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 16 #include <rte_common.h> 17 #include <rte_vect.h> 18 #include <rte_byteorder.h> 19 #include <rte_log.h> 20 #include <rte_memory.h> 21 #include <rte_memcpy.h> 22 #include <rte_eal.h> 23 #include <rte_launch.h> 24 #include <rte_atomic.h> 25 #include <rte_cycles.h> 26 #include <rte_prefetch.h> 27 #include <rte_lcore.h> 28 #include <rte_per_lcore.h> 29 #include <rte_branch_prediction.h> 30 #include <rte_interrupts.h> 31 #include <rte_random.h> 32 #include <rte_debug.h> 33 #include <rte_ether.h> 34 #include <rte_ethdev.h> 35 #include <rte_ring.h> 36 #include <rte_mempool.h> 37 #include <rte_mbuf.h> 38 #include <rte_ip.h> 39 #include <rte_tcp.h> 40 #include <rte_udp.h> 41 #include <rte_string_fns.h> 42 #include <rte_pause.h> 43 #include <rte_timer.h> 44 45 #include <cmdline_parse.h> 46 #include <cmdline_parse_etheraddr.h> 47 48 #include <lthread_api.h> 49 50 #define APP_LOOKUP_EXACT_MATCH 0 51 #define APP_LOOKUP_LPM 1 52 #define DO_RFC_1812_CHECKS 53 54 /* Enable cpu-load stats 0-off, 1-on */ 55 #define APP_CPU_LOAD 1 56 57 #ifndef APP_LOOKUP_METHOD 58 #define APP_LOOKUP_METHOD APP_LOOKUP_LPM 59 #endif 60 61 #ifndef __GLIBC__ /* sched_getcpu() is glibc specific */ 62 #define sched_getcpu() rte_lcore_id() 63 #endif 64 65 static int 66 check_ptype(int portid) 67 { 68 int i, ret; 69 int ipv4 = 0, ipv6 = 0; 70 71 ret = rte_eth_dev_get_supported_ptypes(portid, RTE_PTYPE_L3_MASK, NULL, 72 0); 73 if (ret <= 0) 74 return 0; 75 76 uint32_t ptypes[ret]; 77 78 ret = rte_eth_dev_get_supported_ptypes(portid, RTE_PTYPE_L3_MASK, 79 ptypes, ret); 80 for (i = 0; i < ret; ++i) { 81 if (ptypes[i] & RTE_PTYPE_L3_IPV4) 82 ipv4 = 1; 83 if (ptypes[i] & RTE_PTYPE_L3_IPV6) 84 ipv6 = 1; 85 } 86 87 if (ipv4 && ipv6) 88 return 1; 89 90 return 0; 91 } 92 93 static inline void 94 parse_ptype(struct rte_mbuf *m) 95 { 96 struct ether_hdr *eth_hdr; 97 uint32_t packet_type = RTE_PTYPE_UNKNOWN; 98 uint16_t ether_type; 99 100 eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *); 101 ether_type = eth_hdr->ether_type; 102 if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv4)) 103 packet_type |= RTE_PTYPE_L3_IPV4_EXT_UNKNOWN; 104 else if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv6)) 105 packet_type |= RTE_PTYPE_L3_IPV6_EXT_UNKNOWN; 106 107 m->packet_type = packet_type; 108 } 109 110 static uint16_t 111 cb_parse_ptype(__rte_unused uint16_t port, __rte_unused uint16_t queue, 112 struct rte_mbuf *pkts[], uint16_t nb_pkts, 113 __rte_unused uint16_t max_pkts, __rte_unused void *user_param) 114 { 115 unsigned int i; 116 117 for (i = 0; i < nb_pkts; i++) 118 parse_ptype(pkts[i]); 119 120 return nb_pkts; 121 } 122 123 /* 124 * When set to zero, simple forwaring path is eanbled. 125 * When set to one, optimized forwarding path is enabled. 126 * Note that LPM optimisation path uses SSE4.1 instructions. 127 */ 128 #define ENABLE_MULTI_BUFFER_OPTIMIZE 1 129 130 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH) 131 #include <rte_hash.h> 132 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM) 133 #include <rte_lpm.h> 134 #include <rte_lpm6.h> 135 #else 136 #error "APP_LOOKUP_METHOD set to incorrect value" 137 #endif 138 139 #define RTE_LOGTYPE_L3FWD RTE_LOGTYPE_USER1 140 141 #define MAX_JUMBO_PKT_LEN 9600 142 143 #define IPV6_ADDR_LEN 16 144 145 #define MEMPOOL_CACHE_SIZE 256 146 147 /* 148 * This expression is used to calculate the number of mbufs needed depending on 149 * user input, taking into account memory for rx and tx hardware rings, cache 150 * per lcore and mtable per port per lcore. RTE_MAX is used to ensure that 151 * NB_MBUF never goes below a minimum value of 8192 152 */ 153 154 #define NB_MBUF RTE_MAX(\ 155 (nb_ports*nb_rx_queue*nb_rxd + \ 156 nb_ports*nb_lcores*MAX_PKT_BURST + \ 157 nb_ports*n_tx_queue*nb_txd + \ 158 nb_lcores*MEMPOOL_CACHE_SIZE), \ 159 (unsigned)8192) 160 161 #define MAX_PKT_BURST 32 162 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */ 163 164 /* 165 * Try to avoid TX buffering if we have at least MAX_TX_BURST packets to send. 166 */ 167 #define MAX_TX_BURST (MAX_PKT_BURST / 2) 168 #define BURST_SIZE MAX_TX_BURST 169 170 #define NB_SOCKETS 8 171 172 /* Configure how many packets ahead to prefetch, when reading packets */ 173 #define PREFETCH_OFFSET 3 174 175 /* Used to mark destination port as 'invalid'. */ 176 #define BAD_PORT ((uint16_t)-1) 177 178 #define FWDSTEP 4 179 180 /* 181 * Configurable number of RX/TX ring descriptors 182 */ 183 #define RTE_TEST_RX_DESC_DEFAULT 1024 184 #define RTE_TEST_TX_DESC_DEFAULT 1024 185 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT; 186 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT; 187 188 /* ethernet addresses of ports */ 189 static uint64_t dest_eth_addr[RTE_MAX_ETHPORTS]; 190 static struct ether_addr ports_eth_addr[RTE_MAX_ETHPORTS]; 191 192 static xmm_t val_eth[RTE_MAX_ETHPORTS]; 193 194 /* replace first 12B of the ethernet header. */ 195 #define MASK_ETH 0x3f 196 197 /* mask of enabled ports */ 198 static uint32_t enabled_port_mask; 199 static int promiscuous_on; /**< Set in promiscuous mode off by default. */ 200 static int numa_on = 1; /**< NUMA is enabled by default. */ 201 static int parse_ptype_on; 202 203 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH) 204 static int ipv6; /**< ipv6 is false by default. */ 205 #endif 206 207 #if (APP_CPU_LOAD == 1) 208 209 #define MAX_CPU RTE_MAX_LCORE 210 #define CPU_LOAD_TIMEOUT_US (5 * 1000 * 1000) /**< Timeout for collecting 5s */ 211 212 #define CPU_PROCESS 0 213 #define CPU_POLL 1 214 #define MAX_CPU_COUNTER 2 215 216 struct cpu_load { 217 uint16_t n_cpu; 218 uint64_t counter; 219 uint64_t hits[MAX_CPU_COUNTER][MAX_CPU]; 220 } __rte_cache_aligned; 221 222 static struct cpu_load cpu_load; 223 static int cpu_load_lcore_id = -1; 224 225 #define SET_CPU_BUSY(thread, counter) \ 226 thread->conf.busy[counter] = 1 227 228 #define SET_CPU_IDLE(thread, counter) \ 229 thread->conf.busy[counter] = 0 230 231 #define IS_CPU_BUSY(thread, counter) \ 232 (thread->conf.busy[counter] > 0) 233 234 #else 235 236 #define SET_CPU_BUSY(thread, counter) 237 #define SET_CPU_IDLE(thread, counter) 238 #define IS_CPU_BUSY(thread, counter) 0 239 240 #endif 241 242 struct mbuf_table { 243 uint16_t len; 244 struct rte_mbuf *m_table[MAX_PKT_BURST]; 245 }; 246 247 struct lcore_rx_queue { 248 uint16_t port_id; 249 uint8_t queue_id; 250 } __rte_cache_aligned; 251 252 #define MAX_RX_QUEUE_PER_LCORE 16 253 #define MAX_TX_QUEUE_PER_PORT RTE_MAX_ETHPORTS 254 #define MAX_RX_QUEUE_PER_PORT 128 255 256 #define MAX_LCORE_PARAMS 1024 257 struct rx_thread_params { 258 uint16_t port_id; 259 uint8_t queue_id; 260 uint8_t lcore_id; 261 uint8_t thread_id; 262 } __rte_cache_aligned; 263 264 static struct rx_thread_params rx_thread_params_array[MAX_LCORE_PARAMS]; 265 static struct rx_thread_params rx_thread_params_array_default[] = { 266 {0, 0, 2, 0}, 267 {0, 1, 2, 1}, 268 {0, 2, 2, 2}, 269 {1, 0, 2, 3}, 270 {1, 1, 2, 4}, 271 {1, 2, 2, 5}, 272 {2, 0, 2, 6}, 273 {3, 0, 3, 7}, 274 {3, 1, 3, 8}, 275 }; 276 277 static struct rx_thread_params *rx_thread_params = 278 rx_thread_params_array_default; 279 static uint16_t nb_rx_thread_params = RTE_DIM(rx_thread_params_array_default); 280 281 struct tx_thread_params { 282 uint8_t lcore_id; 283 uint8_t thread_id; 284 } __rte_cache_aligned; 285 286 static struct tx_thread_params tx_thread_params_array[MAX_LCORE_PARAMS]; 287 static struct tx_thread_params tx_thread_params_array_default[] = { 288 {4, 0}, 289 {5, 1}, 290 {6, 2}, 291 {7, 3}, 292 {8, 4}, 293 {9, 5}, 294 {10, 6}, 295 {11, 7}, 296 {12, 8}, 297 }; 298 299 static struct tx_thread_params *tx_thread_params = 300 tx_thread_params_array_default; 301 static uint16_t nb_tx_thread_params = RTE_DIM(tx_thread_params_array_default); 302 303 static struct rte_eth_conf port_conf = { 304 .rxmode = { 305 .mq_mode = ETH_MQ_RX_RSS, 306 .max_rx_pkt_len = ETHER_MAX_LEN, 307 .split_hdr_size = 0, 308 .offloads = DEV_RX_OFFLOAD_CHECKSUM, 309 }, 310 .rx_adv_conf = { 311 .rss_conf = { 312 .rss_key = NULL, 313 .rss_hf = ETH_RSS_TCP, 314 }, 315 }, 316 .txmode = { 317 .mq_mode = ETH_MQ_TX_NONE, 318 }, 319 }; 320 321 static struct rte_mempool *pktmbuf_pool[NB_SOCKETS]; 322 323 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH) 324 325 #include <rte_hash_crc.h> 326 #define DEFAULT_HASH_FUNC rte_hash_crc 327 328 struct ipv4_5tuple { 329 uint32_t ip_dst; 330 uint32_t ip_src; 331 uint16_t port_dst; 332 uint16_t port_src; 333 uint8_t proto; 334 } __attribute__((__packed__)); 335 336 union ipv4_5tuple_host { 337 struct { 338 uint8_t pad0; 339 uint8_t proto; 340 uint16_t pad1; 341 uint32_t ip_src; 342 uint32_t ip_dst; 343 uint16_t port_src; 344 uint16_t port_dst; 345 }; 346 __m128i xmm; 347 }; 348 349 #define XMM_NUM_IN_IPV6_5TUPLE 3 350 351 struct ipv6_5tuple { 352 uint8_t ip_dst[IPV6_ADDR_LEN]; 353 uint8_t ip_src[IPV6_ADDR_LEN]; 354 uint16_t port_dst; 355 uint16_t port_src; 356 uint8_t proto; 357 } __attribute__((__packed__)); 358 359 union ipv6_5tuple_host { 360 struct { 361 uint16_t pad0; 362 uint8_t proto; 363 uint8_t pad1; 364 uint8_t ip_src[IPV6_ADDR_LEN]; 365 uint8_t ip_dst[IPV6_ADDR_LEN]; 366 uint16_t port_src; 367 uint16_t port_dst; 368 uint64_t reserve; 369 }; 370 __m128i xmm[XMM_NUM_IN_IPV6_5TUPLE]; 371 }; 372 373 struct ipv4_l3fwd_route { 374 struct ipv4_5tuple key; 375 uint8_t if_out; 376 }; 377 378 struct ipv6_l3fwd_route { 379 struct ipv6_5tuple key; 380 uint8_t if_out; 381 }; 382 383 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = { 384 {{IPv4(101, 0, 0, 0), IPv4(100, 10, 0, 1), 101, 11, IPPROTO_TCP}, 0}, 385 {{IPv4(201, 0, 0, 0), IPv4(200, 20, 0, 1), 102, 12, IPPROTO_TCP}, 1}, 386 {{IPv4(111, 0, 0, 0), IPv4(100, 30, 0, 1), 101, 11, IPPROTO_TCP}, 2}, 387 {{IPv4(211, 0, 0, 0), IPv4(200, 40, 0, 1), 102, 12, IPPROTO_TCP}, 3}, 388 }; 389 390 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = { 391 {{ 392 {0xfe, 0x80, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0}, 393 {0xfe, 0x80, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 394 0x05}, 395 101, 11, IPPROTO_TCP}, 0}, 396 397 {{ 398 {0xfe, 0x90, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0}, 399 {0xfe, 0x90, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 400 0x05}, 401 102, 12, IPPROTO_TCP}, 1}, 402 403 {{ 404 {0xfe, 0xa0, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0}, 405 {0xfe, 0xa0, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 406 0x05}, 407 101, 11, IPPROTO_TCP}, 2}, 408 409 {{ 410 {0xfe, 0xb0, 0, 0, 0, 0, 0, 0, 0x02, 0x1e, 0x67, 0xff, 0xfe, 0, 0, 0}, 411 {0xfe, 0xb0, 0, 0, 0, 0, 0, 0, 0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 412 0x05}, 413 102, 12, IPPROTO_TCP}, 3}, 414 }; 415 416 typedef struct rte_hash lookup_struct_t; 417 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS]; 418 static lookup_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS]; 419 420 #ifdef RTE_ARCH_X86_64 421 /* default to 4 million hash entries (approx) */ 422 #define L3FWD_HASH_ENTRIES (1024*1024*4) 423 #else 424 /* 32-bit has less address-space for hugepage memory, limit to 1M entries */ 425 #define L3FWD_HASH_ENTRIES (1024*1024*1) 426 #endif 427 #define HASH_ENTRY_NUMBER_DEFAULT 4 428 429 static uint32_t hash_entry_number = HASH_ENTRY_NUMBER_DEFAULT; 430 431 static inline uint32_t 432 ipv4_hash_crc(const void *data, __rte_unused uint32_t data_len, 433 uint32_t init_val) 434 { 435 const union ipv4_5tuple_host *k; 436 uint32_t t; 437 const uint32_t *p; 438 439 k = data; 440 t = k->proto; 441 p = (const uint32_t *)&k->port_src; 442 443 init_val = rte_hash_crc_4byte(t, init_val); 444 init_val = rte_hash_crc_4byte(k->ip_src, init_val); 445 init_val = rte_hash_crc_4byte(k->ip_dst, init_val); 446 init_val = rte_hash_crc_4byte(*p, init_val); 447 return init_val; 448 } 449 450 static inline uint32_t 451 ipv6_hash_crc(const void *data, __rte_unused uint32_t data_len, 452 uint32_t init_val) 453 { 454 const union ipv6_5tuple_host *k; 455 uint32_t t; 456 const uint32_t *p; 457 const uint32_t *ip_src0, *ip_src1, *ip_src2, *ip_src3; 458 const uint32_t *ip_dst0, *ip_dst1, *ip_dst2, *ip_dst3; 459 460 k = data; 461 t = k->proto; 462 p = (const uint32_t *)&k->port_src; 463 464 ip_src0 = (const uint32_t *) k->ip_src; 465 ip_src1 = (const uint32_t *)(k->ip_src + 4); 466 ip_src2 = (const uint32_t *)(k->ip_src + 8); 467 ip_src3 = (const uint32_t *)(k->ip_src + 12); 468 ip_dst0 = (const uint32_t *) k->ip_dst; 469 ip_dst1 = (const uint32_t *)(k->ip_dst + 4); 470 ip_dst2 = (const uint32_t *)(k->ip_dst + 8); 471 ip_dst3 = (const uint32_t *)(k->ip_dst + 12); 472 init_val = rte_hash_crc_4byte(t, init_val); 473 init_val = rte_hash_crc_4byte(*ip_src0, init_val); 474 init_val = rte_hash_crc_4byte(*ip_src1, init_val); 475 init_val = rte_hash_crc_4byte(*ip_src2, init_val); 476 init_val = rte_hash_crc_4byte(*ip_src3, init_val); 477 init_val = rte_hash_crc_4byte(*ip_dst0, init_val); 478 init_val = rte_hash_crc_4byte(*ip_dst1, init_val); 479 init_val = rte_hash_crc_4byte(*ip_dst2, init_val); 480 init_val = rte_hash_crc_4byte(*ip_dst3, init_val); 481 init_val = rte_hash_crc_4byte(*p, init_val); 482 return init_val; 483 } 484 485 #define IPV4_L3FWD_NUM_ROUTES RTE_DIM(ipv4_l3fwd_route_array) 486 #define IPV6_L3FWD_NUM_ROUTES RTE_DIM(ipv6_l3fwd_route_array) 487 488 static uint8_t ipv4_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned; 489 static uint8_t ipv6_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned; 490 491 #endif 492 493 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM) 494 struct ipv4_l3fwd_route { 495 uint32_t ip; 496 uint8_t depth; 497 uint8_t if_out; 498 }; 499 500 struct ipv6_l3fwd_route { 501 uint8_t ip[16]; 502 uint8_t depth; 503 uint8_t if_out; 504 }; 505 506 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = { 507 {IPv4(1, 1, 1, 0), 24, 0}, 508 {IPv4(2, 1, 1, 0), 24, 1}, 509 {IPv4(3, 1, 1, 0), 24, 2}, 510 {IPv4(4, 1, 1, 0), 24, 3}, 511 {IPv4(5, 1, 1, 0), 