examples/l3fwd-power/main.c

   1 /*-
   2  *   BSD LICENSE
   3  *
   4  *   Copyright(c) 2010-2016 Intel Corporation. All rights reserved.
   5  *   All rights reserved.
   6  *
   7  *   Redistribution and use in source and binary forms, with or without
   8  *   modification, are permitted provided that the following conditions
   9  *   are met:
  10  *
  11  *     * Redistributions of source code must retain the above copyright
  12  *       notice, this list of conditions and the following disclaimer.
  13  *     * Redistributions in binary form must reproduce the above copyright
  14  *       notice, this list of conditions and the following disclaimer in
  15  *       the documentation and/or other materials provided with the
  16  *       distribution.
  17  *     * Neither the name of Intel Corporation nor the names of its
  18  *       contributors may be used to endorse or promote products derived
  19  *       from this software without specific prior written permission.
  20  *
  21  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  22  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  23  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
  24  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  25  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  26  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
  27  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  28  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  29  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  30  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
  31  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  32  */
  33
  34 #include <stdio.h>
  35 #include <stdlib.h>
  36 #include <stdint.h>
  37 #include <inttypes.h>
  38 #include <sys/types.h>
  39 #include <string.h>
  40 #include <sys/queue.h>
  41 #include <stdarg.h>
  42 #include <errno.h>
  43 #include <getopt.h>
  44 #include <unistd.h>
  45 #include <signal.h>
  46
  47 #include <rte_common.h>
  48 #include <rte_byteorder.h>
  49 #include <rte_log.h>
  50 #include <rte_malloc.h>
  51 #include <rte_memory.h>
  52 #include <rte_memcpy.h>
  53 #include <rte_memzone.h>
  54 #include <rte_eal.h>
  55 #include <rte_launch.h>
  56 #include <rte_atomic.h>
  57 #include <rte_cycles.h>
  58 #include <rte_prefetch.h>
  59 #include <rte_lcore.h>
  60 #include <rte_per_lcore.h>
  61 #include <rte_branch_prediction.h>
  62 #include <rte_interrupts.h>
  63 #include <rte_pci.h>
  64 #include <rte_random.h>
  65 #include <rte_debug.h>
  66 #include <rte_ether.h>
  67 #include <rte_ethdev.h>
  68 #include <rte_mempool.h>
  69 #include <rte_mbuf.h>
  70 #include <rte_ip.h>
  71 #include <rte_tcp.h>
  72 #include <rte_udp.h>
  73 #include <rte_string_fns.h>
  74 #include <rte_timer.h>
  75 #include <rte_power.h>
  76 #include <rte_spinlock.h>
  77
  78 #define RTE_LOGTYPE_L3FWD_POWER RTE_LOGTYPE_USER1
  79
  80 #define MAX_PKT_BURST 32
  81
  82 #define MIN_ZERO_POLL_COUNT 10
  83
  84 /* around 100ms at 2 Ghz */
  85 #define TIMER_RESOLUTION_CYCLES           200000000ULL
  86 /* 100 ms interval */
  87 #define TIMER_NUMBER_PER_SECOND           10
  88 /* 100000 us */
  89 #define SCALING_PERIOD                    (1000000/TIMER_NUMBER_PER_SECOND)
  90 #define SCALING_DOWN_TIME_RATIO_THRESHOLD 0.25
  91
  92 #define APP_LOOKUP_EXACT_MATCH          0
  93 #define APP_LOOKUP_LPM                  1
  94 #define DO_RFC_1812_CHECKS
  95
  96 #ifndef APP_LOOKUP_METHOD
  97 #define APP_LOOKUP_METHOD             APP_LOOKUP_LPM
  98 #endif
  99
 100 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 101 #include <rte_hash.h>
 102 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 103 #include <rte_lpm.h>
 104 #else
 105 #error "APP_LOOKUP_METHOD set to incorrect value"
 106 #endif
 107
 108 #ifndef IPv6_BYTES
 109 #define IPv6_BYTES_FMT "%02x%02x:%02x%02x:%02x%02x:%02x%02x:"\
 110                        "%02x%02x:%02x%02x:%02x%02x:%02x%02x"
 111 #define IPv6_BYTES(addr) \
 112         addr[0],  addr[1], addr[2],  addr[3], \
 113         addr[4],  addr[5], addr[6],  addr[7], \
 114         addr[8],  addr[9], addr[10], addr[11],\
 115         addr[12], addr[13],addr[14], addr[15]
 116 #endif
 117
 118 #define MAX_JUMBO_PKT_LEN  9600
 119
 120 #define IPV6_ADDR_LEN 16
 121
 122 #define MEMPOOL_CACHE_SIZE 256
 123
 124 /*
 125  * This expression is used to calculate the number of mbufs needed depending on
 126  * user input, taking into account memory for rx and tx hardware rings, cache
 127  * per lcore and mtable per port per lcore. RTE_MAX is used to ensure that
 128  * NB_MBUF never goes below a minimum value of 8192.
 129  */
 130
 131 #define NB_MBUF RTE_MAX ( \
 132         (nb_ports*nb_rx_queue*nb_rxd + \
 133         nb_ports*nb_lcores*MAX_PKT_BURST + \
 134         nb_ports*n_tx_queue*nb_txd + \
 135         nb_lcores*MEMPOOL_CACHE_SIZE), \
 136         (unsigned)8192)
 137
 138 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
 139
 140 #define NB_SOCKETS 8
 141
 142 /* Configure how many packets ahead to prefetch, when reading packets */
 143 #define PREFETCH_OFFSET 3
 144
 145 /*
 146  * Configurable number of RX/TX ring descriptors
 147  */
 148 #define RTE_TEST_RX_DESC_DEFAULT 512
 149 #define RTE_TEST_TX_DESC_DEFAULT 512
 150 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
 151 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
 152
 153 /* ethernet addresses of ports */
 154 static struct ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
 155
 156 /* ethernet addresses of ports */
 157 static rte_spinlock_t locks[RTE_MAX_ETHPORTS];
 158
 159 /* mask of enabled ports */
 160 static uint32_t enabled_port_mask = 0;
 161 /* Ports set in promiscuous mode off by default. */
 162 static int promiscuous_on = 0;
 163 /* NUMA is enabled by default. */
 164 static int numa_on = 1;
 165 static int parse_ptype; /**< Parse packet type using rx callback, and */
 166                         /**< disabled by default */
 167
 168 enum freq_scale_hint_t
 169 {
 170         FREQ_LOWER    =      -1,
 171         FREQ_CURRENT  =       0,
 172         FREQ_HIGHER   =       1,
 173         FREQ_HIGHEST  =       2
 174 };
 175
 176 struct lcore_rx_queue {
 177         uint16_t port_id;
 178         uint8_t queue_id;
 179         enum freq_scale_hint_t freq_up_hint;
 180         uint32_t zero_rx_packet_count;
 181         uint32_t idle_hint;
 182 } __rte_cache_aligned;
 183
 184 #define MAX_RX_QUEUE_PER_LCORE 16
 185 #define MAX_TX_QUEUE_PER_PORT RTE_MAX_ETHPORTS
 186 #define MAX_RX_QUEUE_PER_PORT 128
 187
 188 #define MAX_RX_QUEUE_INTERRUPT_PER_PORT 16
 189
 190
 191 #define MAX_LCORE_PARAMS 1024
 192 struct lcore_params {
 193         uint16_t port_id;
 194         uint8_t queue_id;
 195         uint8_t lcore_id;
 196 } __rte_cache_aligned;
 197
 198 static struct lcore_params lcore_params_array[MAX_LCORE_PARAMS];
 199 static struct lcore_params lcore_params_array_default[] = {
 200         {0, 0, 2},
 201         {0, 1, 2},
 202         {0, 2, 2},
 203         {1, 0, 2},
 204         {1, 1, 2},
 205         {1, 2, 2},
 206         {2, 0, 2},
 207         {3, 0, 3},
 208         {3, 1, 3},
 209 };
 210
 211 static struct lcore_params * lcore_params = lcore_params_array_default;
 212 static uint16_t nb_lcore_params = sizeof(lcore_params_array_default) /
 213                                 sizeof(lcore_params_array_default[0]);
 214
 215 static struct rte_eth_conf port_conf = {
 216         .rxmode = {
 217                 .mq_mode        = ETH_MQ_RX_RSS,
 218                 .max_rx_pkt_len = ETHER_MAX_LEN,
 219                 .split_hdr_size = 0,
 220                 .header_split   = 0, /**< Header Split disabled */
 221                 .hw_ip_checksum = 1, /**< IP checksum offload enabled */
 222                 .hw_vlan_filter = 0, /**< VLAN filtering disabled */
 223                 .jumbo_frame    = 0, /**< Jumbo Frame Support disabled */
 224                 .hw_strip_crc   = 1, /**< CRC stripped by hardware */
 225         },
 226         .rx_adv_conf = {
 227                 .rss_conf = {
 228                         .rss_key = NULL,
 229                         .rss_hf = ETH_RSS_UDP,
 230                 },
 231         },
 232         .txmode = {
 233                 .mq_mode = ETH_MQ_TX_NONE,
 234         },
 235         .intr_conf = {
 236                 .lsc = 1,
 237                 .rxq = 1,
 238         },
 239 };
 240
 241 static struct rte_mempool * pktmbuf_pool[NB_SOCKETS];
 242
 243
 244 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 245
 246 #ifdef RTE_ARCH_X86
 247 #include <rte_hash_crc.h>
 248 #define DEFAULT_HASH_FUNC       rte_hash_crc
