examples/l3fwd-power/main.c

   1 /*-
   2  *   BSD LICENSE
   3  *
   4  *   Copyright(c) 2010-2016 Intel Corporation. All rights reserved.
   5  *   All rights reserved.
   6  *
   7  *   Redistribution and use in source and binary forms, with or without
   8  *   modification, are permitted provided that the following conditions
   9  *   are met:
  10  *
  11  *     * Redistributions of source code must retain the above copyright
  12  *       notice, this list of conditions and the following disclaimer.
  13  *     * Redistributions in binary form must reproduce the above copyright
  14  *       notice, this list of conditions and the following disclaimer in
  15  *       the documentation and/or other materials provided with the
  16  *       distribution.
  17  *     * Neither the name of Intel Corporation nor the names of its
  18  *       contributors may be used to endorse or promote products derived
  19  *       from this software without specific prior written permission.
  20  *
  21  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  22  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  23  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
  24  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  25  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  26  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
  27  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  28  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  29  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  30  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
  31  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  32  */
  33
  34 #include <stdio.h>
  35 #include <stdlib.h>
  36 #include <stdint.h>
  37 #include <inttypes.h>
  38 #include <sys/types.h>
  39 #include <string.h>
  40 #include <sys/queue.h>
  41 #include <stdarg.h>
  42 #include <errno.h>
  43 #include <getopt.h>
  44 #include <unistd.h>
  45 #include <signal.h>
  46
  47 #include <rte_common.h>
  48 #include <rte_byteorder.h>
  49 #include <rte_log.h>
  50 #include <rte_malloc.h>
  51 #include <rte_memory.h>
  52 #include <rte_memcpy.h>
  53 #include <rte_memzone.h>
  54 #include <rte_eal.h>
  55 #include <rte_per_lcore.h>
  56 #include <rte_launch.h>
  57 #include <rte_atomic.h>
  58 #include <rte_cycles.h>
  59 #include <rte_prefetch.h>
  60 #include <rte_lcore.h>
  61 #include <rte_per_lcore.h>
  62 #include <rte_branch_prediction.h>
  63 #include <rte_interrupts.h>
  64 #include <rte_pci.h>
  65 #include <rte_random.h>
  66 #include <rte_debug.h>
  67 #include <rte_ether.h>
  68 #include <rte_ethdev.h>
  69 #include <rte_ring.h>
  70 #include <rte_mempool.h>
  71 #include <rte_mbuf.h>
  72 #include <rte_ip.h>
  73 #include <rte_tcp.h>
  74 #include <rte_udp.h>
  75 #include <rte_string_fns.h>
  76 #include <rte_timer.h>
  77 #include <rte_power.h>
  78 #include <rte_eal.h>
  79 #include <rte_spinlock.h>
  80
  81 #define RTE_LOGTYPE_L3FWD_POWER RTE_LOGTYPE_USER1
  82
  83 #define MAX_PKT_BURST 32
  84
  85 #define MIN_ZERO_POLL_COUNT 10
  86
  87 /* around 100ms at 2 Ghz */
  88 #define TIMER_RESOLUTION_CYCLES           200000000ULL
  89 /* 100 ms interval */
  90 #define TIMER_NUMBER_PER_SECOND           10
  91 /* 100000 us */
  92 #define SCALING_PERIOD                    (1000000/TIMER_NUMBER_PER_SECOND)
  93 #define SCALING_DOWN_TIME_RATIO_THRESHOLD 0.25
  94
  95 #define APP_LOOKUP_EXACT_MATCH          0
  96 #define APP_LOOKUP_LPM                  1
  97 #define DO_RFC_1812_CHECKS
  98
  99 #ifndef APP_LOOKUP_METHOD
 100 #define APP_LOOKUP_METHOD             APP_LOOKUP_LPM
 101 #endif
 102
 103 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 104 #include <rte_hash.h>
 105 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 106 #include <rte_lpm.h>
 107 #else
 108 #error "APP_LOOKUP_METHOD set to incorrect value"
 109 #endif
 110
 111 #ifndef IPv6_BYTES
 112 #define IPv6_BYTES_FMT "%02x%02x:%02x%02x:%02x%02x:%02x%02x:"\
 113                        "%02x%02x:%02x%02x:%02x%02x:%02x%02x"
 114 #define IPv6_BYTES(addr) \
 115         addr[0],  addr[1], addr[2],  addr[3], \
 116         addr[4],  addr[5], addr[6],  addr[7], \
 117         addr[8],  addr[9], addr[10], addr[11],\
 118         addr[12], addr[13],addr[14], addr[15]
 119 #endif
 120
 121 #define MAX_JUMBO_PKT_LEN  9600
 122
 123 #define IPV6_ADDR_LEN 16
 124
 125 #define MEMPOOL_CACHE_SIZE 256
 126
 127 /*
 128  * This expression is used to calculate the number of mbufs needed depending on
 129  * user input, taking into account memory for rx and tx hardware rings, cache
 130  * per lcore and mtable per port per lcore. RTE_MAX is used to ensure that
 131  * NB_MBUF never goes below a minimum value of 8192.
 132  */
 133
 134 #define NB_MBUF RTE_MAX ( \
 135         (nb_ports*nb_rx_queue*RTE_TEST_RX_DESC_DEFAULT + \
 136         nb_ports*nb_lcores*MAX_PKT_BURST + \
 137         nb_ports*n_tx_queue*RTE_TEST_TX_DESC_DEFAULT + \
 138         nb_lcores*MEMPOOL_CACHE_SIZE), \
 139         (unsigned)8192)
 140
 141 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
 142
 143 #define NB_SOCKETS 8
 144
 145 /* Configure how many packets ahead to prefetch, when reading packets */
 146 #define PREFETCH_OFFSET 3
 147
 148 /*
 149  * Configurable number of RX/TX ring descriptors
 150  */
 151 #define RTE_TEST_RX_DESC_DEFAULT 128
 152 #define RTE_TEST_TX_DESC_DEFAULT 512
 153 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
 154 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
 155
 156 /* ethernet addresses of ports */
 157 static struct ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
 158
 159 /* ethernet addresses of ports */
 160 static rte_spinlock_t locks[RTE_MAX_ETHPORTS];
 161
 162 /* mask of enabled ports */
 163 static uint32_t enabled_port_mask = 0;
 164 /* Ports set in promiscuous mode off by default. */
 165 static int promiscuous_on = 0;
 166 /* NUMA is enabled by default. */
 167 static int numa_on = 1;
 168
 169 enum freq_scale_hint_t
 170 {
 171         FREQ_LOWER    =      -1,
 172         FREQ_CURRENT  =       0,
 173         FREQ_HIGHER   =       1,
 174         FREQ_HIGHEST  =       2
 175 };
 176
 177 struct lcore_rx_queue {
 178         uint8_t port_id;
 179         uint8_t queue_id;
 180         enum freq_scale_hint_t freq_up_hint;
 181         uint32_t zero_rx_packet_count;
 182         uint32_t idle_hint;
 183 } __rte_cache_aligned;
 184
 185 #define MAX_RX_QUEUE_PER_LCORE 16
 186 #define MAX_TX_QUEUE_PER_PORT RTE_MAX_ETHPORTS
 187 #define MAX_RX_QUEUE_PER_PORT 128
 188
 189 #define MAX_RX_QUEUE_INTERRUPT_PER_PORT 16
 190
 191
 192 #define MAX_LCORE_PARAMS 1024
 193 struct lcore_params {
 194         uint8_t port_id;
 195         uint8_t queue_id;
 196         uint8_t lcore_id;
 197 } __rte_cache_aligned;
 198
 199 static struct lcore_params lcore_params_array[MAX_LCORE_PARAMS];
 200 static struct lcore_params lcore_params_array_default[] = {
 201         {0, 0, 2},
 202         {0, 1, 2},
 203         {0, 2, 2},
 204         {1, 0, 2},
 205         {1, 1, 2},
 206         {1, 2, 2},
 207         {2, 0, 2},
 208         {3, 0, 3},
 209         {3, 1, 3},
 210 };
 211
 212 static struct lcore_params * lcore_params = lcore_params_array_default;
 213 static uint16_t nb_lcore_params = sizeof(lcore_params_array_default) /
 214                                 sizeof(lcore_params_array_default[0]);
 215
 216 static struct rte_eth_conf port_conf = {
 217         .rxmode = {
 218                 .mq_mode        = ETH_MQ_RX_RSS,
 219                 .max_rx_pkt_len = ETHER_MAX_LEN,
 220                 .split_hdr_size = 0,
 221                 .header_split   = 0, /**< Header Split disabled */
 222                 .hw_ip_checksum = 1, /**< IP checksum offload enabled */
 223                 .hw_vlan_filter = 0, /**< VLAN filtering disabled */