24, 4}, 512 {IPv4(6, 1, 1, 0), 24, 5}, 513 {IPv4(7, 1, 1, 0), 24, 6}, 514 {IPv4(8, 1, 1, 0), 24, 7}, 515 }; 516 517 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = { 518 {{1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 0}, 519 {{2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 1}, 520 {{3, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 2}, 521 {{4, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 3}, 522 {{5, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 4}, 523 {{6, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 5}, 524 {{7, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 6}, 525 {{8, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, 48, 7}, 526 }; 527 528 #define IPV4_L3FWD_NUM_ROUTES RTE_DIM(ipv4_l3fwd_route_array) 529 #define IPV6_L3FWD_NUM_ROUTES RTE_DIM(ipv6_l3fwd_route_array) 530 531 #define IPV4_L3FWD_LPM_MAX_RULES 1024 532 #define IPV6_L3FWD_LPM_MAX_RULES 1024 533 #define IPV6_L3FWD_LPM_NUMBER_TBL8S (1 << 16) 534 535 typedef struct rte_lpm lookup_struct_t; 536 typedef struct rte_lpm6 lookup6_struct_t; 537 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS]; 538 static lookup6_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS]; 539 #endif 540 541 struct lcore_conf { 542 lookup_struct_t *ipv4_lookup_struct; 543 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM) 544 lookup6_struct_t *ipv6_lookup_struct; 545 #else 546 lookup_struct_t *ipv6_lookup_struct; 547 #endif 548 void *data; 549 } __rte_cache_aligned; 550 551 static struct lcore_conf lcore_conf[RTE_MAX_LCORE]; 552 RTE_DEFINE_PER_LCORE(struct lcore_conf *, lcore_conf); 553 554 #define MAX_RX_QUEUE_PER_THREAD 16 555 #define MAX_TX_PORT_PER_THREAD RTE_MAX_ETHPORTS 556 #define MAX_TX_QUEUE_PER_PORT RTE_MAX_ETHPORTS 557 #define MAX_RX_QUEUE_PER_PORT 128 558 559 #define MAX_RX_THREAD 1024 560 #define MAX_TX_THREAD 1024 561 #define MAX_THREAD (MAX_RX_THREAD + MAX_TX_THREAD) 562 563 /** 564 * Producers and consumers threads configuration 565 */ 566 static int lthreads_on = 1; /**< Use lthreads for processing*/ 567 568 rte_atomic16_t rx_counter; /**< Number of spawned rx threads */ 569 rte_atomic16_t tx_counter; /**< Number of spawned tx threads */ 570 571 struct thread_conf { 572 uint16_t lcore_id; /**< Initial lcore for rx thread */ 573 uint16_t cpu_id; /**< Cpu id for cpu load stats counter */ 574 uint16_t thread_id; /**< Thread ID */ 575 576 #if (APP_CPU_LOAD > 0) 577 int busy[MAX_CPU_COUNTER]; 578 #endif 579 }; 580 581 struct thread_rx_conf { 582 struct thread_conf conf; 583 584 uint16_t n_rx_queue; 585 struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE]; 586 587 uint16_t n_ring; /**< Number of output rings */ 588 struct rte_ring *ring[RTE_MAX_LCORE]; 589 struct lthread_cond *ready[RTE_MAX_LCORE]; 590 591 #if (APP_CPU_LOAD > 0) 592 int busy[MAX_CPU_COUNTER]; 593 #endif 594 } __rte_cache_aligned; 595 596 uint16_t n_rx_thread; 597 struct thread_rx_conf rx_thread[MAX_RX_THREAD]; 598 599 struct thread_tx_conf { 600 struct thread_conf conf; 601 602 uint16_t tx_queue_id[RTE_MAX_LCORE]; 603 struct mbuf_table tx_mbufs[RTE_MAX_LCORE]; 604 605 struct rte_ring *ring; 606 struct lthread_cond **ready; 607 608 } __rte_cache_aligned; 609 610 uint16_t n_tx_thread; 611 struct thread_tx_conf tx_thread[MAX_TX_THREAD]; 612 613 /* Send burst of packets on an output interface */ 614 static inline int 615 send_burst(struct thread_tx_conf *qconf, uint16_t n, uint16_t port) 616 { 617 struct rte_mbuf **m_table; 618 int ret; 619 uint16_t queueid; 620 621 queueid = qconf->tx_queue_id[port]; 622 m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table; 623 624 ret = rte_eth_tx_burst(port, queueid, m_table, n); 625 if (unlikely(ret < n)) { 626 do { 627 rte_pktmbuf_free(m_table[ret]); 628 } while (++ret < n); 629 } 630 631 return 0; 632 } 633 634 /* Enqueue a single packet, and send burst if queue is filled */ 635 static inline int 636 send_single_packet(struct rte_mbuf *m, uint16_t port) 637 { 638 uint16_t len; 639 struct thread_tx_conf *qconf; 640 641 if (lthreads_on) 642 qconf = (struct thread_tx_conf *)lthread_get_data(); 643 else 644 qconf = (struct thread_tx_conf *)RTE_PER_LCORE(lcore_conf)->data; 645 646 len = qconf->tx_mbufs[port].len; 647 qconf->tx_mbufs[port].m_table[len] = m; 648 len++; 649 650 /* enough pkts to be sent */ 651 if (unlikely(len == MAX_PKT_BURST)) { 652 send_burst(qconf, MAX_PKT_BURST, port); 653 len = 0; 654 } 655 656 qconf->tx_mbufs[port].len = len; 657 return 0; 658 } 659 660 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_LPM) && \ 661 (ENABLE_MULTI_BUFFER_OPTIMIZE == 1)) 662 static __rte_always_inline void 663 send_packetsx4(uint16_t port, 664 struct rte_mbuf *m[], uint32_t num) 665 { 666 uint32_t len, j, n; 667 struct thread_tx_conf *qconf; 668 669 if (lthreads_on) 670 qconf = (struct thread_tx_conf *)lthread_get_data(); 671 else 672 qconf = (struct thread_tx_conf *)RTE_PER_LCORE(lcore_conf)->data; 673 674 len = qconf->tx_mbufs[port].len; 675 676 /* 677 * If TX buffer for that queue is empty, and we have enough packets, 678 * then send them straightway. 679 */ 680 if (num >= MAX_TX_BURST && len == 0) { 681 n = rte_eth_tx_burst(port, qconf->tx_queue_id[port], m, num); 682 if (unlikely(n < num)) { 683 do { 684 rte_pktmbuf_free(m[n]); 685 } while (++n < num); 686 } 687 return; 688 } 689 690 /* 691 * Put packets into TX buffer for that queue. 692 */ 693 694 n = len + num; 695 n = (n > MAX_PKT_BURST) ? MAX_PKT_BURST - len : num; 696 697 j = 0; 698 switch (n % FWDSTEP) { 699 while (j < n) { 700 case 0: 701 qconf->tx_mbufs[port].m_table[len + j] = m[j]; 702 j++; 703 /* fall-through */ 704 case 3: 705 qconf->tx_mbufs[port].m_table[len + j] = m[j]; 706 j++; 707 /* fall-through */ 708 case 2: 709 qconf->tx_mbufs[port].m_table[len + j] = m[j]; 710 j++; 711 /* fall-through */ 712 case 1: 713 qconf->tx_mbufs[port].m_table[len + j] = m[j]; 714 j++; 715 } 716 } 717 718 len += n; 719 720 /* enough pkts to be sent */ 721 if (unlikely(len == MAX_PKT_BURST)) { 722 723 send_burst(qconf, MAX_PKT_BURST, port); 724 725 /* copy rest of the packets into the TX buffer. */ 726 len = num - n; 727 j = 0; 728 switch (len % FWDSTEP) { 729 while (j < len) { 730 case 0: 731 qconf->tx_mbufs[port].m_table[j] = m[n + j]; 732 j++; 733 /* fall-through */ 734 case 3: 735 qconf->tx_mbufs[port].m_table[j] = m[n + j]; 736 j++; 737 /* fall-through */ 738 case 2: 739 qconf->tx_mbufs[port].m_table[j] = m[n + j]; 740 j++; 741 /* fall-through */ 742 case 1: 743 qconf->tx_mbufs[port].m_table[j] = m[n + j]; 744 j++; 745 } 746 } 747 } 748 749 qconf->tx_mbufs[port].len = len; 750 } 751 #endif /* APP_LOOKUP_LPM */ 752 753 #ifdef DO_RFC_1812_CHECKS 754 static inline int 755 is_valid_ipv4_pkt(struct ipv4_hdr *pkt, uint32_t link_len) 756 { 757 /* From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2 */ 758 /* 759 * 1. The packet length reported by the Link Layer must be large 760 * enough to hold the minimum length legal IP datagram (20 bytes). 761 */ 762 if (link_len < sizeof(struct ipv4_hdr)) 763 return -1; 764 765 /* 2. The IP checksum must be correct. */ 766 /* this is checked in H/W */ 767 768 /* 769 * 3. The IP version number must be 4. If the version number is not 4 770 * then the packet may be another version of IP, such as IPng or 771 * ST-II. 772 */ 773 if (((pkt->version_ihl) >> 4) != 4) 774 return -3; 775 /* 776 * 4. The IP header length field must be large enough to hold the 777 * minimum length legal IP datagram (20 bytes = 5 words). 778 */ 779 if ((pkt->version_ihl & 0xf) < 5) 780 return -4; 781 782 /* 783 * 5. The IP total length field must be large enough to hold the IP 784 * datagram header, whose length is specified in the IP header length 785 * field. 786 */ 787 if (rte_cpu_to_be_16(pkt->total_length) < sizeof(struct ipv4_hdr)) 788 return -5; 789 790 return 0; 791 } 792 #endif 793 794 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH) 795 796 static __m128i mask0; 797 static __m128i mask1; 798 static __m128i mask2; 799 static inline uint16_t 800 get_ipv4_dst_port(void *ipv4_hdr, uint16_t portid, 801 lookup_struct_t *ipv4_l3fwd_lookup_struct) 802 { 803 int ret = 0; 804 union ipv4_5tuple_host key; 805 806 ipv4_hdr = (uint8_t *)ipv4_hdr + offsetof(struct ipv4_hdr, time_to_live); 807 __m128i data = _mm_loadu_si128((__m128i *)(ipv4_hdr)); 808 /* Get 5 tuple: dst port, src port, dst IP address, src IP address and 809 protocol */ 810 key.xmm = _mm_and_si128(data, mask0); 811 /* Find destination port */ 812 ret = rte_hash_lookup(ipv4_l3fwd_lookup_struct, (const void *)&key); 813 return ((ret < 0) ? portid : ipv4_l3fwd_out_if[ret]); 814 } 815 816 static inline uint16_t 817 get_ipv6_dst_port(void *ipv6_hdr, uint16_t portid, 818 lookup_struct_t *ipv6_l3fwd_lookup_struct) 819 { 820 int ret = 0; 821 union ipv6_5tuple_host key; 822 823 ipv6_hdr = (uint8_t *)ipv6_hdr + offsetof(struct ipv6_hdr, payload_len); 824 __m128i data0 = _mm_loadu_si128((__m128i *)(ipv6_hdr)); 825 __m128i data1 = _mm_loadu_si128((__m128i *)(((uint8_t *)ipv6_hdr) + 826 sizeof(__m128i))); 827 __m128i data2 = _mm_loadu_si128((__m128i *)(((uint8_t *)ipv6_hdr) + 828 sizeof(__m128i) + sizeof(__m128i))); 829 /* Get part of 5 tuple: src IP address lower 96 bits and protocol */ 830 key.xmm[0] = _mm_and_si128(data0, mask1); 831 /* Get part of 5 tuple: dst IP address lower 96 bits and src IP address 832 higher 32 bits */ 833 key.xmm[1] = data1; 834 /* Get part of 5 tuple: dst port and src port and dst IP address higher 835 32 bits */ 836 key.xmm[2] = _mm_and_si128(data2, mask2); 837 838 /* Find destination port */ 839 ret = rte_hash_lookup(ipv6_l3fwd_lookup_struct, (const void *)&key); 840 return ((ret < 0) ? portid : ipv6_l3fwd_out_if[ret]); 841 } 842 #endif 843 844 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM) 845 846 static inline uint16_t 847 get_ipv4_dst_port(void *ipv4_hdr, uint16_t portid, 848 lookup_struct_t *ipv4_l3fwd_lookup_struct) 849 { 850 uint32_t next_hop; 851 852 return ((rte_lpm_lookup(ipv4_l3fwd_lookup_struct, 853 rte_be_to_cpu_32(((struct ipv4_hdr *)ipv4_hdr)->dst_addr), 854 &next_hop) == 0) ? next_hop : portid); 855 } 856 857 static inline uint16_t 858 get_ipv6_dst_port(void *ipv6_hdr, uint16_t portid, 859 lookup6_struct_t *ipv6_l3fwd_lookup_struct) 860 { 861 uint32_t next_hop; 862 863 return ((rte_lpm6_lookup(ipv6_l3fwd_lookup_struct, 864 ((struct ipv6_hdr *)ipv6_hdr)->dst_addr, &next_hop) == 0) ? 865 next_hop : portid); 866 } 867 #endif 868 869 static inline void l3fwd_simple_forward(struct rte_mbuf *m, uint16_t portid) 870 __attribute__((unused)); 871 872 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH) && \ 873 (ENABLE_MULTI_BUFFER_OPTIMIZE == 1)) 874 875 #define MASK_ALL_PKTS 0xff 876 #define EXCLUDE_1ST_PKT 0xfe 877 #define EXCLUDE_2ND_PKT 0xfd 878 #define EXCLUDE_3RD_PKT 0xfb 879 #define EXCLUDE_4TH_PKT 0xf7 880 #define EXCLUDE_5TH_PKT 0xef 881 #define EXCLUDE_6TH_PKT 0xdf 882 #define EXCLUDE_7TH_PKT 0xbf 883 #define EXCLUDE_8TH_PKT 0x7f 884 885 static inline void 886 simple_ipv4_fwd_8pkts(struct rte_mbuf *m[8], uint16_t portid) 887 { 888 struct ether_hdr *eth_hdr[8]; 889 struct ipv4_hdr *ipv4_hdr[8]; 890 uint16_t dst_port[8]; 891 int32_t ret[8]; 892 union ipv4_5tuple_host key[8]; 893 __m128i data[8]; 894 895 eth_hdr[0] = rte_pktmbuf_mtod(m[0], struct ether_hdr *); 896 eth_hdr[1] = rte_pktmbuf_mtod(m[1], struct ether_hdr *); 897 eth_hdr[2] = rte_pktmbuf_mtod(m[2], struct ether_hdr *); 898 eth_hdr[3] = rte_pktmbuf_mtod(m[3], struct ether_hdr *); 899 eth_hdr[4] = rte_pktmbuf_mtod(m[4], struct ether_hdr *); 900 eth_hdr[5] = rte_pktmbuf_mtod(m[5], struct ether_hdr *); 901 eth_hdr[6] = rte_pktmbuf_mtod(m[6], struct ether_hdr *); 902 eth_hdr[7] = rte_pktmbuf_mtod(m[7], struct ether_hdr *); 903 904 /* Handle IPv4 headers.*/ 905 ipv4_hdr[0] = rte_pktmbuf_mtod_offset(m[0], struct ipv4_hdr *, 906 sizeof(struct ether_hdr)); 907 ipv4_hdr[1] = rte_pktmbuf_mtod_offset(m[1], struct ipv4_hdr *, 908 sizeof(struct ether_hdr)); 909 ipv4_hdr[2] = rte_pktmbuf_mtod_offset(m[2], struct ipv4_hdr *, 910 sizeof(struct ether_hdr)); 911 ipv4_hdr[3] = rte_pktmbuf_mtod_offset(m[3], struct ipv4_hdr *, 912 sizeof(struct ether_hdr)); 913 ipv4_hdr[4] = rte_pktmbuf_mtod_offset(m[4], struct ipv4_hdr *, 914 sizeof(struct ether_hdr)); 915 ipv4_hdr[5] = rte_pktmbuf_mtod_offset(m[5], struct ipv4_hdr *, 916 sizeof(struct ether_hdr)); 917 ipv4_hdr[6] = rte_pktmbuf_mtod_offset(m[6], struct ipv4_hdr *, 918 sizeof(struct ether_hdr)); 919 ipv4_hdr[7] = rte_pktmbuf_mtod_offset(m[7], struct ipv4_hdr *, 920 sizeof(struct ether_hdr)); 921 922 #ifdef DO_RFC_1812_CHECKS 923 /* Check to make sure the packet is valid (RFC1812) */ 924 uint8_t valid_mask = MASK_ALL_PKTS; 925 926 if (is_valid_ipv4_pkt(ipv4_hdr[0], m[0]->pkt_len) < 0) { 927 rte_pktmbuf_free(m[0]); 928 valid_mask &= EXCLUDE_1ST_PKT; 929 } 930 if (is_valid_ipv4_pkt(ipv4_hdr[1], m[1]->pkt_len) < 0) { 931 rte_pktmbuf_free(m[1]); 932 valid_mask &= EXCLUDE_2ND_PKT; 933 } 934 if (is_valid_ipv4_pkt(ipv4_hdr[2], m[2]->pkt_len) < 0) { 935 rte_pktmbuf_free(m[2]); 936 valid_mask &= EXCLUDE_3RD_PKT; 937 } 938 if (is_valid_ipv4_pkt(ipv4_hdr[3], m[3]->pkt_len) < 0) { 939 rte_pktmbuf_free(m[3]); 940 valid_mask &= EXCLUDE_4TH_PKT; 941 } 942 if (is_valid_ipv4_pkt(ipv4_hdr[4], m[4]->pkt_len) < 0) { 943 rte_pktmbuf_free(m[4]); 944 valid_mask &= EXCLUDE_5TH_PKT; 945 } 946 if (is_valid_ipv4_pkt(ipv4_hdr[5], m[5]->pkt_len) < 0) { 947 rte_pktmbuf_free(m[5]); 948 valid_mask &= EXCLUDE_6TH_PKT; 949 } 950 if (is_valid_ipv4_pkt(ipv4_hdr[6], m[6]->pkt_len) < 0) { 951 rte_pktmbuf_free(m[6]); 952 valid_mask &= EXCLUDE_7TH_PKT; 953 } 954 if (is_valid_ipv4_pkt(ipv4_hdr[7], m[7]->pkt_len) < 0) { 955 rte_pktmbuf_free(m[7]); 956 valid_mask &= EXCLUDE_8TH_PKT; 957 } 958 if (unlikely(valid_mask != MASK_ALL_PKTS)) { 959 if (valid_mask == 0) 960 return; 961 962 uint8_t i = 0; 963 964 for (i = 0; i < 8; i++) 965 if ((0x1 << i) & valid_mask) 966 l3fwd_simple_forward(m[i], portid); 967 } 968 #endif /* End of #ifdef DO_RFC_1812_CHECKS */ 969 970 data[0] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[0], __m128i *, 971 sizeof(struct ether_hdr) + 972 offsetof(struct ipv4_hdr, time_to_live))); 973 data[1] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[1], __m128i *, 974 sizeof(struct ether_hdr) + 975 offsetof(struct ipv4_hdr, time_to_live))); 976 data[2] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[2], __m128i *, 977 sizeof(struct ether_hdr) + 978 offsetof(struct ipv4_hdr, time_to_live))); 979 data[3] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[3], __m128i *, 980 sizeof(struct ether_hdr) + 981 offsetof(struct ipv4_hdr, time_to_live))); 982 data[4] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[4], __m128i *, 983 sizeof(struct ether_hdr) + 984 offsetof(struct ipv4_hdr, time_to_live))); 985 data[5] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[5], __m128i *, 986 sizeof(struct ether_hdr) + 987 offsetof(struct ipv4_hdr, time_to_live))); 988 data[6] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[6], __m128i *, 989 sizeof(struct ether_hdr) + 990 offsetof(struct ipv4_hdr, time_to_live))); 991 data[7] = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m[7], __m128i *, 992 sizeof(struct ether_hdr) + 993 offsetof(struct ipv4_hdr, time_to_live))); 994 995 key[0].xmm = _mm_and_si128(data[0], mask0); 996 key[1].xmm = _mm_and_si128(data[1], mask0); 997 key[2].xmm = _mm_and_si128(data[2], mask0); 998 key[3].xmm = _mm_and_si128(data[3], mask0); 999 key[4].xmm = _mm_and_si128(data[4], mask0); 1000 key[5].xmm = _mm_and_si128(data[5], mask0); 1001 key[6].xmm = _mm_and_si128(data[6], mask0); 1002 key[7].xmm = _mm_and_si128(data[7], mask0); 1003 1004 const void *key_array[8] = {&key[0], &key[1], &key[2], &key[3], 1005 &key[4], &key[5], &key[6], &key[7]}; 1006 1007 rte_hash_lookup_bulk(RTE_PER_LCORE(lcore_conf)->ipv4_lookup_struct, 1008 &key_array[0], 8, ret); 1009 dst_port[0] = ((ret[0] < 0) ? portid : ipv4_l3fwd_out_if[ret[0]]); 1010 dst_port[1] = ((ret[1] < 0) ? portid : ipv4_l3fwd_out_if[ret[1]]); 1011 dst_port[2] = ((ret[2] < 0) ? portid : ipv4_l3fwd_out_if[ret[2]]); 1012 dst_port[3] = ((ret[3] < 0) ? portid : ipv4_l3fwd_out_if[ret[3]]); 1013 dst_port[4] = ((ret[4] < 0) ? portid : ipv4_l3fwd_out_if[ret[4]]); 1014 dst_port[5] = ((ret[5] < 0) ? portid : ipv4_l3fwd_out_if[ret[5]]); 1015 dst_port[6] = ((ret[6] < 0) ? portid : ipv4_l3fwd_out_if[ret[6]]); 1016 dst_port[7] = ((ret[7] < 0) ? portid : ipv4_l3fwd_out_if[ret[7]]); 1017 1018 if (dst_port[0] >= RTE_MAX_ETHPORTS || 1019 (enabled_port_mask & 1 << dst_port[0]) == 0) 1020 dst_port[0] = portid; 1021 if (dst_port[1] >= RTE_MAX_ETHPORTS || 1022 (enabled_port_mask & 1 << dst_port[1]) == 0) 1023 dst_port[1] = portid; 1024 if (dst_port[2] >= RTE_MAX_ETHPORTS || 1025 (enabled_port_mask & 1 << dst_port[2]) == 0) 1026 dst_port[2] = portid; 1027 if (dst_port[3] >= RTE_MAX_ETHPORTS || 1028 (enabled_port_mask & 1 << dst_port[3]) == 0) 1029 dst_port[3] = portid; 1030 if (dst_port[4] >= RTE_MAX_ETHPORTS || 1031 (enabled_port_mask & 1 << dst_port[4]) == 0) 1032 dst_port[4] = portid; 1033 if (dst_port[5] >= RTE_MAX_ETHPORTS || 1034 (enabled_port_mask & 1 << dst_port[5]) == 0) 1035 dst_port[5] = portid; 1036 if (dst_port[6] >= RTE_MAX_ETHPORTS || 1037 (enabled_port_mask & 1 << dst_port[6]) == 0) 1038 dst_port[6] = portid; 1039 if (dst_port[7] >= RTE_MAX_ETHPORTS || 1040 (enabled_port_mask & 1 << dst_port[7]) == 0) 1041 dst_port[7] = portid; 1042 1043 #ifdef DO_RFC_1812_CHECKS 1044 /* Update time to live and header checksum */ 1045 --(ipv4_hdr[0]->time_to_live); 1046 --(ipv4_hdr[1]->time_to_live); 1047 --(ipv4_hdr[2]->time_to_live); 1048 --(ipv4_hdr[3]->time_to_live); 1049 ++(ipv4_hdr[0]->hdr_checksum); 1050 ++(ipv4_hdr[1]->hdr_checksum); 1051 ++(ipv4_hdr[2]->hdr_checksum); 1052 ++(ipv4_hdr[3]->hdr_checksum); 1053 --(ipv4_hdr[4]->time_to_live); 1054 --(ipv4_hdr[5]->time_to_live); 1055 --(ipv4_hdr[6]->time_to_live); 1056 --(ipv4_hdr[7]->time_to_live); 1057 ++(ipv4_hdr[4]->hdr_checksum); 1058 ++(ipv4_hdr[5]->hdr_checksum); 1059 ++(ipv4_hdr[6]->hdr_checksum); 1060 ++(ipv4_hdr[7]->hdr_checksum); 1061 #endif 1062 1063 /* dst addr */ 1064 *(uint64_t *)ð_hdr[0]->d_addr = dest_eth_addr[dst_port[0]]; 1065 *(uint64_t *)ð_hdr[1]->d_addr = dest_eth_addr[dst_port[1]]; 1066 *(uint64_t *)ð_hdr[2]->d_addr = dest_eth_addr[dst_port[2]]; 1067 *(uint64_t *)ð_hdr[3]->d_addr = dest_eth_addr[dst_port[3]]; 1068 *(uint64_t *)ð_hdr[4]->d_addr = dest_eth_addr[dst_port[4]]; 1069 *(uint64_t *)ð_hdr[5]->d_addr = dest_eth_addr[dst_port[5]]; 1070 *(uint64_t *)ð_hdr[6]->d_addr = dest_eth_addr[dst_port[6]]; 1071 *(uint64_t *)ð_hdr[7]->d_addr = dest_eth_addr[dst_port[7]]; 1072 1073 /* src addr */ 1074 ether_addr_copy(&ports_eth_addr[dst_port[0]], ð_hdr[0]->s_addr); 1075 ether_addr_copy(&ports_eth_addr[dst_port[1]], ð_hdr[1]->s_addr); 1076 ether_addr_copy(&ports_eth_addr[dst_port[2]], ð_hdr[2]->s_addr); 1077 ether_addr_copy(&ports_eth_addr[dst_port[3]], ð_hdr[3]->s_addr); 1078 ether_addr_copy(&ports_eth_addr[dst_port[4]], ð_hdr[4]->s_addr); 1079 ether_addr_copy(&ports_eth_addr[dst_port[5]], ð_hdr[5]->s_addr); 1080 ether_addr_copy(&ports_eth_addr[dst_port[6]], ð_hdr[6]->s_addr); 1081 ether_addr_copy(&ports_eth_addr[dst_port[7]], ð_hdr[7]->s_addr); 1082 1083 send_single_packet(m[0], (uint8_t)dst_port[0]); 1084 send_single_packet(m[1], (uint8_t)dst_port[1]); 1085 send_single_packet(m[2], (uint8_t)dst_port[2]); 1086 send_single_packet(m[3], (uint8_t)dst_port[3]); 1087 send_single_packet(m[4], (uint8_t)dst_port[4]); 1088 send_single_packet(m[5], (uint8_t)dst_port[5]); 1089 send_single_packet(m[6], (uint8_t)dst_port[6]); 1090 send_single_packet(m[7], (uint8_t)dst_port[7]); 1091 1092 } 1093 1094 static inline void get_ipv6_5tuple(struct rte_mbuf *m0, __m128i mask0, 1095 __m128i mask1, union ipv6_5tuple_host *key) 1096 { 1097 __m128i tmpdata0 = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m0, 1098 __m128i *, sizeof(struct ether_hdr) + 1099 offsetof(struct ipv6_hdr, payload_len))); 1100 __m128i tmpdata1 = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m0, 1101 __m128i *, sizeof(struct ether_hdr) + 1102 offsetof(struct ipv6_hdr, payload_len) + sizeof(__m128i))); 1103 __m128i tmpdata2 = _mm_loadu_si128(rte_pktmbuf_mtod_offset(m0, 1104 __m128i *, sizeof(struct ether_hdr) + 1105 offsetof(struct ipv6_hdr, payload_len) + sizeof(__m128i) + 1106 sizeof(__m128i))); 1107 key->xmm[0] = _mm_and_si128(tmpdata0, mask0); 1108 key->xmm[1] = tmpdata1; 1109 key->xmm[2] = _mm_and_si128(tmpdata2, mask1); 1110 } 1111 1112 static inline void 1113 simple_ipv6_fwd_8pkts(struct rte_mbuf *m[8], uint16_t portid) 1114 { 1115 int32_t ret[8]; 1116 uint16_t dst_port[8]; 1117 struct ether_hdr *eth_hdr[8]; 1118 union ipv6_5tuple_host key[8]; 1119 1120 __attribute__((unused)) struct ipv6_hdr *ipv6_hdr[8]; 1121 1122 eth_hdr[0] = rte_pktmbuf_mtod(m[0], struct ether_hdr *); 1123 eth_hdr[1] = rte_pktmbuf_mtod(m[1], struct ether_hdr *); 1124 eth_hdr[2] = rte_pktmbuf_mtod(m[2], struct ether_hdr *); 1125 eth_hdr[3] = rte_pktmbuf_mtod(m[3], struct ether_hdr *); 1126 eth_hdr[4] = rte_pktmbuf_mtod(m[4], struct ether_hdr *); 1127 eth_hdr[5] = rte_pktmbuf_mtod(m[5], struct ether_hdr *); 1128 eth_hdr[6] = rte_pktmbuf_mtod(m[6], struct ether_hdr *); 1129 eth_hdr[7] = rte_pktmbuf_mtod(m[7], struct ether_hdr *); 1130 1131 /* Handle IPv6 headers.*/ 1132 ipv6_hdr[0] = rte_pktmbuf_mtod_offset(m[0], struct ipv6_hdr *, 1133 sizeof(struct ether_hdr)); 1134 ipv6_hdr[1] = rte_pktmbuf_mtod_offset(m[1], struct ipv6_hdr *, 1135 sizeof(struct ether_hdr)); 1136 ipv6_hdr[2] = rte_pktmbuf_mtod_offset(m[2], struct ipv6_hdr *, 1137 sizeof(struct ether_hdr)); 1138 ipv6_hdr[3] = rte_pktmbuf_mtod_offset(m[3], struct ipv6_hdr *, 1139 sizeof(struct ether_hdr)); 1140 ipv6_hdr[4] = rte_pktmbuf_mtod_offset(m[4], struct ipv6_hdr *, 1141 sizeof(struct ether_hdr)); 1142 ipv6_hdr[5] = rte_pktmbuf_mtod_offset(m[5], struct ipv6_hdr *, 1143 sizeof(struct ether_hdr)); 1144 ipv6_hdr[6] = rte_pktmbuf_mtod_offset(m[6], struct ipv6_hdr *, 1145 sizeof(struct ether_hdr)); 1146 ipv6_hdr[7] = rte_pktmbuf_mtod_offset(m[7], struct ipv6_hdr *, 1147 sizeof(struct ether_hdr)); 1148 1149 get_ipv6_5tuple(m[0], mask1, mask2, &key[0]); 1150 get_ipv6_5tuple(m[1], mask1, mask2, &key[1]); 1151 get_ipv6_5tuple(m[2], mask1, mask2, &key[2]); 1152 get_ipv6_5tuple(m[3], mask1, mask2, &key[3]); 1153 get_ipv6_5tuple(m[4], mask1, mask2, &key[4]); 1154 get_ipv6_5tuple(m[5], mask1, mask2, &key[5]); 1155 get_ipv6_5tuple(m[6], mask1, mask2, &key[6]); 1156 get_ipv6_5tuple(m[7], mask1, mask2, &key[7]); 1157 1158 const void *key_array[8] = {&key[0], &key[1], &key[2], &key[3], 1159 &key[4], &key[5], &key[6], &key[7]}; 1160 1161 rte_hash_lookup_bulk(RTE_PER_LCORE(lcore_conf)->ipv6_lookup_struct, 1162 &key_array[0], 4, ret); 1163 dst_port[0] = ((ret[0] < 0) ? portid : ipv6_l3fwd_out_if[ret[0]]); 1164 dst_port[1] = ((ret[1] < 0) ? portid : ipv6_l3fwd_out_if[ret[1]]); 1165 dst_port[2] = ((ret[2] < 0) ? portid : ipv6_l3fwd_out_if[ret[2]]); 1166 dst_port[3] = ((ret[3] < 0) ? portid : ipv6_l3fwd_out_if[ret[3]]); 1167 dst_port[4] = ((ret[4] < 0) ? portid : ipv6_l3fwd_out_if[ret[4]]); 1168 dst_port[5] = ((ret[5] < 0) ? portid : ipv6_l3fwd_out_if[ret[5]]); 1169 dst_port[6] = ((ret[6] < 0) ? portid : ipv6_l3fwd_out_if[ret[6]]); 1170 dst_port[7] = ((ret[7] < 0) ? portid : ipv6_l3fwd_out_if[ret[7]]); 1171 1172 if (dst_port[0] >= RTE_MAX_ETHPORTS || 1173 (enabled_port_mask & 1 << dst_port[0]) == 0) 1174 dst_port[0] = portid; 1175 if (dst_port[1] >= RTE_MAX_ETHPORTS || 1176 (enabled_port_mask & 1 << dst_port[1]) == 0) 1177 dst_port[1] = portid; 1178 if (dst_port[2] >= RTE_MAX_ETHPORTS || 1179 (enabled_port_mask & 1 << dst_port[2]) == 0) 1180 dst_port[2] = portid; 1181 if (dst_port[3] >= RTE_MAX_ETHPORTS || 1182 (enabled_port_mask & 1 << dst_port[3]) == 0) 1183 dst_port[3] = portid; 1184 if (dst_port[4] >= RTE_MAX_ETHPORTS || 1185 (enabled_port_mask & 1 << dst_port[4]) == 0) 1186 dst_port[4] = portid; 1187 if (dst_port[5] >= RTE_MAX_ETHPORTS || 1188 (enabled_port_mask & 1 << dst_port[5]) == 0) 1189 dst_port[5] = portid; 1190 if (dst_port[6] >= RTE_MAX_ETHPORTS || 1191 (enabled_port_mask & 1 << dst_port[6]) == 0) 1192 dst_port[6] = portid; 1193 if (dst_port[7] >= RTE_MAX_ETHPORTS || 1194 (enabled_port_mask & 1 << dst_port[7]) == 0) 1195 dst_port[7] = portid; 1196 1197 /* dst addr */ 1198 *(uint64_t *)ð_hdr[0]->d_addr = dest_eth_addr[dst_port[0]]; 1199 *(uint64_t *)ð_hdr[1]->d_addr = dest_eth_addr[dst_port[1]]; 1200 *(uint64_t *)ð_hdr[2]->d_addr = dest_eth_addr[dst_port[2]]; 1201 *(uint64_t *)ð_hdr[3]->d_addr = dest_eth_addr[dst_port[3]]; 1202 *(uint64_t *)ð_hdr[4]->d_addr = dest_eth_addr[dst_port[4]]; 1203 *(uint64_t *)ð_hdr[5]->d_addr = dest_eth_addr[dst_port[5]]; 1204 *(uint64_t *)ð_hdr[6]->d_addr = dest_eth_addr[dst_port[6]]; 1205 *(uint64_t *)ð_hdr[7]->d_addr = dest_eth_addr[dst_port[7]]; 1206 1207 /* src addr */ 1208 ether_addr_copy(&ports_eth_addr[dst_port[0]], ð_hdr[0]->s_addr); 1209 ether_addr_copy(&ports_eth_addr[dst_port[1]], ð_hdr[1]->s_addr); 1210 ether_addr_copy(&ports_eth_addr[dst_port[2]], ð_hdr[2]->s_addr); 1211 ether_addr_copy(&ports_eth_addr[dst_port[3]], ð_hdr[3]->s_addr); 1212 ether_addr_copy(&ports_eth_addr[dst_port[4]], ð_hdr[4]->s_addr); 1213 ether_addr_copy(&ports_eth_addr[dst_port[5]], ð_hdr[5]->s_addr); 1214 ether_addr_copy(&ports_eth_addr[dst_port[6]], ð_hdr[6]->s_addr); 1215 ether_addr_copy(&ports_eth_addr[dst_port[7]], ð_hdr[7]->s_addr); 1216 1217 send_single_packet(m[0], dst_port[0]); 1218 send_single_packet(m[1], dst_port[1]); 1219 send_single_packet(m[2], dst_port[2]); 1220 send_single_packet(m[3], dst_port[3]); 1221 send_single_packet(m[4], dst_port[4]); 1222 send_single_packet(m[5], dst_port[5]); 1223 send_single_packet(m[6], dst_port[6]); 1224 send_single_packet(m[7], dst_port[7]); 1225 1226 } 1227 #endif /* APP_LOOKUP_METHOD */ 1228 1229 static __rte_always_inline void 1230 l3fwd_simple_forward(struct rte_mbuf *m, uint16_t portid) 1231 { 1232 struct ether_hdr *eth_hdr; 1233 struct ipv4_hdr *ipv4_hdr; 1234 uint16_t dst_port; 1235 1236 eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *); 1237 1238 if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) { 1239 /* Handle IPv4 headers.*/ 1240 ipv4_hdr = rte_pktmbuf_mtod_offset(m, struct ipv4_hdr *, 1241 sizeof(struct ether_hdr)); 1242 1243 #ifdef DO_RFC_1812_CHECKS 1244 /* Check to make sure the packet is valid (RFC1812) */ 1245 if (is_valid_ipv4_pkt(ipv4_hdr, m->pkt_len) < 0) { 1246 rte_pktmbuf_free(m); 1247 return; 1248 } 1249 #endif 1250 1251 dst_port = get_ipv4_dst_port(ipv4_hdr, portid, 1252 RTE_PER_LCORE(lcore_conf)->ipv4_lookup_struct); 1253 if (dst_port >= RTE_MAX_ETHPORTS || 1254 (enabled_port_mask & 1 << dst_port) == 0) 1255 dst_port = portid; 1256 1257 #ifdef DO_RFC_1812_CHECKS 1258 /* Update time to live and header checksum */ 1259 --(ipv4_hdr->time_to_live); 1260 ++(ipv4_hdr->hdr_checksum); 1261 #endif 1262 /* dst addr */ 1263 *(uint64_t *)ð_hdr->d_addr = dest_eth_addr[dst_port]; 1264 1265 /* src addr */ 1266 ether_addr_copy(&ports_eth_addr[dst_port], ð_hdr->s_addr); 1267 1268 send_single_packet(m, dst_port); 1269 } else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) { 1270 /* Handle IPv6 headers.*/ 1271 struct ipv6_hdr *ipv6_hdr; 1272 1273 ipv6_hdr = rte_pktmbuf_mtod_offset(m, struct ipv6_hdr *, 1274 sizeof(struct ether_hdr)); 1275 1276 dst_port = get_ipv6_dst_port(ipv6_hdr, portid, 1277 RTE_PER_LCORE(lcore_conf)->ipv6_lookup_struct); 1278 1279 if (dst_port >= RTE_MAX_ETHPORTS || 1280 (enabled_port_mask & 1 << dst_port) == 0) 1281 dst_port = portid; 1282 1283 /* dst addr */ 1284 *(uint64_t *)ð_hdr->d_addr = dest_eth_addr[dst_port]; 1285 1286 /* src addr */ 1287 ether_addr_copy(&ports_eth_addr[dst_port], ð_hdr->s_addr); 1288 1289 send_single_packet(m, dst_port); 1290 } else 1291 /* Free the mbuf that contains non-IPV4/IPV6 packet */ 1292 rte_pktmbuf_free(m); 1293 } 1294 1295 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_LPM) && \ 1296 (ENABLE_MULTI_BUFFER_OPTIMIZE == 1)) 1297 #ifdef DO_RFC_1812_CHECKS 1298 1299 #define IPV4_MIN_VER_IHL 0x45 1300 #define IPV4_MAX_VER_IHL 0x4f 1301 #define IPV4_MAX_VER_IHL_DIFF (IPV4_MAX_VER_IHL - IPV4_MIN_VER_IHL) 1302 1303 /* Minimum value of IPV4 total length (20B) in network byte order. */ 1304 #define IPV4_MIN_LEN_BE (sizeof(struct ipv4_hdr) << 8) 1305 1306 /* 1307 * From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2: 1308 * - The IP version number must be 4. 