 249 #else
 250 #include <rte_jhash.h>
 251 #define DEFAULT_HASH_FUNC       rte_jhash
 252 #endif
 253
 254 struct ipv4_5tuple {
 255         uint32_t ip_dst;
 256         uint32_t ip_src;
 257         uint16_t port_dst;
 258         uint16_t port_src;
 259         uint8_t  proto;
 260 } __attribute__((__packed__));
 261
 262 struct ipv6_5tuple {
 263         uint8_t  ip_dst[IPV6_ADDR_LEN];
 264         uint8_t  ip_src[IPV6_ADDR_LEN];
 265         uint16_t port_dst;
 266         uint16_t port_src;
 267         uint8_t  proto;
 268 } __attribute__((__packed__));
 269
 270 struct ipv4_l3fwd_route {
 271         struct ipv4_5tuple key;
 272         uint8_t if_out;
 273 };
 274
 275 struct ipv6_l3fwd_route {
 276         struct ipv6_5tuple key;
 277         uint8_t if_out;
 278 };
 279
 280 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
 281         {{IPv4(100,10,0,1), IPv4(200,10,0,1), 101, 11, IPPROTO_TCP}, 0},
 282         {{IPv4(100,20,0,2), IPv4(200,20,0,2), 102, 12, IPPROTO_TCP}, 1},
 283         {{IPv4(100,30,0,3), IPv4(200,30,0,3), 103, 13, IPPROTO_TCP}, 2},
 284         {{IPv4(100,40,0,4), IPv4(200,40,0,4), 104, 14, IPPROTO_TCP}, 3},
 285 };
 286
 287 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
 288         {
 289                 {
 290                         {0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
 291                          0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
 292                         {0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
 293                          0x02, 0x1e, 0x67, 0xff, 0xfe, 0x0d, 0xb6, 0x0a},
 294                          1, 10, IPPROTO_UDP
 295                 }, 4
 296         },
 297 };
 298
 299 typedef struct rte_hash lookup_struct_t;
 300 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
 301 static lookup_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
 302
 303 #define L3FWD_HASH_ENTRIES      1024
 304
 305 #define IPV4_L3FWD_NUM_ROUTES \
 306         (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
 307
 308 #define IPV6_L3FWD_NUM_ROUTES \
 309         (sizeof(ipv6_l3fwd_route_array) / sizeof(ipv6_l3fwd_route_array[0]))
 310
 311 static uint16_t ipv4_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 312 static uint16_t ipv6_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 313 #endif
 314
 315 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 316 struct ipv4_l3fwd_route {
 317         uint32_t ip;
 318         uint8_t  depth;
 319         uint8_t  if_out;
 320 };
 321
 322 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
 323         {IPv4(1,1,1,0), 24, 0},
 324         {IPv4(2,1,1,0), 24, 1},
 325         {IPv4(3,1,1,0), 24, 2},
 326         {IPv4(4,1,1,0), 24, 3},
 327         {IPv4(5,1,1,0), 24, 4},
 328         {IPv4(6,1,1,0), 24, 5},
 329         {IPv4(7,1,1,0), 24, 6},
 330         {IPv4(8,1,1,0), 24, 7},
 331 };
 332
 333 #define IPV4_L3FWD_NUM_ROUTES \
 334         (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
 335
 336 #define IPV4_L3FWD_LPM_MAX_RULES     1024
 337
 338 typedef struct rte_lpm lookup_struct_t;
 339 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
 340 #endif
 341
 342 struct lcore_conf {
 343         uint16_t n_rx_queue;
 344         struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
 345         uint16_t n_tx_port;
 346         uint16_t tx_port_id[RTE_MAX_ETHPORTS];
 347         uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
 348         struct rte_eth_dev_tx_buffer *tx_buffer[RTE_MAX_ETHPORTS];
 349         lookup_struct_t * ipv4_lookup_struct;
 350         lookup_struct_t * ipv6_lookup_struct;
 351 } __rte_cache_aligned;
 352
 353 struct lcore_stats {
 354         /* total sleep time in ms since last frequency scaling down */
 355         uint32_t sleep_time;
 356         /* number of long sleep recently */
 357         uint32_t nb_long_sleep;
 358         /* freq. scaling up trend */
 359         uint32_t trend;
 360         /* total packet processed recently */
 361         uint64_t nb_rx_processed;
 362         /* total iterations looped recently */
 363         uint64_t nb_iteration_looped;
 364         uint32_t padding[9];
 365 } __rte_cache_aligned;
 366
 367 static struct lcore_conf lcore_conf[RTE_MAX_LCORE] __rte_cache_aligned;
 368 static struct lcore_stats stats[RTE_MAX_LCORE] __rte_cache_aligned;
 369 static struct rte_timer power_timers[RTE_MAX_LCORE];
 370
 371 static inline uint32_t power_idle_heuristic(uint32_t zero_rx_packet_count);
 372 static inline enum freq_scale_hint_t power_freq_scaleup_heuristic( \
 373                 unsigned int lcore_id, uint16_t port_id, uint16_t queue_id);
 374
 375 /* exit signal handler */
 376 static void
 377 signal_exit_now(int sigtype)
 378 {
 379         unsigned lcore_id;
 380         unsigned int portid, nb_ports;
 381         int ret;
 382
 383         if (sigtype == SIGINT) {
 384                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
 385                         if (rte_lcore_is_enabled(lcore_id) == 0)
 386                                 continue;
 387
 388                         /* init power management library */
 389                         ret = rte_power_exit(lcore_id);
 390                         if (ret)
 391                                 rte_exit(EXIT_FAILURE, "Power management "
 392                                         "library de-initialization failed on "
 393                                                         "core%u\n", lcore_id);
 394                 }
 395
 396                 nb_ports = rte_eth_dev_count();
 397                 for (portid = 0; portid < nb_ports; portid++) {
 398                         if ((enabled_port_mask & (1 << portid)) == 0)
 399                                 continue;
 400
 401                         rte_eth_dev_stop(portid);
 402                         rte_eth_dev_close(portid);
 403                 }
 404         }
 405
 406         rte_exit(EXIT_SUCCESS, "User forced exit\n");
 407 }
 408
 409 /*  Freqency scale down timer callback */
 410 static void
 411 power_timer_cb(__attribute__((unused)) struct rte_timer *tim,
 412                           __attribute__((unused)) void *arg)
 413 {
 414         uint64_t hz;
 415         float sleep_time_ratio;
 416         unsigned lcore_id = rte_lcore_id();
 417
 418         /* accumulate total execution time in us when callback is invoked */
 419         sleep_time_ratio = (float)(stats[lcore_id].sleep_time) /
 420                                         (float)SCALING_PERIOD;
 421         /**
 422          * check whether need to scale down frequency a step if it sleep a lot.
 423          */
 424         if (sleep_time_ratio >= SCALING_DOWN_TIME_RATIO_THRESHOLD) {
 425                 if (rte_power_freq_down)
 426                         rte_power_freq_down(lcore_id);
 427         }
 428         else if ( (unsigned)(stats[lcore_id].nb_rx_processed /
 429                 stats[lcore_id].nb_iteration_looped) < MAX_PKT_BURST) {
 430                 /**
 431                  * scale down a step if average packet per iteration less
 432                  * than expectation.
 433                  */
 434                 if (rte_power_freq_down)
 435                         rte_power_freq_down(lcore_id);
 436         }
 437
 438         /**
 439          * initialize another timer according to current frequency to ensure
 440          * timer interval is relatively fixed.
 441          */
 442         hz = rte_get_timer_hz();
 443         rte_timer_reset(&power_timers[lcore_id], hz/TIMER_NUMBER_PER_SECOND,
 444                                 SINGLE, lcore_id, power_timer_cb, NULL);
 445
 446         stats[lcore_id].nb_rx_processed = 0;
 447         stats[lcore_id].nb_iteration_looped = 0;
 448
 449         stats[lcore_id].sleep_time = 0;
 450 }
 451
 452 /* Enqueue a single packet, and send burst if queue is filled */
 453 static inline int
 454 send_single_packet(struct rte_mbuf *m, uint16_t port)
 455 {
 456         uint32_t lcore_id;
 457         struct lcore_conf *qconf;
 458
 459         lcore_id = rte_lcore_id();
 460         qconf = &lcore_conf[lcore_id];
 461
 462         rte_eth_tx_buffer(port, qconf->tx_queue_id[port],
 463                         qconf->tx_buffer[port], m);
 464
 465         return 0;
 466 }
 467
 468 #ifdef DO_RFC_1812_CHECKS
 469 static inline int
 470 is_valid_ipv4_pkt(struct ipv4_hdr *pkt, uint32_t link_len)
 471 {
 472         /* From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2 */
 473         /*
 474          * 1. The packet length reported by the Link Layer must be large
 475          * enough to hold the minimum length legal IP datagram (20 bytes).
