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 1024
 116 #define RTE_TEST_TX_DESC_DEFAULT 1024
 117 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
 118 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
 119
 120 /* ethernet addresses of ports */
 121 static struct ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
 122
 123 /* ethernet addresses of ports */
 124 static rte_spinlock_t locks[RTE_MAX_ETHPORTS];
 125
 126 /* mask of enabled ports */
 127 static uint32_t enabled_port_mask = 0;
 128 /* Ports set in promiscuous mode off by default. */
 129 static int promiscuous_on = 0;
 130 /* NUMA is enabled by default. */
 131 static int numa_on = 1;
 132 static int parse_ptype; /**< Parse packet type using rx callback, and */
 133                         /**< disabled by default */
 134
 135 enum freq_scale_hint_t
 136 {
 137         FREQ_LOWER    =      -1,
 138         FREQ_CURRENT  =       0,
 139         FREQ_HIGHER   =       1,
 140         FREQ_HIGHEST  =       2
 141 };
 142
 143 struct lcore_rx_queue {
 144         uint16_t port_id;
 145         uint8_t queue_id;
 146         enum freq_scale_hint_t freq_up_hint;
 147         uint32_t zero_rx_packet_count;
 148         uint32_t idle_hint;
 149 } __rte_cache_aligned;
 150
 151 #define MAX_RX_QUEUE_PER_LCORE 16
 152 #define MAX_TX_QUEUE_PER_PORT RTE_MAX_ETHPORTS
 153 #define MAX_RX_QUEUE_PER_PORT 128
 154
 155 #define MAX_RX_QUEUE_INTERRUPT_PER_PORT 16
 156
 157
 158 #define MAX_LCORE_PARAMS 1024
 159 struct lcore_params {
 160         uint16_t port_id;
 161         uint8_t queue_id;
 162         uint8_t lcore_id;
 163 } __rte_cache_aligned;
 164
 165 static struct lcore_params lcore_params_array[MAX_LCORE_PARAMS];
 166 static struct lcore_params lcore_params_array_default[] = {
 167         {0, 0, 2},
 168         {0, 1, 2},
 169         {0, 2, 2},
 170         {1, 0, 2},
 171         {1, 1, 2},
 172         {1, 2, 2},
 173         {2, 0, 2},
 174         {3, 0, 3},
 175         {3, 1, 3},
 176 };
 177
 178 static struct lcore_params * lcore_params = lcore_params_array_default;
 179 static uint16_t nb_lcore_params = sizeof(lcore_params_array_default) /
 180                                 sizeof(lcore_params_array_default[0]);
 181
 182 static struct rte_eth_conf port_conf = {
 183         .rxmode = {
 184                 .mq_mode        = ETH_MQ_RX_RSS,
 185                 .max_rx_pkt_len = ETHER_MAX_LEN,
 186                 .split_hdr_size = 0,
 187                 .offloads = (DEV_RX_OFFLOAD_CRC_STRIP |
 188                              DEV_RX_OFFLOAD_CHECKSUM),
 189         },
 190         .rx_adv_conf = {
 191                 .rss_conf = {
 192                         .rss_key = NULL,
 193                         .rss_hf = ETH_RSS_UDP,
 194                 },
 195         },
 196         .txmode = {
 197                 .mq_mode = ETH_MQ_TX_NONE,
 198         },
 199         .intr_conf = {
 200                 .rxq = 1,
 201         },
 202 };
 203
 204 static struct rte_mempool * pktmbuf_pool[NB_SOCKETS];
 205
 206
 207 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 208
 209 #ifdef RTE_ARCH_X86
 210 #include <rte_hash_crc.h>
 211 #define DEFAULT_HASH_FUNC       rte_hash_crc
 212 #else
 213 #include <rte_jhash.h>
 214 #define DEFAULT_HASH_FUNC       rte_jhash
 215 #endif
 216
 217 struct ipv4_5tuple {
 218         uint32_t ip_dst;
 219         uint32_t ip_src;
 220         uint16_t port_dst;
 221         uint16_t port_src;
 222         uint8_t  proto;
 223 } __attribute__((__packed__));
 224
 225 struct ipv6_5tuple {
 226         uint8_t  ip_dst[IPV6_ADDR_LEN];
 227         uint8_t  ip_src[IPV6_ADDR_LEN];
 228         uint16_t port_dst;
 229         uint16_t port_src;
 230         uint8_t  proto;
 231 } __attribute__((__packed__));
 232
 233 struct ipv4_l3fwd_route {
 234         struct ipv4_5tuple key;
 235         uint8_t if_out;
 236 };
 237
 238 struct ipv6_l3fwd_route {
 239         struct ipv6_5tuple key;
 240         uint8_t if_out;
 241 };
 242
 243 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
 244         {{IPv4(100,10,0,1), IPv4(200,10,0,1), 101, 11, IPPROTO_TCP}, 0},
 245         {{IPv4(100,20,0,2), IPv4(200,20,0,2), 102, 12, IPPROTO_TCP}, 1},
 246         {{IPv4(100,30,0,3), IPv4(200,30,0,3), 103, 13, IPPROTO_TCP}, 2},
 247         {{IPv4(100,40,0,4), IPv4(200,40,0,4), 104, 14, IPPROTO_TCP}, 3},
 248 };
 249
 250 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
 251         {
 252                 {
 253                         {0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
 254                          0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
 255                         {0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
 256                          0x02, 0x1e, 0x67, 0xff, 0xfe, 0x0d, 0xb6, 0x0a},
 257                          1, 10, IPPROTO_UDP
 258                 }, 4
 259         },
 260 };
 261
 262 typedef struct rte_hash lookup_struct_t;
 263 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
 264 static lookup_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
 265
 266 #define L3FWD_HASH_ENTRIES      1024
 267
 268 #define IPV4_L3FWD_NUM_ROUTES \
 269         (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
 270
 271 #define IPV6_L3FWD_NUM_ROUTES \
 272         (sizeof(ipv6_l3fwd_route_array) / sizeof(ipv6_l3fwd_route_array[0]))
 273
 274 static uint16_t ipv4_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 275 static uint16_t ipv6_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 276 #endif
 277
 278 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 279 struct ipv4_l3fwd_route {
 280         uint32_t ip;
 281         uint8_t  depth;
 282         uint8_t  if_out;
 283 };
 284
 285 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
 286         {IPv4(1,1,1,0), 24, 0},
 287         {IPv4(2,1,1,0), 24, 1},
 288         {IPv4(3,1,1,0), 24, 2},
 289         {IPv4(4,1,1,0), 24, 3},
 290         {IPv4(5,1,1,0), 24, 4},
 291         {IPv4(6,1,1,0), 24, 5},
 292         {IPv4(7,1,1,0), 24, 6},
 293         {IPv4(8,1,1,0), 24, 7},
 294 };
 295
 296 #define IPV4_L3FWD_NUM_ROUTES \
 297         (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
 298
 299 #define IPV4_L3FWD_LPM_MAX_RULES     1024
 300
 301 typedef struct rte_lpm lookup_struct_t;
 302 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
 303 #endif
 304
 305 struct lcore_conf {
 306         uint16_t n_rx_queue;
 307         struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
 308         uint16_t n_tx_port;
 309         uint16_t tx_port_id[RTE_MAX_ETHPORTS];
 310         uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
 311         struct rte_eth_dev_tx_buffer *tx_buffer[RTE_MAX_ETHPORTS];
 312         lookup_struct_t * ipv4_lookup_struct;
 313         lookup_struct_t * ipv6_lookup_struct;
 314 } __rte_cache_aligned;
 315
 316 struct lcore_stats {
 317         /* total sleep time in ms since last frequency scaling down */
 318         uint32_t sleep_time;
 319         /* number of long sleep recently */
 320         uint32_t nb_long_sleep;
 321         /* freq. scaling up trend */
 322         uint32_t trend;
 323         /* total packet processed recently */
 324         uint64_t nb_rx_processed;
 325         /* total iterations looped recently */
 326         uint64_t nb_iteration_looped;
 327         uint32_t padding[9];
 328 } __rte_cache_aligned;
 329
 330 static struct lcore_conf lcore_conf[RTE_MAX_LCORE] __rte_cache_aligned;
 331 static struct lcore_stats stats[RTE_MAX_LCORE] __rte_cache_aligned;
 332 static struct rte_timer power_timers[RTE_MAX_LCORE];
 333
 334 static inline uint32_t power_idle_heuristic(uint32_t zero_rx_packet_count);
 335 static inline enum freq_scale_hint_t power_freq_scaleup_heuristic( \
 336                 unsigned int lcore_id, uint16_t port_id, uint16_t queue_id);
 337
 338 /* exit signal handler */
 339 static void
 340 signal_exit_now(int sigtype)
 341 {
 342         unsigned lcore_id;
 343         unsigned int portid;
 344         int ret;
 345
 346         if (sigtype == SIGINT) {
 347                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
 348                         if (rte_lcore_is_enabled(lcore_id) == 0)
 349                                 continue;
 350
 351                         /* init power management library */
 352                         ret = rte_power_exit(lcore_id);
 353                         if (ret)
 354                                 rte_exit(EXIT_FAILURE, "Power management "
 355                                         "library de-initialization failed on "
 356                                                         "core%u\n", lcore_id);
 357                 }
 358
 359                 RTE_ETH_FOREACH_DEV(portid) {
 360                         if ((enabled_port_mask & (1 << portid)) == 0)
 361                                 continue;
 362
 363                         rte_eth_dev_stop(portid);
 364                         rte_eth_dev_close(portid);
 365                 }
 366         }
 367
 368         rte_exit(EXIT_SUCCESS, "User forced exit\n");
 369 }
 370
 371 /*  Freqency scale down timer callback */
 372 static void
 373 power_timer_cb(__attribute__((unused)) struct rte_timer *tim,
 374                           __attribute__((unused)) void *arg)
 375 {
 376         uint64_t hz;
 377         float sleep_time_ratio;
 378         unsigned lcore_id = rte_lcore_id();
 379
 380         /* accumulate total execution time in us when callback is invoked */
 381         sleep_time_ratio = (float)(stats[lcore_id].sleep_time) /
 382                                         (float)SCALING_PERIOD;
 383         /**
 384          * check whether need to scale down frequency a step if it sleep a lot.
