[lcore_id].nb_rx_processed = 0; 411 stats[lcore_id].nb_iteration_looped = 0; 412 413 stats[lcore_id].sleep_time = 0; 414 } 415 416 /* Enqueue a single packet, and send burst if queue is filled */ 417 static inline int 418 send_single_packet(struct rte_mbuf *m, uint16_t port) 419 { 420 uint32_t lcore_id; 421 struct lcore_conf *qconf; 422 423 lcore_id = rte_lcore_id(); 424 qconf = &lcore_conf[lcore_id]; 425 426 rte_eth_tx_buffer(port, qconf->tx_queue_id[port], 427 qconf->tx_buffer[port], m); 428 429 return 0; 430 } 431 432 #ifdef DO_RFC_1812_CHECKS 433 static inline int 434 is_valid_ipv4_pkt(struct ipv4_hdr *pkt, uint32_t link_len) 435 { 436 /* From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2 */ 437 /* 438 * 1. The packet length reported by the Link Layer must be large 439 * enough to hold the minimum length legal IP datagram (20 bytes). 440 */ 441 if (link_len < sizeof(struct ipv4_hdr)) 442 return -1; 443 444 /* 2. The IP checksum must be correct. */ 445 /* this is checked in H/W */ 446 447 /* 448 * 3. The IP version number must be 4. If the version number is not 4 449 * then the packet may be another version of IP, such as IPng or 450 * ST-II. 451 */ 452 if (((pkt->version_ihl) >> 4) != 4) 453 return -3; 454 /* 455 * 4. The IP header length field must be large enough to hold the 456 * minimum length legal IP datagram (20 bytes = 5 words). 457 */ 458 if ((pkt->version_ihl & 0xf) < 5) 459 return -4; 460 461 /* 462 * 5. The IP total length field must be large enough to hold the IP 463 * datagram header, whose length is specified in the IP header length 464 * field. 465 */ 466 if (rte_cpu_to_be_16(pkt->total_length) < sizeof(struct ipv4_hdr)) 467 return -5; 468 469 return 0; 470 } 471 #endif 472 473 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH) 474 static void 475 print_ipv4_key(struct ipv4_5tuple key) 476 { 477 printf("IP dst = %08x, IP src = %08x, port dst = %d, port src = %d, " 478 "proto = %d\n", (unsigned)key.ip_dst, (unsigned)key.ip_src, 479 key.port_dst, key.port_src, key.proto); 480 } 481 static void 482 print_ipv6_key(struct ipv6_5tuple key) 483 { 484 printf( "IP dst = " IPv6_BYTES_FMT ", IP src = " IPv6_BYTES_FMT ", " 485 "port dst = %d, port src = %d, proto = %d\n", 486 IPv6_BYTES(key.ip_dst), IPv6_BYTES(key.ip_src), 487 key.port_dst, key.port_src, key.proto); 488 } 489 490 static inline uint16_t 491 get_ipv4_dst_port(struct ipv4_hdr *ipv4_hdr, uint16_t portid, 492 lookup_struct_t * ipv4_l3fwd_lookup_struct) 493 { 494 struct ipv4_5tuple key; 495 struct tcp_hdr *tcp; 496 struct udp_hdr *udp; 497 int ret = 0; 498 499 key.ip_dst = rte_be_to_cpu_32(ipv4_hdr->dst_addr); 500 key.ip_src = rte_be_to_cpu_32(ipv4_hdr->src_addr); 501 key.proto = ipv4_hdr->next_proto_id; 502 503 switch (ipv4_hdr->next_proto_id) { 504 case IPPROTO_TCP: 505 tcp = (struct tcp_hdr *)((unsigned char *)ipv4_hdr + 506 sizeof(struct ipv4_hdr)); 507 key.port_dst = rte_be_to_cpu_16(tcp->dst_port); 508 key.port_src = rte_be_to_cpu_16(tcp->src_port); 509 break; 510 511 case IPPROTO_UDP: 512 udp = (struct udp_hdr *)((unsigned char *)ipv4_hdr + 513 sizeof(struct ipv4_hdr)); 514 key.port_dst = rte_be_to_cpu_16(udp->dst_port); 515 key.port_src = rte_be_to_cpu_16(udp->src_port); 516 break; 517 518 default: 519 key.port_dst = 0; 520 key.port_src = 0; 521 break; 522 } 523 524 /* Find destination port */ 525 ret = rte_hash_lookup(ipv4_l3fwd_lookup_struct, (const void *)&key); 526 return ((ret < 0) ? portid : ipv4_l3fwd_out_if[ret]); 527 } 528 529 static inline uint16_t 530 get_ipv6_dst_port(struct ipv6_hdr *ipv6_hdr, uint16_t portid, 531 lookup_struct_t *ipv6_l3fwd_lookup_struct) 532 { 533 struct ipv6_5tuple key; 534 struct tcp_hdr *tcp; 535 struct udp_hdr *udp; 536 int ret = 0; 537 538 memcpy(key.ip_dst, ipv6_hdr->dst_addr, IPV6_ADDR_LEN); 539 memcpy(key.ip_src, ipv6_hdr->src_addr, IPV6_ADDR_LEN); 540 541 key.proto = ipv6_hdr->proto; 542 543 switch (ipv6_hdr->proto) { 544 case IPPROTO_TCP: 545 tcp = (struct tcp_hdr *)((unsigned char *) ipv6_hdr + 546 sizeof(struct ipv6_hdr)); 547 key.port_dst = rte_be_to_cpu_16(tcp->dst_port); 548 key.port_src = rte_be_to_cpu_16(tcp->src_port); 549 break; 550 551 case IPPROTO_UDP: 552 udp = (struct udp_hdr *)((unsigned char *) ipv6_hdr + 553 sizeof(struct ipv6_hdr)); 554 key.port_dst = rte_be_to_cpu_16(udp->dst_port); 555 key.port_src = rte_be_to_cpu_16(udp->src_port); 556 break; 557 558 default: 559 key.port_dst = 0; 560 key.port_src = 0; 561 break; 562 } 563 564 /* Find destination port */ 565 ret = rte_hash_lookup(ipv6_l3fwd_lookup_struct, (const void *)&key); 566 return ((ret < 0) ? portid : ipv6_l3fwd_out_if[ret]); 567 } 568 #endif 569 570 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM) 571 static inline uint16_t 572 get_ipv4_dst_port(struct ipv4_hdr *ipv4_hdr, uint16_t portid, 573 lookup_struct_t *ipv4_l3fwd_lookup_struct) 574 { 575 uint32_t next_hop; 576 577 return ((rte_lpm_lookup(ipv4_l3fwd_lookup_struct, 578 rte_be_to_cpu_32(ipv4_hdr->dst_addr), &next_hop) == 0)? 