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_eal.h>
  54 #include <rte_launch.h>
  55 #include <rte_atomic.h>
  56 #include <rte_cycles.h>
  57 #include <rte_prefetch.h>
  58 #include <rte_lcore.h>
  59 #include <rte_per_lcore.h>
  60 #include <rte_branch_prediction.h>
  61 #include <rte_interrupts.h>
  62 #include <rte_pci.h>
  63 #include <rte_random.h>
  64 #include <rte_debug.h>
  65 #include <rte_ether.h>
  66 #include <rte_ethdev.h>
  67 #include <rte_mempool.h>
  68 #include <rte_mbuf.h>
  69 #include <rte_ip.h>
  70 #include <rte_tcp.h>
  71 #include <rte_udp.h>
  72 #include <rte_string_fns.h>
  73 #include <rte_timer.h>
  74 #include <rte_power.h>
  75 #include <rte_spinlock.h>
  76
  77 #define RTE_LOGTYPE_L3FWD_POWER RTE_LOGTYPE_USER1
  78
  79 #define MAX_PKT_BURST 32
  80
  81 #define MIN_ZERO_POLL_COUNT 10
  82
  83 /* around 100ms at 2 Ghz */
  84 #define TIMER_RESOLUTION_CYCLES           200000000ULL
  85 /* 100 ms interval */
  86 #define TIMER_NUMBER_PER_SECOND           10
  87 /* 100000 us */
  88 #define SCALING_PERIOD                    (1000000/TIMER_NUMBER_PER_SECOND)
  89 #define SCALING_DOWN_TIME_RATIO_THRESHOLD 0.25
  90
  91 #define APP_LOOKUP_EXACT_MATCH          0
  92 #define APP_LOOKUP_LPM                  1
  93 #define DO_RFC_1812_CHECKS
  94
  95 #ifndef APP_LOOKUP_METHOD
  96 #define APP_LOOKUP_METHOD             APP_LOOKUP_LPM
  97 #endif
  98
  99 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 100 #include <rte_hash.h>
 101 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 102 #include <rte_lpm.h>
 103 #else
 104 #error "APP_LOOKUP_METHOD set to incorrect value"
 105 #endif
 106
 107 #ifndef IPv6_BYTES
 108 #define IPv6_BYTES_FMT "%02x%02x:%02x%02x:%02x%02x:%02x%02x:"\
 109                        "%02x%02x:%02x%02x:%02x%02x:%02x%02x"
 110 #define IPv6_BYTES(addr) \
 111         addr[0],  addr[1], addr[2],  addr[3], \
 112         addr[4],  addr[5], addr[6],  addr[7], \
 113         addr[8],  addr[9], addr[10], addr[11],\
 114         addr[12], addr[13],addr[14], addr[15]
 115 #endif
 116
 117 #define MAX_JUMBO_PKT_LEN  9600
 118
 119 #define IPV6_ADDR_LEN 16
 120
 121 #define MEMPOOL_CACHE_SIZE 256
 122
 123 /*
 124  * This expression is used to calculate the number of mbufs needed depending on
 125  * user input, taking into account memory for rx and tx hardware rings, cache
 126  * per lcore and mtable per port per lcore. RTE_MAX is used to ensure that
 127  * NB_MBUF never goes below a minimum value of 8192.
 128  */
 129
 130 #define NB_MBUF RTE_MAX ( \
 131         (nb_ports*nb_rx_queue*nb_rxd + \
 132         nb_ports*nb_lcores*MAX_PKT_BURST + \
 133         nb_ports*n_tx_queue*nb_txd + \
 134         nb_lcores*MEMPOOL_CACHE_SIZE), \
 135         (unsigned)8192)
 136
 137 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
 138
 139 #define NB_SOCKETS 8
 140
 141 /* Configure how many packets ahead to prefetch, when reading packets */
 142 #define PREFETCH_OFFSET 3
 143
 144 /*
 145  * Configurable number of RX/TX ring descriptors
 146  */
 147 #define RTE_TEST_RX_DESC_DEFAULT 512
 148 #define RTE_TEST_TX_DESC_DEFAULT 512
 149 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
 150 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
 151
 152 /* ethernet addresses of ports */
 153 static struct ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
 154
 155 /* ethernet addresses of ports */
 156 static rte_spinlock_t locks[RTE_MAX_ETHPORTS];
 157
 158 /* mask of enabled ports */
 159 static uint32_t enabled_port_mask = 0;
 160 /* Ports set in promiscuous mode off by default. */
 161 static int promiscuous_on = 0;
 162 /* NUMA is enabled by default. */
 163 static int numa_on = 1;
 164 static int parse_ptype; /**< Parse packet type using rx callback, and */
 165                         /**< disabled by default */
 166
 167 enum freq_scale_hint_t
 168 {
 169         FREQ_LOWER    =      -1,
 170         FREQ_CURRENT  =       0,
 171         FREQ_HIGHER   =       1,
 172         FREQ_HIGHEST  =       2
 173 };
 174
 175 struct lcore_rx_queue {
 176         uint16_t port_id;
 177         uint8_t queue_id;
 178         enum freq_scale_hint_t freq_up_hint;
 179         uint32_t zero_rx_packet_count;
 180         uint32_t idle_hint;
 181 } __rte_cache_aligned;
 182
 183 #define MAX_RX_QUEUE_PER_LCORE 16
 184 #define MAX_TX_QUEUE_PER_PORT RTE_MAX_ETHPORTS