 385          */
 386         if (sleep_time_ratio >= SCALING_DOWN_TIME_RATIO_THRESHOLD) {
 387                 if (rte_power_freq_down)
 388                         rte_power_freq_down(lcore_id);
 389         }
 390         else if ( (unsigned)(stats[lcore_id].nb_rx_processed /
 391                 stats[lcore_id].nb_iteration_looped) < MAX_PKT_BURST) {
 392                 /**
 393                  * scale down a step if average packet per iteration less
 394                  * than expectation.
 395                  */
 396                 if (rte_power_freq_down)
 397                         rte_power_freq_down(lcore_id);
 398         }
 399
 400         /**
 401          * initialize another timer according to current frequency to ensure
 402          * timer interval is relatively fixed.
 403          */
 404         hz = rte_get_timer_hz();
 405         rte_timer_reset(&power_timers[lcore_id], hz/TIMER_NUMBER_PER_SECOND,
 406                                 SINGLE, lcore_id, power_timer_cb, NULL);
 407
 408         stats[lcore_id].nb_rx_processed = 0;
 409         stats[lcore_id].nb_iteration_looped = 0;
 410
 411         stats[lcore_id].sleep_time = 0;
 412 }
 413
 414 /* Enqueue a single packet, and send burst if queue is filled */
 415 static inline int
 416 send_single_packet(struct rte_mbuf *m, uint16_t port)
 417 {
 418         uint32_t lcore_id;
 419         struct lcore_conf *qconf;
 420
 421         lcore_id = rte_lcore_id();
 422         qconf = &lcore_conf[lcore_id];
 423
 424         rte_eth_tx_buffer(port, qconf->tx_queue_id[port],
 425                         qconf->tx_buffer[port], m);
 426
 427         return 0;
 428 }
 429
 430 #ifdef DO_RFC_1812_CHECKS
 431 static inline int
 432 is_valid_ipv4_pkt(struct ipv4_hdr *pkt, uint32_t link_len)
 433 {
 434         /* From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2 */
 435         /*
 436          * 1. The packet length reported by the Link Layer must be large
 437          * enough to hold the minimum length legal IP datagram (20 bytes).
 438          */
 439         if (link_len < sizeof(struct ipv4_hdr))
 440                 return -1;
 441
 442         /* 2. The IP checksum must be correct. */
 443         /* this is checked in H/W */
 444
 445         /*
 446          * 3. The IP version number must be 4. If the version number is not 4
 447          * then the packet may be another version of IP, such as IPng or
 448          * ST-II.
 449          */
 450         if (((pkt->version_ihl) >> 4) != 4)
 451                 return -3;
 452         /*
 453          * 4. The IP header length field must be large enough to hold the
 454          * minimum length legal IP datagram (20 bytes = 5 words).
 455          */
 456         if ((pkt->version_ihl & 0xf) < 5)
 457                 return -4;
 458
 459         /*
 460          * 5. The IP total length field must be large enough to hold the IP
 461          * datagram header, whose length is specified in the IP header length
 462          * field.
 463          */
 464         if (rte_cpu_to_be_16(pkt->total_length) < sizeof(struct ipv4_hdr))
 465                 return -5;
 466
 467         return 0;
 468 }
 469 #endif
 470
 471 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 472 static void
 473 print_ipv4_key(struct ipv4_5tuple key)
 474 {
 475         printf("IP dst = %08x, IP src = %08x, port dst = %d, port src = %d, "
 476                 "proto = %d\n", (unsigned)key.ip_dst, (unsigned)key.ip_src,
 477                                 key.port_dst, key.port_src, key.proto);
 478 }
 479 static void
 480 print_ipv6_key(struct ipv6_5tuple key)
 481 {
 482         printf( "IP dst = " IPv6_BYTES_FMT ", IP src = " IPv6_BYTES_FMT ", "
 483                 "port dst = %d, port src = %d, proto = %d\n",
 484                 IPv6_BYTES(key.ip_dst), IPv6_BYTES(key.ip_src),
 485                 key.port_dst, key.port_src, key.proto);
 486 }
 487
 488 static inline uint16_t
 489 get_ipv4_dst_port(struct ipv4_hdr *ipv4_hdr, uint16_t portid,
 490                 lookup_struct_t * ipv4_l3fwd_lookup_struct)
 491 {
 492         struct ipv4_5tuple key;
 493         struct tcp_hdr *tcp;
 494         struct udp_hdr *udp;
 495         int ret = 0;
 496
 497         key.ip_dst = rte_be_to_cpu_32(ipv4_hdr->dst_addr);
 498         key.ip_src = rte_be_to_cpu_32(ipv4_hdr->src_addr);
 499         key.proto = ipv4_hdr->next_proto_id;
 500
 501         switch (ipv4_hdr->next_proto_id) {
 502         case IPPROTO_TCP:
 503                 tcp = (struct tcp_hdr *)((unsigned char *)ipv4_hdr +
 504                                         sizeof(struct ipv4_hdr));
 505                 key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
 506                 key.port_src = rte_be_to_cpu_16(tcp->src_port);
 507                 break;
 508
 509         case IPPROTO_UDP:
 510                 udp = (struct udp_hdr *)((unsigned char *)ipv4_hdr +
 511                                         sizeof(struct ipv4_hdr));
 512                 key.port_dst = rte_be_to_cpu_16(udp->dst_port);
 513                 key.port_src = rte_be_to_cpu_16(udp->src_port);
 514                 break;
 515
 516         default:
 517                 key.port_dst = 0;
 518                 key.port_src = 0;
 519                 break;
 520         }
 521
 522         /* Find destination port */
 523         ret = rte_hash_lookup(ipv4_l3fwd_lookup_struct, (const void *)&key);
 524         return ((ret < 0) ? portid : ipv4_l3fwd_out_if[ret]);
 525 }
 526
 527 static inline uint16_t
 528 get_ipv6_dst_port(struct ipv6_hdr *ipv6_hdr, uint16_t portid,
 529                         lookup_struct_t *ipv6_l3fwd_lookup_struct)
 530 {
 531         struct ipv6_5tuple key;
 532         struct tcp_hdr *tcp;
 533         struct udp_hdr *udp;
 534         int ret = 0;
 535
 536         memcpy(key.ip_dst, ipv6_hdr->dst_addr, IPV6_ADDR_LEN);
 537         memcpy(key.ip_src, ipv6_hdr->src_addr, IPV6_ADDR_LEN);
 538
 539         key.proto = ipv6_hdr->proto;
 540
 541         switch (ipv6_hdr->proto) {
 542         case IPPROTO_TCP:
 543                 tcp = (struct tcp_hdr *)((unsigned char *) ipv6_hdr +
 544                                         sizeof(struct ipv6_hdr));
 545                 key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
 546                 key.port_src = rte_be_to_cpu_16(tcp->src_port);
 547                 break;
 548
 549         case IPPROTO_UDP:
 550                 udp = (struct udp_hdr *)((unsigned char *) ipv6_hdr +
 551                                         sizeof(struct ipv6_hdr));
 552                 key.port_dst = rte_be_to_cpu_16(udp->dst_port);
 553                 key.port_src = rte_be_to_cpu_16(udp->src_port);
 554                 break;
 555
 556         default:
 557                 key.port_dst = 0;
 558                 key.port_src = 0;
 559                 break;
 560         }
 561
 562         /* Find destination port */
 563         ret = rte_hash_lookup(ipv6_l3fwd_lookup_struct, (const void *)&key);
 564         return ((ret < 0) ? portid : ipv6_l3fwd_out_if[ret]);
 565 }
 566 #endif
 567
 568 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 569 static inline uint16_t
 570 get_ipv4_dst_port(struct ipv4_hdr *ipv4_hdr, uint16_t portid,
 571                 lookup_struct_t *ipv4_l3fwd_lookup_struct)
 572 {
 573         uint32_t next_hop;
 574
 575         return ((rte_lpm_lookup(ipv4_l3fwd_lookup_struct,
 576                         rte_be_to_cpu_32(ipv4_hdr->dst_addr), &next_hop) == 0)?