579 next_hop : portid); 580 } 581 #endif 582 583 static inline void 584 parse_ptype_one(struct rte_mbuf *m) 585 { 586 struct ether_hdr *eth_hdr; 587 uint32_t packet_type = RTE_PTYPE_UNKNOWN; 588 uint16_t ether_type; 589 590 eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *); 591 ether_type = eth_hdr->ether_type; 592 if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv4)) 593 packet_type |= RTE_PTYPE_L3_IPV4_EXT_UNKNOWN; 594 else if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv6)) 595 packet_type |= RTE_PTYPE_L3_IPV6_EXT_UNKNOWN; 596 597 m->packet_type = packet_type; 598 } 599 600 static uint16_t 601 cb_parse_ptype(uint16_t port __rte_unused, uint16_t queue __rte_unused, 602 struct rte_mbuf *pkts[], uint16_t nb_pkts, 603 uint16_t max_pkts __rte_unused, 604 void *user_param __rte_unused) 605 { 606 unsigned int i; 607 608 for (i = 0; i < nb_pkts; ++i) 609 parse_ptype_one(pkts[i]); 610 611 return nb_pkts; 612 } 613 614 static int 615 add_cb_parse_ptype(uint16_t portid, uint16_t queueid) 616 { 617 printf("Port %d: softly parse packet type info\n", portid); 618 if (rte_eth_add_rx_callback(portid, queueid, cb_parse_ptype, NULL)) 619 return 0; 620 621 printf("Failed to add rx callback: port=%d\n", portid); 622 return -1; 623 } 624 625 static inline void 626 l3fwd_simple_forward(struct rte_mbuf *m, uint16_t portid, 627 struct lcore_conf *qconf) 628 { 629 struct ether_hdr *eth_hdr; 630 struct ipv4_hdr *ipv4_hdr; 631 void *d_addr_bytes; 632 uint16_t dst_port; 633 634 eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *); 635 636 if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) { 637 /* Handle IPv4 headers.*/ 638 ipv4_hdr = 639 rte_pktmbuf_mtod_offset(m, struct ipv4_hdr *, 640 sizeof(struct ether_hdr)); 641 642 #ifdef DO_RFC_1812_CHECKS 643 /* Check to make sure the packet is valid (RFC1812) */ 644 if (is_valid_ipv4_pkt(ipv4_hdr, m->pkt_len) < 0) { 645 rte_pktmbuf_free(m); 646 return; 647 } 648 #endif 649 650 dst_port = get_ipv4_dst_port(ipv4_hdr, portid, 651 qconf->ipv4_lookup_struct); 652 if (dst_port >= RTE_MAX_ETHPORTS || 653 (enabled_port_mask & 1 << dst_port) == 0) 654 dst_port = portid; 655 656 /* 02:00:00:00:00:xx */ 657 d_addr_bytes = ð_hdr->d_addr.addr_bytes[0]; 658 *((uint64_t *)d_addr_bytes) = 659 0x000000000002 + ((uint64_t)dst_port << 40); 660 661 #ifdef DO_RFC_1812_CHECKS 662 /* Update time to live and header checksum */ 663 --(ipv4_hdr->time_to_live); 664 ++(ipv4_hdr->hdr_checksum); 665 #endif 666 667 /* src addr */ 668 ether_addr_copy(&ports_eth_addr[dst_port], ð_hdr->s_addr); 669 670 send_single_packet(m, dst_port); 671 } else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) { 672 /* Handle IPv6 headers.*/ 673 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH) 674 struct ipv6_hdr *ipv6_hdr; 675 676 ipv6_hdr = 677 rte_pktmbuf_mtod_offset(m, struct ipv6_hdr *, 678 sizeof(struct ether_hdr)); 679 680 dst_port = get_ipv6_dst_port(ipv6_hdr, portid, 681 qconf->ipv6_lookup_struct); 682 683 if (dst_port >= RTE_MAX_ETHPORTS || 684 (enabled_port_mask & 1 << dst_port) == 0) 685 dst_port = portid; 686 687 /* 02:00:00:00:00:xx */ 688 d_addr_bytes = ð_hdr->d_addr.addr_bytes[0]; 689 *((uint64_t *)d_addr_bytes) = 690 0x000000000002 + ((uint64_t)dst_port << 40); 691 692 /* src addr */ 693 ether_addr_copy(&ports_eth_addr[dst_port], ð_hdr->s_addr); 694 695 send_single_packet(m, dst_port); 696 #else 697 /* We don't currently handle IPv6 packets in LPM mode. */ 698 rte_pktmbuf_free(m); 699 #endif 700 } else 701 rte_pktmbuf_free(m); 702 703 } 704 705 #define MINIMUM_SLEEP_TIME 1 706 #define SUSPEND_THRESHOLD 300 707 708 static inline uint32_t 709 power_idle_heuristic(uint32_t zero_rx_packet_count) 710 { 711 /* If zero count is less than 100, sleep 1us */ 712 if (zero_rx_packet_count < SUSPEND_THRESHOLD) 713 return MINIMUM_SLEEP_TIME; 714 /* If zero count is less than 1000, sleep 100 us which is the 715 minimum latency switching from C3/C6 to C0 716 */ 717 else 718 return SUSPEND_THRESHOLD; 719 } 720 721 static inline enum freq_scale_hint_t 722 power_freq_scaleup_heuristic(unsigned lcore_id, 723 uint16_t port_id, 724 uint16_t queue_id) 725 { 726 uint32_t rxq_count = rte_eth_rx_queue_count(port_id, queue_id); 727 /** 728 * HW Rx queue size is 128 by default, Rx burst read at maximum 32 entries 729 * per iteration 730 */ 731 #define FREQ_GEAR1_RX_PACKET_THRESHOLD MAX_PKT_BURST 732 #define FREQ_GEAR2_RX_PACKET_THRESHOLD (MAX_PKT_BURST*2) 733 #define FREQ_GEAR3_RX_PACKET_THRESHOLD (MAX_PKT_BURST*3) 734 #define FREQ_UP_TREND1_ACC 1 735 #define FREQ_UP_TREND2_ACC 100 736 #define FREQ_UP_THRESHOLD 10000 737 738 if (likely(rxq_count > FREQ_GEAR3_RX_PACKET_THRESHOLD)) { 739 stats[lcore_id].trend = 0; 740 return FREQ_HIGHEST; 741 } else if (likely(rxq_count > FREQ_GEAR2_RX_PACKET_THRESHOLD)) 742 stats[lcore_id].trend += FREQ_UP_TREND2_ACC; 743 else if (likely(rxq_count > FREQ_GEAR1_RX_PACKET_THRESHOLD)) 744 stats[lcore_id].trend += FREQ_UP_TREND1_ACC; 745 746 if (likely(stats[lcore_id].trend > FREQ_UP_THRESHOLD)) { 747 stats[lcore_id].trend = 0; 748 return FREQ_HIGHER; 749 } 750 751 return FREQ_CURRENT; 752 } 753 754 /** 755 * force polling thread sleep until one-shot rx interrupt triggers 756 * @param port_id 757 * Port id. 758 * @param queue_id 759 * Rx queue id. 760 * @return 761 * 0 on success 762 */ 763 static int 764 sleep_until_rx_interrupt(int