 224                 .jumbo_frame    = 0, /**< Jumbo Frame Support disabled */
 225                 .hw_strip_crc   = 0, /**< CRC stripped by hardware */
 226         },
 227         .rx_adv_conf = {
 228                 .rss_conf = {
 229                         .rss_key = NULL,
 230                         .rss_hf = ETH_RSS_UDP,
 231                 },
 232         },
 233         .txmode = {
 234                 .mq_mode = ETH_MQ_TX_NONE,
 235         },
 236         .intr_conf = {
 237                 .lsc = 1,
 238                 .rxq = 1,
 239         },
 240 };
 241
 242 static struct rte_mempool * pktmbuf_pool[NB_SOCKETS];
 243
 244
 245 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 246
 247 #ifdef RTE_MACHINE_CPUFLAG_SSE4_2
 248 #include <rte_hash_crc.h>
 249 #define DEFAULT_HASH_FUNC       rte_hash_crc
 250 #else
 251 #include <rte_jhash.h>
 252 #define DEFAULT_HASH_FUNC       rte_jhash
 253 #endif
 254
 255 struct ipv4_5tuple {
 256         uint32_t ip_dst;
 257         uint32_t ip_src;
 258         uint16_t port_dst;
 259         uint16_t port_src;
 260         uint8_t  proto;
 261 } __attribute__((__packed__));
 262
 263 struct ipv6_5tuple {
 264         uint8_t  ip_dst[IPV6_ADDR_LEN];
 265         uint8_t  ip_src[IPV6_ADDR_LEN];
 266         uint16_t port_dst;
 267         uint16_t port_src;
 268         uint8_t  proto;
 269 } __attribute__((__packed__));
 270
 271 struct ipv4_l3fwd_route {
 272         struct ipv4_5tuple key;
 273         uint8_t if_out;
 274 };
 275
 276 struct ipv6_l3fwd_route {
 277         struct ipv6_5tuple key;
 278         uint8_t if_out;
 279 };
 280
 281 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
 282         {{IPv4(100,10,0,1), IPv4(200,10,0,1), 101, 11, IPPROTO_TCP}, 0},
 283         {{IPv4(100,20,0,2), IPv4(200,20,0,2), 102, 12, IPPROTO_TCP}, 1},
 284         {{IPv4(100,30,0,3), IPv4(200,30,0,3), 103, 13, IPPROTO_TCP}, 2},
 285         {{IPv4(100,40,0,4), IPv4(200,40,0,4), 104, 14, IPPROTO_TCP}, 3},
 286 };
 287
 288 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
 289         {
 290                 {
 291                         {0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
 292                          0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
 293                         {0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
 294                          0x02, 0x1e, 0x67, 0xff, 0xfe, 0x0d, 0xb6, 0x0a},
 295                          1, 10, IPPROTO_UDP
 296                 }, 4
 297         },
 298 };
 299
 300 typedef struct rte_hash lookup_struct_t;
 301 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
 302 static lookup_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
 303
 304 #define L3FWD_HASH_ENTRIES      1024
 305
 306 #define IPV4_L3FWD_NUM_ROUTES \
 307         (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
 308
 309 #define IPV6_L3FWD_NUM_ROUTES \
 310         (sizeof(ipv6_l3fwd_route_array) / sizeof(ipv6_l3fwd_route_array[0]))
 311
 312 static uint8_t ipv4_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 313 static uint8_t ipv6_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 314 #endif
 315
 316 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 317 struct ipv4_l3fwd_route {
 318         uint32_t ip;
 319         uint8_t  depth;
 320         uint8_t  if_out;
 321 };
 322
 323 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
 324         {IPv4(1,1,1,0), 24, 0},
 325         {IPv4(2,1,1,0), 24, 1},
 326         {IPv4(3,1,1,0), 24, 2},
 327         {IPv4(4,1,1,0), 24, 3},
 328         {IPv4(5,1,1,0), 24, 4},
 329         {IPv4(6,1,1,0), 24, 5},
 330         {IPv4(7,1,1,0), 24, 6},
 331         {IPv4(8,1,1,0), 24, 7},
 332 };
 333
 334 #define IPV4_L3FWD_NUM_ROUTES \
 335         (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
 336
 337 #define IPV4_L3FWD_LPM_MAX_RULES     1024
 338
 339 typedef struct rte_lpm lookup_struct_t;
 340 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
 341 #endif
 342
 343 struct lcore_conf {
 344         uint16_t n_rx_queue;
 345         struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
 346         uint16_t n_tx_port;
 347         uint16_t tx_port_id[RTE_MAX_ETHPORTS];
 348         uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
 349         struct rte_eth_dev_tx_buffer *tx_buffer[RTE_MAX_ETHPORTS];
 350         lookup_struct_t * ipv4_lookup_struct;
 351         lookup_struct_t * ipv6_lookup_struct;
 352 } __rte_cache_aligned;
 353
 354 struct lcore_stats {
 355         /* total sleep time in ms since last frequency scaling down */
 356         uint32_t sleep_time;
 357         /* number of long sleep recently */
 358         uint32_t nb_long_sleep;
 359         /* freq. scaling up trend */
 360         uint32_t trend;
 361         /* total packet processed recently */
 362         uint64_t nb_rx_processed;
 363         /* total iterations looped recently */
 364         uint64_t nb_iteration_looped;
 365         uint32_t padding[9];
 366 } __rte_cache_aligned;
 367
 368 static struct lcore_conf lcore_conf[RTE_MAX_LCORE] __rte_cache_aligned;
 369 static struct lcore_stats stats[RTE_MAX_LCORE] __rte_cache_aligned;
 370 static struct rte_timer power_timers[RTE_MAX_LCORE];
 371
 372 static inline uint32_t power_idle_heuristic(uint32_t zero_rx_packet_count);
 373 static inline enum freq_scale_hint_t power_freq_scaleup_heuristic( \
 374                         unsigned lcore_id, uint8_t port_id, uint16_t queue_id);
 375
 376 /* exit signal handler */
 377 static void
 378 signal_exit_now(int sigtype)
 379 {
 380         unsigned lcore_id;
 381         int ret;
 382
 383         if (sigtype == SIGINT) {
 384                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
 385                         if (rte_lcore_is_enabled(lcore_id) == 0)
 386                                 continue;
 387
 388                         /* init power management library */
 389                         ret = rte_power_exit(lcore_id);
 390                         if (ret)
 391                                 rte_exit(EXIT_FAILURE, "Power management "
 392                                         "library de-initialization failed on "
 393                                                         "core%u\n", lcore_id);
 394                 }
 395         }
 396
 397         rte_exit(EXIT_SUCCESS, "User forced exit\n");
 398 }
 399
 400 /*  Freqency scale down timer callback */
 401 static void
 402 power_timer_cb(__attribute__((unused)) struct rte_timer *tim,
 403                           __attribute__((unused)) void *arg)
 404 {
 405         uint64_t hz;
 406         float sleep_time_ratio;
 407         unsigned lcore_id = rte_lcore_id();
 408
 409         /* accumulate total execution time in us when callback is invoked */
 410         sleep_time_ratio = (float)(stats[lcore_id].sleep_time) /
 411                                         (float)SCALING_PERIOD;
 412         /**
 413          * check whether need to scale down frequency a step if it sleep a lot.
 414          */
 415         if (sleep_time_ratio >= SCALING_DOWN_TIME_RATIO_THRESHOLD) {
 416                 if (rte_power_freq_down)
 417                         rte_power_freq_down(lcore_id);
 418         }
 419         else if ( (unsigned)(stats[lcore_id].nb_rx_processed /
 420                 stats[lcore_id].nb_iteration_looped) < MAX_PKT_BURST) {
 421                 /**
 422                  * scale down a step if average packet per iteration less
 423                  * than expectation.
 424                  */
 425                 if (rte_power_freq_down)
 426                         rte_power_freq_down(lcore_id);
 427         }
 428
 429         /**
 430          * initialize another timer according to current frequency to ensure
 431          * timer interval is relatively fixed.