 476          */
 477         if (link_len < sizeof(struct ipv4_hdr))
 478                 return -1;
 479
 480         /* 2. The IP checksum must be correct. */
 481         /* this is checked in H/W */
 482
 483         /*
 484          * 3. The IP version number must be 4. If the version number is not 4
 485          * then the packet may be another version of IP, such as IPng or
 486          * ST-II.
 487          */
 488         if (((pkt->version_ihl) >> 4) != 4)
 489                 return -3;
 490         /*
 491          * 4. The IP header length field must be large enough to hold the
 492          * minimum length legal IP datagram (20 bytes = 5 words).
 493          */
 494         if ((pkt->version_ihl & 0xf) < 5)
 495                 return -4;
 496
 497         /*
 498          * 5. The IP total length field must be large enough to hold the IP
 499          * datagram header, whose length is specified in the IP header length
 500          * field.
 501          */
 502         if (rte_cpu_to_be_16(pkt->total_length) < sizeof(struct ipv4_hdr))
 503                 return -5;
 504
 505         return 0;
 506 }
 507 #endif
 508
 509 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 510 static void
 511 print_ipv4_key(struct ipv4_5tuple key)
 512 {
 513         printf("IP dst = %08x, IP src = %08x, port dst = %d, port src = %d, "
 514                 "proto = %d\n", (unsigned)key.ip_dst, (unsigned)key.ip_src,
 515                                 key.port_dst, key.port_src, key.proto);
 516 }
 517 static void
 518 print_ipv6_key(struct ipv6_5tuple key)
 519 {
 520         printf( "IP dst = " IPv6_BYTES_FMT ", IP src = " IPv6_BYTES_FMT ", "
 521                 "port dst = %d, port src = %d, proto = %d\n",
 522                 IPv6_BYTES(key.ip_dst), IPv6_BYTES(key.ip_src),
 523                 key.port_dst, key.port_src, key.proto);
 524 }
 525
 526 static inline uint16_t
 527 get_ipv4_dst_port(struct ipv4_hdr *ipv4_hdr, uint16_t portid,
 528                 lookup_struct_t * ipv4_l3fwd_lookup_struct)
 529 {
 530         struct ipv4_5tuple key;
 531         struct tcp_hdr *tcp;
 532         struct udp_hdr *udp;
 533         int ret = 0;
 534
 535         key.ip_dst = rte_be_to_cpu_32(ipv4_hdr->dst_addr);
 536         key.ip_src = rte_be_to_cpu_32(ipv4_hdr->src_addr);
 537         key.proto = ipv4_hdr->next_proto_id;
 538
 539         switch (ipv4_hdr->next_proto_id) {
 540         case IPPROTO_TCP:
 541                 tcp = (struct tcp_hdr *)((unsigned char *)ipv4_hdr +
 542                                         sizeof(struct ipv4_hdr));
 543                 key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
 544                 key.port_src = rte_be_to_cpu_16(tcp->src_port);
 545                 break;
 546
 547         case IPPROTO_UDP:
 548                 udp = (struct udp_hdr *)((unsigned char *)ipv4_hdr +
 549                                         sizeof(struct ipv4_hdr));
 550                 key.port_dst = rte_be_to_cpu_16(udp->dst_port);
 551                 key.port_src = rte_be_to_cpu_16(udp->src_port);
 552                 break;
 553
 554         default:
 555                 key.port_dst = 0;
 556                 key.port_src = 0;
 557                 break;
 558         }
 559
 560         /* Find destination port */
 561         ret = rte_hash_lookup(ipv4_l3fwd_lookup_struct, (const void *)&key);
 562         return ((ret < 0) ? portid : ipv4_l3fwd_out_if[ret]);
 563 }
 564
 565 static inline uint16_t
 566 get_ipv6_dst_port(struct ipv6_hdr *ipv6_hdr, uint16_t portid,
 567                         lookup_struct_t *ipv6_l3fwd_lookup_struct)
 568 {
 569         struct ipv6_5tuple key;
 570         struct tcp_hdr *tcp;
 571         struct udp_hdr *udp;
 572         int ret = 0;
 573
 574         memcpy(key.ip_dst, ipv6_hdr->dst_addr, IPV6_ADDR_LEN);
 575         memcpy(key.ip_src, ipv6_hdr->src_addr, IPV6_ADDR_LEN);
 576
 577         key.proto = ipv6_hdr->proto;
 578
 579         switch (ipv6_hdr->proto) {
 580         case IPPROTO_TCP:
 581                 tcp = (struct tcp_hdr *)((unsigned char *) ipv6_hdr +
 582                                         sizeof(struct ipv6_hdr));
 583                 key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
 584                 key.port_src = rte_be_to_cpu_16(tcp->src_port);
 585                 break;
 586
 587         case IPPROTO_UDP:
 588                 udp = (struct udp_hdr *)((unsigned char *) ipv6_hdr +
 589                                         sizeof(struct ipv6_hdr));
 590                 key.port_dst = rte_be_to_cpu_16(udp->dst_port);
 591                 key.port_src = rte_be_to_cpu_16(udp->src_port);
 592                 break;
 593
 594         default:
 595                 key.port_dst = 0;
 596                 key.port_src = 0;
 597                 break;
 598         }
 599
 600         /* Find destination port */
 601         ret = rte_hash_lookup(ipv6_l3fwd_lookup_struct, (const void *)&key);
 602         return ((ret < 0) ? portid : ipv6_l3fwd_out_if[ret]);
 603 }
 604 #endif
 605
 606 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 607 static inline uint16_t
 608 get_ipv4_dst_port(struct ipv4_hdr *ipv4_hdr, uint16_t portid,
 609                 lookup_struct_t *ipv4_l3fwd_lookup_struct)
 610 {
 611         uint32_t next_hop;
 612
 613         return ((rte_lpm_lookup(ipv4_l3fwd_lookup_struct,
 614                         rte_be_to_cpu_32(ipv4_hdr->dst_addr), &next_hop) == 0)?
 615                         next_hop : portid);
 616 }
 617 #endif
 618
 619 static inline void
 620 parse_ptype_one(struct rte_mbuf *m)
 621 {
 622         struct ether_hdr *eth_hdr;
 623         uint32_t packet_type = RTE_PTYPE_UNKNOWN;
 624         uint16_t ether_type;
 625
 626         eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
 627         ether_type = eth_hdr->ether_type;
 628         if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv4))
 629                 packet_type |= RTE_PTYPE_L3_IPV4_EXT_UNKNOWN;
 630         else if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv6))
 631                 packet_type |= RTE_PTYPE_L3_IPV6_EXT_UNKNOWN;
 632
 633         m->packet_type = packet_type;
 634 }
 635
 636 static uint16_t
 637 cb_parse_ptype(uint16_t port __rte_unused, uint16_t queue __rte_unused,
 638                struct rte_mbuf *pkts[], uint16_t nb_pkts,
 639                uint16_t max_pkts __rte_unused,
 640                void *user_param __rte_unused)
 641 {
 642         unsigned int i;
 643
 644         for (i = 0; i < nb_pkts; ++i)
 645                 parse_ptype_one(pkts[i]);
 646
 647         return nb_pkts;
 648 }
 649
 650 static int
 651 add_cb_parse_ptype(uint16_t portid, uint16_t queueid)
 652 {
 653         printf("Port %d: softly parse packet type info\n", portid);
 654         if (rte_eth_add_rx_callback(portid, queueid, cb_parse_ptype, NULL))
 655                 return 0;
 656
 657         printf("Failed to add rx callback: port=%d\n", portid);
 658         return -1;
 659 }
 660
 661 static inline void
 662 l3fwd_simple_forward(struct rte_mbuf *m, uint16_t portid,
 663                                 struct lcore_conf *qconf)
 664 {
 665         struct ether_hdr *eth_hdr;
 666         struct ipv4_hdr *ipv4_hdr;
 667         void *d_addr_bytes;
 668         uint16_t dst_port;
 669
 670         eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
 671
 672         if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) {
 673                 /* Handle IPv4 headers.*/
 674                 ipv4_hdr =
 675                         rte_pktmbuf_mtod_offset(m, struct ipv4_hdr *,
 676                                                 sizeof(struct ether_hdr));
 677
 678 #ifdef DO_RFC_1812_CHECKS
 679                 /* Check to make sure the packet is valid (RFC1812) */
 680                 if (is_valid_ipv4_pkt(ipv4_hdr, m->pkt_len) < 0) {
 681                         rte_pktmbuf_free(m);
 682                         return;
 683                 }
 684 #endif
 685
 686                 dst_port = get_ipv4_dst_port(ipv4_hdr, portid,
 687                                         qconf->ipv4_lookup_struct);
 688                 if (dst_port >= RTE_MAX_ETHPORTS ||
 689                                 (enabled_port_mask & 1 << dst_port) == 0)
 690                         dst_port = portid;
 691
 692                 /* 02:00:00:00:00:xx */
 693                 d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
 694                 *((uint64_t *)d_addr_bytes) =
 695                         0x000000000002 + ((uint64_t)dst_port << 40);
 696
 697 #ifdef DO_RFC_1812_CHECKS
 698                 /* Update time to live and header checksum */
 699                 --(ipv4_hdr->time_to_live);
 700                 ++(ipv4_hdr->hdr_checksum);
 701 #endif
 702
 703                 /* src addr */
 704                 ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
 705
 706                 send_single_packet(m, dst_port);
 707         } else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) {
 708                 /* Handle IPv6 headers.*/
 709 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 710                 struct ipv6_hdr *ipv6_hdr;
 711
 712                 ipv6_hdr =
 713                         rte_pktmbuf_mtod_offset(m, struct ipv6_hdr *,
 714                                                 sizeof(struct ether_hdr));
 715
 716                 dst_port = get_ipv6_dst_port(ipv6_hdr, portid,
 717                                         qconf->ipv6_lookup_struct);
 718
 719                 if (dst_port >= RTE_MAX_ETHPORTS ||
 720                                 (enabled_port_mask & 1 << dst_port) == 0)