num) 765 { 766 struct rte_epoll_event event[num]; 767 int n, i; 768 uint16_t port_id; 769 uint8_t queue_id; 770 void *data; 771 772 RTE_LOG(INFO, L3FWD_POWER, 773 "lcore %u sleeps until interrupt triggers\n", 774 rte_lcore_id()); 775 776 n = rte_epoll_wait(RTE_EPOLL_PER_THREAD, event, num, -1); 777 for (i = 0; i < n; i++) { 778 data = event[i].epdata.data; 779 port_id = ((uintptr_t)data) >> CHAR_BIT; 780 queue_id = ((uintptr_t)data) & 781 RTE_LEN2MASK(CHAR_BIT, uint8_t); 782 rte_eth_dev_rx_intr_disable(port_id, queue_id); 783 RTE_LOG(INFO, L3FWD_POWER, 784 "lcore %u is waked up from rx interrupt on" 785 " port %d queue %d\n", 786 rte_lcore_id(), port_id, queue_id); 787 } 788 789 return 0; 790 } 791 792 static void turn_on_intr(struct lcore_conf *qconf) 793 { 794 int i; 795 struct lcore_rx_queue *rx_queue; 796 uint8_t queue_id; 797 uint16_t port_id; 798 799 for (i = 0; i < qconf->n_rx_queue; ++i) { 800 rx_queue = &(qconf->rx_queue_list[i]); 801 port_id = rx_queue->port_id; 802 queue_id = rx_queue->queue_id; 803 804 rte_spinlock_lock(&(locks[port_id])); 805 rte_eth_dev_rx_intr_enable(port_id, queue_id); 806 rte_spinlock_unlock(&(locks[port_id])); 807 } 808 } 809 810 static int event_register(struct lcore_conf *qconf) 811 { 812 struct lcore_rx_queue *rx_queue; 813 uint8_t queueid; 814 uint16_t portid; 815 uint32_t data; 816 int ret; 817 int i; 818 819 for (i = 0; i < qconf->n_rx_queue; ++i) { 820 rx_queue = &(qconf->rx_queue_list[i]); 821 portid = rx_queue->port_id; 822 queueid = rx_queue->queue_id; 823 data = portid << CHAR_BIT | queueid; 824 825 ret = rte_eth_dev_rx_intr_ctl_q(portid, queueid, 826 RTE_EPOLL_PER_THREAD, 827 RTE_INTR_EVENT_ADD, 828 (void *)((uintptr_t)data)); 829 if (ret) 830 return ret; 831 } 832 833 return 0; 834 } 835 836 /* main processing loop */ 837 static int 838 main_loop(__attribute__((unused)) void *dummy) 839 { 840 struct rte_mbuf *pkts_burst[MAX_PKT_BURST]; 841 unsigned lcore_id; 842 uint64_t prev_tsc, diff_tsc, cur_tsc, tim_res_tsc, hz; 843 uint64_t prev_tsc_power = 0, cur_tsc_power, diff_tsc_power; 844 int i, j, nb_rx; 845 uint8_t queueid; 846 uint16_t portid; 847 struct lcore_conf *qconf; 848 struct lcore_rx_queue *rx_queue; 849 enum freq_scale_hint_t lcore_scaleup_hint; 850 uint32_t lcore_rx_idle_count = 0; 851 uint32_t lcore_idle_hint = 0; 852 int intr_en = 0; 853 854 const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US; 855 856 prev_tsc = 0; 857 hz = rte_get_timer_hz(); 858 tim_res_tsc = hz/TIMER_NUMBER_PER_SECOND; 859 860 lcore_id = rte_lcore_id(); 861 qconf = &lcore_conf[lcore_id]; 862 863 if (qconf->n_rx_queue == 0) { 864 RTE_LOG(INFO, L3FWD_POWER, "lcore %u has nothing to do\n", lcore_id); 865 return 0; 866 } 867 868 RTE_LOG(INFO, L3FWD_POWER, "entering main loop on lcore %u\n", lcore_id); 869 870 for (i = 0; i < qconf->n_rx_queue; i++) { 871 portid = qconf->rx_queue_list[i].port_id; 872 queueid = qconf->rx_queue_list[i].queue_id; 873 RTE_LOG(INFO, L3FWD_POWER, " -- lcoreid=%u portid=%u " 874 "rxqueueid=%hhu\n", lcore_id, portid, queueid); 875 } 876 877 /* add into event wait list */ 878 if (event_register(qconf) == 0) 879 intr_en = 1; 880 else 881 RTE_LOG(INFO, L3FWD_POWER, "RX interrupt won't enable.\n"); 882 883 while (1) { 884 stats[lcore_id].nb_iteration_looped++; 885 886 cur_tsc = rte_rdtsc(); 887 cur_tsc_power = cur_tsc; 888 889 /* 890 * TX burst queue drain 891 */ 892 diff_tsc = cur_tsc - prev_tsc; 893 if (unlikely(diff_tsc > drain_tsc)) { 894 for (i = 0; i < qconf->n_tx_port; ++i) { 895 portid = qconf->tx_port_id[i]; 896 rte_eth_tx_buffer_flush(portid, 897 qconf->tx_queue_id[portid], 898 qconf->tx_buffer[portid]); 899 } 900 prev_tsc = cur_tsc; 901 } 902 903 diff_tsc_power = cur_tsc_power - prev_tsc_power; 904 if (diff_tsc_power > tim_res_tsc) { 905 rte_timer_manage(); 906 prev_tsc_power = cur_tsc_power; 907 } 908 909 start_rx: 910 /* 911 * Read packet from RX queues 912 */ 913 lcore_scaleup_hint = FREQ_CURRENT; 914 lcore_rx_idle_count = 0; 915 for (i = 0; i < qconf->n_rx_queue; ++i) { 916 rx_queue = &(qconf->rx_queue_list[i]); 917 rx_queue->idle_hint = 0; 918 portid = rx_queue->port_id; 919 queueid = rx_queue->queue_id; 920 921 nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst, 922 MAX_PKT_BURST); 923 924 stats[lcore_id].nb_rx_processed += nb_rx; 925 if (unlikely(nb_rx == 0)) { 926 /** 927 * no packet received from rx queue, try to 928 * sleep for a while forcing CPU enter deeper 929 * C states. 930 */ 931 rx_queue->zero_rx_packet_count++; 932 933 if (rx_queue->zero_rx_packet_count <= 934 MIN_ZERO_POLL_COUNT) 935 continue; 936 937 rx_queue->idle_hint = power_idle_heuristic(\ 938 rx_queue->zero_rx_packet_count); 939 lcore_rx_idle_count++; 940 } else { 941 rx_queue->zero_rx_packet_count = 0; 942 943 /** 944 * do not scale up frequency immediately as 945 * user to kernel space communication is costly 946 * which might impact packet I/O for received 947 * packets. 