examples/l3fwd-power/main.c

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