1309 * - The IP header length field must be large enough to hold the 1310 * minimum length legal IP datagram (20 bytes = 5 words). 1311 * - The IP total length field must be large enough to hold the IP 1312 * datagram header, whose length is specified in the IP header length 1313 * field. 1314 * If we encounter invalid IPV4 packet, then set destination port for it 1315 * to BAD_PORT value. 1316 */ 1317 static __rte_always_inline void 1318 rfc1812_process(struct ipv4_hdr *ipv4_hdr, uint16_t *dp, uint32_t ptype) 1319 { 1320 uint8_t ihl; 1321 1322 if (RTE_ETH_IS_IPV4_HDR(ptype)) { 1323 ihl = ipv4_hdr->version_ihl - IPV4_MIN_VER_IHL; 1324 1325 ipv4_hdr->time_to_live--; 1326 ipv4_hdr->hdr_checksum++; 1327 1328 if (ihl > IPV4_MAX_VER_IHL_DIFF || 1329 ((uint8_t)ipv4_hdr->total_length == 0 && 1330 ipv4_hdr->total_length < IPV4_MIN_LEN_BE)) { 1331 dp[0] = BAD_PORT; 1332 } 1333 } 1334 } 1335 1336 #else 1337 #define rfc1812_process(mb, dp, ptype) do { } while (0) 1338 #endif /* DO_RFC_1812_CHECKS */ 1339 #endif /* APP_LOOKUP_LPM && ENABLE_MULTI_BUFFER_OPTIMIZE */ 1340 1341 1342 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_LPM) && \ 1343 (ENABLE_MULTI_BUFFER_OPTIMIZE == 1)) 1344 1345 static __rte_always_inline uint16_t 1346 get_dst_port(struct rte_mbuf *pkt, uint32_t dst_ipv4, uint16_t portid) 1347 { 1348 uint32_t next_hop; 1349 struct ipv6_hdr *ipv6_hdr; 1350 struct ether_hdr *eth_hdr; 1351 1352 if (RTE_ETH_IS_IPV4_HDR(pkt->packet_type)) { 1353 return (uint16_t) ((rte_lpm_lookup( 1354 RTE_PER_LCORE(lcore_conf)->ipv4_lookup_struct, dst_ipv4, 1355 &next_hop) == 0) ? next_hop : portid); 1356 1357 } else if (RTE_ETH_IS_IPV6_HDR(pkt->packet_type)) { 1358 1359 eth_hdr = rte_pktmbuf_mtod(pkt, struct ether_hdr *); 1360 ipv6_hdr = (struct ipv6_hdr *)(eth_hdr + 1); 1361 1362 return (uint16_t) ((rte_lpm6_lookup( 1363 RTE_PER_LCORE(lcore_conf)->ipv6_lookup_struct, 1364 ipv6_hdr->dst_addr, &next_hop) == 0) ? 1365 next_hop : portid); 1366 1367 } 1368 1369 return portid; 1370 } 1371 1372 static inline void 1373 process_packet(struct rte_mbuf *pkt, uint16_t *dst_port, uint16_t portid) 1374 { 1375 struct ether_hdr *eth_hdr; 1376 struct ipv4_hdr *ipv4_hdr; 1377 uint32_t dst_ipv4; 1378 uint16_t dp; 1379 __m128i te, ve; 1380 1381 eth_hdr = rte_pktmbuf_mtod(pkt, struct ether_hdr *); 1382 ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1); 1383 1384 dst_ipv4 = ipv4_hdr->dst_addr; 1385 dst_ipv4 = rte_be_to_cpu_32(dst_ipv4); 1386 dp = get_dst_port(pkt, dst_ipv4, portid); 1387 1388 te = _mm_load_si128((__m128i *)eth_hdr); 1389 ve = val_eth[dp]; 1390 1391 dst_port[0] = dp; 1392 rfc1812_process(ipv4_hdr, dst_port, pkt->packet_type); 1393 1394 te = _mm_blend_epi16(te, ve, MASK_ETH); 1395 _mm_store_si128((__m128i *)eth_hdr, te); 1396 } 1397 1398 /* 1399 * Read packet_type and destination IPV4 addresses from 4 mbufs. 1400 */ 1401 static inline void 1402 processx4_step1(struct rte_mbuf *pkt[FWDSTEP], 1403 __m128i *dip, 1404 uint32_t *ipv4_flag) 1405 { 1406 struct ipv4_hdr *ipv4_hdr; 1407 struct ether_hdr *eth_hdr; 1408 uint32_t x0, x1, x2, x3; 1409 1410 eth_hdr = rte_pktmbuf_mtod(pkt[0], struct ether_hdr *); 1411 ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1); 1412 x0 = ipv4_hdr->dst_addr; 1413 ipv4_flag[0] = pkt[0]->packet_type & RTE_PTYPE_L3_IPV4; 1414 1415 eth_hdr = rte_pktmbuf_mtod(pkt[1], struct ether_hdr *); 1416 ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1); 1417 x1 = ipv4_hdr->dst_addr; 1418 ipv4_flag[0] &= pkt[1]->packet_type; 1419 1420 eth_hdr = rte_pktmbuf_mtod(pkt[2], struct ether_hdr *); 1421 ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1); 1422 x2 = ipv4_hdr->dst_addr; 1423 ipv4_flag[0] &= pkt[2]->packet_type; 1424 1425 eth_hdr = rte_pktmbuf_mtod(pkt[3], struct ether_hdr *); 1426 ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1); 1427 x3 = ipv4_hdr->dst_addr; 1428 ipv4_flag[0] &= pkt[3]->packet_type; 1429 1430 dip[0] = _mm_set_epi32(x3, x2, x1, x0); 1431 } 1432 1433 /* 1434 * Lookup into LPM for destination port. 1435 * If lookup fails, use incoming port (portid) as destination port. 1436 */ 1437 static inline void 1438 processx4_step2(__m128i dip, 1439 uint32_t ipv4_flag, 1440 uint16_t portid, 1441 struct rte_mbuf *pkt[FWDSTEP], 1442 uint16_t dprt[FWDSTEP]) 1443 { 1444 rte_xmm_t dst; 1445 const __m128i bswap_mask = _mm_set_epi8(12, 13, 14, 15, 8, 9, 10, 11, 1446 4, 5, 6, 7, 0, 1, 2, 3); 1447 1448 /* Byte swap 4 IPV4 addresses. */ 1449 dip = _mm_shuffle_epi8(dip, bswap_mask); 1450 1451 /* if all 4 packets are IPV4. */ 1452 if (likely(ipv4_flag)) { 1453 rte_lpm_lookupx4(RTE_PER_LCORE(lcore_conf)->ipv4_lookup_struct, dip, 1454 dst.u32, portid); 1455 1456 /* get rid of unused upper 16 bit for each dport. */ 1457 dst.x = _mm_packs_epi32(dst.x, dst.x); 1458 *(uint64_t *)dprt = dst.u64[0]; 1459 } else { 1460 dst.x = dip; 1461 dprt[0] = get_dst_port(pkt[0], dst.u32[0], portid); 1462 dprt[1] = get_dst_port(pkt[1], dst.u32[1], portid); 1463 dprt[2] = get_dst_port(pkt[2], dst.u32[2], portid); 1464 dprt[3] = get_dst_port(pkt[3], dst.u32[3], portid); 1465 } 1466 } 1467 1468 /* 1469 * Update source and destination MAC addresses in the ethernet header. 1470 * Perform RFC1812 checks and updates for IPV4 packets. 1471 */ 1472 static inline void 1473 processx4_step3(struct rte_mbuf *pkt[FWDSTEP], uint16_t dst_port[FWDSTEP]) 1474 { 1475 __m128i te[FWDSTEP]; 1476 __m128i ve[FWDSTEP]; 1477 __m128i *p[FWDSTEP]; 1478 1479 p[0] = rte_pktmbuf_mtod(pkt[0], __m128i *); 1480 p[1] = rte_pktmbuf_mtod(pkt[1], __m128i *); 1481 p[2] = rte_pktmbuf_mtod(pkt[2], __m128i *); 1482 p[3] = rte_pktmbuf_mtod(pkt[3], __m128i *); 1483 1484 ve[0] = val_eth[dst_port[0]]; 1485 te[0] = _mm_load_si128(p[0]); 1486 1487 ve[1] = val_eth[dst_port[1]]; 1488 te[1] = _mm_load_si128(p[1]); 1489 1490 ve[2] = val_eth[dst_port[2]]; 1491 te[2] = _mm_load_si128(p[2]); 1492 1493 ve[3] = val_eth[dst_port[3]]; 1494 te[3] = _mm_load_si128(p[3]); 1495 1496 /* Update first 12 bytes, keep rest bytes intact. */ 1497 te[0] = _mm_blend_epi16(te[0], ve[0], MASK_ETH); 1498 te[1] = _mm_blend_epi16(te[1], ve[1], MASK_ETH); 1499 te[2] = _mm_blend_epi16(te[2], ve[2], MASK_ETH); 1500 te[3] = _mm_blend_epi16(te[3], ve[3], MASK_ETH); 1501 1502 _mm_store_si128(p[0], te[0]); 1503 _mm_store_si128(p[1], te[1]); 1504 _mm_store_si128(p[2], te[2]); 1505 _mm_store_si128(p[3], te[3]); 1506 1507 rfc1812_process((struct ipv4_hdr *)((struct ether_hdr *)p[0] + 1), 1508 &dst_port[0], pkt[0]->packet_type); 1509 rfc1812_process((struct ipv4_hdr *)((struct ether_hdr *)p[1] + 1), 1510 &dst_port[1], pkt[1]->packet_type); 1511 rfc1812_process((struct ipv4_hdr *)((struct ether_hdr *)p[2] + 1), 1512 &dst_port[2], pkt[2]->packet_type); 1513 rfc1812_process((struct ipv4_hdr *)((struct ether_hdr *)p[3] + 1), 1514 &dst_port[3], pkt[3]->packet_type); 1515 } 1516 1517 /* 1518 * We group consecutive packets with the same destionation port into one burst. 1519 * To avoid extra latency this is done together with some other packet 1520 * processing, but after we made a final decision about packet's destination. 1521 * To do this we maintain: 1522 * pnum - array of number of consecutive packets with the same dest port for 1523 * each packet in the input burst. 1524 * lp - pointer to the last updated element in the pnum. 1525 * dlp - dest port value lp corresponds to. 1526 */ 1527 1528 #define GRPSZ (1 << FWDSTEP) 1529 #define GRPMSK (GRPSZ - 1) 1530 1531 #define GROUP_PORT_STEP(dlp, dcp, lp, pn, idx) do { \ 1532 if (likely((dlp) == (dcp)[(idx)])) { \ 1533 (lp)[0]++; \ 1534 } else { \ 1535 (dlp) = (dcp)[idx]; \ 1536 (lp) = (pn) + (idx); \ 1537 (lp)[0] = 1; \ 1538 } \ 1539 } while (0) 1540 1541 /* 1542 * Group consecutive packets with the same destination port in bursts of 4. 1543 * Suppose we have array of destionation ports: 1544 * dst_port[] = {a, b, c, d,, e, ... } 1545 * dp1 should contain: <a, b, c, d>, dp2: <b, c, d, e>. 1546 * We doing 4 comparisons at once and the result is 4 bit mask. 1547 * This mask is used as an index into prebuild array of pnum values. 1548 */ 1549 static inline uint16_t * 1550 port_groupx4(uint16_t pn[FWDSTEP + 1], uint16_t *lp, __m128i dp1, __m128i dp2) 1551 { 1552 static const struct { 1553 uint64_t pnum; /* prebuild 4 values for pnum[]. */ 1554 int32_t idx; /* index for new last updated elemnet. */ 1555 uint16_t lpv; /* add value to the last updated element. */ 1556 } gptbl[GRPSZ] = { 1557 { 1558 /* 0: a != b, b != c, c != d, d != e */ 1559 .pnum = UINT64_C(0x0001000100010001), 1560 .idx = 4, 1561 .lpv = 0, 1562 }, 1563 { 1564 /* 1: a == b, b != c, c != d, d != e */ 1565 .pnum = UINT64_C(0x0001000100010002), 1566 .idx = 4, 1567 .lpv = 1, 1568 }, 1569 { 1570 /* 2: a != b, b == c, c != d, d != e */ 1571 .pnum = UINT64_C(0x0001000100020001), 1572 .idx = 4, 1573 .lpv = 0, 1574 }, 1575 { 1576 /* 3: a == b, b == c, c != d, d != e */ 1577 .pnum = UINT64_C(0x0001000100020003), 1578 .idx = 4, 1579 .lpv = 2, 1580 }, 1581 { 1582 /* 4: a != b, b != c, c == d, d != e */ 1583 .pnum = UINT64_C(0x0001000200010001), 1584 .idx = 4, 1585 .lpv = 0, 1586 }, 1587 { 1588 /* 5: a == b, b != c, c == d, d != e */ 1589 .pnum = UINT64_C(0x0001000200010002), 1590 .idx = 4, 1591 .lpv = 1, 1592 }, 1593 { 1594 /* 6: a != b, b == c, c == d, d != e */ 1595 .pnum = UINT64_C(0x0001000200030001), 1596 .idx = 4, 1597 .lpv = 0, 1598 }, 1599 { 1600 /* 7: a == b, b == c, c == d, d != e */ 1601 .pnum = UINT64_C(0x0001000200030004), 1602 .idx = 4, 1603 .lpv = 3, 1604 }, 1605 { 1606 /* 8: a != b, b != c, c != d, d == e */ 1607 .pnum = UINT64_C(0x0002000100010001), 1608 .idx = 3, 1609 .lpv = 0, 1610 }, 1611 { 1612 /* 9: a == b, b != c, c != d, d == e */ 1613 .pnum = UINT64_C(0x0002000100010002), 1614 .idx = 3, 1615 .lpv = 1, 1616 }, 1617 { 1618 /* 0xa: a != b, b == c, c != d, d == e */ 1619 .pnum = UINT64_C(0x0002000100020001), 1620 .idx = 3, 1621 .lpv = 0, 1622 }, 1623 { 1624 /* 0xb: a == b, b == c, c != d, d == e */ 1625 .pnum = UINT64_C(0x0002000100020003), 1626 .idx = 3, 1627 .lpv = 2, 1628 }, 1629 { 1630 /* 0xc: a != b, b != c, c == d, d == e */ 1631 .pnum = UINT64_C(0x0002000300010001), 1632 .idx = 2, 1633 .lpv = 0, 1634 }, 1635 { 1636 /* 0xd: a == b, b != c, c == d, d == e */ 1637 .pnum = UINT64_C(0x0002000300010002), 1638 .idx = 2, 1639 .lpv = 1, 1640 }, 1641 { 1642 /* 0xe: a != b, b == c, c == d, d == e */ 1643 .pnum = UINT64_C(0x0002000300040001), 1644 .idx = 1, 1645 .lpv = 0, 1646 }, 1647 { 1648 /* 0xf: a == b, b == c, c == d, d == e */ 1649 .pnum = UINT64_C(0x0002000300040005), 1650 .idx = 0, 1651 .lpv = 4, 1652 }, 1653 }; 1654 1655 union { 1656 uint16_t u16[FWDSTEP + 1]; 1657 uint64_t u64; 1658 } *pnum = (void *)pn; 1659 1660 int32_t v; 1661 1662 dp1 = _mm_cmpeq_epi16(dp1, dp2); 1663 dp1 = _mm_unpacklo_epi16(dp1, dp1); 1664 v = _mm_movemask_ps((__m128)dp1); 1665 1666 /* update last port counter. */ 1667 lp[0] += gptbl[v].lpv; 1668 1669 /* if dest port value has changed. */ 1670 if (v != GRPMSK) { 1671 pnum->u64 = gptbl[v].pnum; 1672 pnum->u16[FWDSTEP] = 1; 1673 lp = pnum->u16 + gptbl[v].idx; 1674 } 1675 1676 return lp; 1677 } 1678 1679 #endif /* APP_LOOKUP_METHOD */ 1680 1681 static void 1682 process_burst(struct rte_mbuf *pkts_burst[MAX_PKT_BURST], int nb_rx, 1683 uint16_t portid) 1684 { 1685 1686 int j; 1687 1688 #if ((APP_LOOKUP_METHOD == APP_LOOKUP_LPM) && \ 1689 (ENABLE_MULTI_BUFFER_OPTIMIZE == 1)) 1690 int32_t k; 1691 uint16_t dlp; 1692 uint16_t *lp; 1693 uint16_t dst_port[MAX_PKT_BURST]; 1694 __m128i dip[MAX_PKT_BURST / FWDSTEP]; 1695 uint32_t ipv4_flag[MAX_PKT_BURST / FWDSTEP]; 1696 uint16_t pnum[MAX_PKT_BURST + 1]; 1697 #endif 1698 1699 1700 #if (ENABLE_MULTI_BUFFER_OPTIMIZE == 1) 1701 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH) 1702 { 1703 /* 1704 * Send nb_rx - nb_rx%8 packets 1705 * in groups of 8. 