 432          */
 433         hz = rte_get_timer_hz();
 434         rte_timer_reset(&power_timers[lcore_id], hz/TIMER_NUMBER_PER_SECOND,
 435                                 SINGLE, lcore_id, power_timer_cb, NULL);
 436
 437         stats[lcore_id].nb_rx_processed = 0;
 438         stats[lcore_id].nb_iteration_looped = 0;
 439
 440         stats[lcore_id].sleep_time = 0;
 441 }
 442
 443 /* Enqueue a single packet, and send burst if queue is filled */
 444 static inline int
 445 send_single_packet(struct rte_mbuf *m, uint8_t port)
 446 {
 447         uint32_t lcore_id;
 448         struct lcore_conf *qconf;
 449
 450         lcore_id = rte_lcore_id();
 451         qconf = &lcore_conf[lcore_id];
 452
 453         rte_eth_tx_buffer(port, qconf->tx_queue_id[port],
 454                         qconf->tx_buffer[port], m);
 455
 456         return 0;
 457 }
 458
 459 #ifdef DO_RFC_1812_CHECKS
 460 static inline int
 461 is_valid_ipv4_pkt(struct ipv4_hdr *pkt, uint32_t link_len)
 462 {
 463         /* From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2 */
 464         /*
 465          * 1. The packet length reported by the Link Layer must be large
 466          * enough to hold the minimum length legal IP datagram (20 bytes).
 467          */
 468         if (link_len < sizeof(struct ipv4_hdr))
 469                 return -1;
 470
 471         /* 2. The IP checksum must be correct. */
 472         /* this is checked in H/W */
 473
 474         /*
 475          * 3. The IP version number must be 4. If the version number is not 4
 476          * then the packet may be another version of IP, such as IPng or
 477          * ST-II.
 478          */
 479         if (((pkt->version_ihl) >> 4) != 4)
 480                 return -3;
 481         /*
 482          * 4. The IP header length field must be large enough to hold the
 483          * minimum length legal IP datagram (20 bytes = 5 words).
 484          */
 485         if ((pkt->version_ihl & 0xf) < 5)
 486                 return -4;
 487
 488         /*
 489          * 5. The IP total length field must be large enough to hold the IP
 490          * datagram header, whose length is specified in the IP header length
 491          * field.
 492          */
 493         if (rte_cpu_to_be_16(pkt->total_length) < sizeof(struct ipv4_hdr))
 494                 return -5;
 495
 496         return 0;
 497 }
 498 #endif
 499
 500 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 501 static void
 502 print_ipv4_key(struct ipv4_5tuple key)
 503 {
 504         printf("IP dst = %08x, IP src = %08x, port dst = %d, port src = %d, "
 505                 "proto = %d\n", (unsigned)key.ip_dst, (unsigned)key.ip_src,
 506                                 key.port_dst, key.port_src, key.proto);
 507 }
 508 static void
 509 print_ipv6_key(struct ipv6_5tuple key)
 510 {
 511         printf( "IP dst = " IPv6_BYTES_FMT ", IP src = " IPv6_BYTES_FMT ", "
 512                 "port dst = %d, port src = %d, proto = %d\n",
 513                 IPv6_BYTES(key.ip_dst), IPv6_BYTES(key.ip_src),
 514                 key.port_dst, key.port_src, key.proto);
 515 }
 516
 517 static inline uint8_t
 518 get_ipv4_dst_port(struct ipv4_hdr *ipv4_hdr, uint8_t portid,
 519                 lookup_struct_t * ipv4_l3fwd_lookup_struct)
 520 {
 521         struct ipv4_5tuple key;
 522         struct tcp_hdr *tcp;
 523         struct udp_hdr *udp;
 524         int ret = 0;
 525
 526         key.ip_dst = rte_be_to_cpu_32(ipv4_hdr->dst_addr);
 527         key.ip_src = rte_be_to_cpu_32(ipv4_hdr->src_addr);
 528         key.proto = ipv4_hdr->next_proto_id;
 529
 530         switch (ipv4_hdr->next_proto_id) {
 531         case IPPROTO_TCP:
 532                 tcp = (struct tcp_hdr *)((unsigned char *)ipv4_hdr +
 533                                         sizeof(struct ipv4_hdr));
 534                 key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
 535                 key.port_src = rte_be_to_cpu_16(tcp->src_port);
 536                 break;
 537
 538         case IPPROTO_UDP:
 539                 udp = (struct udp_hdr *)((unsigned char *)ipv4_hdr +
 540                                         sizeof(struct ipv4_hdr));
 541                 key.port_dst = rte_be_to_cpu_16(udp->dst_port);
 542                 key.port_src = rte_be_to_cpu_16(udp->src_port);
 543                 break;
 544
 545         default:
 546                 key.port_dst = 0;
 547                 key.port_src = 0;
 548                 break;
 549         }
 550
 551         /* Find destination port */
 552         ret = rte_hash_lookup(ipv4_l3fwd_lookup_struct, (const void *)&key);
 553         return (uint8_t)((ret < 0)? portid : ipv4_l3fwd_out_if[ret]);
 554 }
 555
 556 static inline uint8_t
 557 get_ipv6_dst_port(struct ipv6_hdr *ipv6_hdr,  uint8_t portid,
 558                         lookup_struct_t *ipv6_l3fwd_lookup_struct)
 559 {
 560         struct ipv6_5tuple key;
 561         struct tcp_hdr *tcp;
 562         struct udp_hdr *udp;
 563         int ret = 0;
 564
 565         memcpy(key.ip_dst, ipv6_hdr->dst_addr, IPV6_ADDR_LEN);
 566         memcpy(key.ip_src, ipv6_hdr->src_addr, IPV6_ADDR_LEN);
 567
 568         key.proto = ipv6_hdr->proto;
 569
 570         switch (ipv6_hdr->proto) {
 571         case IPPROTO_TCP:
 572                 tcp = (struct tcp_hdr *)((unsigned char *) ipv6_hdr +
 573                                         sizeof(struct ipv6_hdr));
 574                 key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
 575                 key.port_src = rte_be_to_cpu_16(tcp->src_port);
 576                 break;
 577
 578         case IPPROTO_UDP:
 579                 udp = (struct udp_hdr *)((unsigned char *) ipv6_hdr +
 580                                         sizeof(struct ipv6_hdr));
 581                 key.port_dst = rte_be_to_cpu_16(udp->dst_port);
 582                 key.port_src = rte_be_to_cpu_16(udp->src_port);
 583                 break;
 584
 585         default:
 586                 key.port_dst = 0;
 587                 key.port_src = 0;
 588                 break;
 589         }
 590
 591         /* Find destination port */
 592         ret = rte_hash_lookup(ipv6_l3fwd_lookup_struct, (const void *)&key);
 593         return (uint8_t)((ret < 0)? portid : ipv6_l3fwd_out_if[ret]);
 594 }
 595 #endif
 596
 597 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 598 static inline uint8_t
 599 get_ipv4_dst_port(struct ipv4_hdr *ipv4_hdr, uint8_t portid,
 600                 lookup_struct_t *ipv4_l3fwd_lookup_struct)
 601 {
 602         uint32_t next_hop;
 603
 604         return (uint8_t) ((rte_lpm_lookup(ipv4_l3fwd_lookup_struct,
 605                         rte_be_to_cpu_32(ipv4_hdr->dst_addr), &next_hop) == 0)?