948 */ 949 rx_queue->freq_up_hint = 950 power_freq_scaleup_heuristic(lcore_id, 951 portid, queueid); 952 } 953 954 /* Prefetch first packets */ 955 for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) { 956 rte_prefetch0(rte_pktmbuf_mtod( 957 pkts_burst[j], void *)); 958 } 959 960 /* Prefetch and forward already prefetched packets */ 961 for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) { 962 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[ 963 j + PREFETCH_OFFSET], void *)); 964 l3fwd_simple_forward(pkts_burst[j], portid, 965 qconf); 966 } 967 968 /* Forward remaining prefetched packets */ 969 for (; j < nb_rx; j++) { 970 l3fwd_simple_forward(pkts_burst[j], portid, 971 qconf); 972 } 973 } 974 975 if (likely(lcore_rx_idle_count != qconf->n_rx_queue)) { 976 for (i = 1, lcore_scaleup_hint = 977 qconf->rx_queue_list[0].freq_up_hint; 978 i < qconf->n_rx_queue; ++i) { 979 rx_queue = &(qconf->rx_queue_list[i]); 980 if (rx_queue->freq_up_hint > 981 lcore_scaleup_hint) 982 lcore_scaleup_hint = 983 rx_queue->freq_up_hint; 984 } 985 986 if (lcore_scaleup_hint == FREQ_HIGHEST) { 987 if (rte_power_freq_max) 988 rte_power_freq_max(lcore_id); 989 } else if (lcore_scaleup_hint == FREQ_HIGHER) { 990 if (rte_power_freq_up) 991 rte_power_freq_up(lcore_id); 992 } 993 } else { 994 /** 995 * All Rx queues empty in recent consecutive polls, 996 * sleep in a conservative manner, meaning sleep as 997 * less as possible. 998 */ 999 for (i = 1, lcore_idle_hint = 1000 qconf->rx_queue_list[0].idle_hint; 1001 i < qconf->n_rx_queue; ++i) { 1002 rx_queue = &(qconf->rx_queue_list[i]); 1003 if (rx_queue->idle_hint < lcore_idle_hint) 1004 lcore_idle_hint = rx_queue->idle_hint; 1005 } 1006 1007 if (lcore_idle_hint < SUSPEND_THRESHOLD) 1008 /** 1009 * execute "pause" instruction to avoid context 1010 * switch which generally take hundred of 1011 * microseconds for short sleep. 1012 */ 1013 rte_delay_us(lcore_idle_hint); 1014 else { 1015 /* suspend until rx interrupt trigges */ 1016 if (intr_en) { 1017 turn_on_intr(qconf); 1018 sleep_until_rx_interrupt( 1019 qconf->n_rx_queue); 1020 /** 1021 * start receiving packets immediately 1022 */ 1023 goto start_rx; 1024 } 1025 } 1026 stats[lcore_id].sleep_time += lcore_idle_hint; 1027 } 1028 } 1029 } 1030 1031 static int 1032 check_lcore_params(void) 1033 { 1034 uint8_t queue, lcore; 1035 uint16_t i; 1036 int socketid; 1037 1038 for (i = 0; i < nb_lcore_params; ++i) { 1039 queue = lcore_params[i].queue_id; 1040 if (queue >= MAX_RX_QUEUE_PER_PORT) { 1041 printf("invalid queue number: %hhu\n", queue); 1042 return -1; 1043 } 1044 lcore = lcore_params[i].lcore_id; 1045 if (!rte_lcore_is_enabled(lcore)) { 1046 printf("error: lcore %hhu is not enabled in lcore " 1047 "mask\n", lcore); 1048 return -1; 1049 } 1050 if ((socketid = rte_lcore_to_socket_id(lcore) != 0) && 1051 (numa_on == 0)) { 1052 printf("warning: lcore %hhu is on socket %d with numa " 1053 "off\n", lcore, socketid); 1054 } 1055 } 1056 return 0; 1057 } 1058 1059 static int 1060 check_port_config(const unsigned nb_ports) 1061 { 1062 unsigned portid; 1063 uint16_t i; 1064 1065 for (i = 0; i < nb_lcore_params; ++i) { 1066 portid = lcore_params[i].port_id; 1067 if ((enabled_port_mask & (1 << portid)) == 0) { 1068 printf("port %u is not enabled in port mask\n", 1069 portid); 1070 return -1; 1071 } 1072 if (portid >= nb_ports) { 1073 printf("port %u is not present on the board\n", 1074 portid); 1075 return -1; 1076 } 1077 } 1078 return 0; 1079 } 1080 1081 static uint8_t 1082 get_port_n_rx_queues(const uint16_t port) 1083 { 1084 int queue = -1; 1085 uint16_t i; 1086 1087 for (i = 0; i < nb_lcore_params; ++i) { 1088 if (lcore_params[i].port_id == port && 1089 lcore_params[i].queue_id > queue) 1090 queue = lcore_params[i].queue_id; 1091 } 1092 return (uint8_t)(++queue); 1093 } 1094 1095 static int 1096 init_lcore_rx_queues(void) 1097 { 1098 uint16_t i, nb_rx_queue; 1099 uint8_t lcore; 1100 1101 for (i = 0; i < nb_lcore_params; ++i) { 1102 lcore = lcore_params[i].lcore_id; 1103 nb_rx_queue = lcore_conf[lcore].n_rx_queue; 1104 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) { 1105 printf("error: too many queues (%u) for lcore: %u\n", 1106 (unsigned)nb_rx_queue + 1, (unsigned)lcore); 1107 return -1; 1108 } else { 1109 lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id = 1110 lcore_params[i].port_id; 1111 lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id = 1112 lcore_params[i].queue_id; 1113 lcore_conf[lcore].n_rx_queue++; 1114 } 1115 } 1116 return 0; 1117 } 1118 1119 /* display usage */ 1120 static void 1121 print_usage(const char *prgname) 1122 { 1123 printf ("%s [EAL options] -- -p PORTMASK -P" 1124 " [--config (port,queue,lcore)[,(port,queue,lcore]]" 1125 " [--enable-jumbo [--max-pkt-len PKTLEN]]\n" 1126 " -p PORTMASK: hexadecimal bitmask of ports to configure\n" 1127 " -P : enable promiscuous mode\n" 1128 " --config (port,queue,lcore): rx queues configuration\n" 1129 " --no-numa: optional, disable numa awareness\n" 1130 " --enable-jumbo: enable jumbo frame" 1131 " which max packet len is PKTLEN in decimal (64-9600)\n" 1132 " --parse-ptype: parse packet type by software\n", 1133 prgname); 1134 } 1135 1136 static int parse_max_pkt_len(const char *pktlen) 1137 { 1138 char *end = NULL; 1139 unsigned long len; 1140 1141 /* parse decimal string */ 1142 len = strtoul(pktlen, &end, 10); 1143 if ((pktlen[0] == '\0') || (end == NULL) || (*end != '\0')) 