 577                         next_hop : portid);
 578 }
 579 #endif
 580
 581 static inline void
 582 parse_ptype_one(struct rte_mbuf *m)
 583 {
 584         struct ether_hdr *eth_hdr;
 585         uint32_t packet_type = RTE_PTYPE_UNKNOWN;
 586         uint16_t ether_type;
 587
 588         eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
 589         ether_type = eth_hdr->ether_type;
 590         if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv4))
 591                 packet_type |= RTE_PTYPE_L3_IPV4_EXT_UNKNOWN;
 592         else if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv6))
 593                 packet_type |= RTE_PTYPE_L3_IPV6_EXT_UNKNOWN;
 594
 595         m->packet_type = packet_type;
 596 }
 597
 598 static uint16_t
 599 cb_parse_ptype(uint16_t port __rte_unused, uint16_t queue __rte_unused,
 600                struct rte_mbuf *pkts[], uint16_t nb_pkts,
 601                uint16_t max_pkts __rte_unused,
 602                void *user_param __rte_unused)
 603 {
 604         unsigned int i;
 605
 606         for (i = 0; i < nb_pkts; ++i)
 607                 parse_ptype_one(pkts[i]);
 608
 609         return nb_pkts;
 610 }
 611
 612 static int
 613 add_cb_parse_ptype(uint16_t portid, uint16_t queueid)
 614 {
 615         printf("Port %d: softly parse packet type info\n", portid);
 616         if (rte_eth_add_rx_callback(portid, queueid, cb_parse_ptype, NULL))
 617                 return 0;
 618
 619         printf("Failed to add rx callback: port=%d\n", portid);
 620         return -1;
 621 }
 622
 623 static inline void
 624 l3fwd_simple_forward(struct rte_mbuf *m, uint16_t portid,
 625                                 struct lcore_conf *qconf)
 626 {
 627         struct ether_hdr *eth_hdr;
 628         struct ipv4_hdr *ipv4_hdr;
 629         void *d_addr_bytes;
 630         uint16_t dst_port;
 631
 632         eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
 633
 634         if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) {
 635                 /* Handle IPv4 headers.*/
 636                 ipv4_hdr =
 637                         rte_pktmbuf_mtod_offset(m, struct ipv4_hdr *,
 638                                                 sizeof(struct ether_hdr));
 639
 640 #ifdef DO_RFC_1812_CHECKS
 641                 /* Check to make sure the packet is valid (RFC1812) */
 642                 if (is_valid_ipv4_pkt(ipv4_hdr, m->pkt_len) < 0) {
 643                         rte_pktmbuf_free(m);
 644                         return;
 645                 }
 646 #endif
 647
 648                 dst_port = get_ipv4_dst_port(ipv4_hdr, portid,
 649                                         qconf->ipv4_lookup_struct);
 650                 if (dst_port >= RTE_MAX_ETHPORTS ||
 651                                 (enabled_port_mask & 1 << dst_port) == 0)
 652                         dst_port = portid;
 653
 654                 /* 02:00:00:00:00:xx */
 655                 d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
 656                 *((uint64_t *)d_addr_bytes) =
 657                         0x000000000002 + ((uint64_t)dst_port << 40);
 658
 659 #ifdef DO_RFC_1812_CHECKS
 660                 /* Update time to live and header checksum */
 661                 --(ipv4_hdr->time_to_live);
 662                 ++(ipv4_hdr->hdr_checksum);
 663 #endif
 664
 665                 /* src addr */
 666                 ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
 667
 668                 send_single_packet(m, dst_port);
 669         } else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) {
 670                 /* Handle IPv6 headers.*/
 671 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 672                 struct ipv6_hdr *ipv6_hdr;
 673
 674                 ipv6_hdr =
 675                         rte_pktmbuf_mtod_offset(m, struct ipv6_hdr *,
 676                                                 sizeof(struct ether_hdr));
 677
 678                 dst_port = get_ipv6_dst_port(ipv6_hdr, portid,
 679                                         qconf->ipv6_lookup_struct);
 680
 681                 if (dst_port >= RTE_MAX_ETHPORTS ||
 682                                 (enabled_port_mask & 1 << dst_port) == 0)
 683                         dst_port = portid;
 684
 685                 /* 02:00:00:00:00:xx */
 686                 d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
 687                 *((uint64_t *)d_addr_bytes) =
 688                         0x000000000002 + ((uint64_t)dst_port << 40);
 689
 690                 /* src addr */
 691                 ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
 692
 693                 send_single_packet(m, dst_port);
 694 #else
 695                 /* We don't currently handle IPv6 packets in LPM mode. */
 696                 rte_pktmbuf_free(m);
 697 #endif
 698         } else
 699                 rte_pktmbuf_free(m);
 700
 701 }
 702
 703 #define MINIMUM_SLEEP_TIME         1
 704 #define SUSPEND_THRESHOLD          300
 705
 706 static inline uint32_t
 707 power_idle_heuristic(uint32_t zero_rx_packet_count)
 708 {
 709         /* If zero count is less than 100,  sleep 1us */
 710         if (zero_rx_packet_count < SUSPEND_THRESHOLD)
 711                 return MINIMUM_SLEEP_TIME;
 712         /* If zero count is less than 1000, sleep 100 us which is the
 713                 minimum latency switching from C3/C6 to C0
 714         */
 715         else
 716                 return SUSPEND_THRESHOLD;
 717 }
 718
 719 static inline enum freq_scale_hint_t
 720 power_freq_scaleup_heuristic(unsigned lcore_id,
 721                              uint16_t port_id,
 722                              uint16_t queue_id)
 723 {
 724         uint32_t rxq_count = rte_eth_rx_queue_count(port_id, queue_id);
 725 /**
 726  * HW Rx queue size is 128 by default, Rx burst read at maximum 32 entries
 727  * per iteration
 728  */
 729 #define FREQ_GEAR1_RX_PACKET_THRESHOLD             MAX_PKT_BURST
 730 #define FREQ_GEAR2_RX_PACKET_THRESHOLD             (MAX_PKT_BURST*2)
 731 #define FREQ_GEAR3_RX_PACKET_THRESHOLD             (MAX_PKT_BURST*3)
 732 #define FREQ_UP_TREND1_ACC   1
 733 #define FREQ_UP_TREND2_ACC   100
 734 #define FREQ_UP_THRESHOLD    10000
 735
 736         if (likely(rxq_count > FREQ_GEAR3_RX_PACKET_THRESHOLD)) {
 737                 stats[lcore_id].trend = 0;
 738                 return FREQ_HIGHEST;
 739         } else if (likely(rxq_count > FREQ_GEAR2_RX_PACKET_THRESHOLD))
 740                 stats[lcore_id].trend += FREQ_UP_TREND2_ACC;
 741         else if (likely(rxq_count > FREQ_GEAR1_RX_PACKET_THRESHOLD))
 742                 stats[lcore_id].trend += FREQ_UP_TREND1_ACC;
 743
 744         if (likely(stats[lcore_id].trend > FREQ_UP_THRESHOLD)) {
 745                 stats[lcore_id].trend = 0;
 746                 return FREQ_HIGHER;
 747         }
 748
 749         return FREQ_CURRENT;
 750 }
 751
 752 /**
 753  * force polling thread sleep until one-shot rx interrupt triggers
 754  * @param port_id
 755  *  Port id.