 606                         next_hop : portid);
 607 }
 608 #endif
 609
 610 static inline void
 611 l3fwd_simple_forward(struct rte_mbuf *m, uint8_t portid,
 612                                 struct lcore_conf *qconf)
 613 {
 614         struct ether_hdr *eth_hdr;
 615         struct ipv4_hdr *ipv4_hdr;
 616         void *d_addr_bytes;
 617         uint8_t dst_port;
 618
 619         eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
 620
 621         if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) {
 622                 /* Handle IPv4 headers.*/
 623                 ipv4_hdr =
 624                         rte_pktmbuf_mtod_offset(m, struct ipv4_hdr *,
 625                                                 sizeof(struct ether_hdr));
 626
 627 #ifdef DO_RFC_1812_CHECKS
 628                 /* Check to make sure the packet is valid (RFC1812) */
 629                 if (is_valid_ipv4_pkt(ipv4_hdr, m->pkt_len) < 0) {
 630                         rte_pktmbuf_free(m);
 631                         return;
 632                 }
 633 #endif
 634
 635                 dst_port = get_ipv4_dst_port(ipv4_hdr, portid,
 636                                         qconf->ipv4_lookup_struct);
 637                 if (dst_port >= RTE_MAX_ETHPORTS ||
 638                                 (enabled_port_mask & 1 << dst_port) == 0)
 639                         dst_port = portid;
 640
 641                 /* 02:00:00:00:00:xx */
 642                 d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
 643                 *((uint64_t *)d_addr_bytes) =
 644                         0x000000000002 + ((uint64_t)dst_port << 40);
 645
 646 #ifdef DO_RFC_1812_CHECKS
 647                 /* Update time to live and header checksum */
 648                 --(ipv4_hdr->time_to_live);
 649                 ++(ipv4_hdr->hdr_checksum);
 650 #endif
 651
 652                 /* src addr */
 653                 ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
 654
 655                 send_single_packet(m, dst_port);
 656         } else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) {
 657                 /* Handle IPv6 headers.*/
 658 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 659                 struct ipv6_hdr *ipv6_hdr;
 660
 661                 ipv6_hdr =
 662                         rte_pktmbuf_mtod_offset(m, struct ipv6_hdr *,
 663                                                 sizeof(struct ether_hdr));
 664
 665                 dst_port = get_ipv6_dst_port(ipv6_hdr, portid,
 666                                         qconf->ipv6_lookup_struct);
 667
 668                 if (dst_port >= RTE_MAX_ETHPORTS ||
 669                                 (enabled_port_mask & 1 << dst_port) == 0)
 670                         dst_port = portid;
 671
 672                 /* 02:00:00:00:00:xx */
 673                 d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
 674                 *((uint64_t *)d_addr_bytes) =
 675                         0x000000000002 + ((uint64_t)dst_port << 40);
 676
 677                 /* src addr */
 678                 ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
 679
 680                 send_single_packet(m, dst_port);
 681 #else
 682                 /* We don't currently handle IPv6 packets in LPM mode. */
 683                 rte_pktmbuf_free(m);
 684 #endif
 685         }
 686
 687 }
 688
 689 #define MINIMUM_SLEEP_TIME         1
 690 #define SUSPEND_THRESHOLD          300
 691
 692 static inline uint32_t
 693 power_idle_heuristic(uint32_t zero_rx_packet_count)
 694 {
 695         /* If zero count is less than 100,  sleep 1us */
 696         if (zero_rx_packet_count < SUSPEND_THRESHOLD)
 697                 return MINIMUM_SLEEP_TIME;
 698         /* If zero count is less than 1000, sleep 100 us which is the
 699                 minimum latency switching from C3/C6 to C0
 700         */
 701         else
 702                 return SUSPEND_THRESHOLD;
 703
 704         return 0;
 705 }
 706
 707 static inline enum freq_scale_hint_t
 708 power_freq_scaleup_heuristic(unsigned lcore_id,
 709                              uint8_t port_id,
 710                              uint16_t queue_id)
 711 {
 712 /**
 713  * HW Rx queue size is 128 by default, Rx burst read at maximum 32 entries
 714  * per iteration
 715  */
 716 #define FREQ_GEAR1_RX_PACKET_THRESHOLD             MAX_PKT_BURST
 717 #define FREQ_GEAR2_RX_PACKET_THRESHOLD             (MAX_PKT_BURST*2)
 718 #define FREQ_GEAR3_RX_PACKET_THRESHOLD             (MAX_PKT_BURST*3)
 719 #define FREQ_UP_TREND1_ACC   1
 720 #define FREQ_UP_TREND2_ACC   100
 721 #define FREQ_UP_THRESHOLD    10000
 722
 723         if (likely(rte_eth_rx_descriptor_done(port_id, queue_id,
 724                         FREQ_GEAR3_RX_PACKET_THRESHOLD) > 0)) {
 725                 stats[lcore_id].trend = 0;
 726                 return FREQ_HIGHEST;
 727         } else if (likely(rte_eth_rx_descriptor_done(port_id, queue_id,
 728                         FREQ_GEAR2_RX_PACKET_THRESHOLD) > 0))
 729                 stats[lcore_id].trend += FREQ_UP_TREND2_ACC;
 730         else if (likely(rte_eth_rx_descriptor_done(port_id, queue_id,
 731                         FREQ_GEAR1_RX_PACKET_THRESHOLD) > 0))
 732                 stats[lcore_id].trend += FREQ_UP_TREND1_ACC;
 733
 734         if (likely(stats[lcore_id].trend > FREQ_UP_THRESHOLD)) {
 735                 stats[lcore_id].trend = 0;
 736                 return FREQ_HIGHER;
 737         }
 738
 739         return FREQ_CURRENT;
 740 }
 741
 742 /**
 743  * force polling thread sleep until one-shot rx interrupt triggers
 744  * @param port_id
 745  *  Port id.
 746  * @param queue_id
 747  *  Rx queue id.
 748  * @return
 749  *  0 on success
 750  */
 751 static int
 752 sleep_until_rx_interrupt(int num)
 753 {
 754         struct rte_epoll_event event[num];
 755         int n, i;
 756         uint8_t port_id, queue_id;
 757         void *data;
 758
 759         RTE_LOG(INFO, L3FWD_POWER,
 760                 "lcore %u sleeps until interrupt triggers\n",
 761                 rte_lcore_id());
 762
 763         n = rte_epoll_wait(RTE_EPOLL_PER_THREAD, event, num, -1);
 764         for (i = 0; i < n; i++) {
 765                 data = event[i].epdata.data;
 766                 port_id = ((uintptr_t)data) >> CHAR_BIT;
 767                 queue_id = ((uintptr_t)data) &
 768                         RTE_LEN2MASK(CHAR_BIT, uint8_t);
 769                 rte_eth_dev_rx_intr_disable(port_id, queue_id);
 770                 RTE_LOG(INFO, L3FWD_POWER,
 771                         "lcore %u is waked up from rx interrupt on"
 772                         " port %d queue %d\n",
 773                         rte_lcore_id(), port_id, queue_id);
 774         }
 775
 776         return 0;
 777 }
 778
 779 static void turn_on_intr(struct lcore_conf *qconf)
 780 {
 781         int i;
 782         struct lcore_rx_queue *rx_queue;
 783         uint8_t port_id, queue_id;
 784
 785         for (i = 0; i < qconf->n_rx_queue; ++i) {
 786                 rx_queue = &(qconf->rx_queue_list[i]);
 787                 port_id = rx_queue->port_id;
 788                 queue_id = rx_queue->queue_id;
 789
 790                 rte_spinlock_lock(&(locks[port_id]));
 791                 rte_eth_dev_rx_intr_enable(port_id, queue_id);
 792                 rte_spinlock_unlock(&(locks[port_id]));
 793         }
 794 }
 795
 796 static int event_register(struct lcore_conf *qconf)
 797 {
 798         struct lcore_rx_queue *rx_queue;
 799         uint8_t portid, queueid;
 800         uint32_t data;
 801         int ret;
 802         int i;
 803
 804         for (i = 0; i < qconf->n_rx_queue; ++i) {
 805                 rx_queue = &(qconf->rx_queue_list[i]);
 806                 portid = rx_queue->port_id;
 807                 queueid = rx_queue->queue_id;
 808                 data = portid << CHAR_BIT | queueid;
 809
 810                 ret = rte_eth_dev_rx_intr_ctl_q(portid, queueid,
 811                                                 RTE_EPOLL_PER_THREAD,
 812                                                 RTE_INTR_EVENT_ADD,
 813                                                 (void *)((uintptr_t)data));
 814                 if (ret)
 815                         return ret;
 816         }
 817
 818         return 0;
 819 }
 820
 821 /* main processing loop */
 822 static int
 823 main_loop(__attribute__((unused)) void *dummy)
 824 {
 825         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
 826         unsigned lcore_id;
 827         uint64_t prev_tsc, diff_tsc, cur_tsc;
 828         uint64_t prev_tsc_power = 0, cur_tsc_power, diff_tsc_power;
 829         int i, j, nb_rx;
 830         uint8_t portid, queueid;
 831         struct lcore_conf *qconf;
 832         struct lcore_rx_queue *rx_queue;
 833         enum freq_scale_hint_t lcore_scaleup_hint;
 834         uint32_t lcore_rx_idle_count = 0;
 835         uint32_t lcore_idle_hint = 0;
 836         int intr_en = 0;
 837
 838         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
 839
 840         prev_tsc = 0;
 841
 842         lcore_id = rte_lcore_id();
 843         qconf = &lcore_conf[lcore_id];
 844
 845         if (qconf->n_rx_queue == 0) {
 846                 RTE_LOG(INFO, L3FWD_POWER, "lcore %u has nothing to do\n", lcore_id);
 847                 return 0;
 848         }
 849
 850         RTE_LOG(INFO, L3FWD_POWER, "entering main loop on lcore %u\n", lcore_id);
 851
 852         for (i = 0; i < qconf->n_rx_queue; i++) {
 853                 portid = qconf->rx_queue_list[i].port_id;
 854                 queueid = qconf->rx_queue_list[i].queue_id;
 855                 RTE_LOG(INFO, L3FWD_POWER, " -- lcoreid=%u portid=%hhu "
 856                         "rxqueueid=%hhu\n", lcore_id, portid, queueid);
 857         }
 858
 859         /* add into event wait list */
 860         if (event_register(qconf) == 0)
 861                 intr_en = 1;
 862         else
 863                 RTE_LOG(INFO, L3FWD_POWER, "RX interrupt won't enable.\n");
 864
 865         while (1) {
 866                 stats[lcore_id].nb_iteration_looped++;
 867
 868                 cur_tsc = rte_rdtsc();
 869                 cur_tsc_power = cur_tsc;
 870
 871                 /*
 872                  * TX burst queue drain
 873                  */
 874                 diff_tsc = cur_tsc - prev_tsc;
 875                 if (unlikely(diff_tsc > drain_tsc)) {
 876                         for (i = 0; i < qconf->n_tx_port; ++i) {
 877                                 portid = qconf->tx_port_id[i];
 878                                 rte_eth_tx_buffer_flush(portid,
 879                                                 qconf->tx_queue_id[portid],
 880                                                 qconf->tx_buffer[portid]);
 881                         }
 882                         prev_tsc = cur_tsc;
 883                 }
 884
 885                 diff_tsc_power = cur_tsc_power - prev_tsc_power;
 886                 if (diff_tsc_power > TIMER_RESOLUTION_CYCLES) {
 887                         rte_timer_manage();
 888                         prev_tsc_power = cur_tsc_power;
 889                 }
 890
 891 start_rx:
 892                 /*
 893                  * Read packet from RX queues
 894                  */
 895                 lcore_scaleup_hint = FREQ_CURRENT;
 896                 lcore_rx_idle_count = 0;
 897                 for (i = 0; i < qconf->n_rx_queue; ++i) {
 898                         rx_queue = &(qconf->rx_queue_list[i]);
 899                         rx_queue->idle_hint = 0;
 900                         portid = rx_queue->port_id;
 901                         queueid = rx_queue->queue_id;
 902
 903                         nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
 904                                                                 MAX_PKT_BURST);
 905
 906                         stats[lcore_id].nb_rx_processed += nb_rx;
 907                         if (unlikely(nb_rx == 0)) {
 908                                 /**
 909                                  * no packet received from rx queue, try to
 910                                  * sleep for a while forcing CPU enter deeper
 911                                  * C states.