 721                         dst_port = portid;
 722
 723                 /* 02:00:00:00:00:xx */
 724                 d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
 725                 *((uint64_t *)d_addr_bytes) =
 726                         0x000000000002 + ((uint64_t)dst_port << 40);
 727
 728                 /* src addr */
 729                 ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
 730
 731                 send_single_packet(m, dst_port);
 732 #else
 733                 /* We don't currently handle IPv6 packets in LPM mode. */
 734                 rte_pktmbuf_free(m);
 735 #endif
 736         } else
 737                 rte_pktmbuf_free(m);
 738
 739 }
 740
 741 #define MINIMUM_SLEEP_TIME         1
 742 #define SUSPEND_THRESHOLD          300
 743
 744 static inline uint32_t
 745 power_idle_heuristic(uint32_t zero_rx_packet_count)
 746 {
 747         /* If zero count is less than 100,  sleep 1us */
 748         if (zero_rx_packet_count < SUSPEND_THRESHOLD)
 749                 return MINIMUM_SLEEP_TIME;
 750         /* If zero count is less than 1000, sleep 100 us which is the
 751                 minimum latency switching from C3/C6 to C0
 752         */
 753         else
 754                 return SUSPEND_THRESHOLD;
 755 }
 756
 757 static inline enum freq_scale_hint_t
 758 power_freq_scaleup_heuristic(unsigned lcore_id,
 759                              uint16_t port_id,
 760                              uint16_t queue_id)
 761 {
 762 /**
 763  * HW Rx queue size is 128 by default, Rx burst read at maximum 32 entries
 764  * per iteration
 765  */
 766 #define FREQ_GEAR1_RX_PACKET_THRESHOLD             MAX_PKT_BURST
 767 #define FREQ_GEAR2_RX_PACKET_THRESHOLD             (MAX_PKT_BURST*2)
 768 #define FREQ_GEAR3_RX_PACKET_THRESHOLD             (MAX_PKT_BURST*3)
 769 #define FREQ_UP_TREND1_ACC   1
 770 #define FREQ_UP_TREND2_ACC   100
 771 #define FREQ_UP_THRESHOLD    10000
 772
 773         if (likely(rte_eth_rx_descriptor_done(port_id, queue_id,
 774                         FREQ_GEAR3_RX_PACKET_THRESHOLD) > 0)) {
 775                 stats[lcore_id].trend = 0;
 776                 return FREQ_HIGHEST;
 777         } else if (likely(rte_eth_rx_descriptor_done(port_id, queue_id,
 778                         FREQ_GEAR2_RX_PACKET_THRESHOLD) > 0))
 779                 stats[lcore_id].trend += FREQ_UP_TREND2_ACC;
 780         else if (likely(rte_eth_rx_descriptor_done(port_id, queue_id,
 781                         FREQ_GEAR1_RX_PACKET_THRESHOLD) > 0))
 782                 stats[lcore_id].trend += FREQ_UP_TREND1_ACC;
 783
 784         if (likely(stats[lcore_id].trend > FREQ_UP_THRESHOLD)) {
 785                 stats[lcore_id].trend = 0;
 786                 return FREQ_HIGHER;
 787         }
 788
 789         return FREQ_CURRENT;
 790 }
 791
 792 /**
 793  * force polling thread sleep until one-shot rx interrupt triggers
 794  * @param port_id
 795  *  Port id.
 796  * @param queue_id
 797  *  Rx queue id.
 798  * @return
 799  *  0 on success
 800  */
 801 static int
 802 sleep_until_rx_interrupt(int num)
 803 {
 804         struct rte_epoll_event event[num];
 805         int n, i;
 806         uint16_t port_id;
 807         uint8_t queue_id;
 808         void *data;
 809
 810         RTE_LOG(INFO, L3FWD_POWER,
 811                 "lcore %u sleeps until interrupt triggers\n",
 812                 rte_lcore_id());
 813
 814         n = rte_epoll_wait(RTE_EPOLL_PER_THREAD, event, num, -1);
 815         for (i = 0; i < n; i++) {
 816                 data = event[i].epdata.data;
 817                 port_id = ((uintptr_t)data) >> CHAR_BIT;
 818                 queue_id = ((uintptr_t)data) &
 819                         RTE_LEN2MASK(CHAR_BIT, uint8_t);
 820                 rte_eth_dev_rx_intr_disable(port_id, queue_id);
 821                 RTE_LOG(INFO, L3FWD_POWER,
 822                         "lcore %u is waked up from rx interrupt on"
 823                         " port %d queue %d\n",
 824                         rte_lcore_id(), port_id, queue_id);
 825         }
 826
 827         return 0;
 828 }
 829
 830 static void turn_on_intr(struct lcore_conf *qconf)
 831 {
 832         int i;
 833         struct lcore_rx_queue *rx_queue;
 834         uint8_t queue_id;
 835         uint16_t port_id;
 836
 837         for (i = 0; i < qconf->n_rx_queue; ++i) {
 838                 rx_queue = &(qconf->rx_queue_list[i]);
 839                 port_id = rx_queue->port_id;
 840                 queue_id = rx_queue->queue_id;
 841
 842                 rte_spinlock_lock(&(locks[port_id]));
 843                 rte_eth_dev_rx_intr_enable(port_id, queue_id);
 844                 rte_spinlock_unlock(&(locks[port_id]));
 845         }
 846 }
 847
 848 static int event_register(struct lcore_conf *qconf)
 849 {
 850         struct lcore_rx_queue *rx_queue;
 851         uint8_t queueid;
 852         uint16_t portid;
 853         uint32_t data;
 854         int ret;
 855         int i;
 856
 857         for (i = 0; i < qconf->n_rx_queue; ++i) {
 858                 rx_queue = &(qconf->rx_queue_list[i]);
 859                 portid = rx_queue->port_id;
 860                 queueid = rx_queue->queue_id;
 861                 data = portid << CHAR_BIT | queueid;
 862
 863                 ret = rte_eth_dev_rx_intr_ctl_q(portid, queueid,
 864                                                 RTE_EPOLL_PER_THREAD,
 865                                                 RTE_INTR_EVENT_ADD,
 866                                                 (void *)((uintptr_t)data));
 867                 if (ret)
 868                         return ret;
 869         }
 870
 871         return 0;
 872 }
 873
 874 /* main processing loop */
 875 static int
 876 main_loop(__attribute__((unused)) void *dummy)
 877 {
 878         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
 879         unsigned lcore_id;
 880         uint64_t prev_tsc, diff_tsc, cur_tsc;
 881         uint64_t prev_tsc_power = 0, cur_tsc_power, diff_tsc_power;
 882         int i, j, nb_rx;
 883         uint8_t queueid;
 884         uint16_t portid;
 885         struct lcore_conf *qconf;
 886         struct lcore_rx_queue *rx_queue;
 887         enum freq_scale_hint_t lcore_scaleup_hint;
 888         uint32_t lcore_rx_idle_count = 0;
 889         uint32_t lcore_idle_hint = 0;
 890         int intr_en = 0;
 891
 892         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
 893
 894         prev_tsc = 0;
 895
 896         lcore_id = rte_lcore_id();
 897         qconf = &lcore_conf[lcore_id];
 898
 899         if (qconf->n_rx_queue == 0) {
 900                 RTE_LOG(INFO, L3FWD_POWER, "lcore %u has nothing to do\n", lcore_id);
 901                 return 0;
 902         }
 903
 904         RTE_LOG(INFO, L3FWD_POWER, "entering main loop on lcore %u\n", lcore_id);
 905
 906         for (i = 0; i < qconf->n_rx_queue; i++) {
 907                 portid = qconf->rx_queue_list[i].port_id;
 908                 queueid = qconf->rx_queue_list[i].queue_id;
 909                 RTE_LOG(INFO, L3FWD_POWER, " -- lcoreid=%u portid=%u "
 910                         "rxqueueid=%hhu\n", lcore_id, portid, queueid);
 911         }
 912
 913         /* add into event wait list */
 914         if (event_register(qconf) == 0)
 915                 intr_en = 1;
 916         else
 917                 RTE_LOG(INFO, L3FWD_POWER, "RX interrupt won't enable.\n");
 918
 919         while (1) {
 920                 stats[lcore_id].nb_iteration_looped++;
 921
 922                 cur_tsc = rte_rdtsc();
 923                 cur_tsc_power = cur_tsc;
 924
 925                 /*
 926                  * TX burst queue drain
 927                  */
 928                 diff_tsc = cur_tsc - prev_tsc;
 929                 if (unlikely(diff_tsc > drain_tsc)) {
 930                         for (i = 0; i < qconf->n_tx_port; ++i) {
 931                                 portid = qconf->tx_port_id[i];
 932                                 rte_eth_tx_buffer_flush(portid,
 933                                                 qconf->tx_queue_id[portid],
 934                                                 qconf->tx_buffer[portid]);
 935                         }
 936                         prev_tsc = cur_tsc;
 937                 }
 938
 939                 diff_tsc_power = cur_tsc_power - prev_tsc_power;
 940                 if (diff_tsc_power > TIMER_RESOLUTION_CYCLES) {
 941                         rte_timer_manage();
 942                         prev_tsc_power = cur_tsc_power;
 943                 }
 944
 945 start_rx:
 946                 /*
 947                  * Read packet from RX queues
 948                  */
 949                 lcore_scaleup_hint = FREQ_CURRENT;
 950                 lcore_rx_idle_count = 0;
 951                 for (i = 0; i < qconf->n_rx_queue; ++i) {
 952                         rx_queue = &(qconf->rx_queue_list[i]);
 953                         rx_queue->idle_hint = 0;
 954                         portid = rx_queue->port_id;
 955                         queueid = rx_queue->queue_id;
 956
 957                         nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
 958                                                                 MAX_PKT_BURST);
 959
 960                         stats[lcore_id].nb_rx_processed += nb_rx;
 961                         if (unlikely(nb_rx == 0)) {
 962                                 /**
 963                                  * no packet received from rx queue, try to
 964                                  * sleep for a while forcing CPU enter deeper
 965                                  * C states.