1706 */ 1707 int32_t n = RTE_ALIGN_FLOOR(nb_rx, 8); 1708 1709 for (j = 0; j < n; j += 8) { 1710 uint32_t pkt_type = 1711 pkts_burst[j]->packet_type & 1712 pkts_burst[j+1]->packet_type & 1713 pkts_burst[j+2]->packet_type & 1714 pkts_burst[j+3]->packet_type & 1715 pkts_burst[j+4]->packet_type & 1716 pkts_burst[j+5]->packet_type & 1717 pkts_burst[j+6]->packet_type & 1718 pkts_burst[j+7]->packet_type; 1719 if (pkt_type & RTE_PTYPE_L3_IPV4) { 1720 simple_ipv4_fwd_8pkts(&pkts_burst[j], portid); 1721 } else if (pkt_type & 1722 RTE_PTYPE_L3_IPV6) { 1723 simple_ipv6_fwd_8pkts(&pkts_burst[j], portid); 1724 } else { 1725 l3fwd_simple_forward(pkts_burst[j], portid); 1726 l3fwd_simple_forward(pkts_burst[j+1], portid); 1727 l3fwd_simple_forward(pkts_burst[j+2], portid); 1728 l3fwd_simple_forward(pkts_burst[j+3], portid); 1729 l3fwd_simple_forward(pkts_burst[j+4], portid); 1730 l3fwd_simple_forward(pkts_burst[j+5], portid); 1731 l3fwd_simple_forward(pkts_burst[j+6], portid); 1732 l3fwd_simple_forward(pkts_burst[j+7], portid); 1733 } 1734 } 1735 for (; j < nb_rx ; j++) 1736 l3fwd_simple_forward(pkts_burst[j], portid); 1737 } 1738 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM) 1739 1740 k = RTE_ALIGN_FLOOR(nb_rx, FWDSTEP); 1741 for (j = 0; j != k; j += FWDSTEP) 1742 processx4_step1(&pkts_burst[j], &dip[j / FWDSTEP], 1743 &ipv4_flag[j / FWDSTEP]); 1744 1745 k = RTE_ALIGN_FLOOR(nb_rx, FWDSTEP); 1746 for (j = 0; j != k; j += FWDSTEP) 1747 processx4_step2(dip[j / FWDSTEP], ipv4_flag[j / FWDSTEP], 1748 portid, &pkts_burst[j], &dst_port[j]); 1749 1750 /* 1751 * Finish packet processing and group consecutive 1752 * packets with the same destination port. 1753 */ 1754 k = RTE_ALIGN_FLOOR(nb_rx, FWDSTEP); 1755 if (k != 0) { 1756 __m128i dp1, dp2; 1757 1758 lp = pnum; 1759 lp[0] = 1; 1760 1761 processx4_step3(pkts_burst, dst_port); 1762 1763 /* dp1: <d[0], d[1], d[2], d[3], ... > */ 1764 dp1 = _mm_loadu_si128((__m128i *)dst_port); 1765 1766 for (j = FWDSTEP; j != k; j += FWDSTEP) { 1767 processx4_step3(&pkts_burst[j], &dst_port[j]); 1768 1769 /* 1770 * dp2: 1771 * <d[j-3], d[j-2], d[j-1], d[j], ... > 1772 */ 1773 dp2 = _mm_loadu_si128( 1774 (__m128i *)&dst_port[j - FWDSTEP + 1]); 1775 lp = port_groupx4(&pnum[j - FWDSTEP], lp, dp1, dp2); 1776 1777 /* 1778 * dp1: 1779 * <d[j], d[j+1], d[j+2], d[j+3], ... > 1780 */ 1781 dp1 = _mm_srli_si128(dp2, (FWDSTEP - 1) * 1782 sizeof(dst_port[0])); 1783 } 1784 1785 /* 1786 * dp2: <d[j-3], d[j-2], d[j-1], d[j-1], ... > 1787 */ 1788 dp2 = _mm_shufflelo_epi16(dp1, 0xf9); 1789 lp = port_groupx4(&pnum[j - FWDSTEP], lp, dp1, dp2); 1790 1791 /* 1792 * remove values added by the last repeated 1793 * dst port. 1794 */ 1795 lp[0]--; 1796 dlp = dst_port[j - 1]; 1797 } else { 1798 /* set dlp and lp to the never used values. */ 1799 dlp = BAD_PORT - 1; 1800 lp = pnum + MAX_PKT_BURST; 1801 } 1802 1803 /* Process up to last 3 packets one by one. */ 1804 switch (nb_rx % FWDSTEP) { 1805 case 3: 1806 process_packet(pkts_burst[j], dst_port + j, portid); 1807 GROUP_PORT_STEP(dlp, dst_port, lp, pnum, j); 1808 j++; 1809 /* fall-through */ 1810 case 2: 1811 process_packet(pkts_burst[j], dst_port + j, portid); 1812 GROUP_PORT_STEP(dlp, dst_port, lp, pnum, j); 1813 j++; 1814 /* fall-through */ 1815 case 1: 1816 process_packet(pkts_burst[j], dst_port + j, portid); 1817 GROUP_PORT_STEP(dlp, dst_port, lp, pnum, j); 1818 j++; 1819 } 1820 1821 /* 1822 * Send packets out, through destination port. 1823 * Consecuteve pacekts with the same destination port 1824 * are already grouped together. 1825 * If destination port for the packet equals BAD_PORT, 1826 * then free the packet without sending it out. 1827 */ 1828 for (j = 0; j < nb_rx; j += k) { 1829 1830 int32_t m; 1831 uint16_t pn; 1832 1833 pn = dst_port[j]; 1834 k = pnum[j]; 1835 1836 if (likely(pn != BAD_PORT)) 1837 send_packetsx4(pn, pkts_burst + j, k); 1838 else 1839 for (m = j; m != j + k; m++) 1840 rte_pktmbuf_free(pkts_burst[m]); 1841 1842 } 1843 1844 #endif /* APP_LOOKUP_METHOD */ 1845 #else /* ENABLE_MULTI_BUFFER_OPTIMIZE == 0 */ 1846 1847 /* Prefetch first packets */ 1848 for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) 1849 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[j], void *)); 1850 1851 /* Prefetch and forward already prefetched packets */ 1852 for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) { 1853 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[ 1854 j + PREFETCH_OFFSET], void *)); 1855 l3fwd_simple_forward(pkts_burst[j], portid); 1856 } 1857 1858 /* Forward remaining prefetched packets */ 1859 for (; j < nb_rx; j++) 1860 l3fwd_simple_forward(pkts_burst[j], portid); 1861 1862 #endif /* ENABLE_MULTI_BUFFER_OPTIMIZE */ 1863 1864 } 1865 1866 #if (APP_CPU_LOAD > 0) 1867 1868 /* 1869 * CPU-load stats collector 1870 */ 1871 static int 1872 cpu_load_collector(__rte_unused void *arg) { 1873 unsigned i, j, k; 1874 uint64_t hits; 1875 uint64_t prev_tsc, diff_tsc, cur_tsc; 1876 uint64_t total[MAX_CPU] = { 0 }; 1877 unsigned min_cpu = MAX_CPU; 1878 unsigned max_cpu = 0; 1879 unsigned cpu_id; 1880 int busy_total = 0; 1881 int busy_flag = 0; 1882 1883 unsigned int n_thread_per_cpu[MAX_CPU] = { 0 }; 1884 struct thread_conf *thread_per_cpu[MAX_CPU][MAX_THREAD]; 1885 1886 struct thread_conf *thread_conf; 1887 1888 const uint64_t interval_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / 1889 US_PER_S * CPU_LOAD_TIMEOUT_US; 1890 1891 prev_tsc = 0; 1892 /* 1893 * Wait for all threads 1894 */ 1895 1896 printf("Waiting for %d rx threads and %d tx threads\n", n_rx_thread, 1897 n_tx_thread); 1898 1899 while (rte_atomic16_read(&rx_counter) < n_rx_thread) 1900 rte_pause(); 1901 1902 while (rte_atomic16_read(&tx_counter) < n_tx_thread) 1903 rte_pause(); 1904 1905 for (i = 0; i < n_rx_thread; i++) { 1906 1907 thread_conf = &rx_thread[i].conf; 1908 cpu_id = thread_conf->cpu_id; 1909 thread_per_cpu[cpu_id][n_thread_per_cpu[cpu_id]++] = thread_conf; 1910 1911 if (cpu_id > max_cpu) 1912 max_cpu = cpu_id; 1913 if (cpu_id < min_cpu) 1914 min_cpu = cpu_id; 1915 } 1916 for (i = 0; i < n_tx_thread; i++) { 1917 1918 thread_conf = &tx_thread[i].conf; 1919 cpu_id = thread_conf->cpu_id; 1920 thread_per_cpu[cpu_id][n_thread_per_cpu[cpu_id]++] = thread_conf; 1921 1922 if (thread_conf->cpu_id > max_cpu) 1923 max_cpu = thread_conf->cpu_id; 1924 if (thread_conf->cpu_id < min_cpu) 1925 min_cpu = thread_conf->cpu_id; 1926 } 1927 1928 while (1) { 1929 1930 cpu_load.counter++; 1931 for (i = min_cpu; i <= max_cpu; i++) { 1932 for (j = 0; j < MAX_CPU_COUNTER; j++) { 1933 for (k = 0; k < n_thread_per_cpu[i]; k++) 1934 if (thread_per_cpu[i][k]->busy[j]) { 1935 busy_flag = 1; 1936 break; 1937 } 1938 if (busy_flag) { 1939 cpu_load.hits[j][i]++; 1940 busy_total = 1; 1941 busy_flag = 0; 1942 } 1943 } 1944 1945 if (busy_total) { 1946 total[i]++; 1947 busy_total = 0; 1948 } 1949 } 1950 1951 cur_tsc = rte_rdtsc(); 1952 1953 diff_tsc = cur_tsc - prev_tsc; 1954 if (unlikely(diff_tsc > interval_tsc)) { 1955 1956 printf("\033c"); 1957 1958 printf("Cpu usage for %d rx threads and %d tx threads:\n\n", 1959 n_rx_thread, n_tx_thread); 1960 1961 printf("cpu# proc%% poll%% overhead%%\n\n"); 1962 1963 for (i = min_cpu; i <= max_cpu; i++) { 1964 hits = 0; 1965 printf("CPU %d:", i); 1966 for (j = 0; j < MAX_CPU_COUNTER; j++) { 1967 printf("%7" PRIu64 "", 1968 cpu_load.hits[j][i] * 100 / cpu_load.counter); 1969 hits += cpu_load.hits[j][i]; 1970 cpu_load.hits[j][i] = 0; 1971 } 1972 printf("%7" PRIu64 "\n", 1973 100 - total[i] * 100 / cpu_load.counter); 1974 total[i] = 0; 1975 } 1976 cpu_load.counter = 0; 1977 1978 prev_tsc = cur_tsc; 1979 } 1980 1981 } 1982 } 1983 #endif /* APP_CPU_LOAD */ 1984 1985 /* 1986 * Null processing lthread loop 1987 * 1988 * This loop is used to start empty scheduler on lcore. 1989 */ 1990 static void * 1991 lthread_null(__rte_unused void *args) 1992 { 1993 int lcore_id = rte_lcore_id(); 1994 1995 RTE_LOG(INFO, L3FWD, "Starting scheduler on lcore %d.\n", lcore_id); 1996 lthread_exit(NULL); 1997 return NULL; 1998 } 1999 2000 /* main processing loop */ 2001 static void * 2002 lthread_tx_per_ring(void *dummy) 2003 { 2004 int nb_rx; 2005 uint16_t portid; 2006 struct rte_ring *ring; 2007 struct thread_tx_conf *tx_conf; 2008 struct rte_mbuf *pkts_burst[MAX_PKT_BURST]; 2009 struct lthread_cond *ready; 2010 2011 tx_conf = (struct thread_tx_conf *)dummy; 2012 ring = tx_conf->ring; 2013 ready = *tx_conf->ready; 2014 2015 lthread_set_data((void *)tx_conf); 2016 2017 /* 2018 * Move this lthread to lcore 2019 */ 2020 lthread_set_affinity(tx_conf->conf.lcore_id); 2021 2022 RTE_LOG(INFO, L3FWD, "entering main tx loop on lcore %u\n", rte_lcore_id()); 2023 2024 nb_rx = 0; 2025 rte_atomic16_inc(&tx_counter); 2026 while (1) { 2027 2028 /* 2029 * Read packet from ring 2030 */ 2031 SET_CPU_BUSY(tx_conf, CPU_POLL); 2032 nb_rx = rte_ring_sc_dequeue_burst(ring, (void **)pkts_burst, 2033 MAX_PKT_BURST, NULL); 2034 SET_CPU_IDLE(tx_conf, CPU_POLL); 2035 2036 if (nb_rx > 0) { 2037 SET_CPU_BUSY(tx_conf, CPU_PROCESS); 2038 portid = pkts_burst[0]->port; 2039 process_burst(pkts_burst, nb_rx, portid); 2040 SET_CPU_IDLE(tx_conf, CPU_PROCESS); 2041 lthread_yield(); 2042 } else 2043 lthread_cond_wait(ready, 0); 2044 2045 } 2046 return NULL; 2047 } 2048 2049 /* 2050 * Main tx-lthreads spawner lthread. 2051 * 2052 * This lthread is used to spawn one new lthread per ring from producers. 2053 * 2054 */ 2055 static void * 2056 lthread_tx(void *args) 2057 { 2058 struct lthread *lt; 2059 2060 unsigned lcore_id; 2061 uint16_t portid; 2062 struct thread_tx_conf *tx_conf; 2063 2064 tx_conf = (struct thread_tx_conf *)args; 2065 lthread_set_data((void *)tx_conf); 2066 2067 /* 2068 * Move this lthread to the selected lcore 2069 */ 2070 lthread_set_affinity(tx_conf->conf.lcore_id); 2071 2072 /* 2073 * Spawn tx readers (one per input ring) 2074 */ 2075 lthread_create(<, tx_conf->conf.lcore_id, lthread_tx_per_ring, 2076 (void *)tx_conf); 2077 2078 lcore_id = rte_lcore_id(); 2079 2080 RTE_LOG(INFO, L3FWD, "Entering Tx main loop on lcore %u\n", lcore_id); 2081 2082 tx_conf->conf.cpu_id = sched_getcpu(); 2083 while (1) { 2084 2085 lthread_sleep(BURST_TX_DRAIN_US * 1000); 2086 2087 /* 2088 * TX burst queue drain 2089 */ 2090 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) { 2091 if (tx_conf->tx_mbufs[portid].len == 0) 2092 continue; 2093 SET_CPU_BUSY(tx_conf, CPU_PROCESS); 2094 send_burst(tx_conf, tx_conf->tx_mbufs[portid].len, portid); 2095 SET_CPU_IDLE(tx_conf, CPU_PROCESS); 2096 tx_conf->tx_mbufs[portid].len = 0; 2097 } 2098 2099 } 2100 return NULL; 2101 } 2102 2103 static void * 2104 lthread_rx(void *dummy) 2105 { 2106 int ret; 2107 uint16_t nb_rx; 2108 int i; 2109 uint16_t portid; 2110 uint8_t queueid; 2111 int worker_id; 2112 int len[RTE_MAX_LCORE] = { 0 }; 2113 int old_len, new_len; 2114 struct rte_mbuf *pkts_burst[MAX_PKT_BURST]; 2115 struct thread_rx_conf *rx_conf; 2116 2117 rx_conf = (struct thread_rx_conf *)dummy; 2118 lthread_set_data((void *)rx_conf); 2119 2120 /* 2121 * Move this lthread to lcore 2122 */ 2123 lthread_set_affinity(rx_conf->conf.lcore_id); 2124 2125 if (rx_conf->n_rx_queue == 0) { 2126 RTE_LOG(INFO, L3FWD, "lcore %u has nothing to do\n", rte_lcore_id()); 2127 return NULL; 2128 } 2129 2130 RTE_LOG(INFO, L3FWD, "Entering main Rx loop on lcore %u\n", rte_lcore_id()); 2131 2132 for (i = 0; i < rx_conf->n_rx_queue; i++) { 2133 2134 portid = rx_conf->rx_queue_list[i].port_id; 2135 queueid = rx_conf->rx_queue_list[i].queue_id; 2136 RTE_LOG(INFO, L3FWD, 2137 " -- lcoreid=%u portid=%u rxqueueid=%hhu\n", 2138 rte_lcore_id(), portid, queueid); 2139 } 2140 2141 /* 2142 * Init all condition variables (one per rx thread) 2143 */ 2144 for (i = 0; i < rx_conf->n_rx_queue; i++) 2145 lthread_cond_init(NULL, &rx_conf->ready[i], NULL); 2146 2147 worker_id = 0; 2148 2149 rx_conf->conf.cpu_id = sched_getcpu(); 2150 rte_atomic16_inc(&rx_counter); 2151 while (1) { 2152 2153 /* 2154 * Read packet from RX queues 2155 */ 2156 for (i = 0; i < rx_conf->n_rx_queue; ++i) { 2157 portid = rx_conf->rx_queue_list[i].port_id; 2158 queueid = rx_conf->rx_queue_list[i].queue_id; 2159 2160 SET_CPU_BUSY(rx_conf, CPU_POLL); 2161 nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst, 2162 MAX_PKT_BURST); 2163 SET_CPU_IDLE(rx_conf, CPU_POLL); 2164 2165 if (nb_rx != 0) { 2166 worker_id = (worker_id + 1) % rx_conf->n_ring; 2167 old_len = len[worker_id]; 2168 2169 SET_CPU_BUSY(rx_conf, CPU_PROCESS); 2170 ret = rte_ring_sp_enqueue_burst( 2171 rx_conf->ring[worker_id], 2172 (void **) pkts_burst, 2173 nb_rx, NULL); 2174 2175 new_len = old_len + ret; 2176 2177 if (new_len >= BURST_SIZE) { 2178 lthread_cond_signal(rx_conf->ready[worker_id]); 2179 new_len = 0; 2180 } 2181 2182 len[worker_id] = new_len; 2183 2184 if (unlikely(ret < nb_rx)) { 2185 uint32_t k; 2186 2187 for (k = ret; k < nb_rx; k++) { 2188 struct rte_mbuf *m = pkts_burst[k]; 2189 2190 rte_pktmbuf_free(m); 2191 } 2192 } 2193 SET_CPU_IDLE(rx_conf, CPU_PROCESS); 2194 } 2195 2196 lthread_yield(); 2197 } 2198 } 2199 return NULL; 2200 } 2201 2202 /* 2203 * Start scheduler with initial lthread on lcore 2204 * 2205 * This lthread loop spawns all rx and tx lthreads on master lcore 2206 */ 2207 2208 static void * 2209 lthread_spawner(__rte_unused void *arg) 2210 { 2211 struct lthread *lt[MAX_THREAD]; 2212 int i; 2213 int n_thread = 0; 2214 2215 printf("Entering lthread_spawner\n"); 2216 2217 /* 2218 * Create producers (rx threads) on default lcore 2219 */ 2220 for (i = 0; i < n_rx_thread; i++) { 2221 rx_thread[i].conf.thread_id = i; 2222 lthread_create(<[n_thread], -1, lthread_rx, 2223 (void *)&rx_thread[i]); 2224 n_thread++; 2225 } 2226 2227 /* 2228 * Wait for all producers. Until some producers can be started on the same 2229 * scheduler as this lthread, yielding is required to let them to run and 2230 * prevent deadlock here. 2231 */ 2232 while (rte_atomic16_read(&rx_counter) < n_rx_thread) 2233 lthread_sleep(100000); 2234 2235 /* 2236 * Create consumers (tx threads) on default lcore_id 2237 */ 2238 for (i = 0; i < n_tx_thread; i++) { 2239 tx_thread[i].conf.thread_id = i; 2240 lthread_create(<[n_thread], -1, lthread_tx, 2241 (void *)&tx_thread[i]); 2242 n_thread++; 2243 } 2244 2245 /* 2246 * Wait for all threads finished 2247 */ 2248 for (i = 0; i < n_thread; i++) 2249 lthread_join(lt[i], NULL); 2250 2251 return NULL; 2252 } 2253 2254 /* 2255 * Start master scheduler with initial lthread spawning rx and tx lthreads 2256 * (main_lthread_master). 2257 */ 2258 static int 2259 lthread_master_spawner(__rte_unused void *arg) { 2260 struct lthread *lt; 2261 int lcore_id = rte_lcore_id(); 2262 2263 RTE_PER_LCORE(lcore_conf) = &lcore_conf[lcore_id]; 2264 lthread_create(<, -1, lthread_spawner, NULL); 2265 lthread_run(); 2266 2267 return 0; 2268 } 2269 2270 /* 2271 * Start scheduler on lcore. 