 185 #define MAX_RX_QUEUE_PER_PORT 128
 186
 187 #define MAX_RX_QUEUE_INTERRUPT_PER_PORT 16
 188
 189
 190 #define MAX_LCORE_PARAMS 1024
 191 struct lcore_params {
 192         uint16_t port_id;
 193         uint8_t queue_id;
 194         uint8_t lcore_id;
 195 } __rte_cache_aligned;
 196
 197 static struct lcore_params lcore_params_array[MAX_LCORE_PARAMS];
 198 static struct lcore_params lcore_params_array_default[] = {
 199         {0, 0, 2},
 200         {0, 1, 2},
 201         {0, 2, 2},
 202         {1, 0, 2},
 203         {1, 1, 2},
 204         {1, 2, 2},
 205         {2, 0, 2},
 206         {3, 0, 3},
 207         {3, 1, 3},
 208 };
 209
 210 static struct lcore_params * lcore_params = lcore_params_array_default;
 211 static uint16_t nb_lcore_params = sizeof(lcore_params_array_default) /
 212                                 sizeof(lcore_params_array_default[0]);
 213
 214 static struct rte_eth_conf port_conf = {
 215         .rxmode = {
 216                 .mq_mode        = ETH_MQ_RX_RSS,
 217                 .max_rx_pkt_len = ETHER_MAX_LEN,
 218                 .split_hdr_size = 0,
 219                 .header_split   = 0, /**< Header Split disabled */
 220                 .hw_ip_checksum = 1, /**< IP checksum offload enabled */
 221                 .hw_vlan_filter = 0, /**< VLAN filtering disabled */
 222                 .jumbo_frame    = 0, /**< Jumbo Frame Support disabled */
 223                 .hw_strip_crc   = 1, /**< CRC stripped by hardware */
 224         },
 225         .rx_adv_conf = {
 226                 .rss_conf = {
 227                         .rss_key = NULL,
 228                         .rss_hf = ETH_RSS_UDP,
 229                 },
 230         },
 231         .txmode = {
 232                 .mq_mode = ETH_MQ_TX_NONE,
 233         },
 234         .intr_conf = {
 235                 .lsc = 1,
 236                 .rxq = 1,
 237         },
 238 };
 239
 240 static struct rte_mempool * pktmbuf_pool[NB_SOCKETS];
 241
 242
 243 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 244
 245 #ifdef RTE_ARCH_X86
 246 #include <rte_hash_crc.h>
 247 #define DEFAULT_HASH_FUNC       rte_hash_crc
 248 #else
 249 #include <rte_jhash.h>
 250 #define DEFAULT_HASH_FUNC       rte_jhash
 251 #endif
 252
 253 struct ipv4_5tuple {
 254         uint32_t ip_dst;
 255         uint32_t ip_src;
 256         uint16_t port_dst;
 257         uint16_t port_src;
 258         uint8_t  proto;
 259 } __attribute__((__packed__));
 260
 261 struct ipv6_5tuple {
 262         uint8_t  ip_dst[IPV6_ADDR_LEN];
 263         uint8_t  ip_src[IPV6_ADDR_LEN];
 264         uint16_t port_dst;
 265         uint16_t port_src;
 266         uint8_t  proto;
 267 } __attribute__((__packed__));
 268
 269 struct ipv4_l3fwd_route {
 270         struct ipv4_5tuple key;
 271         uint8_t if_out;
 272 };
 273
 274 struct ipv6_l3fwd_route {
 275         struct ipv6_5tuple key;
 276         uint8_t if_out;
 277 };
 278
 279 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
 280         {{IPv4(100,10,0,1), IPv4(200,10,0,1), 101, 11, IPPROTO_TCP}, 0},
 281         {{IPv4(100,20,0,2), IPv4(200,20,0,2), 102, 12, IPPROTO_TCP}, 1},
 282         {{IPv4(100,30,0,3), IPv4(200,30,0,3), 103, 13, IPPROTO_TCP}, 2},
 283         {{IPv4(100,40,0,4), IPv4(200,40,0,4), 104, 14, IPPROTO_TCP}, 3},
 284 };
 285
 286 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
 287         {
 288                 {
 289                         {0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
 290                          0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
 291                         {0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
 292                          0x02, 0x1e, 0x67, 0xff, 0xfe, 0x0d, 0xb6, 0x0a},
 293                          1, 10, IPPROTO_UDP
 294                 }, 4
 295         },
 296 };
 297
 298 typedef struct rte_hash lookup_struct_t;
 299 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
 300 static lookup_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
 301
 302 #define L3FWD_HASH_ENTRIES      1024
 303
 304 #define IPV4_L3FWD_NUM_ROUTES \
 305         (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
 306
 307 #define IPV6_L3FWD_NUM_ROUTES \
 308         (sizeof(ipv6_l3fwd_route_array) / sizeof(ipv6_l3fwd_route_array[0]))
 309