 966                                  */
 967                                 rx_queue->zero_rx_packet_count++;
 968
 969                                 if (rx_queue->zero_rx_packet_count <=
 970                                                         MIN_ZERO_POLL_COUNT)
 971                                         continue;
 972
 973                                 rx_queue->idle_hint = power_idle_heuristic(\
 974                                         rx_queue->zero_rx_packet_count);
 975                                 lcore_rx_idle_count++;
 976                         } else {
 977                                 rx_queue->zero_rx_packet_count = 0;
 978
 979                                 /**
 980                                  * do not scale up frequency immediately as
 981                                  * user to kernel space communication is costly
 982                                  * which might impact packet I/O for received
 983                                  * packets.
 984                                  */
 985                                 rx_queue->freq_up_hint =
 986                                         power_freq_scaleup_heuristic(lcore_id,
 987                                                         portid, queueid);
 988                         }
 989
 990                         /* Prefetch first packets */
 991                         for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
 992                                 rte_prefetch0(rte_pktmbuf_mtod(
 993                                                 pkts_burst[j], void *));
 994                         }
 995
 996                         /* Prefetch and forward already prefetched packets */
 997                         for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
 998                                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
 999                                                 j + PREFETCH_OFFSET], void *));
1000                                 l3fwd_simple_forward(pkts_burst[j], portid,
1001                                                                 qconf);
1002                         }
1003
1004                         /* Forward remaining prefetched packets */
1005                         for (; j < nb_rx; j++) {
1006                                 l3fwd_simple_forward(pkts_burst[j], portid,
1007                                                                 qconf);
1008                         }
1009                 }
1010
1011                 if (likely(lcore_rx_idle_count != qconf->n_rx_queue)) {
1012                         for (i = 1, lcore_scaleup_hint =
1013                                 qconf->rx_queue_list[0].freq_up_hint;
1014                                         i < qconf->n_rx_queue; ++i) {
1015                                 rx_queue = &(qconf->rx_queue_list[i]);
1016                                 if (rx_queue->freq_up_hint >
1017                                                 lcore_scaleup_hint)
1018                                         lcore_scaleup_hint =
1019                                                 rx_queue->freq_up_hint;
1020                         }
1021
1022                         if (lcore_scaleup_hint == FREQ_HIGHEST) {
1023                                 if (rte_power_freq_max)
1024                                         rte_power_freq_max(lcore_id);
1025                         } else if (lcore_scaleup_hint == FREQ_HIGHER) {
1026                                 if (rte_power_freq_up)
1027                                         rte_power_freq_up(lcore_id);
1028                         }
1029                 } else {
1030                         /**
1031                          * All Rx queues empty in recent consecutive polls,
1032                          * sleep in a conservative manner, meaning sleep as
1033                          * less as possible.
1034                          */
1035                         for (i = 1, lcore_idle_hint =
1036                                 qconf->rx_queue_list[0].idle_hint;
1037                                         i < qconf->n_rx_queue; ++i) {
1038                                 rx_queue = &(qconf->rx_queue_list[i]);
1039                                 if (rx_queue->idle_hint < lcore_idle_hint)
1040                                         lcore_idle_hint = rx_queue->idle_hint;
1041                         }
1042
1043                         if (lcore_idle_hint < SUSPEND_THRESHOLD)
1044                                 /**
1045                                  * execute "pause" instruction to avoid context
1046                                  * switch which generally take hundred of
1047                                  * microseconds for short sleep.
1048                                  */
1049                                 rte_delay_us(lcore_idle_hint);
1050                         else {
1051                                 /* suspend until rx interrupt trigges */
1052                                 if (intr_en) {
1053                                         turn_on_intr(qconf);
1054                                         sleep_until_rx_interrupt(
1055                                                 qconf->n_rx_queue);
1056                                 }
1057                                 /* start receiving packets immediately */
1058                                 goto start_rx;
1059                         }
1060                         stats[lcore_id].sleep_time += lcore_idle_hint;
1061                 }
1062         }
1063 }
1064
1065 static int
1066 check_lcore_params(void)
1067 {
1068         uint8_t queue, lcore;
1069         uint16_t i;
1070         int socketid;
1071
1072         for (i = 0; i < nb_lcore_params; ++i) {
1073                 queue = lcore_params[i].queue_id;
1074                 if (queue >= MAX_RX_QUEUE_PER_PORT) {
1075                         printf("invalid queue number: %hhu\n", queue);
1076                         return -1;
1077                 }
1078                 lcore = lcore_params[i].lcore_id;
1079                 if (!rte_lcore_is_enabled(lcore)) {
1080                         printf("error: lcore %hhu is not enabled in lcore "
1081                                                         "mask\n", lcore);
1082                         return -1;
1083                 }
1084                 if ((socketid = rte_lcore_to_socket_id(lcore) != 0) &&
1085                                                         (numa_on == 0)) {
1086                         printf("warning: lcore %hhu is on socket %d with numa "
1087                                                 "off\n", lcore, socketid);
1088                 }
1089         }
1090         return 0;
1091 }
1092
1093 static int
1094 check_port_config(const unsigned nb_ports)
1095 {
1096         unsigned portid;
1097         uint16_t i;
1098
1099         for (i = 0; i < nb_lcore_params; ++i) {
1100                 portid = lcore_params[i].port_id;
1101                 if ((enabled_port_mask & (1 << portid)) == 0) {
1102                         printf("port %u is not enabled in port mask\n",
1103                                                                 portid);
1104                         return -1;
1105                 }
1106                 if (portid >= nb_ports) {
1107                         printf("port %u is not present on the board\n",
1108                                                                 portid);
1109                         return -1;
1110                 }
1111         }
1112         return 0;
1113 }
1114
1115 static uint8_t
1116 get_port_n_rx_queues(const uint16_t port)
1117 {
1118         int queue = -1;
1119         uint16_t i;
1120
1121         for (i = 0; i < nb_lcore_params; ++i) {
1122                 if (lcore_params[i].port_id == port &&
1123                                 lcore_params[i].queue_id > queue)
1124                         queue = lcore_params[i].queue_id;
1125         }
1126         return (uint8_t)(++queue);
1127 }
1128
1129 static int
1130 init_lcore_rx_queues(void)
1131 {
1132         uint16_t i, nb_rx_queue;
1133         uint8_t lcore;
1134
1135         for (i = 0; i < nb_lcore_params; ++i) {
1136                 lcore = lcore_params[i].lcore_id;
1137                 nb_rx_queue = lcore_conf[lcore].n_rx_queue;
1138                 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
1139                         printf("error: too many queues (%u) for lcore: %u\n",
1140                                 (unsigned)nb_rx_queue + 1, (unsigned)lcore);
1141                         return -1;
1142                 } else {
1143                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
1144                                 lcore_params[i].port_id;
1145                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
1146                                 lcore_params[i].queue_id;
1147                         lcore_conf[lcore].n_rx_queue++;
1148                 }
1149         }
1150         return 0;
1151 }
1152
1153 /* display usage */
1154 static void
1155 print_usage(const char *prgname)
1156 {
1157         printf ("%s [EAL options] -- -p PORTMASK -P"
1158                 "  [--config (port,queue,lcore)[,(port,queue,lcore]]"
1159                 "  [--enable-jumbo [--max-pkt-len PKTLEN]]\n"
1160                 "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
1161                 "  -P : enable promiscuous mode\n"
1162                 "  --config (port,queue,lcore): rx queues configuration\n"
1163                 "  --no-numa: optional, disable numa awareness\n"
1164                 "  --enable-jumbo: enable jumbo frame"
1165                 " which max packet len is PKTLEN in decimal (64-9600)\n"
1166                 "  --parse-ptype: parse packet type by software\n",
1167                 prgname);
1168 }
1169
1170 static int parse_max_pkt_len(const char *pktlen)
1171 {
1172         char *end = NULL;
1173         unsigned long len;
1174
1175         /* parse decimal string */
1176         len = strtoul(pktlen, &end, 10);
1177         if ((pktlen[0] == '\0') || (end == NULL) || (*end != '\0'))
1178                 return -1;
1179
1180         if (len == 0)
1181                 return -1;
1182