 756  * @param queue_id
 757  *  Rx queue id.
 758  * @return
 759  *  0 on success
 760  */
 761 static int
 762 sleep_until_rx_interrupt(int num)
 763 {
 764         struct rte_epoll_event event[num];
 765         int n, i;
 766         uint16_t port_id;
 767         uint8_t queue_id;
 768         void *data;
 769
 770         RTE_LOG(INFO, L3FWD_POWER,
 771                 "lcore %u sleeps until interrupt triggers\n",
 772                 rte_lcore_id());
 773
 774         n = rte_epoll_wait(RTE_EPOLL_PER_THREAD, event, num, -1);
 775         for (i = 0; i < n; i++) {
 776                 data = event[i].epdata.data;
 777                 port_id = ((uintptr_t)data) >> CHAR_BIT;
 778                 queue_id = ((uintptr_t)data) &
 779                         RTE_LEN2MASK(CHAR_BIT, uint8_t);
 780                 rte_eth_dev_rx_intr_disable(port_id, queue_id);
 781                 RTE_LOG(INFO, L3FWD_POWER,
 782                         "lcore %u is waked up from rx interrupt on"
 783                         " port %d queue %d\n",
 784                         rte_lcore_id(), port_id, queue_id);
 785         }
 786
 787         return 0;
 788 }
 789
 790 static void turn_on_intr(struct lcore_conf *qconf)
 791 {
 792         int i;
 793         struct lcore_rx_queue *rx_queue;
 794         uint8_t queue_id;
 795         uint16_t port_id;
 796
 797         for (i = 0; i < qconf->n_rx_queue; ++i) {
 798                 rx_queue = &(qconf->rx_queue_list[i]);
 799                 port_id = rx_queue->port_id;
 800                 queue_id = rx_queue->queue_id;
 801
 802                 rte_spinlock_lock(&(locks[port_id]));
 803                 rte_eth_dev_rx_intr_enable(port_id, queue_id);
 804                 rte_spinlock_unlock(&(locks[port_id]));
 805         }
 806 }
 807
 808 static int event_register(struct lcore_conf *qconf)
 809 {
 810         struct lcore_rx_queue *rx_queue;
 811         uint8_t queueid;
 812         uint16_t portid;
 813         uint32_t data;
 814         int ret;
 815         int i;
 816
 817         for (i = 0; i < qconf->n_rx_queue; ++i) {
 818                 rx_queue = &(qconf->rx_queue_list[i]);
 819                 portid = rx_queue->port_id;
 820                 queueid = rx_queue->queue_id;
 821                 data = portid << CHAR_BIT | queueid;
 822
 823                 ret = rte_eth_dev_rx_intr_ctl_q(portid, queueid,
 824                                                 RTE_EPOLL_PER_THREAD,
 825                                                 RTE_INTR_EVENT_ADD,
 826                                                 (void *)((uintptr_t)data));
 827                 if (ret)
 828                         return ret;
 829         }
 830
 831         return 0;
 832 }
 833
 834 /* main processing loop */
 835 static int
 836 main_loop(__attribute__((unused)) void *dummy)
 837 {
 838         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
 839         unsigned lcore_id;
 840         uint64_t prev_tsc, diff_tsc, cur_tsc, tim_res_tsc, hz;
 841         uint64_t prev_tsc_power = 0, cur_tsc_power, diff_tsc_power;
 842         int i, j, nb_rx;
 843         uint8_t queueid;
 844         uint16_t portid;
 845         struct lcore_conf *qconf;
 846         struct lcore_rx_queue *rx_queue;
 847         enum freq_scale_hint_t lcore_scaleup_hint;
 848         uint32_t lcore_rx_idle_count = 0;
 849         uint32_t lcore_idle_hint = 0;
 850         int intr_en = 0;
 851
 852         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
 853
 854         prev_tsc = 0;
 855         hz = rte_get_timer_hz();
 856         tim_res_tsc = hz/TIMER_NUMBER_PER_SECOND;
 857
 858         lcore_id = rte_lcore_id();
 859         qconf = &lcore_conf[lcore_id];
 860
 861         if (qconf->n_rx_queue == 0) {
 862                 RTE_LOG(INFO, L3FWD_POWER, "lcore %u has nothing to do\n", lcore_id);
 863                 return 0;
 864         }
 865
 866         RTE_LOG(INFO, L3FWD_POWER, "entering main loop on lcore %u\n", lcore_id);
 867
 868         for (i = 0; i < qconf->n_rx_queue; i++) {
 869                 portid = qconf->rx_queue_list[i].port_id;
 870                 queueid = qconf->rx_queue_list[i].queue_id;
 871                 RTE_LOG(INFO, L3FWD_POWER, " -- lcoreid=%u portid=%u "
 872                         "rxqueueid=%hhu\n", lcore_id, portid, queueid);
 873         }
 874
 875         /* add into event wait list */
 876         if (event_register(qconf) == 0)
 877                 intr_en = 1;
 878         else
 879                 RTE_LOG(INFO, L3FWD_POWER, "RX interrupt won't enable.\n");
 880
 881         while (1) {
 882                 stats[lcore_id].nb_iteration_looped++;
 883
 884                 cur_tsc = rte_rdtsc();
 885                 cur_tsc_power = cur_tsc;
 886
 887                 /*
 888                  * TX burst queue drain
 889                  */
 890                 diff_tsc = cur_tsc - prev_tsc;
 891                 if (unlikely(diff_tsc > drain_tsc)) {
 892                         for (i = 0; i < qconf->n_tx_port; ++i) {
 893                                 portid = qconf->tx_port_id[i];
 894                                 rte_eth_tx_buffer_flush(portid,
 895                                                 qconf->tx_queue_id[portid],
 896                                                 qconf->tx_buffer[portid]);
 897                         }
 898                         prev_tsc = cur_tsc;
 899                 }
 900
 901                 diff_tsc_power = cur_tsc_power - prev_tsc_power;
 902                 if (diff_tsc_power > tim_res_tsc) {
 903                         rte_timer_manage();
 904                         prev_tsc_power = cur_tsc_power;
 905                 }
 906
 907 start_rx:
 908                 /*
 909                  * Read packet from RX queues
 910                  */
 911                 lcore_scaleup_hint = FREQ_CURRENT;
 912                 lcore_rx_idle_count = 0;
 913                 for (i = 0; i < qconf->n_rx_queue; ++i) {
 914                         rx_queue = &(qconf->rx_queue_list[i]);
 915                         rx_queue->idle_hint = 0;
 916                         portid = rx_queue->port_id;
 917                         queueid = rx_queue->queue_id;
 918
 919                         nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
 920                                                                 MAX_PKT_BURST);
 921
 922                         stats[lcore_id].nb_rx_processed += nb_rx;
 923                         if (unlikely(nb_rx == 0)) {
 924                                 /**
 925                                  * no packet received from rx queue, try to
 926                                  * sleep for a while forcing CPU enter deeper
 927                                  * C states.
 928                                  */
 929                                 rx_queue->zero_rx_packet_count++;
 930
 931                                 if (rx_queue->zero_rx_packet_count <=
 932                                                         MIN_ZERO_POLL_COUNT)
 933                                         continue;
 934
 935                                 rx_queue->idle_hint = power_idle_heuristic(\
 936                                         rx_queue->zero_rx_packet_count);
 937                                 lcore_rx_idle_count++;
 938                         } else {
 939                                 rx_queue->zero_rx_packet_count = 0;
 940
 941                                 /**
 942                                  * do not scale up frequency immediately as
 943                                  * user to kernel space communication is costly
 944                                  * which might impact packet I/O for received
 945                                  * packets.