 912                                  */
 913                                 rx_queue->zero_rx_packet_count++;
 914
 915                                 if (rx_queue->zero_rx_packet_count <=
 916                                                         MIN_ZERO_POLL_COUNT)
 917                                         continue;
 918
 919                                 rx_queue->idle_hint = power_idle_heuristic(\
 920                                         rx_queue->zero_rx_packet_count);
 921                                 lcore_rx_idle_count++;
 922                         } else {
 923                                 rx_queue->zero_rx_packet_count = 0;
 924
 925                                 /**
 926                                  * do not scale up frequency immediately as
 927                                  * user to kernel space communication is costly
 928                                  * which might impact packet I/O for received
 929                                  * packets.
 930                                  */
 931                                 rx_queue->freq_up_hint =
 932                                         power_freq_scaleup_heuristic(lcore_id,
 933                                                         portid, queueid);
 934                         }
 935
 936                         /* Prefetch first packets */
 937                         for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
 938                                 rte_prefetch0(rte_pktmbuf_mtod(
 939                                                 pkts_burst[j], void *));
 940                         }
 941
 942                         /* Prefetch and forward already prefetched packets */
 943                         for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
 944                                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
 945                                                 j + PREFETCH_OFFSET], void *));
 946                                 l3fwd_simple_forward(pkts_burst[j], portid,
 947                                                                 qconf);
 948                         }
 949
 950                         /* Forward remaining prefetched packets */
 951                         for (; j < nb_rx; j++) {
 952                                 l3fwd_simple_forward(pkts_burst[j], portid,
 953                                                                 qconf);
 954                         }
 955                 }
 956
 957                 if (likely(lcore_rx_idle_count != qconf->n_rx_queue)) {
 958                         for (i = 1, lcore_scaleup_hint =
 959                                 qconf->rx_queue_list[0].freq_up_hint;
 960                                         i < qconf->n_rx_queue; ++i) {
 961                                 rx_queue = &(qconf->rx_queue_list[i]);
 962                                 if (rx_queue->freq_up_hint >
 963                                                 lcore_scaleup_hint)
 964                                         lcore_scaleup_hint =
 965                                                 rx_queue->freq_up_hint;
 966                         }
 967
 968                         if (lcore_scaleup_hint == FREQ_HIGHEST) {
 969                                 if (rte_power_freq_max)
 970                                         rte_power_freq_max(lcore_id);
 971                         } else if (lcore_scaleup_hint == FREQ_HIGHER) {
 972                                 if (rte_power_freq_up)
 973                                         rte_power_freq_up(lcore_id);
 974                         }
 975                 } else {
 976                         /**
 977                          * All Rx queues empty in recent consecutive polls,
 978                          * sleep in a conservative manner, meaning sleep as
 979                          * less as possible.
 980                          */
 981                         for (i = 1, lcore_idle_hint =
 982                                 qconf->rx_queue_list[0].idle_hint;
 983                                         i < qconf->n_rx_queue; ++i) {
 984                                 rx_queue = &(qconf->rx_queue_list[i]);
 985                                 if (rx_queue->idle_hint < lcore_idle_hint)
 986                                         lcore_idle_hint = rx_queue->idle_hint;
 987                         }
 988
 989                         if (lcore_idle_hint < SUSPEND_THRESHOLD)
 990                                 /**
 991                                  * execute "pause" instruction to avoid context
 992                                  * switch which generally take hundred of
 993                                  * microseconds for short sleep.
 994                                  */
 995                                 rte_delay_us(lcore_idle_hint);
 996                         else {
 997                                 /* suspend until rx interrupt trigges */
 998                                 if (intr_en) {
 999                                         turn_on_intr(qconf);
1000                                         sleep_until_rx_interrupt(
1001                                                 qconf->n_rx_queue);
1002                                 }
1003                                 /* start receiving packets immediately */
1004                                 goto start_rx;
1005                         }
1006                         stats[lcore_id].sleep_time += lcore_idle_hint;
1007                 }
1008         }
1009 }
1010
1011 static int
1012 check_lcore_params(void)
1013 {
1014         uint8_t queue, lcore;
1015         uint16_t i;
1016         int socketid;
1017
1018         for (i = 0; i < nb_lcore_params; ++i) {
1019                 queue = lcore_params[i].queue_id;
1020                 if (queue >= MAX_RX_QUEUE_PER_PORT) {
1021                         printf("invalid queue number: %hhu\n", queue);
1022                         return -1;
1023                 }
1024                 lcore = lcore_params[i].lcore_id;
1025                 if (!rte_lcore_is_enabled(lcore)) {
1026                         printf("error: lcore %hhu is not enabled in lcore "
1027                                                         "mask\n", lcore);
1028                         return -1;
1029                 }
1030                 if ((socketid = rte_lcore_to_socket_id(lcore) != 0) &&
1031                                                         (numa_on == 0)) {
1032                         printf("warning: lcore %hhu is on socket %d with numa "
1033                                                 "off\n", lcore, socketid);
1034                 }
1035         }
1036         return 0;
1037 }
1038
1039 static int
1040 check_port_config(const unsigned nb_ports)
1041 {
1042         unsigned portid;
1043         uint16_t i;
1044
1045         for (i = 0; i < nb_lcore_params; ++i) {
1046                 portid = lcore_params[i].port_id;
1047                 if ((enabled_port_mask & (1 << portid)) == 0) {
1048                         printf("port %u is not enabled in port mask\n",
1049                                                                 portid);
1050                         return -1;
1051                 }
1052                 if (portid >= nb_ports) {
1053                         printf("port %u is not present on the board\n",
1054                                                                 portid);
1055                         return -1;
1056                 }
1057         }
1058         return 0;
1059 }
1060
1061 static uint8_t
1062 get_port_n_rx_queues(const uint8_t port)
1063 {
1064         int queue = -1;
1065         uint16_t i;
1066
1067         for (i = 0; i < nb_lcore_params; ++i) {
1068                 if (lcore_params[i].port_id == port &&
1069                                 lcore_params[i].queue_id > queue)
1070                         queue = lcore_params[i].queue_id;
1071         }
1072         return (uint8_t)(++queue);
1073 }
1074
1075 static int
1076 init_lcore_rx_queues(void)
1077 {
1078         uint16_t i, nb_rx_queue;
1079         uint8_t lcore;
1080
1081         for (i = 0; i < nb_lcore_params; ++i) {
1082                 lcore = lcore_params[i].lcore_id;
1083                 nb_rx_queue = lcore_conf[lcore].n_rx_queue;
1084                 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
1085                         printf("error: too many queues (%u) for lcore: %u\n",
1086                                 (unsigned)nb_rx_queue + 1, (unsigned)lcore);
1087                         return -1;
1088                 } else {
1089                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
1090                                 lcore_params[i].port_id;
1091                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
1092                                 lcore_params[i].queue_id;
1093                         lcore_conf[lcore].n_rx_queue++;
1094                 }
1095         }
1096         return 0;
1097 }
1098
1099 /* display usage */
1100 static void
1101 print_usage(const char *prgname)
1102 {
1103         printf ("%s [EAL options] -- -p PORTMASK -P"
1104                 "  [--config (port,queue,lcore)[,(port,queue,lcore]]"
1105                 "  [--enable-jumbo [--max-pkt-len PKTLEN]]\n"
1106                 "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
1107                 "  -P : enable promiscuous mode\n"
1108                 "  --config (port,queue,lcore): rx queues configuration\n"