2272 */ 2273 static int 2274 sched_spawner(__rte_unused void *arg) { 2275 struct lthread *lt; 2276 int lcore_id = rte_lcore_id(); 2277 2278 #if (APP_CPU_LOAD) 2279 if (lcore_id == cpu_load_lcore_id) { 2280 cpu_load_collector(arg); 2281 return 0; 2282 } 2283 #endif /* APP_CPU_LOAD */ 2284 2285 RTE_PER_LCORE(lcore_conf) = &lcore_conf[lcore_id]; 2286 lthread_create(<, -1, lthread_null, NULL); 2287 lthread_run(); 2288 2289 return 0; 2290 } 2291 2292 /* main processing loop */ 2293 static int 2294 pthread_tx(void *dummy) 2295 { 2296 struct rte_mbuf *pkts_burst[MAX_PKT_BURST]; 2297 uint64_t prev_tsc, diff_tsc, cur_tsc; 2298 int nb_rx; 2299 uint16_t portid; 2300 struct thread_tx_conf *tx_conf; 2301 2302 const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / 2303 US_PER_S * BURST_TX_DRAIN_US; 2304 2305 prev_tsc = 0; 2306 2307 tx_conf = (struct thread_tx_conf *)dummy; 2308 2309 RTE_LOG(INFO, L3FWD, "Entering main Tx loop on lcore %u\n", rte_lcore_id()); 2310 2311 tx_conf->conf.cpu_id = sched_getcpu(); 2312 rte_atomic16_inc(&tx_counter); 2313 while (1) { 2314 2315 cur_tsc = rte_rdtsc(); 2316 2317 /* 2318 * TX burst queue drain 2319 */ 2320 diff_tsc = cur_tsc - prev_tsc; 2321 if (unlikely(diff_tsc > drain_tsc)) { 2322 2323 /* 2324 * This could be optimized (use queueid instead of 2325 * portid), but it is not called so often 2326 */ 2327 SET_CPU_BUSY(tx_conf, CPU_PROCESS); 2328 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) { 2329 if (tx_conf->tx_mbufs[portid].len == 0) 2330 continue; 2331 send_burst(tx_conf, tx_conf->tx_mbufs[portid].len, portid); 2332 tx_conf->tx_mbufs[portid].len = 0; 2333 } 2334 SET_CPU_IDLE(tx_conf, CPU_PROCESS); 2335 2336 prev_tsc = cur_tsc; 2337 } 2338 2339 /* 2340 * Read packet from ring 2341 */ 2342 SET_CPU_BUSY(tx_conf, CPU_POLL); 2343 nb_rx = rte_ring_sc_dequeue_burst(tx_conf->ring, 2344 (void **)pkts_burst, MAX_PKT_BURST, NULL); 2345 SET_CPU_IDLE(tx_conf, CPU_POLL); 2346 2347 if (unlikely(nb_rx == 0)) { 2348 sched_yield(); 2349 continue; 2350 } 2351 2352 SET_CPU_BUSY(tx_conf, CPU_PROCESS); 2353 portid = pkts_burst[0]->port; 2354 process_burst(pkts_burst, nb_rx, portid); 2355 SET_CPU_IDLE(tx_conf, CPU_PROCESS); 2356 2357 } 2358 } 2359 2360 static int 2361 pthread_rx(void *dummy) 2362 { 2363 int i; 2364 int worker_id; 2365 uint32_t n; 2366 uint32_t nb_rx; 2367 unsigned lcore_id; 2368 uint8_t queueid; 2369 uint16_t portid; 2370 struct rte_mbuf *pkts_burst[MAX_PKT_BURST]; 2371 2372 struct thread_rx_conf *rx_conf; 2373 2374 lcore_id = rte_lcore_id(); 2375 rx_conf = (struct thread_rx_conf *)dummy; 2376 2377 if (rx_conf->n_rx_queue == 0) { 2378 RTE_LOG(INFO, L3FWD, "lcore %u has nothing to do\n", lcore_id); 2379 return 0; 2380 } 2381 2382 RTE_LOG(INFO, L3FWD, "entering main rx loop on lcore %u\n", lcore_id); 2383 2384 for (i = 0; i < rx_conf->n_rx_queue; i++) { 2385 2386 portid = rx_conf->rx_queue_list[i].port_id; 2387 queueid = rx_conf->rx_queue_list[i].queue_id; 2388 RTE_LOG(INFO, L3FWD, 2389 " -- lcoreid=%u portid=%u rxqueueid=%hhu\n", 2390 lcore_id, portid, queueid); 2391 } 2392 2393 worker_id = 0; 2394 rx_conf->conf.cpu_id = sched_getcpu(); 2395 rte_atomic16_inc(&rx_counter); 2396 while (1) { 2397 2398 /* 2399 * Read packet from RX queues 2400 */ 2401 for (i = 0; i < rx_conf->n_rx_queue; ++i) { 2402 portid = rx_conf->rx_queue_list[i].port_id; 2403 queueid = rx_conf->rx_queue_list[i].queue_id; 2404 2405 SET_CPU_BUSY(rx_conf, CPU_POLL); 2406 nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst, 2407 MAX_PKT_BURST); 2408 SET_CPU_IDLE(rx_conf, CPU_POLL); 2409 2410 if (nb_rx == 0) { 2411 sched_yield(); 2412 continue; 2413 } 2414 2415 SET_CPU_BUSY(rx_conf, CPU_PROCESS); 2416 worker_id = (worker_id + 1) % rx_conf->n_ring; 2417 n = rte_ring_sp_enqueue_burst(rx_conf->ring[worker_id], 2418 (void **)pkts_burst, nb_rx, NULL); 2419 2420 if (unlikely(n != nb_rx)) { 2421 uint32_t k; 2422 2423 for (k = n; k < nb_rx; k++) { 2424 struct rte_mbuf *m = pkts_burst[k]; 2425 2426 rte_pktmbuf_free(m); 2427 } 2428 } 2429 2430 SET_CPU_IDLE(rx_conf, CPU_PROCESS); 2431 2432 } 2433 } 2434 } 2435 2436 /* 2437 * P-Thread spawner. 2438 */ 2439 static int 2440 pthread_run(__rte_unused void *arg) { 2441 int lcore_id = rte_lcore_id(); 2442 int i; 2443 2444 for (i = 0; i < n_rx_thread; i++) 2445 if (rx_thread[i].conf.lcore_id == lcore_id) { 2446 printf("Start rx thread on %d...\n", lcore_id); 2447 RTE_PER_LCORE(lcore_conf) = &lcore_conf[lcore_id]; 2448 RTE_PER_LCORE(lcore_conf)->data = (void *)&rx_thread[i]; 2449 pthread_rx((void *)&rx_thread[i]); 2450 return 0; 2451 } 2452 2453 for (i = 0; i < n_tx_thread; i++) 2454 if (tx_thread[i].conf.lcore_id == lcore_id) { 2455 printf("Start tx thread on %d...\n", lcore_id); 2456 RTE_PER_LCORE(lcore_conf) = &lcore_conf[lcore_id]; 2457 RTE_PER_LCORE(lcore_conf)->data = (void *)&tx_thread[i]; 2458 pthread_tx((void *)&tx_thread[i]); 2459 return 0; 2460 } 2461 2462 #if (APP_CPU_LOAD) 2463 if (lcore_id == cpu_load_lcore_id) 2464 cpu_load_collector(arg); 2465 #endif /* APP_CPU_LOAD */ 2466 2467 return 0; 2468 } 2469 2470 static int 2471 check_lcore_params(void) 2472 { 2473 uint8_t queue, lcore; 2474 uint16_t i; 2475 int socketid; 2476 2477 for (i = 0; i < nb_rx_thread_params; ++i) { 2478 queue = rx_thread_params[i].queue_id; 2479 if (queue >= MAX_RX_QUEUE_PER_PORT) { 2480 printf("invalid queue number: %hhu\n", queue); 2481 return -1; 2482 } 2483 lcore = rx_thread_params[i].lcore_id; 2484 if (!rte_lcore_is_enabled(lcore)) { 2485 printf("error: lcore %hhu is not enabled in lcore mask\n", lcore); 2486 return -1; 2487 } 2488 socketid = rte_lcore_to_socket_id(lcore); 2489 if ((socketid != 0) && (numa_on == 0)) 2490 printf("warning: lcore %hhu is on socket %d with numa off\n", 2491 lcore, socketid); 2492 } 2493 return 0; 2494 } 2495 2496 static int 2497 check_port_config(void) 2498 { 2499 unsigned portid; 2500 uint16_t i; 2501 2502 for (i = 0; i < nb_rx_thread_params; ++i) { 2503 portid = rx_thread_params[i].port_id; 2504 if ((enabled_port_mask & (1 << portid)) == 0) { 2505 printf("port %u is not enabled in port mask\n", portid); 2506 return -1; 2507 } 2508 if (!rte_eth_dev_is_valid_port(portid)) { 2509 printf("port %u is not present on the board\n", portid); 2510 return -1; 2511 } 2512 } 2513 return 0; 2514 } 2515 2516 static uint8_t 2517 get_port_n_rx_queues(const uint16_t port) 2518 { 2519 int queue = -1; 2520 uint16_t i; 2521 2522 for (i = 0; i < nb_rx_thread_params; ++i) 2523 if (rx_thread_params[i].port_id == port && 2524 rx_thread_params[i].queue_id > queue) 2525 queue = rx_thread_params[i].queue_id; 2526 2527 return (uint8_t)(++queue); 2528 } 2529 2530 static int 2531 init_rx_rings(void) 2532 { 2533 unsigned socket_io; 2534 struct thread_rx_conf *rx_conf; 2535 struct thread_tx_conf *tx_conf; 2536 unsigned rx_thread_id, tx_thread_id; 2537 char name[256]; 2538 struct rte_ring *ring = NULL; 2539 2540 for (tx_thread_id = 0; tx_thread_id < n_tx_thread; tx_thread_id++) { 2541 2542 tx_conf = &tx_thread[tx_thread_id]; 2543 2544 printf("Connecting tx-thread %d with rx-thread %d\n", tx_thread_id, 2545 tx_conf->conf.thread_id); 2546 2547 rx_thread_id = tx_conf->conf.thread_id; 2548 if (rx_thread_id > n_tx_thread) { 2549 printf("connection from tx-thread %u to rx-thread %u fails " 2550 "(rx-thread not defined)\n", tx_thread_id, rx_thread_id); 2551 return -1; 2552 } 2553 2554 rx_conf = &rx_thread[rx_thread_id]; 2555 socket_io = rte_lcore_to_socket_id(rx_conf->conf.lcore_id); 2556 2557 snprintf(name, sizeof(name), "app_ring_s%u_rx%u_tx%u", 2558 socket_io, rx_thread_id, tx_thread_id); 2559 2560 ring = rte_ring_create(name, 1024 * 4, socket_io, 2561 RING_F_SP_ENQ | RING_F_SC_DEQ); 2562 2563 if (ring == NULL) { 2564 rte_panic("Cannot create ring to connect rx-thread %u " 2565 "with tx-thread %u\n", rx_thread_id, tx_thread_id); 2566 } 2567 2568 rx_conf->ring[rx_conf->n_ring] = ring; 2569 2570 tx_conf->ring = ring; 2571 tx_conf->ready = &rx_conf->ready[rx_conf->n_ring]; 2572 2573 rx_conf->n_ring++; 2574 } 2575 return 0; 2576 } 2577 2578 static int 2579 init_rx_queues(void) 2580 { 2581 uint16_t i, nb_rx_queue; 2582 uint8_t thread; 2583 2584 n_rx_thread = 0; 2585 2586 for (i = 0; i < nb_rx_thread_params; ++i) { 2587 thread = rx_thread_params[i].thread_id; 2588 nb_rx_queue = rx_thread[thread].n_rx_queue; 2589 2590 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) { 2591 printf("error: too many queues (%u) for thread: %u\n", 2592 (unsigned)nb_rx_queue + 1, (unsigned)thread); 2593 return -1; 2594 } 2595 2596 rx_thread[thread].conf.thread_id = thread; 2597 rx_thread[thread].conf.lcore_id = rx_thread_params[i].lcore_id; 2598 rx_thread[thread].rx_queue_list[nb_rx_queue].port_id = 2599 rx_thread_params[i].port_id; 2600 rx_thread[thread].rx_queue_list[nb_rx_queue].queue_id = 2601 rx_thread_params[i].queue_id; 2602 rx_thread[thread].n_rx_queue++; 2603 2604 if (thread >= n_rx_thread) 2605 n_rx_thread = thread + 1; 2606 2607 } 2608 return 0; 2609 } 2610 2611 static int 2612 init_tx_threads(void) 2613 { 2614 int i; 2615 2616 n_tx_thread = 0; 2617 for (i = 0; i < nb_tx_thread_params; ++i) { 2618 tx_thread[n_tx_thread].conf.thread_id = tx_thread_params[i].thread_id; 2619 tx_thread[n_tx_thread].conf.lcore_id = tx_thread_params[i].lcore_id; 2620 n_tx_thread++; 2621 } 2622 return 0; 2623 } 2624 2625 /* display usage */ 2626 static void 2627 print_usage(const char *prgname) 2628 { 2629 printf("%s [EAL options] -- -p PORTMASK -P" 2630 " [--rx (port,queue,lcore,thread)[,(port,queue,lcore,thread]]" 2631 " [--tx (lcore,thread)[,(lcore,thread]]" 2632 " [--enable-jumbo [--max-pkt-len PKTLEN]]\n" 2633 " [--parse-ptype]\n\n" 2634 " -p PORTMASK: hexadecimal bitmask of ports to configure\n" 2635 " -P : enable promiscuous mode\n" 2636 " --rx (port,queue,lcore,thread): rx queues configuration\n" 2637 " --tx (lcore,thread): tx threads configuration\n" 2638 " --stat-lcore LCORE: use lcore for stat collector\n" 2639 " --eth-dest=X,MM:MM:MM:MM:MM:MM: optional, ethernet destination for port X\n" 2640 " --no-numa: optional, disable numa awareness\n" 2641 " --ipv6: optional, specify it if running ipv6 packets\n" 2642 " --enable-jumbo: enable jumbo frame" 2643 " which max packet len is PKTLEN in decimal (64-9600)\n" 2644 " --hash-entry-num: specify the hash entry number in hexadecimal to be setup\n" 2645 " --no-lthreads: turn off lthread model\n" 2646 " --parse-ptype: set to use software to analyze packet type\n\n", 2647 prgname); 2648 } 2649 2650 static int parse_max_pkt_len(const char *pktlen) 2651 { 2652 char *end = NULL; 2653 unsigned long len; 2654 2655 /* parse decimal string */ 2656 len = strtoul(pktlen, &end, 10); 2657 if ((pktlen[0] == '\0') || (end == NULL) || (*end != '\0')) 2658 return -1; 2659 2660 if (len == 0) 2661 return -1; 2662 2663 return len; 2664 } 2665 2666 static int 2667 parse_portmask(const char *portmask) 2668 { 2669 char *end = NULL; 2670 unsigned long pm; 2671 2672 /* parse hexadecimal string */ 2673 pm = strtoul(portmask, &end, 16); 2674 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0')) 2675 return -1; 2676 2677 if (pm == 0) 2678 return -1; 2679 2680 return pm; 2681 } 2682 2683 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH) 2684 static int 2685 parse_hash_entry_number(const char *hash_entry_num) 2686 { 2687 char *end = NULL; 2688 unsigned long hash_en; 2689 2690 /* parse hexadecimal string */ 2691 hash_en = strtoul(hash_entry_num, &end, 16); 2692 if ((hash_entry_num[0] == '\0') || (end == NULL) || (*end != '\0')) 2693 return -1; 2694 2695 if (hash_en == 0) 2696 return -1; 2697 2698 return hash_en; 2699 } 2700 #endif 2701 2702 static int 2703 parse_rx_config(const char *q_arg) 2704 { 2705 char s[256]; 2706 const char *p, *p0 = q_arg; 2707 char *end; 2708 enum fieldnames { 2709 FLD_PORT = 0, 2710 FLD_QUEUE, 2711 FLD_LCORE, 2712 FLD_THREAD, 2713 _NUM_FLD 2714 }; 2715 unsigned long int_fld[_NUM_FLD]; 2716 char *str_fld[_NUM_FLD]; 2717 int i; 2718 unsigned size; 2719 2720 nb_rx_thread_params = 0; 2721 2722 while ((p = strchr(p0, '(')) != NULL) { 2723 ++p; 2724 p0 = strchr(p, ')'); 2725 if (p0 == NULL) 2726 return -1; 2727 2728 size = p0 - p; 2729 if (size >= sizeof(s)) 2730 return -1; 2731 2732 snprintf(s, sizeof(s), "%.*s", size, p); 2733 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') != _NUM_FLD) 2734 return -1; 2735 for (i = 0; i < _NUM_FLD; i++) { 2736 errno = 0; 2737 int_fld[i] = strtoul(str_fld[i], &end, 0); 2738 if (errno != 0 || end == str_fld[i] || int_fld[i] > 255) 2739 return -1; 2740 } 2741 if (nb_rx_thread_params >= MAX_LCORE_PARAMS) { 2742 printf("exceeded max number of rx params: %hu\n", 2743 nb_rx_thread_params); 2744 return -1; 2745 } 2746 rx_thread_params_array[nb_rx_thread_params].port_id = 2747 int_fld[FLD_PORT]; 2748 rx_thread_params_array[nb_rx_thread_params].queue_id = 2749 (uint8_t)int_fld[FLD_QUEUE]; 2750 rx_thread_params_array[nb_rx_thread_params].lcore_id = 2751 (uint8_t)int_fld[FLD_LCORE]; 2752 rx_thread_params_array[nb_rx_thread_params].thread_id = 2753 (uint8_t)int_fld[FLD_THREAD]; 2754 ++nb_rx_thread_params; 2755 } 2756 rx_thread_params = rx_thread_params_array; 2757 return 0; 2758 } 2759 2760 static int 2761 parse_tx_config(const char *q_arg) 2762 { 2763 char s[256]; 2764 const char *p, *p0 = q_arg; 2765 char *end; 2766 enum fieldnames { 2767 FLD_LCORE = 0, 2768 FLD_THREAD, 2769 _NUM_FLD 2770 }; 2771 unsigned long int_fld[_NUM_FLD]; 2772 char *str_fld[_NUM_FLD]; 2773 int i; 2774 unsigned size; 2775 2776 nb_tx_thread_params = 0; 2777 2778 while ((p = strchr(p0, '(')) != NULL) { 2779 ++p; 2780 p0 = strchr(p, ')'); 2781 if (p0 == NULL) 2782 return -1; 2783 2784 size = p0 - p; 2785 if (size >= sizeof(s)) 2786 return -1; 2787 2788 snprintf(s, sizeof(s), "%.