 946                                  */
 947                                 rx_queue->freq_up_hint =
 948                                         power_freq_scaleup_heuristic(lcore_id,
 949                                                         portid, queueid);
 950                         }
 951
 952                         /* Prefetch first packets */
 953                         for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
 954                                 rte_prefetch0(rte_pktmbuf_mtod(
 955                                                 pkts_burst[j], void *));
 956                         }
 957
 958                         /* Prefetch and forward already prefetched packets */
 959                         for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
 960                                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
 961                                                 j + PREFETCH_OFFSET], void *));
 962                                 l3fwd_simple_forward(pkts_burst[j], portid,
 963                                                                 qconf);
 964                         }
 965
 966                         /* Forward remaining prefetched packets */
 967                         for (; j < nb_rx; j++) {
 968                                 l3fwd_simple_forward(pkts_burst[j], portid,
 969                                                                 qconf);
 970                         }
 971                 }
 972
 973                 if (likely(lcore_rx_idle_count != qconf->n_rx_queue)) {
 974                         for (i = 1, lcore_scaleup_hint =
 975                                 qconf->rx_queue_list[0].freq_up_hint;
 976                                         i < qconf->n_rx_queue; ++i) {
 977                                 rx_queue = &(qconf->rx_queue_list[i]);
 978                                 if (rx_queue->freq_up_hint >
 979                                                 lcore_scaleup_hint)
 980                                         lcore_scaleup_hint =
 981                                                 rx_queue->freq_up_hint;
 982                         }
 983
 984                         if (lcore_scaleup_hint == FREQ_HIGHEST) {
 985                                 if (rte_power_freq_max)
 986                                         rte_power_freq_max(lcore_id);
 987                         } else if (lcore_scaleup_hint == FREQ_HIGHER) {
 988                                 if (rte_power_freq_up)
 989                                         rte_power_freq_up(lcore_id);
 990                         }
 991                 } else {
 992                         /**
 993                          * All Rx queues empty in recent consecutive polls,
 994                          * sleep in a conservative manner, meaning sleep as
 995                          * less as possible.
 996                          */
 997                         for (i = 1, lcore_idle_hint =
 998                                 qconf->rx_queue_list[0].idle_hint;
 999                                         i < qconf->n_rx_queue; ++i) {
1000                                 rx_queue = &(qconf->rx_queue_list[i]);
1001                                 if (rx_queue->idle_hint < lcore_idle_hint)
1002                                         lcore_idle_hint = rx_queue->idle_hint;
1003                         }
1004
1005                         if (lcore_idle_hint < SUSPEND_THRESHOLD)
1006                                 /**
1007                                  * execute "pause" instruction to avoid context
1008                                  * switch which generally take hundred of
1009                                  * microseconds for short sleep.
1010                                  */
1011                                 rte_delay_us(lcore_idle_hint);
1012                         else {
1013                                 /* suspend until rx interrupt trigges */
1014                                 if (intr_en) {
1015                                         turn_on_intr(qconf);
1016                                         sleep_until_rx_interrupt(
1017                                                 qconf->n_rx_queue);
1018                                         /**
1019                                          * start receiving packets immediately
1020                                          */
1021                                         goto start_rx;
1022                                 }
1023                         }
1024                         stats[lcore_id].sleep_time += lcore_idle_hint;
1025                 }
1026         }
1027 }
1028
1029 static int
1030 check_lcore_params(void)
1031 {
1032         uint8_t queue, lcore;
1033         uint16_t i;
1034         int socketid;
1035
1036         for (i = 0; i < nb_lcore_params; ++i) {
1037                 queue = lcore_params[i].queue_id;
1038                 if (queue >= MAX_RX_QUEUE_PER_PORT) {
1039                         printf("invalid queue number: %hhu\n", queue);
1040                         return -1;
1041                 }
1042                 lcore = lcore_params[i].lcore_id;
1043                 if (!rte_lcore_is_enabled(lcore)) {
1044                         printf("error: lcore %hhu is not enabled in lcore "
1045                                                         "mask\n", lcore);
1046                         return -1;
1047                 }
1048                 if ((socketid = rte_lcore_to_socket_id(lcore) != 0) &&
1049                                                         (numa_on == 0)) {
1050                         printf("warning: lcore %hhu is on socket %d with numa "
1051                                                 "off\n", lcore, socketid);
1052                 }
1053         }
1054         return 0;
1055 }
1056
1057 static int
1058 check_port_config(void)
1059 {
1060         unsigned portid;
1061         uint16_t i;
1062
1063         for (i = 0; i < nb_lcore_params; ++i) {
1064                 portid = lcore_params[i].port_id;
1065                 if ((enabled_port_mask & (1 << portid)) == 0) {
1066                         printf("port %u is not enabled in port mask\n",
1067                                                                 portid);
1068                         return -1;
1069                 }
1070                 if (!rte_eth_dev_is_valid_port(portid)) {
1071                         printf("port %u is not present on the board\n",
1072                                                                 portid);
1073                         return -1;
1074                 }
1075         }
1076         return 0;
1077 }
1078
1079 static uint8_t
1080 get_port_n_rx_queues(const uint16_t port)
1081 {
1082         int queue = -1;
1083         uint16_t i;
1084
1085         for (i = 0; i < nb_lcore_params; ++i) {
1086                 if (lcore_params[i].port_id == port &&
1087                                 lcore_params[i].queue_id > queue)
1088                         queue = lcore_params[i].queue_id;
1089         }
1090         return (uint8_t)(++queue);
1091 }
1092
1093 static int
1094 init_lcore_rx_queues(void)
1095 {
1096         uint16_t i, nb_rx_queue;
1097         uint8_t lcore;
1098
1099         for (i = 0; i < nb_lcore_params; ++i) {
1100                 lcore = lcore_params[i].lcore_id;
1101                 nb_rx_queue = lcore_conf[lcore].n_rx_queue;
1102                 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
1103                         printf("error: too many queues (%u) for lcore: %u\n",
1104                                 (unsigned)nb_rx_queue + 1, (unsigned)lcore);
1105                         return -1;
1106                 } else {
1107                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
1108                                 lcore_params[i].port_id;
1109                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
1110                                 lcore_params[i].queue_id;
1111                         lcore_conf[lcore].n_rx_queue++;
1112                 }
1113         }
1114         return 0;
1115 }
1116
1117 /* display usage */
1118 static void
1119 print_usage(const char *prgname)
1120 {
1121         printf ("%s [EAL options] -- -p PORTMASK -P"
1122                 "  [--config (port,queue,lcore)[,(port,queue,lcore]]"
1123                 "  [--enable-jumbo [--max-pkt-len PKTLEN]]\n"
1124                 "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
1125                 "  -P : enable promiscuous mode\n"
1126                 "  --config (port,queue,lcore): rx queues configuration\n"
1127                 "  --no-numa: optional, disable numa awareness\n"
1128                 "  --enable-jumbo: enable jumbo frame"
1129                 " which max packet len is PKTLEN in decimal (64-9600)\n"
1130                 "  --parse-ptype: parse packet type by software\n",
1131                 prgname);
1132 }
1133
1134 static int parse_max_pkt_len(const char *pktlen)
1135 {
1136         char *end = NULL;
1137         unsigned long len;
1138
1139         /* parse decimal string */
1140         len = strtoul(pktlen, &end, 10);
1141         if ((pktlen[0] == '\0') || (end == NULL) || (*end != '\0'))
1142                 return -1;
1143
1144         if (len == 0)
1145                 return -1;
1146
1147         return len;
1148 }
1149
1150 static int
1151 parse_portmask(const char *portmask)
1152 {
1153         char *end = NULL;
1154         unsigned long pm;
1155
1156         /* parse hexadecimal string */
1157         pm = strtoul(portmask, &end, 16);
1158         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
1159                 return -1;
1160
1161         if (pm == 0)
1162                 return -1;
1163
1164         return pm;
1165 }
1166
1167 static int
1168 parse_config(const char *q_arg)
1169 {
1170         char s[256];
1171         const char *p, *p0 = q_arg;
1172         char *end;
1173         enum fieldnames {
1174                 FLD_PORT = 0,
1175                 FLD_QUEUE,
1176                 FLD_LCORE,