 310 static uint16_t ipv4_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 311 static uint16_t ipv6_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 312 #endif
 313
 314 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 315 struct ipv4_l3fwd_route {
 316         uint32_t ip;
 317         uint8_t  depth;
 318         uint8_t  if_out;
 319 };
 320
 321 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
 322         {IPv4(1,1,1,0), 24, 0},
 323         {IPv4(2,1,1,0), 24, 1},
 324         {IPv4(3,1,1,0), 24, 2},
 325         {IPv4(4,1,1,0), 24, 3},
 326         {IPv4(5,1,1,0), 24, 4},
 327         {IPv4(6,1,1,0), 24, 5},
 328         {IPv4(7,1,1,0), 24, 6},
 329         {IPv4(8,1,1,0), 24, 7},
 330 };
 331
 332 #define IPV4_L3FWD_NUM_ROUTES \
 333         (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
 334
 335 #define IPV4_L3FWD_LPM_MAX_RULES     1024
 336
 337 typedef struct rte_lpm lookup_struct_t;
 338 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
 339 #endif
 340
 341 struct lcore_conf {
 342         uint16_t n_rx_queue;
 343         struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
 344         uint16_t n_tx_port;
 345         uint16_t tx_port_id[RTE_MAX_ETHPORTS];
 346         uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
 347         struct rte_eth_dev_tx_buffer *tx_buffer[RTE_MAX_ETHPORTS];
 348         lookup_struct_t * ipv4_lookup_struct;
 349         lookup_struct_t * ipv6_lookup_struct;
 350 } __rte_cache_aligned;
 351
 352 struct lcore_stats {
 353         /* total sleep time in ms since last frequency scaling down */
 354         uint32_t sleep_time;
 355         /* number of long sleep recently */
 356         uint32_t nb_long_sleep;
 357         /* freq. scaling up trend */
 358         uint32_t trend;
 359         /* total packet processed recently */
 360         uint64_t nb_rx_processed;
 361         /* total iterations looped recently */
 362         uint64_t nb_iteration_looped;
 363         uint32_t padding[9];
 364 } __rte_cache_aligned;
 365
 366 static struct lcore_conf lcore_conf[RTE_MAX_LCORE] __rte_cache_aligned;
 367 static struct lcore_stats stats[RTE_MAX_LCORE] __rte_cache_aligned;
 368 static struct rte_timer power_timers[RTE_MAX_LCORE];
 369
 370 static inline uint32_t power_idle_heuristic(uint32_t zero_rx_packet_count);
 371 static inline enum freq_scale_hint_t power_freq_scaleup_heuristic( \
 372                 unsigned int lcore_id, uint16_t port_id, uint16_t queue_id);
 373
 374 /* exit signal handler */
 375 static void
 376 signal_exit_now(int sigtype)
 377 {
 378         unsigned lcore_id;
 379         unsigned int portid, nb_ports;
 380         int ret;
 381
 382         if (sigtype == SIGINT) {
 383                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
 384                         if (rte_lcore_is_enabled(lcore_id) == 0)
 385                                 continue;
 386
 387                         /* init power management library */
 388                         ret = rte_power_exit(lcore_id);
 389                         if (ret)
 390                                 rte_exit(EXIT_FAILURE, "Power management "
 391                                         "library de-initialization failed on "
 392                                                         "core%u\n", lcore_id);
 393                 }
 394
 395                 nb_ports = rte_eth_dev_count();
 396                 for (portid = 0; portid < nb_ports; portid++) {
 397                         if ((enabled_port_mask & (1 << portid)) == 0)
 398                                 continue;
 399
 400                         rte_eth_dev_stop(portid);
 401                         rte_eth_dev_close(portid);
 402                 }
 403         }
 404
 405         rte_exit(EXIT_SUCCESS, "User forced exit\n");
 406 }
 407
 408 /*  Freqency scale down timer callback */
 409 static void
 410 power_timer_cb(__attribute__((unused)) struct rte_timer *tim,
 411                           __attribute__((unused)) void *arg)
 412 {
 413         uint64_t hz;
 414         float sleep_time_ratio;
 415         unsigned lcore_id = rte_lcore_id();
 416
 417         /* accumulate total execution time in us when callback is invoked */
 418         sleep_time_ratio = (float)(stats[lcore_id].sleep_time) /
 419                                         (float)SCALING_PERIOD;
 420         /**
 421          * check whether need to scale down frequency a step if it sleep a lot.