1183         return len;
1184 }
1185
1186 static int
1187 parse_portmask(const char *portmask)
1188 {
1189         char *end = NULL;
1190         unsigned long pm;
1191
1192         /* parse hexadecimal string */
1193         pm = strtoul(portmask, &end, 16);
1194         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
1195                 return -1;
1196
1197         if (pm == 0)
1198                 return -1;
1199
1200         return pm;
1201 }
1202
1203 static int
1204 parse_config(const char *q_arg)
1205 {
1206         char s[256];
1207         const char *p, *p0 = q_arg;
1208         char *end;
1209         enum fieldnames {
1210                 FLD_PORT = 0,
1211                 FLD_QUEUE,
1212                 FLD_LCORE,
1213                 _NUM_FLD
1214         };
1215         unsigned long int_fld[_NUM_FLD];
1216         char *str_fld[_NUM_FLD];
1217         int i;
1218         unsigned size;
1219
1220         nb_lcore_params = 0;
1221
1222         while ((p = strchr(p0,'(')) != NULL) {
1223                 ++p;
1224                 if((p0 = strchr(p,')')) == NULL)
1225                         return -1;
1226
1227                 size = p0 - p;
1228                 if(size >= sizeof(s))
1229                         return -1;
1230
1231                 snprintf(s, sizeof(s), "%.*s", size, p);
1232                 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') !=
1233                                                                 _NUM_FLD)
1234                         return -1;
1235                 for (i = 0; i < _NUM_FLD; i++){
1236                         errno = 0;
1237                         int_fld[i] = strtoul(str_fld[i], &end, 0);
1238                         if (errno != 0 || end == str_fld[i] || int_fld[i] >
1239                                                                         255)
1240                                 return -1;
1241                 }
1242                 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
1243                         printf("exceeded max number of lcore params: %hu\n",
1244                                 nb_lcore_params);
1245                         return -1;
1246                 }
1247                 lcore_params_array[nb_lcore_params].port_id =
1248                                 (uint8_t)int_fld[FLD_PORT];
1249                 lcore_params_array[nb_lcore_params].queue_id =
1250                                 (uint8_t)int_fld[FLD_QUEUE];
1251                 lcore_params_array[nb_lcore_params].lcore_id =
1252                                 (uint8_t)int_fld[FLD_LCORE];
1253                 ++nb_lcore_params;
1254         }
1255         lcore_params = lcore_params_array;
1256
1257         return 0;
1258 }
1259
1260 #define CMD_LINE_OPT_PARSE_PTYPE "parse-ptype"
1261
1262 /* Parse the argument given in the command line of the application */
1263 static int
1264 parse_args(int argc, char **argv)
1265 {
1266         int opt, ret;
1267         char **argvopt;
1268         int option_index;
1269         char *prgname = argv[0];
1270         static struct option lgopts[] = {
1271                 {"config", 1, 0, 0},
1272                 {"no-numa", 0, 0, 0},
1273                 {"enable-jumbo", 0, 0, 0},
1274                 {CMD_LINE_OPT_PARSE_PTYPE, 0, 0, 0},
1275                 {NULL, 0, 0, 0}
1276         };
1277
1278         argvopt = argv;
1279
1280         while ((opt = getopt_long(argc, argvopt, "p:P",
1281                                 lgopts, &option_index)) != EOF) {
1282
1283                 switch (opt) {
1284                 /* portmask */
1285                 case 'p':
1286                         enabled_port_mask = parse_portmask(optarg);
1287                         if (enabled_port_mask == 0) {
1288                                 printf("invalid portmask\n");
1289                                 print_usage(prgname);
1290                                 return -1;
1291                         }
1292                         break;
1293                 case 'P':
1294                         printf("Promiscuous mode selected\n");
1295                         promiscuous_on = 1;
1296                         break;
1297
1298                 /* long options */
1299                 case 0:
1300                         if (!strncmp(lgopts[option_index].name, "config", 6)) {
1301                                 ret = parse_config(optarg);
1302                                 if (ret) {
1303                                         printf("invalid config\n");
1304                                         print_usage(prgname);
1305                                         return -1;
1306                                 }
1307                         }
1308
1309                         if (!strncmp(lgopts[option_index].name,
1310                                                 "no-numa", 7)) {
1311                                 printf("numa is disabled \n");
1312                                 numa_on = 0;
1313                         }
1314
1315                         if (!strncmp(lgopts[option_index].name,
1316                                         "enable-jumbo", 12)) {
1317                                 struct option lenopts =
1318                                         {"max-pkt-len", required_argument, \
1319                                                                         0, 0};
1320
1321                                 printf("jumbo frame is enabled \n");
1322                                 port_conf.rxmode.jumbo_frame = 1;
1323
1324                                 /**
1325                                  * if no max-pkt-len set, use the default value
1326                                  * ETHER_MAX_LEN
1327                                  */
1328                                 if (0 == getopt_long(argc, argvopt, "",
1329                                                 &lenopts, &option_index)) {
1330                                         ret = parse_max_pkt_len(optarg);
1331                                         if ((ret < 64) ||
1332                                                 (ret > MAX_JUMBO_PKT_LEN)){
1333                                                 printf("invalid packet "
1334                                                                 "length\n");
1335                                                 print_usage(prgname);
1336                                                 return -1;
1337                                         }
1338                                         port_conf.rxmode.max_rx_pkt_len = ret;
1339                                 }
1340                                 printf("set jumbo frame "
1341                                         "max packet length to %u\n",
1342                                 (unsigned int)port_conf.rxmode.max_rx_pkt_len);
1343                         }
1344
1345                         if (!strncmp(lgopts[option_index].name,
1346                                      CMD_LINE_OPT_PARSE_PTYPE,
1347                                      sizeof(CMD_LINE_OPT_PARSE_PTYPE))) {
1348                                 printf("soft parse-ptype is enabled\n");
1349                                 parse_ptype = 1;
1350                         }
1351
1352                         break;
1353
1354                 default:
1355                         print_usage(prgname);
1356                         return -1;
1357                 }
1358         }
1359
1360         if (optind >= 0)
1361                 argv[optind-1] = prgname;
1362
1363         ret = optind-1;
1364         optind = 1; /* reset getopt lib */
1365         return ret;
1366 }
1367
1368 static void
1369 print_ethaddr(const char *name, const struct ether_addr *eth_addr)
1370 {
1371         char buf[ETHER_ADDR_FMT_SIZE];
1372         ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr);
1373         printf("%s%s", name, buf);
1374 }
1375
1376 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1377 static void
1378 setup_hash(int socketid)
1379 {
1380         struct rte_hash_parameters ipv4_l3fwd_hash_params = {
1381                 .name = NULL,
1382                 .entries = L3FWD_HASH_ENTRIES,
1383                 .key_len = sizeof(struct ipv4_5tuple),
1384                 .hash_func = DEFAULT_HASH_FUNC,
1385                 .hash_func_init_val = 0,
1386         };
1387
1388         struct rte_hash_parameters ipv6_l3fwd_hash_params = {
1389                 .name = NULL,
1390                 .entries = L3FWD_HASH_ENTRIES,
1391                 .key_len = sizeof(struct ipv6_5tuple),
1392                 .hash_func = DEFAULT_HASH_FUNC,
1393                 .hash_func_init_val = 0,
1394         };
1395
1396         unsigned i;
1397         int ret;
1398         char s[64];
1399
1400         /* create ipv4 hash */
1401         snprintf(s, sizeof(s), "ipv4_l3fwd_hash_%d", socketid);
1402         ipv4_l3fwd_hash_params.name = s;
1403         ipv4_l3fwd_hash_params.socket_id = socketid;
1404         ipv4_l3fwd_lookup_struct[socketid] =
1405                 rte_hash_create(&ipv4_l3fwd_hash_params);
1406         if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
1407                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
1408                                 "socket %d\n", socketid);
1409
1410         /* create ipv6 hash */
1411         snprintf(s, sizeof(s), "ipv6_l3fwd_hash_%d", socketid);
1412         ipv6_l3fwd_hash_params.name = s;
1413         ipv6_l3fwd_hash_params.socket_id = socketid;
1414         ipv6_l3fwd_lookup_struct[socketid] =
1415                 rte_hash_create(&ipv6_l3fwd_hash_params);
1416         if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
1417                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
1418                                 "socket %d\n", socketid);
1419
1420
1421         /* populate the ipv4 hash */
1422         for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
1423                 ret = rte_hash_add_key (ipv4_l3fwd_lookup_struct[socketid],
1424                                 (void *) &ipv4_l3fwd_route_array[i].key);
1425                 if (ret < 0) {
1426                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
1427                                 "l3fwd hash on socket %d\n", i, socketid);