1177                 _NUM_FLD
1178         };
1179         unsigned long int_fld[_NUM_FLD];
1180         char *str_fld[_NUM_FLD];
1181         int i;
1182         unsigned size;
1183
1184         nb_lcore_params = 0;
1185
1186         while ((p = strchr(p0,'(')) != NULL) {
1187                 ++p;
1188                 if((p0 = strchr(p,')')) == NULL)
1189                         return -1;
1190
1191                 size = p0 - p;
1192                 if(size >= sizeof(s))
1193                         return -1;
1194
1195                 snprintf(s, sizeof(s), "%.*s", size, p);
1196                 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') !=
1197                                                                 _NUM_FLD)
1198                         return -1;
1199                 for (i = 0; i < _NUM_FLD; i++){
1200                         errno = 0;
1201                         int_fld[i] = strtoul(str_fld[i], &end, 0);
1202                         if (errno != 0 || end == str_fld[i] || int_fld[i] >
1203                                                                         255)
1204                                 return -1;
1205                 }
1206                 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
1207                         printf("exceeded max number of lcore params: %hu\n",
1208                                 nb_lcore_params);
1209                         return -1;
1210                 }
1211                 lcore_params_array[nb_lcore_params].port_id =
1212                                 (uint8_t)int_fld[FLD_PORT];
1213                 lcore_params_array[nb_lcore_params].queue_id =
1214                                 (uint8_t)int_fld[FLD_QUEUE];
1215                 lcore_params_array[nb_lcore_params].lcore_id =
1216                                 (uint8_t)int_fld[FLD_LCORE];
1217                 ++nb_lcore_params;
1218         }
1219         lcore_params = lcore_params_array;
1220
1221         return 0;
1222 }
1223
1224 #define CMD_LINE_OPT_PARSE_PTYPE "parse-ptype"
1225
1226 /* Parse the argument given in the command line of the application */
1227 static int
1228 parse_args(int argc, char **argv)
1229 {
1230         int opt, ret;
1231         char **argvopt;
1232         int option_index;
1233         char *prgname = argv[0];
1234         static struct option lgopts[] = {
1235                 {"config", 1, 0, 0},
1236                 {"no-numa", 0, 0, 0},
1237                 {"enable-jumbo", 0, 0, 0},
1238                 {CMD_LINE_OPT_PARSE_PTYPE, 0, 0, 0},
1239                 {NULL, 0, 0, 0}
1240         };
1241
1242         argvopt = argv;
1243
1244         while ((opt = getopt_long(argc, argvopt, "p:P",
1245                                 lgopts, &option_index)) != EOF) {
1246
1247                 switch (opt) {
1248                 /* portmask */
1249                 case 'p':
1250                         enabled_port_mask = parse_portmask(optarg);
1251                         if (enabled_port_mask == 0) {
1252                                 printf("invalid portmask\n");
1253                                 print_usage(prgname);
1254                                 return -1;
1255                         }
1256                         break;
1257                 case 'P':
1258                         printf("Promiscuous mode selected\n");
1259                         promiscuous_on = 1;
1260                         break;
1261
1262                 /* long options */
1263                 case 0:
1264                         if (!strncmp(lgopts[option_index].name, "config", 6)) {
1265                                 ret = parse_config(optarg);
1266                                 if (ret) {
1267                                         printf("invalid config\n");
1268                                         print_usage(prgname);
1269                                         return -1;
1270                                 }
1271                         }
1272
1273                         if (!strncmp(lgopts[option_index].name,
1274                                                 "no-numa", 7)) {
1275                                 printf("numa is disabled \n");
1276                                 numa_on = 0;
1277                         }
1278
1279                         if (!strncmp(lgopts[option_index].name,
1280                                         "enable-jumbo", 12)) {
1281                                 struct option lenopts =
1282                                         {"max-pkt-len", required_argument, \
1283                                                                         0, 0};
1284
1285                                 printf("jumbo frame is enabled \n");
1286                                 port_conf.rxmode.offloads |=
1287                                                 DEV_RX_OFFLOAD_JUMBO_FRAME;
1288                                 port_conf.txmode.offloads |=
1289                                                 DEV_TX_OFFLOAD_MULTI_SEGS;
1290
1291                                 /**
1292                                  * if no max-pkt-len set, use the default value
1293                                  * ETHER_MAX_LEN
1294                                  */
1295                                 if (0 == getopt_long(argc, argvopt, "",
1296                                                 &lenopts, &option_index)) {
1297                                         ret = parse_max_pkt_len(optarg);
1298                                         if ((ret < 64) ||
1299                                                 (ret > MAX_JUMBO_PKT_LEN)){
1300                                                 printf("invalid packet "
1301                                                                 "length\n");
1302                                                 print_usage(prgname);
1303                                                 return -1;
1304                                         }
1305                                         port_conf.rxmode.max_rx_pkt_len = ret;
1306                                 }
1307                                 printf("set jumbo frame "
1308                                         "max packet length to %u\n",
1309                                 (unsigned int)port_conf.rxmode.max_rx_pkt_len);
1310                         }
1311
1312                         if (!strncmp(lgopts[option_index].name,
1313                                      CMD_LINE_OPT_PARSE_PTYPE,
1314                                      sizeof(CMD_LINE_OPT_PARSE_PTYPE))) {
1315                                 printf("soft parse-ptype is enabled\n");
1316                                 parse_ptype = 1;
1317                         }
1318
1319                         break;
1320
1321                 default:
1322                         print_usage(prgname);
1323                         return -1;
1324                 }
1325         }
1326
1327         if (optind >= 0)
1328                 argv[optind-1] = prgname;
1329
1330         ret = optind-1;
1331         optind = 1; /* reset getopt lib */
1332         return ret;
1333 }
1334
1335 static void
1336 print_ethaddr(const char *name, const struct ether_addr *eth_addr)
1337 {
1338         char buf[ETHER_ADDR_FMT_SIZE];
1339         ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr);
1340         printf("%s%s", name, buf);
1341 }
1342
1343 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1344 static void
1345 setup_hash(int socketid)
1346 {
1347         struct rte_hash_parameters ipv4_l3fwd_hash_params = {
1348                 .name = NULL,
1349                 .entries = L3FWD_HASH_ENTRIES,
1350                 .key_len = sizeof(struct ipv4_5tuple),
1351                 .hash_func = DEFAULT_HASH_FUNC,
1352                 .hash_func_init_val = 0,
1353         };
1354
1355         struct rte_hash_parameters ipv6_l3fwd_hash_params = {
1356                 .name = NULL,
1357                 .entries = L3FWD_HASH_ENTRIES,
1358                 .key_len = sizeof(struct ipv6_5tuple),
1359                 .hash_func = DEFAULT_HASH_FUNC,
1360                 .hash_func_init_val = 0,
1361         };
1362
1363         unsigned i;
1364         int ret;
1365         char s[64];
1366
1367         /* create ipv4 hash */
1368         snprintf(s, sizeof(s), "ipv4_l3fwd_hash_%d", socketid);
1369         ipv4_l3fwd_hash_params.name = s;
1370         ipv4_l3fwd_hash_params.socket_id = socketid;
1371         ipv4_l3fwd_lookup_struct[socketid] =
1372                 rte_hash_create(&ipv4_l3fwd_hash_params);
1373         if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
1374                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
1375                                 "socket %d\n", socketid);
1376
1377         /* create ipv6 hash */
1378         snprintf(s, sizeof(s), "ipv6_l3fwd_hash_%d", socketid);
1379         ipv6_l3fwd_hash_params.name = s;
1380         ipv6_l3fwd_hash_params.socket_id = socketid;
1381         ipv6_l3fwd_lookup_struct[socketid] =
1382                 rte_hash_create(&ipv6_l3fwd_hash_params);
1383         if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
1384                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
1385                                 "socket %d\n", socketid);
1386
1387
1388         /* populate the ipv4 hash */
1389         for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
1390                 ret = rte_hash_add_key (ipv4_l3fwd_lookup_struct[socketid],
1391                                 (void *) &ipv4_l3fwd_route_array[i].key);
1392                 if (ret < 0) {
1393                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
1394                                 "l3fwd hash on socket %d\n", i, socketid);
1395                 }
1396                 ipv4_l3fwd_out_if[ret] = ipv4_l3fwd_route_array[i].if_out;
1397                 printf("Hash: Adding key\n");
1398                 print_ipv4_key(ipv4_l3fwd_route_array[i].key);
1399         }
1400
1401         /* populate the ipv6 hash */
1402         for (i = 0; i < IPV6_L3FWD_NUM_ROUTES; i++) {