*s", size, p); 2789 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') != _NUM_FLD) 2790 return -1; 2791 for (i = 0; i < _NUM_FLD; i++) { 2792 errno = 0; 2793 int_fld[i] = strtoul(str_fld[i], &end, 0); 2794 if (errno != 0 || end == str_fld[i] || int_fld[i] > 255) 2795 return -1; 2796 } 2797 if (nb_tx_thread_params >= MAX_LCORE_PARAMS) { 2798 printf("exceeded max number of tx params: %hu\n", 2799 nb_tx_thread_params); 2800 return -1; 2801 } 2802 tx_thread_params_array[nb_tx_thread_params].lcore_id = 2803 (uint8_t)int_fld[FLD_LCORE]; 2804 tx_thread_params_array[nb_tx_thread_params].thread_id = 2805 (uint8_t)int_fld[FLD_THREAD]; 2806 ++nb_tx_thread_params; 2807 } 2808 tx_thread_params = tx_thread_params_array; 2809 2810 return 0; 2811 } 2812 2813 #if (APP_CPU_LOAD > 0) 2814 static int 2815 parse_stat_lcore(const char *stat_lcore) 2816 { 2817 char *end = NULL; 2818 unsigned long lcore_id; 2819 2820 lcore_id = strtoul(stat_lcore, &end, 10); 2821 if ((stat_lcore[0] == '\0') || (end == NULL) || (*end != '\0')) 2822 return -1; 2823 2824 return lcore_id; 2825 } 2826 #endif 2827 2828 static void 2829 parse_eth_dest(const char *optarg) 2830 { 2831 uint16_t portid; 2832 char *port_end; 2833 uint8_t c, *dest, peer_addr[6]; 2834 2835 errno = 0; 2836 portid = strtoul(optarg, &port_end, 10); 2837 if (errno != 0 || port_end == optarg || *port_end++ != ',') 2838 rte_exit(EXIT_FAILURE, 2839 "Invalid eth-dest: %s", optarg); 2840 if (portid >= RTE_MAX_ETHPORTS) 2841 rte_exit(EXIT_FAILURE, 2842 "eth-dest: port %d >= RTE_MAX_ETHPORTS(%d)\n", 2843 portid, RTE_MAX_ETHPORTS); 2844 2845 if (cmdline_parse_etheraddr(NULL, port_end, 2846 &peer_addr, sizeof(peer_addr)) < 0) 2847 rte_exit(EXIT_FAILURE, 2848 "Invalid ethernet address: %s\n", 2849 port_end); 2850 dest = (uint8_t *)&dest_eth_addr[portid]; 2851 for (c = 0; c < 6; c++) 2852 dest[c] = peer_addr[c]; 2853 *(uint64_t *)(val_eth + portid) = dest_eth_addr[portid]; 2854 } 2855 2856 #define CMD_LINE_OPT_RX_CONFIG "rx" 2857 #define CMD_LINE_OPT_TX_CONFIG "tx" 2858 #define CMD_LINE_OPT_STAT_LCORE "stat-lcore" 2859 #define CMD_LINE_OPT_ETH_DEST "eth-dest" 2860 #define CMD_LINE_OPT_NO_NUMA "no-numa" 2861 #define CMD_LINE_OPT_IPV6 "ipv6" 2862 #define CMD_LINE_OPT_ENABLE_JUMBO "enable-jumbo" 2863 #define CMD_LINE_OPT_HASH_ENTRY_NUM "hash-entry-num" 2864 #define CMD_LINE_OPT_NO_LTHREADS "no-lthreads" 2865 #define CMD_LINE_OPT_PARSE_PTYPE "parse-ptype" 2866 2867 /* Parse the argument given in the command line of the application */ 2868 static int 2869 parse_args(int argc, char **argv) 2870 { 2871 int opt, ret; 2872 char **argvopt; 2873 int option_index; 2874 char *prgname = argv[0]; 2875 static struct option lgopts[] = { 2876 {CMD_LINE_OPT_RX_CONFIG, 1, 0, 0}, 2877 {CMD_LINE_OPT_TX_CONFIG, 1, 0, 0}, 2878 {CMD_LINE_OPT_STAT_LCORE, 1, 0, 0}, 2879 {CMD_LINE_OPT_ETH_DEST, 1, 0, 0}, 2880 {CMD_LINE_OPT_NO_NUMA, 0, 0, 0}, 2881 {CMD_LINE_OPT_IPV6, 0, 0, 0}, 2882 {CMD_LINE_OPT_ENABLE_JUMBO, 0, 0, 0}, 2883 {CMD_LINE_OPT_HASH_ENTRY_NUM, 1, 0, 0}, 2884 {CMD_LINE_OPT_NO_LTHREADS, 0, 0, 0}, 2885 {CMD_LINE_OPT_PARSE_PTYPE, 0, 0, 0}, 2886 {NULL, 0, 0, 0} 2887 }; 2888 2889 argvopt = argv; 2890 2891 while ((opt = getopt_long(argc, argvopt, "p:P", 2892 lgopts, &option_index)) != EOF) { 2893 2894 switch (opt) { 2895 /* portmask */ 2896 case 'p': 2897 enabled_port_mask = parse_portmask(optarg); 2898 if (enabled_port_mask == 0) { 2899 printf("invalid portmask\n"); 2900 print_usage(prgname); 2901 return -1; 2902 } 2903 break; 2904 case 'P': 2905 printf("Promiscuous mode selected\n"); 2906 promiscuous_on = 1; 2907 break; 2908 2909 /* long options */ 2910 case 0: 2911 if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_RX_CONFIG, 2912 sizeof(CMD_LINE_OPT_RX_CONFIG))) { 2913 ret = parse_rx_config(optarg); 2914 if (ret) { 2915 printf("invalid rx-config\n"); 2916 print_usage(prgname); 2917 return -1; 2918 } 2919 } 2920 2921 if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_TX_CONFIG, 2922 sizeof(CMD_LINE_OPT_TX_CONFIG))) { 2923 ret = parse_tx_config(optarg); 2924 if (ret) { 2925 printf("invalid tx-config\n"); 2926 print_usage(prgname); 2927 return -1; 2928 } 2929 } 2930 2931 #if (APP_CPU_LOAD > 0) 2932 if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_STAT_LCORE, 2933 sizeof(CMD_LINE_OPT_STAT_LCORE))) { 2934 cpu_load_lcore_id = parse_stat_lcore(optarg); 2935 } 2936 #endif 2937 2938 if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_ETH_DEST, 2939 sizeof(CMD_LINE_OPT_ETH_DEST))) 2940 parse_eth_dest(optarg); 2941 2942 if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_NO_NUMA, 2943 sizeof(CMD_LINE_OPT_NO_NUMA))) { 2944 printf("numa is disabled\n"); 2945 numa_on = 0; 2946 } 2947 2948 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH) 2949 if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_IPV6, 2950 sizeof(CMD_LINE_OPT_IPV6))) { 2951 printf("ipv6 is specified\n"); 2952 ipv6 = 1; 2953 } 2954 #endif 2955 2956 if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_NO_LTHREADS, 2957 sizeof(CMD_LINE_OPT_NO_LTHREADS))) { 2958 printf("l-threads model is disabled\n"); 2959 lthreads_on = 0; 2960 } 2961 2962 if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_PARSE_PTYPE, 2963 sizeof(CMD_LINE_OPT_PARSE_PTYPE))) { 2964 printf("software packet type parsing enabled\n"); 2965 parse_ptype_on = 1; 2966 } 2967 2968 if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_ENABLE_JUMBO, 2969 sizeof(CMD_LINE_OPT_ENABLE_JUMBO))) { 2970 struct option lenopts = {"max-pkt-len", required_argument, 0, 2971 0}; 2972 2973 printf("jumbo frame is enabled - disabling simple TX path\n"); 2974 port_conf.rxmode.offloads |= 2975 DEV_RX_OFFLOAD_JUMBO_FRAME; 2976 port_conf.txmode.offloads |= 2977 DEV_TX_OFFLOAD_MULTI_SEGS; 2978 2979 /* if no max-pkt-len set, use the default value ETHER_MAX_LEN */ 2980 if (0 == getopt_long(argc, argvopt, "", &lenopts, 2981 &option_index)) { 2982 2983 ret = parse_max_pkt_len(optarg); 2984 if ((ret < 64) || (ret > MAX_JUMBO_PKT_LEN)) { 2985 printf("invalid packet length\n"); 2986 print_usage(prgname); 2987 return -1; 2988 } 2989 port_conf.rxmode.max_rx_pkt_len = ret; 2990 } 2991 printf("set jumbo frame max packet length to %u\n", 2992 (unsigned int)port_conf.rxmode.max_rx_pkt_len); 2993 } 2994 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH) 2995 if (!strncmp(lgopts[option_index].name, CMD_LINE_OPT_HASH_ENTRY_NUM, 2996 sizeof(CMD_LINE_OPT_HASH_ENTRY_NUM))) { 2997 ret = parse_hash_entry_number(optarg); 2998 if ((ret > 0) && (ret <= L3FWD_HASH_ENTRIES)) { 2999 hash_entry_number = ret; 3000 } else { 3001 printf("invalid hash entry number\n"); 3002 print_usage(prgname); 3003 return -1; 3004 } 3005 } 3006 #endif 3007 break; 3008 3009 default: 3010 print_usage(prgname); 3011 return -1; 3012 } 3013 } 3014 3015 if (optind >= 0) 3016 argv[optind-1] = prgname; 3017 3018 ret = optind-1; 3019 optind = 1; /* reset getopt lib */ 3020 return ret; 3021 } 3022 3023 static void 3024 print_ethaddr(const char *name, const struct ether_addr *eth_addr) 3025 { 3026 char buf[ETHER_ADDR_FMT_SIZE]; 3027 3028 ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr); 3029 printf("%s%s", name, buf); 3030 } 3031 3032 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH) 3033 3034 static void convert_ipv4_5tuple(struct ipv4_5tuple *key1, 3035 union ipv4_5tuple_host *key2) 3036 { 3037 key2->ip_dst = rte_cpu_to_be_32(key1->ip_dst); 3038 key2->ip_src = rte_cpu_to_be_32(key1->ip_src); 3039 key2->port_dst = rte_cpu_to_be_16(key1->port_dst); 3040 key2->port_src = rte_cpu_to_be_16(key1->port_src); 3041 key2->proto = key1->proto; 3042 key2->pad0 = 0; 3043 key2->pad1 = 0; 3044 } 3045 3046 static void convert_ipv6_5tuple(struct ipv6_5tuple *key1, 3047 union ipv6_5tuple_host *key2) 3048 { 3049 uint32_t i; 3050 3051 for (i = 0; i < 16; i++) { 3052 key2->ip_dst[i] = key1->ip_dst[i]; 3053 key2->ip_src[i] = key1->ip_src[i]; 3054 } 3055 key2->port_dst = rte_cpu_to_be_16(key1->port_dst); 3056 key2->port_src = rte_cpu_to_be_16(key1->port_src); 3057 key2->proto = key1->proto; 3058 key2->pad0 = 0; 3059 key2->pad1 = 0; 3060 key2->reserve = 0; 3061 } 3062 3063 #define BYTE_VALUE_MAX 256 3064 #define ALL_32_BITS 0xffffffff 3065 #define BIT_8_TO_15 0x0000ff00 3066 static inline void 3067 populate_ipv4_few_flow_into_table(const struct rte_hash *h) 3068 { 3069 uint32_t i; 3070 int32_t ret; 3071 uint32_t array_len = RTE_DIM(ipv4_l3fwd_route_array); 3072 3073 mask0 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_8_TO_15); 3074 for (i = 0; i < array_len; i++) { 3075 struct ipv4_l3fwd_route entry; 3076 union ipv4_5tuple_host newkey; 3077 3078 entry = ipv4_l3fwd_route_array[i]; 3079 convert_ipv4_5tuple(&entry.key, &newkey); 3080 ret = rte_hash_add_key(h, (void *)&newkey); 3081 if (ret < 0) { 3082 rte_exit(EXIT_FAILURE, "Unable to add entry %" PRIu32 3083 " to the l3fwd hash.\n", i); 3084 } 3085 ipv4_l3fwd_out_if[ret] = entry.if_out; 3086 } 3087 printf("Hash: Adding 0x%" PRIx32 " keys\n", array_len); 3088 } 3089 3090 #define BIT_16_TO_23 0x00ff0000 3091 static inline void 3092 populate_ipv6_few_flow_into_table(const struct rte_hash *h) 3093 { 3094 uint32_t i; 3095 int32_t ret; 3096 uint32_t array_len = RTE_DIM(ipv6_l3fwd_route_array); 3097 3098 mask1 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_16_TO_23); 3099 mask2 = _mm_set_epi32(0, 0, ALL_32_BITS, ALL_32_BITS); 3100 for (i = 0; i < array_len; i++) { 3101 struct ipv6_l3fwd_route entry; 3102 union ipv6_5tuple_host newkey; 3103 3104 entry = ipv6_l3fwd_route_array[i]; 3105 convert_ipv6_5tuple(&entry.key, &newkey); 3106 ret = rte_hash_add_key(h, (void *)&newkey); 3107 if (ret < 0) { 3108 rte_exit(EXIT_FAILURE, "Unable to add entry %" PRIu32 3109 " to the l3fwd hash.\n", i); 3110 } 3111 ipv6_l3fwd_out_if[ret] = entry.if_out; 3112 } 3113 printf("Hash: Adding 0x%" PRIx32 "keys\n", array_len); 3114 } 3115 3116 #define NUMBER_PORT_USED 4 3117 static inline void 3118 populate_ipv4_many_flow_into_table(const struct rte_hash *h, 3119 unsigned int nr_flow) 3120 { 3121 unsigned i; 3122 3123 mask0 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_8_TO_15); 3124 3125 for (i = 0; i < nr_flow; i++) { 3126 struct ipv4_l3fwd_route entry; 3127 union ipv4_5tuple_host newkey; 3128 uint8_t a = (uint8_t)((i / NUMBER_PORT_USED) % BYTE_VALUE_MAX); 3129 uint8_t b = (uint8_t)(((i / NUMBER_PORT_USED) / BYTE_VALUE_MAX) % 3130 BYTE_VALUE_MAX); 3131 uint8_t c = (uint8_t)((i / NUMBER_PORT_USED) / (BYTE_VALUE_MAX * 3132 BYTE_VALUE_MAX)); 3133 /* Create the ipv4 exact match flow */ 3134 memset(&entry, 0, sizeof(entry)); 3135 switch (i & (NUMBER_PORT_USED - 1)) { 3136 case 0: 3137 entry = ipv4_l3fwd_route_array[0]; 3138 entry.key.ip_dst = IPv4(101, c, b, a); 3139 break; 3140 case 1: 3141 entry = ipv4_l3fwd_route_array[1]; 3142 entry.key.ip_dst = IPv4(201, c, b, a); 3143 break; 3144 case 2: 3145 entry = ipv4_l3fwd_route_array[2]; 3146 entry.key.ip_dst = IPv4(111, c, b, a); 3147 break; 3148 case 3: 3149 entry = ipv4_l3fwd_route_array[3]; 3150 entry.key.ip_dst = IPv4(211, c, b, a); 3151 break; 3152 }; 3153 convert_ipv4_5tuple(&entry.key, &newkey); 3154 int32_t ret = rte_hash_add_key(h, (void *)&newkey); 3155 3156 if (ret < 0) 3157 rte_exit(EXIT_FAILURE, "Unable to add entry %u\n", i); 3158 3159 ipv4_l3fwd_out_if[ret] = (uint8_t)entry.if_out; 3160 3161 } 3162 printf("Hash: Adding 0x%x keys\n", nr_flow); 3163 } 3164 3165 static inline void 3166 populate_ipv6_many_flow_into_table(const struct rte_hash *h, 3167 unsigned int nr_flow) 3168 { 3169 unsigned i; 3170 3171 mask1 = _mm_set_epi32(ALL_32_BITS, ALL_32_BITS, ALL_32_BITS, BIT_16_TO_23); 3172 mask2 = _mm_set_epi32(0, 0, ALL_32_BITS, ALL_32_BITS); 3173 for (i = 0; i < nr_flow; i++) { 3174 struct ipv6_l3fwd_route entry; 3175 union ipv6_5tuple_host newkey; 3176 3177 uint8_t a = (uint8_t) ((i / NUMBER_PORT_USED) % BYTE_VALUE_MAX); 3178 uint8_t b = (uint8_t) (((i / NUMBER_PORT_USED) / BYTE_VALUE_MAX) % 3179 BYTE_VALUE_MAX); 3180 uint8_t c = (uint8_t) ((i / NUMBER_PORT_USED) / (BYTE_VALUE_MAX * 3181 BYTE_VALUE_MAX)); 3182 3183 /* Create the ipv6 exact match flow */ 3184 memset(&entry, 0, sizeof(entry)); 3185 switch (i & (NUMBER_PORT_USED - 1)) { 3186 case 0: 3187 entry = ipv6_l3fwd_route_array[0]; 3188 break; 3189 case 1: 3190 entry = ipv6_l3fwd_route_array[1]; 3191 break; 3192 case 2: 3193 entry = ipv6_l3fwd_route_array[2]; 3194 break; 3195 case 3: 3196 entry = ipv6_l3fwd_route_array[3]; 3197 break; 3198 }; 3199 entry.key.ip_dst[13] = c; 3200 entry.key.ip_dst[14] = b; 3201 entry.key.ip_dst[15] = a; 3202 convert_ipv6_5tuple(&entry.key, &newkey); 3203 int32_t ret = rte_hash_add_key(h, (void *)&newkey); 3204 3205 if (ret < 0) 3206 rte_exit(EXIT_FAILURE, "Unable to add entry %u\n", i); 3207 3208 ipv6_l3fwd_out_if[ret] = (uint8_t) entry.if_out; 3209 3210 } 3211 printf("Hash: Adding 0x%x keys\n", nr_flow); 3212 } 3213 3214 static void 3215 setup_hash(int socketid) 3216 { 3217 struct rte_hash_parameters ipv4_l3fwd_hash_params = { 3218 .name = NULL, 3219 .entries = L3FWD_HASH_ENTRIES, 3220 .key_len = sizeof(union ipv4_5tuple_host), 3221 .hash_func = ipv4_hash_crc, 3222 .hash_func_init_val = 0, 3223 }; 3224 3225 struct rte_hash_parameters ipv6_l3fwd_hash_params = { 3226 .name = NULL, 3227 .entries = L3FWD_HASH_ENTRIES, 3228 .key_len = sizeof(union ipv6_5tuple_host), 3229 .hash_func = ipv6_hash_crc, 3230 .hash_func_init_val = 0, 3231 }; 3232 3233 char s[64]; 3234 3235 /* create ipv4 hash */ 3236 snprintf(s, sizeof(s), "ipv4_l3fwd_hash_%d", socketid); 3237 ipv4_l3fwd_hash_params.name = s; 3238 ipv4_l3fwd_hash_params.socket_id = socketid; 3239 ipv4_l3fwd_lookup_struct[socketid] = 3240 rte_hash_create(&ipv4_l3fwd_hash_params); 3241 if (ipv4_l3fwd_lookup_struct[socketid] == NULL) 3242 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on " 3243 "socket %d\n", socketid); 3244 3245 /* create ipv6 hash */ 3246 snprintf(s, sizeof(s), "ipv6_l3fwd_hash_%d", socketid); 3247 ipv6_l3fwd_hash_params.name = s; 3248 ipv6_l3fwd_hash_params.socket_id = socketid; 3249 ipv6_l3fwd_lookup_struct[socketid] = 3250 rte_hash_create(&ipv6_l3fwd_hash_params); 3251 if (ipv6_l3fwd_lookup_struct[socketid] == NULL) 3252 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on " 3253 "socket %d\n", socketid); 3254 3255 if (hash_entry_number != HASH_ENTRY_NUMBER_DEFAULT) { 3256 /* For testing hash matching with a large number of flows we 3257 * generate millions of IP 5-tuples with an incremented dst 3258 * address to initialize the hash table. */ 3259 if (ipv6 == 0) { 3260 /* populate the ipv4 hash */ 3261 populate_ipv4_many_flow_into_table( 3262 ipv4_l3fwd_lookup_struct[socketid], hash_entry_number); 3263 } else { 3264 /* populate the ipv6 hash */ 3265 populate_ipv6_many_flow_into_table( 3266 ipv6_l3fwd_lookup_struct[socketid], hash_entry_number); 3267 } 3268 } else { 3269 /* Use data in ipv4/ipv6 l3fwd lookup table directly to initialize 3270 * the hash table */ 3271 if (ipv6 == 0) { 3272 /* populate the ipv4 hash */ 3273 populate_ipv4_few_flow_into_table( 3274 ipv4_l3fwd_lookup_struct[socketid]); 