1428                 }
1429                 ipv4_l3fwd_out_if[ret] = ipv4_l3fwd_route_array[i].if_out;
1430                 printf("Hash: Adding key\n");
1431                 print_ipv4_key(ipv4_l3fwd_route_array[i].key);
1432         }
1433
1434         /* populate the ipv6 hash */
1435         for (i = 0; i < IPV6_L3FWD_NUM_ROUTES; i++) {
1436                 ret = rte_hash_add_key (ipv6_l3fwd_lookup_struct[socketid],
1437                                 (void *) &ipv6_l3fwd_route_array[i].key);
1438                 if (ret < 0) {
1439                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
1440                                 "l3fwd hash on socket %d\n", i, socketid);
1441                 }
1442                 ipv6_l3fwd_out_if[ret] = ipv6_l3fwd_route_array[i].if_out;
1443                 printf("Hash: Adding key\n");
1444                 print_ipv6_key(ipv6_l3fwd_route_array[i].key);
1445         }
1446 }
1447 #endif
1448
1449 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1450 static void
1451 setup_lpm(int socketid)
1452 {
1453         unsigned i;
1454         int ret;
1455         char s[64];
1456
1457         /* create the LPM table */
1458         struct rte_lpm_config lpm_ipv4_config;
1459
1460         lpm_ipv4_config.max_rules = IPV4_L3FWD_LPM_MAX_RULES;
1461         lpm_ipv4_config.number_tbl8s = 256;
1462         lpm_ipv4_config.flags = 0;
1463
1464         snprintf(s, sizeof(s), "IPV4_L3FWD_LPM_%d", socketid);
1465         ipv4_l3fwd_lookup_struct[socketid] =
1466                         rte_lpm_create(s, socketid, &lpm_ipv4_config);
1467         if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
1468                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
1469                                 " on socket %d\n", socketid);
1470
1471         /* populate the LPM table */
1472         for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
1473                 ret = rte_lpm_add(ipv4_l3fwd_lookup_struct[socketid],
1474                         ipv4_l3fwd_route_array[i].ip,
1475                         ipv4_l3fwd_route_array[i].depth,
1476                         ipv4_l3fwd_route_array[i].if_out);
1477
1478                 if (ret < 0) {
1479                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
1480                                 "l3fwd LPM table on socket %d\n",
1481                                 i, socketid);
1482                 }
1483
1484                 printf("LPM: Adding route 0x%08x / %d (%d)\n",
1485                         (unsigned)ipv4_l3fwd_route_array[i].ip,
1486                         ipv4_l3fwd_route_array[i].depth,
1487                         ipv4_l3fwd_route_array[i].if_out);
1488         }
1489 }
1490 #endif
1491
1492 static int
1493 init_mem(unsigned nb_mbuf)
1494 {
1495         struct lcore_conf *qconf;
1496         int socketid;
1497         unsigned lcore_id;
1498         char s[64];
1499
1500         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1501                 if (rte_lcore_is_enabled(lcore_id) == 0)
1502                         continue;
1503
1504                 if (numa_on)
1505                         socketid = rte_lcore_to_socket_id(lcore_id);
1506                 else
1507                         socketid = 0;
1508
1509                 if (socketid >= NB_SOCKETS) {
1510                         rte_exit(EXIT_FAILURE, "Socket %d of lcore %u is "
1511                                         "out of range %d\n", socketid,
1512                                                 lcore_id, NB_SOCKETS);
1513                 }
1514                 if (pktmbuf_pool[socketid] == NULL) {
1515                         snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
1516                         pktmbuf_pool[socketid] =
1517                                 rte_pktmbuf_pool_create(s, nb_mbuf,
1518                                         MEMPOOL_CACHE_SIZE, 0,
1519                                         RTE_MBUF_DEFAULT_BUF_SIZE,
1520                                         socketid);
1521                         if (pktmbuf_pool[socketid] == NULL)
1522                                 rte_exit(EXIT_FAILURE,
1523                                         "Cannot init mbuf pool on socket %d\n",
1524                                                                 socketid);
1525                         else
1526                                 printf("Allocated mbuf pool on socket %d\n",
1527                                                                 socketid);
1528
1529 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1530                         setup_lpm(socketid);
1531 #else
1532                         setup_hash(socketid);
1533 #endif
1534                 }
1535                 qconf = &lcore_conf[lcore_id];
1536                 qconf->ipv4_lookup_struct = ipv4_l3fwd_lookup_struct[socketid];
1537 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1538                 qconf->ipv6_lookup_struct = ipv6_l3fwd_lookup_struct[socketid];
1539 #endif
1540         }
1541         return 0;
1542 }
1543
1544 /* Check the link status of all ports in up to 9s, and print them finally */
1545 static void
1546 check_all_ports_link_status(uint16_t port_num, uint32_t port_mask)
1547 {
1548 #define CHECK_INTERVAL 100 /* 100ms */
1549 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
1550         uint8_t count, all_ports_up, print_flag = 0;
1551         uint16_t portid;
1552         struct rte_eth_link link;
1553
1554         printf("\nChecking link status");
1555         fflush(stdout);
1556         for (count = 0; count <= MAX_CHECK_TIME; count++) {
1557                 all_ports_up = 1;
1558                 for (portid = 0; portid < port_num; portid++) {
1559                         if ((port_mask & (1 << portid)) == 0)
1560                                 continue;
1561                         memset(&link, 0, sizeof(link));
1562                         rte_eth_link_get_nowait(portid, &link);
1563                         /* print link status if flag set */
1564                         if (print_flag == 1) {
1565                                 if (link.link_status)
1566                                         printf("Port %d Link Up - speed %u "
1567                                                 "Mbps - %s\n", (uint8_t)portid,
1568                                                 (unsigned)link.link_speed,
1569                                 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
1570                                         ("full-duplex") : ("half-duplex\n"));
1571                                 else
1572                                         printf("Port %d Link Down\n",
1573                                                 (uint8_t)portid);
1574                                 continue;
1575                         }
1576                         /* clear all_ports_up flag if any link down */
1577                         if (link.link_status == ETH_LINK_DOWN) {
1578                                 all_ports_up = 0;
1579                                 break;
1580                         }
1581                 }
1582                 /* after finally printing all link status, get out */
1583                 if (print_flag == 1)
1584                         break;
1585
1586                 if (all_ports_up == 0) {
1587                         printf(".");
1588                         fflush(stdout);
1589                         rte_delay_ms(CHECK_INTERVAL);
1590                 }
1591
1592                 /* set the print_flag if all ports up or timeout */
1593                 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
1594                         print_flag = 1;
1595                         printf("done\n");
1596                 }
1597         }
1598 }
1599
1600 static int check_ptype(uint16_t portid)
1601 {
1602         int i, ret;
1603         int ptype_l3_ipv4 = 0;
1604 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1605         int ptype_l3_ipv6 = 0;
1606 #endif
1607         uint32_t ptype_mask = RTE_PTYPE_L3_MASK;
1608
1609         ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, NULL, 0);
1610         if (ret <= 0)
1611                 return 0;
1612
1613         uint32_t ptypes[ret];
1614
1615         ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, ptypes, ret);
1616         for (i = 0; i < ret; ++i) {
1617                 if (ptypes[i] & RTE_PTYPE_L3_IPV4)
1618                         ptype_l3_ipv4 = 1;
1619 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1620                 if (ptypes[i] & RTE_PTYPE_L3_IPV6)
1621                         ptype_l3_ipv6 = 1;
1622 #endif
1623         }
1624
1625         if (ptype_l3_ipv4 == 0)
1626                 printf("port %d cannot parse RTE_PTYPE_L3_IPV4\n", portid);
1627
1628 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1629         if (ptype_l3_ipv6 == 0)
1630                 printf("port %d cannot parse RTE_PTYPE_L3_IPV6\n", portid);
1631 #endif
1632
1633 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1634         if (ptype_l3_ipv4)
1635 #else /* APP_LOOKUP_EXACT_MATCH */
1636         if (ptype_l3_ipv4 && ptype_l3_ipv6)
1637 #endif
1638                 return 1;
1639
1640         return 0;
1641
1642 }
1643
1644 int
1645 main(int argc, char **argv)
1646 {
1647         struct lcore_conf *qconf;
1648         struct rte_eth_dev_info dev_info;
1649         struct rte_eth_txconf *txconf;
1650         int ret;
1651         uint16_t nb_ports;
1652         uint16_t queueid;
1653         unsigned lcore_id;
1654         uint64_t hz;
1655         uint32_t n_tx_queue, nb_lcores;
1656         uint32_t dev_rxq_num, dev_txq_num;
1657         uint8_t nb_rx_queue, queue, socketid;
1658         uint16_t portid;
1659         uint16_t org_rxq_intr = port_conf.intr_conf.rxq;
1660
1661         /* catch SIGINT and restore cpufreq governor to ondemand */
1662         signal(SIGINT, signal_exit_now);
1663
1664         /* init EAL */
1665         ret = rte_eal_init(argc, argv);
1666         if (ret < 0)
1667                 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
1668         argc -= ret;
1669         argv += ret;
1670
1671         /* init RTE timer library to be used late */