1403                 ret = rte_hash_add_key (ipv6_l3fwd_lookup_struct[socketid],
1404                                 (void *) &ipv6_l3fwd_route_array[i].key);
1405                 if (ret < 0) {
1406                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
1407                                 "l3fwd hash on socket %d\n", i, socketid);
1408                 }
1409                 ipv6_l3fwd_out_if[ret] = ipv6_l3fwd_route_array[i].if_out;
1410                 printf("Hash: Adding key\n");
1411                 print_ipv6_key(ipv6_l3fwd_route_array[i].key);
1412         }
1413 }
1414 #endif
1415
1416 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1417 static void
1418 setup_lpm(int socketid)
1419 {
1420         unsigned i;
1421         int ret;
1422         char s[64];
1423
1424         /* create the LPM table */
1425         struct rte_lpm_config lpm_ipv4_config;
1426
1427         lpm_ipv4_config.max_rules = IPV4_L3FWD_LPM_MAX_RULES;
1428         lpm_ipv4_config.number_tbl8s = 256;
1429         lpm_ipv4_config.flags = 0;
1430
1431         snprintf(s, sizeof(s), "IPV4_L3FWD_LPM_%d", socketid);
1432         ipv4_l3fwd_lookup_struct[socketid] =
1433                         rte_lpm_create(s, socketid, &lpm_ipv4_config);
1434         if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
1435                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
1436                                 " on socket %d\n", socketid);
1437
1438         /* populate the LPM table */
1439         for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
1440                 ret = rte_lpm_add(ipv4_l3fwd_lookup_struct[socketid],
1441                         ipv4_l3fwd_route_array[i].ip,
1442                         ipv4_l3fwd_route_array[i].depth,
1443                         ipv4_l3fwd_route_array[i].if_out);
1444
1445                 if (ret < 0) {
1446                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
1447                                 "l3fwd LPM table on socket %d\n",
1448                                 i, socketid);
1449                 }
1450
1451                 printf("LPM: Adding route 0x%08x / %d (%d)\n",
1452                         (unsigned)ipv4_l3fwd_route_array[i].ip,
1453                         ipv4_l3fwd_route_array[i].depth,
1454                         ipv4_l3fwd_route_array[i].if_out);
1455         }
1456 }
1457 #endif
1458
1459 static int
1460 init_mem(unsigned nb_mbuf)
1461 {
1462         struct lcore_conf *qconf;
1463         int socketid;
1464         unsigned lcore_id;
1465         char s[64];
1466
1467         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1468                 if (rte_lcore_is_enabled(lcore_id) == 0)
1469                         continue;
1470
1471                 if (numa_on)
1472                         socketid = rte_lcore_to_socket_id(lcore_id);
1473                 else
1474                         socketid = 0;
1475
1476                 if (socketid >= NB_SOCKETS) {
1477                         rte_exit(EXIT_FAILURE, "Socket %d of lcore %u is "
1478                                         "out of range %d\n", socketid,
1479                                                 lcore_id, NB_SOCKETS);
1480                 }
1481                 if (pktmbuf_pool[socketid] == NULL) {
1482                         snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
1483                         pktmbuf_pool[socketid] =
1484                                 rte_pktmbuf_pool_create(s, nb_mbuf,
1485                                         MEMPOOL_CACHE_SIZE, 0,
1486                                         RTE_MBUF_DEFAULT_BUF_SIZE,
1487                                         socketid);
1488                         if (pktmbuf_pool[socketid] == NULL)
1489                                 rte_exit(EXIT_FAILURE,
1490                                         "Cannot init mbuf pool on socket %d\n",
1491                                                                 socketid);
1492                         else
1493                                 printf("Allocated mbuf pool on socket %d\n",
1494                                                                 socketid);
1495
1496 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1497                         setup_lpm(socketid);
1498 #else
1499                         setup_hash(socketid);
1500 #endif
1501                 }
1502                 qconf = &lcore_conf[lcore_id];
1503                 qconf->ipv4_lookup_struct = ipv4_l3fwd_lookup_struct[socketid];
1504 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1505                 qconf->ipv6_lookup_struct = ipv6_l3fwd_lookup_struct[socketid];
1506 #endif
1507         }
1508         return 0;
1509 }
1510
1511 /* Check the link status of all ports in up to 9s, and print them finally */
1512 static void
1513 check_all_ports_link_status(uint32_t port_mask)
1514 {
1515 #define CHECK_INTERVAL 100 /* 100ms */
1516 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
1517         uint8_t count, all_ports_up, print_flag = 0;
1518         uint16_t portid;
1519         struct rte_eth_link link;
1520
1521         printf("\nChecking link status");
1522         fflush(stdout);
1523         for (count = 0; count <= MAX_CHECK_TIME; count++) {
1524                 all_ports_up = 1;
1525                 RTE_ETH_FOREACH_DEV(portid) {
1526                         if ((port_mask & (1 << portid)) == 0)
1527                                 continue;
1528                         memset(&link, 0, sizeof(link));
1529                         rte_eth_link_get_nowait(portid, &link);
1530                         /* print link status if flag set */
1531                         if (print_flag == 1) {
1532                                 if (link.link_status)
1533                                         printf("Port %d Link Up - speed %u "
1534                                                 "Mbps - %s\n", (uint8_t)portid,
1535                                                 (unsigned)link.link_speed,
1536                                 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
1537                                         ("full-duplex") : ("half-duplex\n"));
1538                                 else
1539                                         printf("Port %d Link Down\n",
1540                                                 (uint8_t)portid);
1541                                 continue;
1542                         }
1543                         /* clear all_ports_up flag if any link down */
1544                         if (link.link_status == ETH_LINK_DOWN) {
1545                                 all_ports_up = 0;
1546                                 break;
1547                         }
1548                 }
1549                 /* after finally printing all link status, get out */
1550                 if (print_flag == 1)
1551                         break;
1552
1553                 if (all_ports_up == 0) {
1554                         printf(".");
1555                         fflush(stdout);
1556                         rte_delay_ms(CHECK_INTERVAL);
1557                 }
1558
1559                 /* set the print_flag if all ports up or timeout */
1560                 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
1561                         print_flag = 1;
1562                         printf("done\n");
1563                 }
1564         }
1565 }
1566
1567 static int check_ptype(uint16_t portid)
1568 {
1569         int i, ret;
1570         int ptype_l3_ipv4 = 0;
1571 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1572         int ptype_l3_ipv6 = 0;
1573 #endif
1574         uint32_t ptype_mask = RTE_PTYPE_L3_MASK;
1575
1576         ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, NULL, 0);
1577         if (ret <= 0)
1578                 return 0;
1579
1580         uint32_t ptypes[ret];
1581
1582         ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, ptypes, ret);
1583         for (i = 0; i < ret; ++i) {
1584                 if (ptypes[i] & RTE_PTYPE_L3_IPV4)
1585                         ptype_l3_ipv4 = 1;
1586 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1587                 if (ptypes[i] & RTE_PTYPE_L3_IPV6)
1588                         ptype_l3_ipv6 = 1;
1589 #endif
1590         }
1591
1592         if (ptype_l3_ipv4 == 0)
1593                 printf("port %d cannot parse RTE_PTYPE_L3_IPV4\n", portid);
1594
1595 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1596         if (ptype_l3_ipv6 == 0)
1597                 printf("port %d cannot parse RTE_PTYPE_L3_IPV6\n", portid);
1598 #endif
1599
1600 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1601         if (ptype_l3_ipv4)
1602 #else /* APP_LOOKUP_EXACT_MATCH */
1603         if (ptype_l3_ipv4 && ptype_l3_ipv6)
1604 #endif
1605                 return 1;
1606
1607         return 0;
1608
1609 }
1610
1611 int
1612 main(int argc, char **argv)
1613 {
1614         struct lcore_conf *qconf;
1615         struct rte_eth_dev_info dev_info;
1616         struct rte_eth_txconf *txconf;
1617         int ret;
1618         uint16_t nb_ports;
1619         uint16_t queueid;
1620         unsigned lcore_id;
1621         uint64_t hz;
1622         uint32_t n_tx_queue, nb_lcores;
1623         uint32_t dev_rxq_num, dev_txq_num;
1624         uint8_t nb_rx_queue, queue, socketid;
1625         uint16_t portid;
1626
1627         /* catch SIGINT and restore cpufreq governor to ondemand */
1628         signal(SIGINT, signal_exit_now);
1629
1630         /* init EAL */
1631         ret = rte_eal_init(argc, argv);
1632         if (ret < 0)
1633                 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
1634         argc -= ret;
1635         argv += ret;
1636
1637         /* init RTE timer library to be used late */
1638         rte_timer_subsystem_init();
1639
1640         /* parse application arguments (after the EAL ones) */
1641         ret = parse_args(argc, argv);
1642         if (ret < 0)
1643                 rte_exit(EXIT_FAILURE, "Invalid L3FWD parameters\n");
1644
1645         if (check_lcore_params() < 0)