1109                 "  --no-numa: optional, disable numa awareness\n"
1110                 "  --enable-jumbo: enable jumbo frame"
1111                 " which max packet len is PKTLEN in decimal (64-9600)\n",
1112                 prgname);
1113 }
1114
1115 static int parse_max_pkt_len(const char *pktlen)
1116 {
1117         char *end = NULL;
1118         unsigned long len;
1119
1120         /* parse decimal string */
1121         len = strtoul(pktlen, &end, 10);
1122         if ((pktlen[0] == '\0') || (end == NULL) || (*end != '\0'))
1123                 return -1;
1124
1125         if (len == 0)
1126                 return -1;
1127
1128         return len;
1129 }
1130
1131 static int
1132 parse_portmask(const char *portmask)
1133 {
1134         char *end = NULL;
1135         unsigned long pm;
1136
1137         /* parse hexadecimal string */
1138         pm = strtoul(portmask, &end, 16);
1139         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
1140                 return -1;
1141
1142         if (pm == 0)
1143                 return -1;
1144
1145         return pm;
1146 }
1147
1148 static int
1149 parse_config(const char *q_arg)
1150 {
1151         char s[256];
1152         const char *p, *p0 = q_arg;
1153         char *end;
1154         enum fieldnames {
1155                 FLD_PORT = 0,
1156                 FLD_QUEUE,
1157                 FLD_LCORE,
1158                 _NUM_FLD
1159         };
1160         unsigned long int_fld[_NUM_FLD];
1161         char *str_fld[_NUM_FLD];
1162         int i;
1163         unsigned size;
1164
1165         nb_lcore_params = 0;
1166
1167         while ((p = strchr(p0,'(')) != NULL) {
1168                 ++p;
1169                 if((p0 = strchr(p,')')) == NULL)
1170                         return -1;
1171
1172                 size = p0 - p;
1173                 if(size >= sizeof(s))
1174                         return -1;
1175
1176                 snprintf(s, sizeof(s), "%.*s", size, p);
1177                 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') !=
1178                                                                 _NUM_FLD)
1179                         return -1;
1180                 for (i = 0; i < _NUM_FLD; i++){
1181                         errno = 0;
1182                         int_fld[i] = strtoul(str_fld[i], &end, 0);
1183                         if (errno != 0 || end == str_fld[i] || int_fld[i] >
1184                                                                         255)
1185                                 return -1;
1186                 }
1187                 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
1188                         printf("exceeded max number of lcore params: %hu\n",
1189                                 nb_lcore_params);
1190                         return -1;
1191                 }
1192                 lcore_params_array[nb_lcore_params].port_id =
1193                                 (uint8_t)int_fld[FLD_PORT];
1194                 lcore_params_array[nb_lcore_params].queue_id =
1195                                 (uint8_t)int_fld[FLD_QUEUE];
1196                 lcore_params_array[nb_lcore_params].lcore_id =
1197                                 (uint8_t)int_fld[FLD_LCORE];
1198                 ++nb_lcore_params;
1199         }
1200         lcore_params = lcore_params_array;
1201
1202         return 0;
1203 }
1204
1205 /* Parse the argument given in the command line of the application */
1206 static int
1207 parse_args(int argc, char **argv)
1208 {
1209         int opt, ret;
1210         char **argvopt;
1211         int option_index;
1212         char *prgname = argv[0];
1213         static struct option lgopts[] = {
1214                 {"config", 1, 0, 0},
1215                 {"no-numa", 0, 0, 0},
1216                 {"enable-jumbo", 0, 0, 0},
1217                 {NULL, 0, 0, 0}
1218         };
1219
1220         argvopt = argv;
1221
1222         while ((opt = getopt_long(argc, argvopt, "p:P",
1223                                 lgopts, &option_index)) != EOF) {
1224
1225                 switch (opt) {
1226                 /* portmask */
1227                 case 'p':
1228                         enabled_port_mask = parse_portmask(optarg);
1229                         if (enabled_port_mask == 0) {
1230                                 printf("invalid portmask\n");
1231                                 print_usage(prgname);
1232                                 return -1;
1233                         }
1234                         break;
1235                 case 'P':
1236                         printf("Promiscuous mode selected\n");
1237                         promiscuous_on = 1;
1238                         break;
1239
1240                 /* long options */
1241                 case 0:
1242                         if (!strncmp(lgopts[option_index].name, "config", 6)) {
1243                                 ret = parse_config(optarg);
1244                                 if (ret) {
1245                                         printf("invalid config\n");
1246                                         print_usage(prgname);
1247                                         return -1;
1248                                 }
1249                         }
1250
1251                         if (!strncmp(lgopts[option_index].name,
1252                                                 "no-numa", 7)) {
1253                                 printf("numa is disabled \n");
1254                                 numa_on = 0;
1255                         }
1256
1257                         if (!strncmp(lgopts[option_index].name,
1258                                         "enable-jumbo", 12)) {
1259                                 struct option lenopts =
1260                                         {"max-pkt-len", required_argument, \
1261                                                                         0, 0};
1262
1263                                 printf("jumbo frame is enabled \n");
1264                                 port_conf.rxmode.jumbo_frame = 1;
1265
1266                                 /**
1267                                  * if no max-pkt-len set, use the default value
1268                                  * ETHER_MAX_LEN
1269                                  */
1270                                 if (0 == getopt_long(argc, argvopt, "",
1271                                                 &lenopts, &option_index)) {
1272                                         ret = parse_max_pkt_len(optarg);
1273                                         if ((ret < 64) ||
1274                                                 (ret > MAX_JUMBO_PKT_LEN)){
1275                                                 printf("invalid packet "
1276                                                                 "length\n");
1277                                                 print_usage(prgname);
1278                                                 return -1;
1279                                         }
1280                                         port_conf.rxmode.max_rx_pkt_len = ret;
1281                                 }
1282                                 printf("set jumbo frame "
1283                                         "max packet length to %u\n",
1284                                 (unsigned int)port_conf.rxmode.max_rx_pkt_len);
1285                         }
1286
1287                         break;
1288
1289                 default:
1290                         print_usage(prgname);
1291                         return -1;
1292                 }
1293         }
1294
1295         if (optind >= 0)
1296                 argv[optind-1] = prgname;
1297
1298         ret = optind-1;
1299         optind = 0; /* reset getopt lib */
1300         return ret;
1301 }
1302
1303 static void
1304 print_ethaddr(const char *name, const struct ether_addr *eth_addr)
1305 {
1306         char buf[ETHER_ADDR_FMT_SIZE];
1307         ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr);
1308         printf("%s%s", name, buf);
1309 }
1310
1311 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1312 static void
1313 setup_hash(int socketid)
1314 {
1315         struct rte_hash_parameters ipv4_l3fwd_hash_params = {
1316                 .name = NULL,
1317                 .entries = L3FWD_HASH_ENTRIES,
1318                 .key_len = sizeof(struct ipv4_5tuple),
1319                 .hash_func = DEFAULT_HASH_FUNC,
1320                 .hash_func_init_val = 0,
1321         };
1322
1323         struct rte_hash_parameters ipv6_l3fwd_hash_params = {
1324                 .name = NULL,
1325                 .entries = L3FWD_HASH_ENTRIES,
1326                 .key_len = sizeof(struct ipv6_5tuple),
1327                 .hash_func = DEFAULT_HASH_FUNC,
1328                 .hash_func_init_val = 0,
1329         };
1330
1331         unsigned i;
1332         int ret;
1333         char s[64];
1334
1335         /* create ipv4 hash */
1336         snprintf(s, sizeof(s), "ipv4_l3fwd_hash_%d", socketid);
1337         ipv4_l3fwd_hash_params.name = s;
1338         ipv4_l3fwd_hash_params.socket_id = socketid;
1339         ipv4_l3fwd_lookup_struct[socketid] =
1340                 rte_hash_create(&ipv4_l3fwd_hash_params);
1341         if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
1342                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
1343                                 "socket %d\n", socketid);
1344
1345         /* create ipv6 hash */
1346         snprintf(s, sizeof(s), "ipv6_l3fwd_hash_%d", socketid);