1144 return -1; 1145 1146 if (len == 0) 1147 return -1; 1148 1149 return len; 1150 } 1151 1152 static int 1153 parse_portmask(const char *portmask) 1154 { 1155 char *end = NULL; 1156 unsigned long pm; 1157 1158 /* parse hexadecimal string */ 1159 pm = strtoul(portmask, &end, 16); 1160 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0')) 1161 return -1; 1162 1163 if (pm == 0) 1164 return -1; 1165 1166 return pm; 1167 } 1168 1169 static int 1170 parse_config(const char *q_arg) 1171 { 1172 char s[256]; 1173 const char *p, *p0 = q_arg; 1174 char *end; 1175 enum fieldnames { 1176 FLD_PORT = 0, 1177 FLD_QUEUE, 1178 FLD_LCORE, 1179 _NUM_FLD 1180 }; 1181 unsigned long int_fld[_NUM_FLD]; 1182 char *str_fld[_NUM_FLD]; 1183 int i; 1184 unsigned size; 1185 1186 nb_lcore_params = 0; 1187 1188 while ((p = strchr(p0,'(')) != NULL) { 1189 ++p; 1190 if((p0 = strchr(p,')')) == NULL) 1191 return -1; 1192 1193 size = p0 - p; 1194 if(size >= sizeof(s)) 1195 return -1; 1196 1197 snprintf(s, sizeof(s), "%.*s", size, p); 1198 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') != 1199 _NUM_FLD) 1200 return -1; 1201 for (i = 0; i < _NUM_FLD; i++){ 1202 errno = 0; 1203 int_fld[i] = strtoul(str_fld[i], &end, 0); 1204 if (errno != 0 || end == str_fld[i] || int_fld[i] > 1205 255) 1206 return -1; 1207 } 1208 if (nb_lcore_params >= MAX_LCORE_PARAMS) { 1209 printf("exceeded max number of lcore params: %hu\n", 1210 nb_lcore_params); 1211 return -1; 1212 } 1213 lcore_params_array[nb_lcore_params].port_id = 1214 (uint8_t)int_fld[FLD_PORT]; 1215 lcore_params_array[nb_lcore_params].queue_id = 1216 (uint8_t)int_fld[FLD_QUEUE]; 1217 lcore_params_array[nb_lcore_params].lcore_id = 1218 (uint8_t)int_fld[FLD_LCORE]; 1219 ++nb_lcore_params; 1220 } 1221 lcore_params = lcore_params_array; 1222 1223 return 0; 1224 } 1225 1226 #define CMD_LINE_OPT_PARSE_PTYPE "parse-ptype" 1227 1228 /* Parse the argument given in the command line of the application */ 1229 static int 1230 parse_args(int argc, char **argv) 1231 { 1232 int opt, ret; 1233 char **argvopt; 1234 int option_index; 1235 char *prgname = argv[0]; 1236 static struct option lgopts[] = { 1237 {"config", 1, 0, 0}, 1238 {"no-numa", 0, 0, 0}, 1239 {"enable-jumbo", 0, 0, 0}, 1240 {CMD_LINE_OPT_PARSE_PTYPE, 0, 0, 0}, 1241 {NULL, 0, 0, 0} 1242 }; 1243 1244 argvopt = argv; 1245 1246 while ((opt = getopt_long(argc, argvopt, "p:P", 1247 lgopts, &option_index)) != EOF) { 1248 1249 switch (opt) { 1250 /* portmask */ 1251 case 'p': 1252 enabled_port_mask = parse_portmask(optarg); 1253 if (enabled_port_mask == 0) { 1254 printf("invalid portmask\n"); 1255 print_usage(prgname); 1256 return -1; 1257 } 1258 break; 1259 case 'P': 1260 printf("Promiscuous mode selected\n"); 1261 promiscuous_on = 1; 1262 break; 1263 1264 /* long options */ 1265 case 0: 1266 if (!strncmp(lgopts[option_index].name, "config", 6)) { 1267 ret = parse_config(optarg); 1268 if (ret) { 1269 printf("invalid config\n"); 1270 print_usage(prgname); 1271 return -1; 1272 } 1273 } 1274 1275 if (!strncmp(lgopts[option_index].name, 1276 "no-numa", 7)) { 1277 printf("numa is disabled \n"); 1278 numa_on = 0; 1279 } 1280 1281 if (!strncmp(lgopts[option_index].name, 1282 "enable-jumbo", 12)) { 1283 struct option lenopts = 1284 {"max-pkt-len", required_argument, \ 1285 0, 0}; 1286 1287 printf("jumbo frame is enabled \n"); 1288 port_conf.rxmode.offloads |= 1289 DEV_RX_OFFLOAD_JUMBO_FRAME; 1290 port_conf.txmode.offloads |= 1291 DEV_TX_OFFLOAD_MULTI_SEGS; 1292 1293 /** 1294 * if no max-pkt-len set, use the default value 1295 * ETHER_MAX_LEN 1296 */ 1297 if (0 == getopt_long(argc, argvopt, "", 1298 &lenopts, &option_index)) { 1299 ret = parse_max_pkt_len(optarg); 1300 if ((ret < 64) || 1301 (ret > MAX_JUMBO_PKT_LEN)){ 1302 printf("invalid packet " 1303 "length\n"); 1304 print_usage(prgname); 1305 return -1; 1306 } 1307 port_conf.rxmode.max_rx_pkt_len = ret; 1308 } 1309 printf("set jumbo frame " 1310 "max packet length to %u\n", 1311 (unsigned int)port_conf.rxmode.max_rx_pkt_len); 1312 } 1313 1314 if (!strncmp(lgopts[option_index].name, 1315 CMD_LINE_OPT_PARSE_PTYPE, 1316 sizeof(CMD_LINE_OPT_PARSE_PTYPE))) { 1317 printf("soft parse-ptype is enabled\n"); 1318 parse_ptype = 1; 1319 } 1320 1321 break; 1322 1323 default: 1324 print_usage(prgname); 1325 return -1; 1326 } 1327 } 1328 1329 if (optind >= 0) 1330 argv[optind-1] = prgname; 1331 1332 ret = optind-1; 1333 optind = 1; /* reset getopt lib */ 1334 return ret; 1335 } 1336 1337 static void 1338 print_ethaddr(const char *name, const struct ether_addr *eth_addr) 1339 { 1340 char buf[ETHER_ADDR_FMT_SIZE]; 1341 ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr); 1342 printf("%s%s", name, buf); 1343 } 1344 1345 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH) 1346 static void 1347 setup_hash(int socketid) 1348 { 1349 struct rte_hash_parameters ipv4_l3fwd_hash_params = { 1350 .name = NULL, 1351 .entries = L3FWD_HASH_ENTRIES, 1352 .key_len = sizeof(struct ipv4_5tuple), 1353 .hash_func = DEFAULT_HASH_FUNC, 1354 .hash_func_init_val = 0, 1355 }; 1356 1357 struct rte_hash_parameters ipv6_l3fwd_hash_params = { 1358 .name = NULL, 1359 .entries = L3FWD_HASH_ENTRIES, 1360 .key_len = sizeof(struct ipv6_5tuple), 1361 .hash_func = DEFAULT_HASH_FUNC, 1362 .hash_func_init_val = 0, 1363 }; 1364 1365 unsigned i; 1366 int ret; 1367 char s[64]; 1368 1369 /* create ipv4 hash */ 1370 snprintf(s, sizeof(s), "ipv4_l3fwd_hash_%d", socketid); 1371 ipv4_l3fwd_hash_params.name = s; 1372 ipv4_l3fwd_hash_params.socket_id = socketid; 1373 ipv4_l3fwd_lookup_struct[socketid] = 1374 rte_hash_create(&ipv4_l3fwd_hash_params); 1375 if (ipv4_l3fwd_lookup_struct[socketid] == NULL) 1376 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on " 1377 "socket %d\n", socketid); 1378 1379 /* create ipv6 hash */ 1380 snprintf(s, sizeof(s), "ipv6_l3fwd_hash_%d", socketid); 1381 ipv6_l3fwd_hash_params.name = s; 1382 ipv6_l3fwd_hash_params.socket_id = socketid; 1383 ipv6_l3fwd_lookup_struct[socketid] = 1384 rte_hash_create(&ipv6_l3fwd_hash_params); 1385 if (ipv6_l3fwd_lookup_struct[socketid] == NULL) 1386 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on " 1387 "socket %d\n", socketid); 1388 1389 1390 /* populate the ipv4 hash */ 1391 for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) { 1392 ret = rte_hash_add_key (ipv4_l3fwd_lookup_struct[socketid], 1393 (void *) &ipv4_l3fwd_route_array[i].key); 1394 if (ret < 0) { 1395 rte_exit(EXIT_FAILURE, "Unable to add entry %u to the" 1396 "l3fwd hash on socket %d\n", i, socketid); 1397 } 1398 ipv4_l3fwd_out_if[ret] = ipv4_l3fwd_route_array[i].if_out; 1399 printf("Hash: Adding key\n"); 1400 print_ipv4_key(ipv4_l3fwd_route_array[i].key); 1401 } 1402 1403 /* populate the ipv6 hash */ 1404 for (i = 0; i < IPV6_L3FWD_NUM_ROUTES; i++) { 1405 ret = rte_hash_add_key (ipv6_l3fwd_lookup_struct[socketid], 1406 (void *) &ipv6_l3fwd_route_array[i].key); 1407 if (ret < 0) { 1408 rte_exit(EXIT_FAILURE, "Unable to add entry %u to the" 1409 "l3fwd hash on socket %d\n", i, socketid); 1410 } 1411 ipv6_l3fwd_out_if[ret] = ipv6_l3fwd_route_array[i].if_out; 1412 printf("Hash: Adding key\n"); 1413 print_ipv6_key(ipv6_l3fwd_route_array[i].key); 1414 } 1415 } 1416 #endif 1417 1418 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM) 1419 static void 1420 setup_lpm(int socketid) 1421 { 1422 unsigned i; 1423 int ret; 1424 char s[64]; 1425 1426 /* create the LPM table */ 1427 struct rte_lpm_config lpm_ipv4_config; 1428 1429 lpm_ipv4_config.max_rules = IPV4_L3FWD_LPM_MAX_RULES; 1430 lpm_ipv4_config.number_tbl8s = 256; 1431 lpm_ipv4_config.flags = 0; 1432 1433 snprintf(s, sizeof(s), "IPV4_L3FWD_LPM_%d", socketid); 1434 ipv4_l3fwd_lookup_struct[socketid] = 1435 rte_lpm_create(s, socketid, &lpm_ipv4_config); 1436 if (ipv4_l3fwd_lookup_struct[socketid] == NULL) 1437 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table" 1438 " on socket %d\n", socketid); 1439 1440 /* populate the LPM table */ 1441 for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) { 1442 ret = rte_lpm_add(ipv4_l3fwd_lookup_struct[socketid], 1443 ipv4_l3fwd_route_array[i].ip, 1444 ipv4_l3fwd_route_array[i].depth, 1445 ipv4_l3fwd_route_array[i].if_out); 1446 1447 if (ret < 0) { 1448 rte_exit(EXIT_FAILURE, "Unable to add entry %u to the " 1449 "l3fwd LPM table on socket %d\n", 1450 i, socketid); 1451 } 1452 1453 printf("LPM: Adding route 0x%08x / %d (%d)\n", 1454 (unsigned)ipv4_l3fwd_route_array[i].ip, 1455 ipv4_l3fwd_route_array[i].depth, 1456 ipv4_l3fwd_route_array[i].if_out); 1457 } 1458 } 1459 #endif 1460 1461 static int 1462 init_mem(unsigned nb_mbuf) 1463 { 1464 struct lcore_conf *qconf; 1465 int socketid; 1466 unsigned lcore_id; 1467 char s[64]; 1468 1469 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) { 1470 if (rte_lcore_is_enabled(lcore_id) == 0) 1471 continue; 1472 1473 if (numa_on) 1474 socketid = rte_lcore_to_socket_id(lcore_id); 1475 else 1476 socketid = 0; 1477 1478 if (socketid >= NB_SOCKETS) { 1479 rte_exit(EXIT_FAILURE, "Socket %d of lcore %u is " 1480 "out of range %d\n", socketid, 1481 lcore_id, NB_SOCKETS); 1482 } 1483 if (pktmbuf_pool[socketid] == NULL) { 1484 snprintf(s, sizeof(s), "mbuf_pool_%d", socketid); 1485 pktmbuf_pool[socketid] = 1486 rte_pktmbuf_pool_create(s, nb_mbuf, 1487 MEMPOOL_CACHE_SIZE, 0, 1488 RTE_MBUF_DEFAULT_BUF_SIZE, 1489 socketid); 1490 if (pktmbuf_pool[socketid] == NULL) 1491 rte_exit(EXIT_FAILURE, 1492 "Cannot init mbuf pool on socket %d\n", 1493 socketid); 1494 else 1495 printf("Allocated mbuf pool on socket %d\n", 1496 socketid); 1497 1498 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM) 1499 setup_lpm(socketid); 1500 #else 1501 setup_hash(socketid); 1502 #endif 1503 } 1504 qconf = &lcore_conf[lcore_id]; 1505 qconf->ipv4_lookup_struct = ipv4_l3fwd_lookup_struct[socketid]; 1506 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH) 1507 qconf->ipv6_lookup_struct = ipv6_l3fwd_lookup_struct[socketid]; 1508 #endif 1509 } 1510 return 0; 1511 } 1512 1513 /* Check the link status of all ports in up to 9s, and print them finally */ 1514 static void 1515 check_all_ports_link_status(uint16_t port_num, uint32_t port_mask) 1516 { 1517 #define CHECK_INTERVAL 100 /* 100ms */ 1518 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */ 1519 uint8_t count, all_ports_up, print_flag = 0; 1520 uint16_t portid; 1521 struct rte_eth_link link; 1522 1523 printf("\nChecking link status"); 1524 fflush(stdout); 1525 for (count = 0; count <= MAX_CHECK_TIME; count++) { 1526 all_ports_up = 1; 1527 for (portid = 0; portid < port_num; portid++) { 1528 if ((port_mask & (1 << portid)) == 0) 1529 continue; 1530 memset(&link, 0, sizeof(link)); 1531 rte_eth_link_get_nowait(portid, &link); 1532 /* print link status if flag set */ 1533 if (print_flag == 1) { 1534 if (link.link_status) 1535 printf("Port %d Link Up - speed %u " 1536 "Mbps - %s\n", (uint8_t)portid, 1537 (unsigned)link.link_speed, 1538 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ? 