1672         rte_timer_subsystem_init();
1673
1674         /* parse application arguments (after the EAL ones) */
1675         ret = parse_args(argc, argv);
1676         if (ret < 0)
1677                 rte_exit(EXIT_FAILURE, "Invalid L3FWD parameters\n");
1678
1679         if (check_lcore_params() < 0)
1680                 rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
1681
1682         ret = init_lcore_rx_queues();
1683         if (ret < 0)
1684                 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
1685
1686         nb_ports = rte_eth_dev_count();
1687
1688         if (check_port_config(nb_ports) < 0)
1689                 rte_exit(EXIT_FAILURE, "check_port_config failed\n");
1690
1691         nb_lcores = rte_lcore_count();
1692
1693         /* initialize all ports */
1694         for (portid = 0; portid < nb_ports; portid++) {
1695                 /* skip ports that are not enabled */
1696                 if ((enabled_port_mask & (1 << portid)) == 0) {
1697                         printf("\nSkipping disabled port %d\n", portid);
1698                         continue;
1699                 }
1700
1701                 /* init port */
1702                 printf("Initializing port %d ... ", portid );
1703                 fflush(stdout);
1704
1705                 rte_eth_dev_info_get(portid, &dev_info);
1706                 dev_rxq_num = dev_info.max_rx_queues;
1707                 dev_txq_num = dev_info.max_tx_queues;
1708
1709                 nb_rx_queue = get_port_n_rx_queues(portid);
1710                 if (nb_rx_queue > dev_rxq_num)
1711                         rte_exit(EXIT_FAILURE,
1712                                 "Cannot configure not existed rxq: "
1713                                 "port=%d\n", portid);
1714
1715                 n_tx_queue = nb_lcores;
1716                 if (n_tx_queue > dev_txq_num)
1717                         n_tx_queue = dev_txq_num;
1718                 printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
1719                         nb_rx_queue, (unsigned)n_tx_queue );
1720                 /* If number of Rx queue is 0, no need to enable Rx interrupt */
1721                 if (nb_rx_queue == 0)
1722                         port_conf.intr_conf.rxq = 0;
1723                 ret = rte_eth_dev_configure(portid, nb_rx_queue,
1724                                         (uint16_t)n_tx_queue, &port_conf);
1725                 /* Revert to original value */
1726                 port_conf.intr_conf.rxq = org_rxq_intr;
1727                 if (ret < 0)
1728                         rte_exit(EXIT_FAILURE, "Cannot configure device: "
1729                                         "err=%d, port=%d\n", ret, portid);
1730
1731                 ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
1732                                                        &nb_txd);
1733                 if (ret < 0)
1734                         rte_exit(EXIT_FAILURE,
1735                                  "Cannot adjust number of descriptors: err=%d, port=%d\n",
1736                                  ret, portid);
1737
1738                 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
1739                 print_ethaddr(" Address:", &ports_eth_addr[portid]);
1740                 printf(", ");
1741
1742                 /* init memory */
1743                 ret = init_mem(NB_MBUF);
1744                 if (ret < 0)
1745                         rte_exit(EXIT_FAILURE, "init_mem failed\n");
1746
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                         /* Initialize TX buffers */
1752                         qconf = &lcore_conf[lcore_id];
1753                         qconf->tx_buffer[portid] = rte_zmalloc_socket("tx_buffer",
1754                                 RTE_ETH_TX_BUFFER_SIZE(MAX_PKT_BURST), 0,
1755                                 rte_eth_dev_socket_id(portid));
1756                         if (qconf->tx_buffer[portid] == NULL)
1757                                 rte_exit(EXIT_FAILURE, "Can't allocate tx buffer for port %u\n",
1758                                                  portid);
1759
1760                         rte_eth_tx_buffer_init(qconf->tx_buffer[portid], MAX_PKT_BURST);
1761                 }
1762
1763                 /* init one TX queue per couple (lcore,port) */
1764                 queueid = 0;
1765                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1766                         if (rte_lcore_is_enabled(lcore_id) == 0)
1767                                 continue;
1768
1769                         if (queueid >= dev_txq_num)
1770                                 continue;
1771
1772                         if (numa_on)
1773                                 socketid = \
1774                                 (uint8_t)rte_lcore_to_socket_id(lcore_id);
1775                         else
1776                                 socketid = 0;
1777
1778                         printf("txq=%u,%d,%d ", lcore_id, queueid, socketid);
1779                         fflush(stdout);
1780
1781                         rte_eth_dev_info_get(portid, &dev_info);
1782                         txconf = &dev_info.default_txconf;
1783                         if (port_conf.rxmode.jumbo_frame)
1784                                 txconf->txq_flags = 0;
1785                         ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
1786                                                      socketid, txconf);
1787                         if (ret < 0)
1788                                 rte_exit(EXIT_FAILURE,
1789                                         "rte_eth_tx_queue_setup: err=%d, "
1790                                                 "port=%d\n", ret, portid);
1791
1792                         qconf = &lcore_conf[lcore_id];
1793                         qconf->tx_queue_id[portid] = queueid;
1794                         queueid++;
1795
1796                         qconf->tx_port_id[qconf->n_tx_port] = portid;
1797                         qconf->n_tx_port++;
1798                 }
1799                 printf("\n");
1800         }
1801
1802         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1803                 if (rte_lcore_is_enabled(lcore_id) == 0)
1804                         continue;
1805
1806                 /* init power management library */
1807                 ret = rte_power_init(lcore_id);
1808                 if (ret)
1809                         RTE_LOG(ERR, POWER,
1810                                 "Library initialization failed on core %u\n", lcore_id);
1811
1812                 /* init timer structures for each enabled lcore */
1813                 rte_timer_init(&power_timers[lcore_id]);
1814                 hz = rte_get_timer_hz();
1815                 rte_timer_reset(&power_timers[lcore_id],
1816                         hz/TIMER_NUMBER_PER_SECOND, SINGLE, lcore_id,
1817                                                 power_timer_cb, NULL);
1818
1819                 qconf = &lcore_conf[lcore_id];
1820                 printf("\nInitializing rx queues on lcore %u ... ", lcore_id );
1821                 fflush(stdout);
1822                 /* init RX queues */
1823                 for(queue = 0; queue < qconf->n_rx_queue; ++queue) {
1824                         portid = qconf->rx_queue_list[queue].port_id;
1825                         queueid = qconf->rx_queue_list[queue].queue_id;
1826
1827                         if (numa_on)
1828                                 socketid = \
1829                                 (uint8_t)rte_lcore_to_socket_id(lcore_id);
1830                         else
1831                                 socketid = 0;
1832
1833                         printf("rxq=%d,%d,%d ", portid, queueid, socketid);
1834                         fflush(stdout);
1835
1836                         ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
1837                                 socketid, NULL,
1838                                 pktmbuf_pool[socketid]);
1839                         if (ret < 0)
1840                                 rte_exit(EXIT_FAILURE,
1841                                         "rte_eth_rx_queue_setup: err=%d, "
1842                                                 "port=%d\n", ret, portid);
1843
1844                         if (parse_ptype) {
1845                                 if (add_cb_parse_ptype(portid, queueid) < 0)
1846                                         rte_exit(EXIT_FAILURE,
1847                                                  "Fail to add ptype cb\n");
1848                         } else if (!check_ptype(portid))
1849                                 rte_exit(EXIT_FAILURE,
1850                                          "PMD can not provide needed ptypes\n");
1851                 }
1852         }
1853
1854         printf("\n");
1855
1856         /* start ports */
1857         for (portid = 0; portid < nb_ports; portid++) {
1858                 if ((enabled_port_mask & (1 << portid)) == 0) {
1859                         continue;
1860                 }
1861                 /* Start device */
1862                 ret = rte_eth_dev_start(portid);
1863                 if (ret < 0)
1864                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, "
1865                                                 "port=%d\n", ret, portid);
1866                 /*
1867                  * If enabled, put device in promiscuous mode.
1868                  * This allows IO forwarding mode to forward packets
1869                  * to itself through 2 cross-connected  ports of the
1870                  * target machine.
1871                  */
1872                 if (promiscuous_on)
1873                         rte_eth_promiscuous_enable(portid);
1874                 /* initialize spinlock for each port */
1875                 rte_spinlock_init(&(locks[portid]));
1876         }
1877
1878         check_all_ports_link_status(nb_ports, enabled_port_mask);
1879
1880         /* launch per-lcore init on every lcore */
1881         rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
1882         RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1883                 if (rte_eal_wait_lcore(lcore_id) < 0)
1884                         return -1;
1885         }
1886
1887         return 0;
1888 }