1347         ipv6_l3fwd_hash_params.name = s;
1348         ipv6_l3fwd_hash_params.socket_id = socketid;
1349         ipv6_l3fwd_lookup_struct[socketid] =
1350                 rte_hash_create(&ipv6_l3fwd_hash_params);
1351         if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
1352                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
1353                                 "socket %d\n", socketid);
1354
1355
1356         /* populate the ipv4 hash */
1357         for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
1358                 ret = rte_hash_add_key (ipv4_l3fwd_lookup_struct[socketid],
1359                                 (void *) &ipv4_l3fwd_route_array[i].key);
1360                 if (ret < 0) {
1361                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
1362                                 "l3fwd hash on socket %d\n", i, socketid);
1363                 }
1364                 ipv4_l3fwd_out_if[ret] = ipv4_l3fwd_route_array[i].if_out;
1365                 printf("Hash: Adding key\n");
1366                 print_ipv4_key(ipv4_l3fwd_route_array[i].key);
1367         }
1368
1369         /* populate the ipv6 hash */
1370         for (i = 0; i < IPV6_L3FWD_NUM_ROUTES; i++) {
1371                 ret = rte_hash_add_key (ipv6_l3fwd_lookup_struct[socketid],
1372                                 (void *) &ipv6_l3fwd_route_array[i].key);
1373                 if (ret < 0) {
1374                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
1375                                 "l3fwd hash on socket %d\n", i, socketid);
1376                 }
1377                 ipv6_l3fwd_out_if[ret] = ipv6_l3fwd_route_array[i].if_out;
1378                 printf("Hash: Adding key\n");
1379                 print_ipv6_key(ipv6_l3fwd_route_array[i].key);
1380         }
1381 }
1382 #endif
1383
1384 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1385 static void
1386 setup_lpm(int socketid)
1387 {
1388         unsigned i;
1389         int ret;
1390         char s[64];
1391
1392         /* create the LPM table */
1393         struct rte_lpm_config lpm_ipv4_config;
1394
1395         lpm_ipv4_config.max_rules = IPV4_L3FWD_LPM_MAX_RULES;
1396         lpm_ipv4_config.number_tbl8s = 256;
1397         lpm_ipv4_config.flags = 0;
1398
1399         snprintf(s, sizeof(s), "IPV4_L3FWD_LPM_%d", socketid);
1400         ipv4_l3fwd_lookup_struct[socketid] =
1401                         rte_lpm_create(s, socketid, &lpm_ipv4_config);
1402         if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
1403                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
1404                                 " on socket %d\n", socketid);
1405
1406         /* populate the LPM table */
1407         for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
1408                 ret = rte_lpm_add(ipv4_l3fwd_lookup_struct[socketid],
1409                         ipv4_l3fwd_route_array[i].ip,
1410                         ipv4_l3fwd_route_array[i].depth,
1411                         ipv4_l3fwd_route_array[i].if_out);
1412
1413                 if (ret < 0) {
1414                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
1415                                 "l3fwd LPM table on socket %d\n",
1416                                 i, socketid);
1417                 }
1418
1419                 printf("LPM: Adding route 0x%08x / %d (%d)\n",
1420                         (unsigned)ipv4_l3fwd_route_array[i].ip,
1421                         ipv4_l3fwd_route_array[i].depth,
1422                         ipv4_l3fwd_route_array[i].if_out);
1423         }
1424 }
1425 #endif
1426
1427 static int
1428 init_mem(unsigned nb_mbuf)
1429 {
1430         struct lcore_conf *qconf;
1431         int socketid;
1432         unsigned lcore_id;
1433         char s[64];
1434
1435         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1436                 if (rte_lcore_is_enabled(lcore_id) == 0)
1437                         continue;
1438
1439                 if (numa_on)
1440                         socketid = rte_lcore_to_socket_id(lcore_id);
1441                 else
1442                         socketid = 0;
1443
1444                 if (socketid >= NB_SOCKETS) {
1445                         rte_exit(EXIT_FAILURE, "Socket %d of lcore %u is "
1446                                         "out of range %d\n", socketid,
1447                                                 lcore_id, NB_SOCKETS);
1448                 }
1449                 if (pktmbuf_pool[socketid] == NULL) {
1450                         snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
1451                         pktmbuf_pool[socketid] =
1452                                 rte_pktmbuf_pool_create(s, nb_mbuf,
1453                                         MEMPOOL_CACHE_SIZE, 0,
1454                                         RTE_MBUF_DEFAULT_BUF_SIZE,
1455                                         socketid);
1456                         if (pktmbuf_pool[socketid] == NULL)
1457                                 rte_exit(EXIT_FAILURE,
1458                                         "Cannot init mbuf pool on socket %d\n",
1459                                                                 socketid);
1460                         else
1461                                 printf("Allocated mbuf pool on socket %d\n",
1462                                                                 socketid);
1463
1464 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1465                         setup_lpm(socketid);
1466 #else
1467                         setup_hash(socketid);
1468 #endif
1469                 }
1470                 qconf = &lcore_conf[lcore_id];
1471                 qconf->ipv4_lookup_struct = ipv4_l3fwd_lookup_struct[socketid];
1472 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1473                 qconf->ipv6_lookup_struct = ipv6_l3fwd_lookup_struct[socketid];
1474 #endif
1475         }
1476         return 0;
1477 }
1478
1479 /* Check the link status of all ports in up to 9s, and print them finally */
1480 static void
1481 check_all_ports_link_status(uint8_t port_num, uint32_t port_mask)
1482 {
1483 #define CHECK_INTERVAL 100 /* 100ms */
1484 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
1485         uint8_t portid, count, all_ports_up, print_flag = 0;
1486         struct rte_eth_link link;
1487
1488         printf("\nChecking link status");
1489         fflush(stdout);
1490         for (count = 0; count <= MAX_CHECK_TIME; count++) {
1491                 all_ports_up = 1;
1492                 for (portid = 0; portid < port_num; portid++) {
1493                         if ((port_mask & (1 << portid)) == 0)
1494                                 continue;
1495                         memset(&link, 0, sizeof(link));
1496                         rte_eth_link_get_nowait(portid, &link);
1497                         /* print link status if flag set */
1498                         if (print_flag == 1) {
1499                                 if (link.link_status)
1500                                         printf("Port %d Link Up - speed %u "
1501                                                 "Mbps - %s\n", (uint8_t)portid,
1502                                                 (unsigned)link.link_speed,
1503                                 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
1504                                         ("full-duplex") : ("half-duplex\n"));
1505                                 else
1506                                         printf("Port %d Link Down\n",
1507                                                 (uint8_t)portid);
1508                                 continue;
1509                         }
1510                         /* clear all_ports_up flag if any link down */
1511                         if (link.link_status == 0) {
1512                                 all_ports_up = 0;
1513                                 break;
1514                         }
1515                 }
1516                 /* after finally printing all link status, get out */
1517                 if (print_flag == 1)
1518                         break;
1519
1520                 if (all_ports_up == 0) {
1521                         printf(".");
1522                         fflush(stdout);
1523                         rte_delay_ms(CHECK_INTERVAL);
1524                 }
1525
1526                 /* set the print_flag if all ports up or timeout */
1527                 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
1528                         print_flag = 1;
1529                         printf("done\n");
1530                 }
1531         }
1532 }
1533
1534 int
1535 main(int argc, char **argv)
1536 {
1537         struct lcore_conf *qconf;
1538         struct rte_eth_dev_info dev_info;
1539         struct rte_eth_txconf *txconf;
1540         int ret;
1541         unsigned nb_ports;
1542         uint16_t queueid;
1543         unsigned lcore_id;
1544         uint64_t hz;
1545         uint32_t n_tx_queue, nb_lcores;
1546         uint32_t dev_rxq_num, dev_txq_num;
1547         uint8_t portid, nb_rx_queue, queue, socketid;
1548         uint8_t nb_tx_port;
1549
1550         /* catch SIGINT and restore cpufreq governor to ondemand */
1551         signal(SIGINT, signal_exit_now);
1552
1553         /* init EAL */
1554         ret = rte_eal_init(argc, argv);
1555         if (ret < 0)
1556                 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
1557         argc -= ret;