1539 ("full-duplex") : ("half-duplex\n")); 1540 else 1541 printf("Port %d Link Down\n", 1542 (uint8_t)portid); 1543 continue; 1544 } 1545 /* clear all_ports_up flag if any link down */ 1546 if (link.link_status == ETH_LINK_DOWN) { 1547 all_ports_up = 0; 1548 break; 1549 } 1550 } 1551 /* after finally printing all link status, get out */ 1552 if (print_flag == 1) 1553 break; 1554 1555 if (all_ports_up == 0) { 1556 printf("."); 1557 fflush(stdout); 1558 rte_delay_ms(CHECK_INTERVAL); 1559 } 1560 1561 /* set the print_flag if all ports up or timeout */ 1562 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) { 1563 print_flag = 1; 1564 printf("done\n"); 1565 } 1566 } 1567 } 1568 1569 static int check_ptype(uint16_t portid) 1570 { 1571 int i, ret; 1572 int ptype_l3_ipv4 = 0; 1573 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH) 1574 int ptype_l3_ipv6 = 0; 1575 #endif 1576 uint32_t ptype_mask = RTE_PTYPE_L3_MASK; 1577 1578 ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, NULL, 0); 1579 if (ret <= 0) 1580 return 0; 1581 1582 uint32_t ptypes[ret]; 1583 1584 ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, ptypes, ret); 1585 for (i = 0; i < ret; ++i) { 1586 if (ptypes[i] & RTE_PTYPE_L3_IPV4) 1587 ptype_l3_ipv4 = 1; 1588 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH) 1589 if (ptypes[i] & RTE_PTYPE_L3_IPV6) 1590 ptype_l3_ipv6 = 1; 1591 #endif 1592 } 1593 1594 if (ptype_l3_ipv4 == 0) 1595 printf("port %d cannot parse RTE_PTYPE_L3_IPV4\n", portid); 1596 1597 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH) 1598 if (ptype_l3_ipv6 == 0) 1599 printf("port %d cannot parse RTE_PTYPE_L3_IPV6\n", portid); 1600 #endif 1601 1602 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM) 1603 if (ptype_l3_ipv4) 1604 #else /* APP_LOOKUP_EXACT_MATCH */ 1605 if (ptype_l3_ipv4 && ptype_l3_ipv6) 1606 #endif 1607 return 1; 1608 1609 return 0; 1610 1611 } 1612 1613 int 1614 main(int argc, char **argv) 1615 { 1616 struct lcore_conf *qconf; 1617 struct rte_eth_dev_info dev_info; 1618 struct rte_eth_txconf *txconf; 1619 int ret; 1620 uint16_t nb_ports; 1621 uint16_t queueid; 1622 unsigned lcore_id; 1623 uint64_t hz; 1624 uint32_t n_tx_queue, nb_lcores; 1625 uint32_t dev_rxq_num, dev_txq_num; 1626 uint8_t nb_rx_queue, queue, socketid; 1627 uint16_t portid; 1628 1629 /* catch SIGINT and restore cpufreq governor to ondemand */ 1630 signal(SIGINT, signal_exit_now); 1631 1632 /* init EAL */ 1633 ret = rte_eal_init(argc, argv); 1634 if (ret < 0) 1635 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n"); 1636 argc -= ret; 1637 argv += ret; 1638 1639 /* init RTE timer library to be used late */ 1640 rte_timer_subsystem_init(); 1641 1642 /* parse application arguments (after the EAL ones) */ 1643 ret = parse_args(argc, argv); 1644 if (ret < 0) 1645 rte_exit(EXIT_FAILURE, "Invalid L3FWD parameters\n"); 1646 1647 if (check_lcore_params() < 0) 1648 rte_exit(EXIT_FAILURE, "check_lcore_params failed\n"); 1649 1650 ret = init_lcore_rx_queues(); 1651 if (ret < 0) 1652 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n"); 1653 1654 nb_ports = rte_eth_dev_count(); 1655 1656 if (check_port_config(nb_ports) < 0) 1657 rte_exit(EXIT_FAILURE, "check_port_config failed\n"); 1658 1659 nb_lcores = rte_lcore_count(); 1660 1661 /* initialize all ports */ 1662 for (portid = 0; portid < nb_ports; portid++) { 1663 struct rte_eth_conf local_port_conf = port_conf; 1664 1665 /* skip ports that are not enabled */ 1666 if ((enabled_port_mask & (1 << portid)) == 0) { 1667 printf("\nSkipping disabled port %d\n", portid); 1668 continue; 1669 } 1670 1671 /* init port */ 1672 printf("Initializing port %d ... ", portid ); 1673 fflush(stdout); 1674 1675 rte_eth_dev_info_get(portid, &dev_info); 1676 dev_rxq_num = dev_info.max_rx_queues; 1677 dev_txq_num = dev_info.max_tx_queues; 1678 1679 nb_rx_queue = get_port_n_rx_queues(portid); 1680 if (nb_rx_queue > dev_rxq_num) 1681 rte_exit(EXIT_FAILURE, 1682 "Cannot configure not existed rxq: " 1683 "port=%d\n", portid); 1684 1685 n_tx_queue = nb_lcores; 1686 if (n_tx_queue > dev_txq_num) 1687 n_tx_queue = dev_txq_num; 1688 printf("Creating queues: nb_rxq=%d nb_txq=%u... ", 1689 nb_rx_queue, (unsigned)n_tx_queue ); 1690 /* If number of Rx queue is 0, no need to enable Rx interrupt */ 1691 if (nb_rx_queue == 0) 1692 local_port_conf.intr_conf.rxq = 0; 1693 rte_eth_dev_info_get(portid, &dev_info); 1694 if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE) 1695 local_port_conf.txmode.offloads |= 1696 DEV_TX_OFFLOAD_MBUF_FAST_FREE; 1697 ret = rte_eth_dev_configure(portid, nb_rx_queue, 