1646                 rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
1647
1648         ret = init_lcore_rx_queues();
1649         if (ret < 0)
1650                 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
1651
1652         nb_ports = rte_eth_dev_count_avail();
1653
1654         if (check_port_config() < 0)
1655                 rte_exit(EXIT_FAILURE, "check_port_config failed\n");
1656
1657         nb_lcores = rte_lcore_count();
1658
1659         /* initialize all ports */
1660         RTE_ETH_FOREACH_DEV(portid) {
1661                 struct rte_eth_conf local_port_conf = port_conf;
1662
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                         local_port_conf.intr_conf.rxq = 0;
1691                 rte_eth_dev_info_get(portid, &dev_info);
1692                 if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
1693                         local_port_conf.txmode.offloads |=
1694                                 DEV_TX_OFFLOAD_MBUF_FAST_FREE;
1695
1696                 local_port_conf.rx_adv_conf.rss_conf.rss_hf &=
1697                         dev_info.flow_type_rss_offloads;
1698                 if (local_port_conf.rx_adv_conf.rss_conf.rss_hf !=
1699                                 port_conf.rx_adv_conf.rss_conf.rss_hf) {
1700                         printf("Port %u modified RSS hash function based on hardware support,"
1701                                 "requested:%#"PRIx64" configured:%#"PRIx64"\n",
1702                                 portid,
1703                                 port_conf.rx_adv_conf.rss_conf.rss_hf,
1704                                 local_port_conf.rx_adv_conf.rss_conf.rss_hf);
1705                 }
1706
1707                 ret = rte_eth_dev_configure(portid, nb_rx_queue,
1708                                         (uint16_t)n_tx_queue, &local_port_conf);
1709                 if (ret < 0)
1710                         rte_exit(EXIT_FAILURE, "Cannot configure device: "
1711                                         "err=%d, port=%d\n", ret, portid);
1712
1713                 ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
1714                                                        &nb_txd);
1715                 if (ret < 0)
1716                         rte_exit(EXIT_FAILURE,
1717                                  "Cannot adjust number of descriptors: err=%d, port=%d\n",
1718                                  ret, portid);
1719
1720                 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
1721                 print_ethaddr(" Address:", &ports_eth_addr[portid]);
1722                 printf(", ");
1723
1724                 /* init memory */
1725                 ret = init_mem(NB_MBUF);
1726                 if (ret < 0)
1727                         rte_exit(EXIT_FAILURE, "init_mem failed\n");
1728
1729                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1730                         if (rte_lcore_is_enabled(lcore_id) == 0)
1731                                 continue;
1732
1733                         /* Initialize TX buffers */
1734                         qconf = &lcore_conf[lcore_id];
1735                         qconf->tx_buffer[portid] = rte_zmalloc_socket("tx_buffer",
1736                                 RTE_ETH_TX_BUFFER_SIZE(MAX_PKT_BURST), 0,
1737                                 rte_eth_dev_socket_id(portid));
1738                         if (qconf->tx_buffer[portid] == NULL)
1739                                 rte_exit(EXIT_FAILURE, "Can't allocate tx buffer for port %u\n",
1740                                                  portid);
1741
1742                         rte_eth_tx_buffer_init(qconf->tx_buffer[portid], MAX_PKT_BURST);
1743                 }
1744
1745                 /* init one TX queue per couple (lcore,port) */
1746                 queueid = 0;
1747                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1748                         if (rte_lcore_is_enabled(lcore_id) == 0)
1749                                 continue;
1750
1751                         if (queueid >= dev_txq_num)
1752                                 continue;
1753
1754                         if (numa_on)
1755                                 socketid = \
1756                                 (uint8_t)rte_lcore_to_socket_id(lcore_id);
1757                         else
1758                                 socketid = 0;
1759
1760                         printf("txq=%u,%d,%d ", lcore_id, queueid, socketid);
1761                         fflush(stdout);
1762
1763                         txconf = &dev_info.default_txconf;
1764                         txconf->offloads = local_port_conf.txmode.offloads;
1765                         ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
1766                                                      socketid, txconf);
1767                         if (ret < 0)
1768                                 rte_exit(EXIT_FAILURE,
1769                                         "rte_eth_tx_queue_setup: err=%d, "
1770                                                 "port=%d\n", ret, portid);
1771
1772                         qconf = &lcore_conf[lcore_id];
1773                         qconf->tx_queue_id[portid] = queueid;
1774                         queueid++;
1775
1776                         qconf->tx_port_id[qconf->n_tx_port] = portid;
1777                         qconf->n_tx_port++;
1778                 }
1779                 printf("\n");
1780         }
1781
1782         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1783                 if (rte_lcore_is_enabled(lcore_id) == 0)
1784                         continue;
1785
1786                 /* init power management library */
1787                 ret = rte_power_init(lcore_id);
1788                 if (ret)
1789                         RTE_LOG(ERR, POWER,
1790                                 "Library initialization failed on core %u\n", lcore_id);
1791
1792                 /* init timer structures for each enabled lcore */
1793                 rte_timer_init(&power_timers[lcore_id]);
1794                 hz = rte_get_timer_hz();
1795                 rte_timer_reset(&power_timers[lcore_id],
1796                         hz/TIMER_NUMBER_PER_SECOND, SINGLE, lcore_id,
1797                                                 power_timer_cb, NULL);
1798
1799                 qconf = &lcore_conf[lcore_id];
1800                 printf("\nInitializing rx queues on lcore %u ... ", lcore_id );
1801                 fflush(stdout);
1802                 /* init RX queues */
1803                 for(queue = 0; queue < qconf->n_rx_queue; ++queue) {
1804                         struct rte_eth_rxconf rxq_conf;
1805                         struct rte_eth_dev *dev;
1806                         struct rte_eth_conf *conf;
1807
1808                         portid = qconf->rx_queue_list[queue].port_id;
1809                         queueid = qconf->rx_queue_list[queue].queue_id;
1810                         dev = &rte_eth_devices[portid];
1811                         conf = &dev->data->dev_conf;
1812
1813                         if (numa_on)
1814                                 socketid = \
1815                                 (uint8_t)rte_lcore_to_socket_id(lcore_id);
1816                         else
1817                                 socketid = 0;
1818
1819                         printf("rxq=%d,%d,%d ", portid, queueid, socketid);
1820                         fflush(stdout);
1821
1822                         rte_eth_dev_info_get(portid, &dev_info);
1823                         rxq_conf = dev_info.default_rxconf;
1824                         rxq_conf.offloads = conf->rxmode.offloads;
1825                         ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
1826                                 socketid, &rxq_conf,
1827                                 pktmbuf_pool[socketid]);
1828                         if (ret < 0)
1829                                 rte_exit(EXIT_FAILURE,
1830                                         "rte_eth_rx_queue_setup: err=%d, "
1831                                                 "port=%d\n", ret, portid);
1832
1833                         if (parse_ptype) {
1834                                 if (add_cb_parse_ptype(portid, queueid) < 0)
1835                                         rte_exit(EXIT_FAILURE,
1836                                                  "Fail to add ptype cb\n");
1837                         } else if (!check_ptype(portid))
1838                                 rte_exit(EXIT_FAILURE,
1839                                          "PMD can not provide needed ptypes\n");
1840                 }
1841         }
1842
1843         printf("\n");
1844
1845         /* start ports */
1846         RTE_ETH_FOREACH_DEV(portid) {
1847                 if ((enabled_port_mask & (1 << portid)) == 0) {
1848                         continue;
1849                 }
1850                 /* Start device */
1851                 ret = rte_eth_dev_start(portid);
1852                 if (ret < 0)
1853                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, "
1854                                                 "port=%d\n", ret, portid);
1855                 /*
1856                  * If enabled, put device in promiscuous mode.
1857                  * This allows IO forwarding mode to forward packets
1858                  * to itself through 2 cross-connected  ports of the
1859                  * target machine.
1860                  */
1861                 if (promiscuous_on)
1862                         rte_eth_promiscuous_enable(portid);
1863                 /* initialize spinlock for each port */
1864                 rte_spinlock_init(&(locks[portid]));
1865         }
1866
1867         check_all_ports_link_status(enabled_port_mask);
1868
1869         /* launch per-lcore init on every lcore */
1870         rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
1871         RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1872                 if (rte_eal_wait_lcore(lcore_id) < 0)
1873                         return -1;
1874         }
1875
1876         return 0;
1877 }