1558         argv += ret;
1559
1560         /* init RTE timer library to be used late */
1561         rte_timer_subsystem_init();
1562
1563         /* parse application arguments (after the EAL ones) */
1564         ret = parse_args(argc, argv);
1565         if (ret < 0)
1566                 rte_exit(EXIT_FAILURE, "Invalid L3FWD parameters\n");
1567
1568         if (check_lcore_params() < 0)
1569                 rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
1570
1571         ret = init_lcore_rx_queues();
1572         if (ret < 0)
1573                 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
1574
1575
1576         nb_ports = rte_eth_dev_count();
1577         if (nb_ports > RTE_MAX_ETHPORTS)
1578                 nb_ports = RTE_MAX_ETHPORTS;
1579
1580         if (check_port_config(nb_ports) < 0)
1581                 rte_exit(EXIT_FAILURE, "check_port_config failed\n");
1582
1583         nb_lcores = rte_lcore_count();
1584         nb_tx_port = 0;
1585
1586         /* initialize all ports */
1587         for (portid = 0; portid < nb_ports; portid++) {
1588                 /* skip ports that are not enabled */
1589                 if ((enabled_port_mask & (1 << portid)) == 0) {
1590                         printf("\nSkipping disabled port %d\n", portid);
1591                         continue;
1592                 }
1593
1594                 /* init port */
1595                 printf("Initializing port %d ... ", portid );
1596                 fflush(stdout);
1597
1598                 rte_eth_dev_info_get(portid, &dev_info);
1599                 dev_rxq_num = dev_info.max_rx_queues;
1600                 dev_txq_num = dev_info.max_tx_queues;
1601
1602                 nb_rx_queue = get_port_n_rx_queues(portid);
1603                 if (nb_rx_queue > dev_rxq_num)
1604                         rte_exit(EXIT_FAILURE,
1605                                 "Cannot configure not existed rxq: "
1606                                 "port=%d\n", portid);
1607
1608                 n_tx_queue = nb_lcores;
1609                 if (n_tx_queue > dev_txq_num)
1610                         n_tx_queue = dev_txq_num;
1611                 printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
1612                         nb_rx_queue, (unsigned)n_tx_queue );
1613                 ret = rte_eth_dev_configure(portid, nb_rx_queue,
1614                                         (uint16_t)n_tx_queue, &port_conf);
1615                 if (ret < 0)
1616                         rte_exit(EXIT_FAILURE, "Cannot configure device: "
1617                                         "err=%d, port=%d\n", ret, portid);
1618
1619                 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
1620                 print_ethaddr(" Address:", &ports_eth_addr[portid]);
1621                 printf(", ");
1622
1623                 /* init memory */
1624                 ret = init_mem(NB_MBUF);
1625                 if (ret < 0)
1626                         rte_exit(EXIT_FAILURE, "init_mem failed\n");
1627
1628                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1629                         if (rte_lcore_is_enabled(lcore_id) == 0)
1630                                 continue;
1631
1632                         /* Initialize TX buffers */
1633                         qconf = &lcore_conf[lcore_id];
1634                         qconf->tx_buffer[portid] = rte_zmalloc_socket("tx_buffer",
1635                                 RTE_ETH_TX_BUFFER_SIZE(MAX_PKT_BURST), 0,
1636                                 rte_eth_dev_socket_id(portid));
1637                         if (qconf->tx_buffer[portid] == NULL)
1638                                 rte_exit(EXIT_FAILURE, "Can't allocate tx buffer for port %u\n",
1639                                                 (unsigned) portid);
1640
1641                         rte_eth_tx_buffer_init(qconf->tx_buffer[portid], MAX_PKT_BURST);
1642                 }
1643
1644                 /* init one TX queue per couple (lcore,port) */
1645                 queueid = 0;
1646                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1647                         if (rte_lcore_is_enabled(lcore_id) == 0)
1648                                 continue;
1649
1650                         if (queueid >= dev_txq_num)
1651                                 continue;
1652
1653                         if (numa_on)
1654                                 socketid = \
1655                                 (uint8_t)rte_lcore_to_socket_id(lcore_id);
1656                         else
1657                                 socketid = 0;
1658
1659                         printf("txq=%u,%d,%d ", lcore_id, queueid, socketid);
1660                         fflush(stdout);
1661
1662                         rte_eth_dev_info_get(portid, &dev_info);
1663                         txconf = &dev_info.default_txconf;
1664                         if (port_conf.rxmode.jumbo_frame)
1665                                 txconf->txq_flags = 0;
1666                         ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
1667                                                      socketid, txconf);
1668                         if (ret < 0)
1669                                 rte_exit(EXIT_FAILURE,
1670                                         "rte_eth_tx_queue_setup: err=%d, "
1671                                                 "port=%d\n", ret, portid);
1672
1673                         qconf = &lcore_conf[lcore_id];
1674                         qconf->tx_queue_id[portid] = queueid;
1675                         queueid++;
1676
1677                         qconf->n_tx_port = nb_tx_port;
1678                         qconf->tx_port_id[qconf->n_tx_port] = portid;
1679                 }
1680                 printf("\n");
1681
1682                 nb_tx_port++;
1683         }
1684
1685         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1686                 if (rte_lcore_is_enabled(lcore_id) == 0)
1687                         continue;
1688
1689                 /* init power management library */
1690                 ret = rte_power_init(lcore_id);
1691                 if (ret)
1692                         RTE_LOG(ERR, POWER,
1693                                 "Library initialization failed on core %u\n", lcore_id);
1694
1695                 /* init timer structures for each enabled lcore */
1696                 rte_timer_init(&power_timers[lcore_id]);
1697                 hz = rte_get_timer_hz();
1698                 rte_timer_reset(&power_timers[lcore_id],
1699                         hz/TIMER_NUMBER_PER_SECOND, SINGLE, lcore_id,
1700                                                 power_timer_cb, NULL);
1701
1702                 qconf = &lcore_conf[lcore_id];
1703                 printf("\nInitializing rx queues on lcore %u ... ", lcore_id );
1704                 fflush(stdout);
1705                 /* init RX queues */
1706                 for(queue = 0; queue < qconf->n_rx_queue; ++queue) {
1707                         portid = qconf->rx_queue_list[queue].port_id;
1708                         queueid = qconf->rx_queue_list[queue].queue_id;
1709
1710                         if (numa_on)
1711                                 socketid = \
1712                                 (uint8_t)rte_lcore_to_socket_id(lcore_id);
1713                         else
1714                                 socketid = 0;
1715
1716                         printf("rxq=%d,%d,%d ", portid, queueid, socketid);
1717                         fflush(stdout);
1718
1719                         ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
1720                                 socketid, NULL,
1721                                 pktmbuf_pool[socketid]);
1722                         if (ret < 0)
1723                                 rte_exit(EXIT_FAILURE,
1724                                         "rte_eth_rx_queue_setup: err=%d, "
1725                                                 "port=%d\n", ret, portid);
1726                 }
1727         }
1728
1729         printf("\n");
1730
1731         /* start ports */
1732         for (portid = 0; portid < nb_ports; portid++) {
1733                 if ((enabled_port_mask & (1 << portid)) == 0) {
1734                         continue;
1735                 }
1736                 /* Start device */
1737                 ret = rte_eth_dev_start(portid);
1738                 if (ret < 0)
1739                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, "
1740                                                 "port=%d\n", ret, portid);
1741                 /*
1742                  * If enabled, put device in promiscuous mode.
1743                  * This allows IO forwarding mode to forward packets
1744                  * to itself through 2 cross-connected  ports of the
1745                  * target machine.
1746                  */
1747                 if (promiscuous_on)
1748                         rte_eth_promiscuous_enable(portid);
1749                 /* initialize spinlock for each port */
1750                 rte_spinlock_init(&(locks[portid]));
1751         }
1752
1753         check_all_ports_link_status((uint8_t)nb_ports, enabled_port_mask);
1754
1755         /* launch per-lcore init on every lcore */
1756         rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
1757         RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1758                 if (rte_eal_wait_lcore(lcore_id) < 0)
1759                         return -1;
1760         }
1761
1762         return 0;
1763 }