3275 } else { 3276 /* populate the ipv6 hash */ 3277 populate_ipv6_few_flow_into_table( 3278 ipv6_l3fwd_lookup_struct[socketid]); 3279 } 3280 } 3281 } 3282 #endif 3283 3284 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM) 3285 static void 3286 setup_lpm(int socketid) 3287 { 3288 struct rte_lpm6_config config; 3289 struct rte_lpm_config lpm_ipv4_config; 3290 unsigned i; 3291 int ret; 3292 char s[64]; 3293 3294 /* create the LPM table */ 3295 snprintf(s, sizeof(s), "IPV4_L3FWD_LPM_%d", socketid); 3296 lpm_ipv4_config.max_rules = IPV4_L3FWD_LPM_MAX_RULES; 3297 lpm_ipv4_config.number_tbl8s = 256; 3298 lpm_ipv4_config.flags = 0; 3299 ipv4_l3fwd_lookup_struct[socketid] = 3300 rte_lpm_create(s, socketid, &lpm_ipv4_config); 3301 if (ipv4_l3fwd_lookup_struct[socketid] == NULL) 3302 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table" 3303 " on socket %d\n", socketid); 3304 3305 /* populate the LPM table */ 3306 for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) { 3307 3308 /* skip unused ports */ 3309 if ((1 << ipv4_l3fwd_route_array[i].if_out & 3310 enabled_port_mask) == 0) 3311 continue; 3312 3313 ret = rte_lpm_add(ipv4_l3fwd_lookup_struct[socketid], 3314 ipv4_l3fwd_route_array[i].ip, 3315 ipv4_l3fwd_route_array[i].depth, 3316 ipv4_l3fwd_route_array[i].if_out); 3317 3318 if (ret < 0) { 3319 rte_exit(EXIT_FAILURE, "Unable to add entry %u to the " 3320 "l3fwd LPM table on socket %d\n", 3321 i, socketid); 3322 } 3323 3324 printf("LPM: Adding route 0x%08x / %d (%d)\n", 3325 (unsigned)ipv4_l3fwd_route_array[i].ip, 3326 ipv4_l3fwd_route_array[i].depth, 3327 ipv4_l3fwd_route_array[i].if_out); 3328 } 3329 3330 /* create the LPM6 table */ 3331 snprintf(s, sizeof(s), "IPV6_L3FWD_LPM_%d", socketid); 3332 3333 config.max_rules = IPV6_L3FWD_LPM_MAX_RULES; 3334 config.number_tbl8s = IPV6_L3FWD_LPM_NUMBER_TBL8S; 3335 config.flags = 0; 3336 ipv6_l3fwd_lookup_struct[socketid] = rte_lpm6_create(s, socketid, 3337 &config); 3338 if (ipv6_l3fwd_lookup_struct[socketid] == NULL) 3339 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table" 3340 " on socket %d\n", socketid); 3341 3342 /* populate the LPM table */ 3343 for (i = 0; i < IPV6_L3FWD_NUM_ROUTES; i++) { 3344 3345 /* skip unused ports */ 3346 if ((1 << ipv6_l3fwd_route_array[i].if_out & 3347 enabled_port_mask) == 0) 3348 continue; 3349 3350 ret = rte_lpm6_add(ipv6_l3fwd_lookup_struct[socketid], 3351 ipv6_l3fwd_route_array[i].ip, 3352 ipv6_l3fwd_route_array[i].depth, 3353 ipv6_l3fwd_route_array[i].if_out); 3354 3355 if (ret < 0) { 3356 rte_exit(EXIT_FAILURE, "Unable to add entry %u to the " 3357 "l3fwd LPM table on socket %d\n", 3358 i, socketid); 3359 } 3360 3361 printf("LPM: Adding route %s / %d (%d)\n", 3362 "IPV6", 3363 ipv6_l3fwd_route_array[i].depth, 3364 ipv6_l3fwd_route_array[i].if_out); 3365 } 3366 } 3367 #endif 3368 3369 static int 3370 init_mem(unsigned nb_mbuf) 3371 { 3372 struct lcore_conf *qconf; 3373 int socketid; 3374 unsigned lcore_id; 3375 char s[64]; 3376 3377 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) { 3378 if (rte_lcore_is_enabled(lcore_id) == 0) 3379 continue; 3380 3381 if (numa_on) 3382 socketid = rte_lcore_to_socket_id(lcore_id); 3383 else 3384 socketid = 0; 3385 3386 if (socketid >= NB_SOCKETS) { 3387 rte_exit(EXIT_FAILURE, "Socket %d of lcore %u is out of range %d\n", 3388 socketid, lcore_id, NB_SOCKETS); 3389 } 3390 if (pktmbuf_pool[socketid] == NULL) { 3391 snprintf(s, sizeof(s), "mbuf_pool_%d", socketid); 3392 pktmbuf_pool[socketid] = 3393 rte_pktmbuf_pool_create(s, nb_mbuf, 3394 MEMPOOL_CACHE_SIZE, 0, 3395 RTE_MBUF_DEFAULT_BUF_SIZE, socketid); 3396 if (pktmbuf_pool[socketid] == NULL) 3397 rte_exit(EXIT_FAILURE, 3398 "Cannot init mbuf pool on socket %d\n", socketid); 3399 else 3400 printf("Allocated mbuf pool on socket %d\n", socketid); 3401 3402 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM) 3403 setup_lpm(socketid); 3404 #else 3405 setup_hash(socketid); 3406 #endif 3407 } 3408 qconf = &lcore_conf[lcore_id]; 3409 qconf->ipv4_lookup_struct = ipv4_l3fwd_lookup_struct[socketid]; 3410 qconf->ipv6_lookup_struct = ipv6_l3fwd_lookup_struct[socketid]; 3411 } 3412 return 0; 3413 } 3414 3415 /* Check the link status of all ports in up to 9s, and print them finally */ 3416 static void 3417 check_all_ports_link_status(uint32_t port_mask) 3418 { 3419 #define CHECK_INTERVAL 100 /* 100ms */ 3420 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */ 3421 uint16_t portid; 3422 uint8_t count, all_ports_up, print_flag = 0; 3423 struct rte_eth_link link; 3424 3425 printf("\nChecking link status"); 3426 fflush(stdout); 3427 for (count = 0; count <= MAX_CHECK_TIME; count++) { 3428 all_ports_up = 1; 3429 RTE_ETH_FOREACH_DEV(portid) { 3430 if ((port_mask & (1 << portid)) == 0) 3431 continue; 3432 memset(&link, 0, sizeof(link)); 3433 rte_eth_link_get_nowait(portid, &link); 3434 /* print link status if flag set */ 3435 if (print_flag == 1) { 3436 if (link.link_status) 3437 printf( 3438 "Port%d Link Up. Speed %u Mbps - %s\n", 3439 portid, link.link_speed, 3440 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ? 3441 ("full-duplex") : ("half-duplex\n")); 3442 else 3443 printf("Port %d Link Down\n", portid); 3444 continue; 3445 } 3446 /* clear all_ports_up flag if any link down */ 3447 if (link.link_status == ETH_LINK_DOWN) { 3448 all_ports_up = 0; 3449 break; 3450 } 3451 } 3452 /* after finally printing all link status, get out */ 3453 if (print_flag == 1) 3454 break; 3455 3456 if (all_ports_up == 0) { 3457 printf("."); 3458 fflush(stdout); 3459 rte_delay_ms(CHECK_INTERVAL); 3460 } 3461 3462 /* set the print_flag if all ports up or timeout */ 3463 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) { 3464 print_flag = 1; 3465 printf("done\n"); 3466 } 3467 } 3468 } 3469 3470 int 3471 main(int argc, char **argv) 3472 { 3473 struct rte_eth_dev_info dev_info; 3474 struct rte_eth_txconf *txconf; 3475 int ret; 3476 int i; 3477 unsigned nb_ports; 3478 uint16_t queueid, portid; 3479 unsigned lcore_id; 3480 uint32_t n_tx_queue, nb_lcores; 3481 uint8_t nb_rx_queue, queue, socketid; 3482 3483 /* init EAL */ 3484 ret = rte_eal_init(argc, argv); 3485 if (ret < 0) 3486 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n"); 3487 argc -= ret; 3488 argv += ret; 3489 3490 rte_timer_subsystem_init(); 3491 3492 /* pre-init dst MACs for all ports to 02:00:00:00:00:xx */ 3493 for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) { 3494 dest_eth_addr[portid] = ETHER_LOCAL_ADMIN_ADDR + 3495 ((uint64_t)portid << 40); 3496 *(uint64_t *)(val_eth + portid) = dest_eth_addr[portid]; 3497 } 3498 3499 /* parse application arguments (after the EAL ones) */ 3500 ret = parse_args(argc, argv); 3501 if (ret < 0) 3502 rte_exit(EXIT_FAILURE, "Invalid L3FWD parameters\n"); 3503 3504 if (check_lcore_params() < 0) 3505 rte_exit(EXIT_FAILURE, "check_lcore_params failed\n"); 3506 3507 printf("Initializing rx-queues...\n"); 3508 ret = init_rx_queues(); 3509 if (ret < 0) 3510 rte_exit(EXIT_FAILURE, "init_rx_queues failed\n"); 3511 3512 printf("Initializing tx-threads...\n"); 3513 ret = init_tx_threads(); 3514 if (ret < 0) 3515 rte_exit(EXIT_FAILURE, "init_tx_threads failed\n"); 3516 3517 printf("Initializing rings...\n"); 3518 ret = init_rx_rings(); 3519 if (ret < 0) 3520 rte_exit(EXIT_FAILURE, "init_rx_rings failed\n"); 3521 3522 nb_ports = rte_eth_dev_count_avail(); 3523 3524 if (check_port_config() < 0) 3525 rte_exit(EXIT_FAILURE, "check_port_config failed\n"); 3526 3527 nb_lcores = rte_lcore_count(); 3528 3529 /* initialize all ports */ 3530 RTE_ETH_FOREACH_DEV(portid) { 3531 struct rte_eth_conf local_port_conf = port_conf; 3532 3533 /* skip ports that are not enabled */ 3534 if ((enabled_port_mask & (1 << portid)) == 0) { 3535 printf("\nSkipping disabled port %d\n", portid); 3536 continue; 3537 } 3538 3539 /* init port */ 3540 printf("Initializing port %d ... ", portid); 3541 fflush(stdout); 3542 3543 nb_rx_queue = get_port_n_rx_queues(portid); 3544 n_tx_queue = nb_lcores; 3545 if (n_tx_queue > MAX_TX_QUEUE_PER_PORT) 3546 n_tx_queue = MAX_TX_QUEUE_PER_PORT; 3547 printf("Creating queues: nb_rxq=%d nb_txq=%u... ", 3548 nb_rx_queue, (unsigned)n_tx_queue); 3549 rte_eth_dev_info_get(portid, &dev_info); 3550 if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE) 3551 local_port_conf.txmode.offloads |= 3552 DEV_TX_OFFLOAD_MBUF_FAST_FREE; 3553 3554 local_port_conf.rx_adv_conf.rss_conf.rss_hf &= 3555 dev_info.flow_type_rss_offloads; 3556 if (local_port_conf.rx_adv_conf.rss_conf.rss_hf != 3557 port_conf.rx_adv_conf.rss_conf.rss_hf) { 3558 printf("Port %u modified RSS hash function based on hardware support," 3559 "requested:%#"PRIx64" configured:%#"PRIx64"\n", 3560 portid, 3561 port_conf.rx_adv_conf.rss_conf.rss_hf, 3562 local_port_conf.rx_adv_conf.rss_conf.rss_hf); 3563 } 3564 3565 ret = rte_eth_dev_configure(portid, nb_rx_queue, 3566 (uint16_t)n_tx_queue, &local_port_conf); 3567 if (ret < 0) 3568 rte_exit(EXIT_FAILURE, "Cannot configure device: err=%d, port=%d\n", 3569 ret, portid); 3570 3571 ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd, 3572 &nb_txd); 3573 if (ret < 0) 3574 rte_exit(EXIT_FAILURE, 3575 "rte_eth_dev_adjust_nb_rx_tx_desc: err=%d, port=%d\n", 3576 ret, portid); 3577 3578 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]); 3579 print_ethaddr(" Address:", &ports_eth_addr[portid]); 3580 printf(", "); 3581 print_ethaddr("Destination:", 3582 (const struct ether_addr *)&dest_eth_addr[portid]); 3583 printf(", "); 3584 3585 /* 3586 * prepare src MACs for each port. 3587 */ 3588 ether_addr_copy(&ports_eth_addr[portid], 3589 (struct ether_addr *)(val_eth + portid) + 1); 3590 3591 /* init memory */ 3592 ret = init_mem(NB_MBUF); 3593 if (ret < 0) 3594 rte_exit(EXIT_FAILURE, "init_mem failed\n"); 3595 3596 /* init one TX queue per couple (lcore,port) */ 3597 queueid = 0; 3598 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) { 3599 if (rte_lcore_is_enabled(lcore_id) == 0) 3600 continue; 3601 3602 if (numa_on) 3603 socketid = (uint8_t)rte_lcore_to_socket_id(lcore_id); 3604 else 3605 socketid = 0; 3606 3607 printf("txq=%u,%d,%d ", lcore_id, queueid, socketid); 3608 fflush(stdout); 3609 3610 txconf = &dev_info.default_txconf; 3611 txconf->offloads = local_port_conf.txmode.offloads; 3612 ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd, 3613 socketid, txconf); 3614 if (ret < 0) 3615 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: err=%d, " 3616 "port=%d\n", ret, portid); 3617 3618 tx_thread[lcore_id].tx_queue_id[portid] = queueid; 3619 queueid++; 3620 } 3621 printf("\n"); 3622 } 3623 3624 for (i = 0; i < n_rx_thread; i++) { 3625 lcore_id = rx_thread[i].conf.lcore_id; 3626 3627 if (rte_lcore_is_enabled(lcore_id) == 0) { 3628 rte_exit(EXIT_FAILURE, 3629 "Cannot start Rx thread on lcore %u: lcore disabled\n", 3630 lcore_id 3631 ); 3632 } 3633 3634 printf("\nInitializing rx queues for Rx thread %d on lcore %u ... ", 3635 i, lcore_id); 3636 fflush(stdout); 3637 3638 /* init RX queues */ 3639 for (queue = 0; queue < rx_thread[i].n_rx_queue; ++queue) { 3640 struct rte_eth_rxconf rxq_conf; 3641 3642 portid = rx_thread[i].rx_queue_list[queue].port_id; 3643 queueid = rx_thread[i].rx_queue_list[queue].queue_id; 3644 3645 if (numa_on) 3646 socketid = (uint8_t)rte_lcore_to_socket_id(lcore_id); 3647 else 3648 socketid = 0; 3649 3650 printf("rxq=%d,%d,%d ", portid, queueid, socketid); 3651 fflush(stdout); 3652 3653 rte_eth_dev_info_get(portid, &dev_info); 3654 rxq_conf = dev_info.default_rxconf; 3655 rxq_conf.offloads = port_conf.rxmode.offloads; 3656 ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd, 3657 socketid, 3658 &rxq_conf, 3659 pktmbuf_pool[socketid]); 3660 if (ret < 0) 3661 rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: err=%d, " 3662 "port=%d\n", ret, portid); 3663 } 3664 } 3665 3666 printf("\n"); 3667 3668 /* start ports */ 3669 RTE_ETH_FOREACH_DEV(portid) { 3670 if ((enabled_port_mask & (1 << portid)) == 0) 3671 continue; 3672 3673 /* Start device */ 3674 ret = rte_eth_dev_start(portid); 3675 if (ret < 0) 3676 rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%d\n", 3677 ret, portid); 3678 3679 /* 3680 * If enabled, put device in promiscuous mode. 3681 * This allows IO forwarding mode to forward packets 3682 * to itself through 2 cross-connected ports of the 3683 * target machine. 3684 */ 3685 if (promiscuous_on) 3686 rte_eth_promiscuous_enable(portid); 3687 } 3688 3689 for (i = 0; i < n_rx_thread; i++) { 3690 lcore_id = rx_thread[i].conf.lcore_id; 3691 if (rte_lcore_is_enabled(lcore_id) == 0) 3692 continue; 3693 3694 /* check if hw packet type is supported */ 3695 for (queue = 0; queue < rx_thread[i].n_rx_queue; ++queue) { 3696 portid = rx_thread[i].rx_queue_list[queue].port_id; 3697 queueid = rx_thread[i].rx_queue_list[queue].queue_id; 3698 3699 if (parse_ptype_on) { 3700 if (!rte_eth_add_rx_callback(portid, queueid, 3701 cb_parse_ptype, NULL)) 3702 rte_exit(EXIT_FAILURE, 3703 "Failed to add rx callback: " 3704 "port=%d\n", portid); 3705 } else if (!check_ptype(portid)) 3706 rte_exit(EXIT_FAILURE, 3707 "Port %d cannot parse packet type.\n\n" 3708 "Please add --parse-ptype to use sw " 3709 "packet type analyzer.\n\n", 3710 portid); 3711 } 3712 } 3713 3714 check_all_ports_link_status(enabled_port_mask); 3715 3716 if (lthreads_on) { 3717 printf("Starting L-Threading Model\n"); 3718 3719 #if (APP_CPU_LOAD > 0) 3720 if (cpu_load_lcore_id > 0) 3721 /* Use one lcore for cpu load collector */ 3722 nb_lcores--; 3723 #endif 3724 3725 lthread_num_schedulers_set(nb_lcores); 3726 rte_eal_mp_remote_launch(sched_spawner, NULL, SKIP_MASTER); 3727 lthread_master_spawner(NULL); 3728 3729 } else { 3730 printf("Starting P-Threading Model\n"); 3731 /* launch per-lcore init on every lcore */ 3732 rte_eal_mp_remote_launch(pthread_run, NULL, CALL_MASTER); 3733 RTE_LCORE_FOREACH_SLAVE(lcore_id) { 3734 if (rte_eal_wait_lcore(lcore_id) < 0) 3735 return -1; 3736 } 3737 } 3738 3739 return 0; 3740 } 3741