1698 (uint16_t)n_tx_queue, &local_port_conf); 1699 if (ret < 0) 1700 rte_exit(EXIT_FAILURE, "Cannot configure device: " 1701 "err=%d, port=%d\n", ret, portid); 1702 1703 ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd, 1704 &nb_txd); 1705 if (ret < 0) 1706 rte_exit(EXIT_FAILURE, 1707 "Cannot adjust number of descriptors: err=%d, port=%d\n", 1708 ret, portid); 1709 1710 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]); 1711 print_ethaddr(" Address:", &ports_eth_addr[portid]); 1712 printf(", "); 1713 1714 /* init memory */ 1715 ret = init_mem(NB_MBUF); 1716 if (ret < 0) 1717 rte_exit(EXIT_FAILURE, "init_mem failed\n"); 1718 1719 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) { 1720 if (rte_lcore_is_enabled(lcore_id) == 0) 1721 continue; 1722 1723 /* Initialize TX buffers */ 1724 qconf = &lcore_conf[lcore_id]; 1725 qconf->tx_buffer[portid] = rte_zmalloc_socket("tx_buffer", 1726 RTE_ETH_TX_BUFFER_SIZE(MAX_PKT_BURST), 0, 1727 rte_eth_dev_socket_id(portid)); 1728 if (qconf->tx_buffer[portid] == NULL) 1729 rte_exit(EXIT_FAILURE, "Can't allocate tx buffer for port %u\n", 1730 portid); 1731 1732 rte_eth_tx_buffer_init(qconf->tx_buffer[portid], MAX_PKT_BURST); 1733 } 1734 1735 /* init one TX queue per couple (lcore,port) */ 1736 queueid = 0; 1737 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) { 1738 if (rte_lcore_is_enabled(lcore_id) == 0) 1739 continue; 1740 1741 if (queueid >= dev_txq_num) 1742 continue; 1743 1744 if (numa_on) 1745 socketid = \ 1746 (uint8_t)rte_lcore_to_socket_id(lcore_id); 1747 else 1748 socketid = 0; 1749 1750 printf("txq=%u,%d,%d ", lcore_id, queueid, socketid); 1751 fflush(stdout); 1752 1753 txconf = &dev_info.default_txconf; 1754 txconf->txq_flags = ETH_TXQ_FLAGS_IGNORE; 1755 txconf->offloads = local_port_conf.txmode.offloads; 1756 ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd, 1757 socketid, txconf); 1758 if (ret < 0) 1759 rte_exit(EXIT_FAILURE, 1760 "rte_eth_tx_queue_setup: err=%d, " 1761 "port=%d\n", ret, portid); 1762 1763 qconf = &lcore_conf[lcore_id]; 1764 qconf->tx_queue_id[portid] = queueid; 1765 queueid++; 1766 1767 qconf->tx_port_id[qconf->n_tx_port] = portid; 1768 qconf->n_tx_port++; 1769 } 1770 printf("\n"); 1771 } 1772 1773 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) { 1774 if (rte_lcore_is_enabled(lcore_id) == 0) 1775 continue; 1776 1777 /* init power management library */ 1778 ret = rte_power_init(lcore_id); 1779 if (ret) 1780 RTE_LOG(ERR, POWER, 1781 "Library initialization failed on core %u\n", lcore_id); 1782 1783 /* init timer structures for each enabled lcore */ 1784 rte_timer_init(&power_timers[lcore_id]); 1785 hz = rte_get_timer_hz(); 1786 rte_timer_reset(&power_timers[lcore_id], 1787 hz/TIMER_NUMBER_PER_SECOND, SINGLE, lcore_id, 1788 power_timer_cb, NULL); 1789 1790 qconf = &lcore_conf[lcore_id]; 1791 printf("\nInitializing rx queues on lcore %u ... ", lcore_id ); 1792 fflush(stdout); 1793 /* init RX queues */ 1794 for(queue = 0; queue < qconf->n_rx_queue; ++queue) { 1795 struct rte_eth_rxconf rxq_conf; 1796 struct rte_eth_dev *dev; 1797 struct rte_eth_conf *conf; 1798 1799 portid = qconf->rx_queue_list[queue].port_id; 1800 queueid = qconf->rx_queue_list[queue].queue_id; 1801 dev = &rte_eth_devices[portid]; 1802 conf = &dev->data->dev_conf; 1803 1804 if (numa_on) 1805 socketid = \ 1806 (uint8_t)rte_lcore_to_socket_id(lcore_id); 1807 else 1808 socketid = 0; 1809 1810 printf("rxq=%d,%d,%d ", portid, queueid, socketid); 1811 fflush(stdout); 1812 1813 rte_eth_dev_info_get(portid, &dev_info); 1814 rxq_conf = dev_info.default_rxconf; 1815 rxq_conf.offloads = conf->rxmode.offloads; 1816 ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd, 1817 socketid, &rxq_conf, 1818 pktmbuf_pool[socketid]); 1819 if (ret < 0) 1820 rte_exit(EXIT_FAILURE, 1821 "rte_eth_rx_queue_setup: err=%d, " 1822 "port=%d\n", ret, portid); 1823 1824 if (parse_ptype) { 1825 if (add_cb_parse_ptype(portid, queueid) < 0) 1826 rte_exit(EXIT_FAILURE, 1827 "Fail to add ptype cb\n"); 1828 } else if (!check_ptype(portid)) 1829 rte_exit(EXIT_FAILURE, 1830 "PMD can not provide needed ptypes\n"); 1831 } 1832 } 1833 1834 printf("\n"); 1835 1836 /* start ports */ 1837 for (portid = 0; portid < nb_ports; portid++) { 1838 if ((enabled_port_mask & (1 << portid)) == 0) { 1839 continue; 1840 } 1841 /* Start device */ 1842 ret = rte_eth_dev_start(portid); 1843 if (ret < 0) 1844 rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, " 1845 "port=%d\n", ret, portid); 1846 /* 1847 * If enabled, put device in promiscuous mode. 1848 * This allows IO forwarding mode to forward packets 1849 * to itself through 2 cross-connected ports of the 1850 * target machine. 1851 */ 1852 if (promiscuous_on) 1853 rte_eth_promiscuous_enable(portid); 1854 /* initialize spinlock for each port */ 1855 rte_spinlock_init(&(locks[portid])); 1856 } 1857 1858 check_all_ports_link_status(nb_ports, enabled_port_mask); 1859 1860 /* launch per-lcore init on every lcore */ 1861 rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER); 1862 RTE_LCORE_FOREACH_SLAVE(lcore_id) { 1863 if (rte_eal_wait_lcore(lcore_id) < 0) 1864 return -1; 1865 } 1866 1867 return 0; 1868 } 1869