 422          */
 423         if (sleep_time_ratio >= SCALING_DOWN_TIME_RATIO_THRESHOLD) {
 424                 if (rte_power_freq_down)
 425                         rte_power_freq_down(lcore_id);
 426         }
 427         else if ( (unsigned)(stats[lcore_id].nb_rx_processed /
 428                 stats[lcore_id].nb_iteration_looped) < MAX_PKT_BURST) {
 429                 /**
 430                  * scale down a step if average packet per iteration less
 431                  * than expectation.
 432                  */
 433                 if (rte_power_freq_down)
 434                         rte_power_freq_down(lcore_id);
 435         }
 436
 437         /**
 438          * initialize another timer according to current frequency to ensure
 439          * timer interval is relatively fixed.
 440          */
 441         hz = rte_get_timer_hz();
 442         rte_timer_reset(&power_timers[lcore_id], hz/TIMER_NUMBER_PER_SECOND,
 443                                 SINGLE, lcore_id, power_timer_cb, NULL);
 444
 445         stats[lcore_id].nb_rx_processed = 0;
 446         stats[lcore_id].nb_iteration_looped = 0;
 447
 448         stats[lcore_id].sleep_time = 0;
 449 }
 450
 451 /* Enqueue a single packet, and send burst if queue is filled */
 452 static inline int
 453 send_single_packet(struct rte_mbuf *m, uint16_t port)
 454 {
 455         uint32_t lcore_id;
 456         struct lcore_conf *qconf;
 457
 458         lcore_id = rte_lcore_id();
 459         qconf = &lcore_conf[lcore_id];
 460
 461         rte_eth_tx_buffer(port, qconf->tx_queue_id[port],
 462                         qconf->tx_buffer[port], m);
 463
 464         return 0;
 465 }
 466
 467 #ifdef DO_RFC_1812_CHECKS
 468 static inline int
 469 is_valid_ipv4_pkt(struct ipv4_hdr *pkt, uint32_t link_len)
 470 {
 471         /* From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2 */
 472         /*
 473          * 1. The packet length reported by the Link Layer must be large
 474          * enough to hold the minimum length legal IP datagram (20 bytes).
 475          */
 476         if (link_len < sizeof(struct ipv4_hdr))
 477                 return -1;
 478
 479         /* 2. The IP checksum must be correct. */
 480         /* this is checked in H/W */
 481
 482         /*
 483          * 3. The IP version number must be 4. If the version number is not 4
 484          * then the packet may be another version of IP, such as IPng or
 485          * ST-II.
 486          */
 487         if (((pkt->version_ihl) >> 4) != 4)
 488                 return -3;
 489         /*
 490          * 4. The IP header length field must be large enough to hold the
 491          * minimum length legal IP datagram (20 bytes = 5 words).
 492          */
 493         if ((pkt->version_ihl & 0xf) < 5)
 494                 return -4;
 495
 496         /*
 497          * 5. The IP total length field must be large enough to hold the IP
 498          * datagram header, whose length is specified in the IP header length
 499          * field.
 500          */
 501         if (rte_cpu_to_be_16(pkt->total_length) < sizeof(struct ipv4_hdr))
 502                 return -5;
 503
 504         return 0;
 505 }
 506 #endif
 507
 508 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 509 static void
 510 print_ipv4_key(struct ipv4_5tuple key)
 511 {
 512         printf("IP dst = %08x, IP src = %08x, port dst = %d, port src = %d, "
 513                 "proto = %d\n", (unsigned)key.ip_dst, (unsigned)key.ip_src,
 514                                 key.port_dst, key.port_src, key.proto);
 515 }
 516 static void
 517 print_ipv6_key(struct ipv6_5tuple key)
 518 {
 519         printf( "IP dst = " IPv6_BYTES_FMT ", IP src = " IPv6_BYTES_FMT ", "
 520                 "port dst = %d, port src = %d, proto = %d\n",
 521                 IPv6_BYTES(key.ip_dst), IPv6_BYTES(key.ip_src),
 522                 key.port_dst, key.port_src, key.proto);
 523 }
 524
 525 static inline uint16_t
 526 get_ipv4_dst_port(struct ipv4_hdr *ipv4_hdr, uint16_t portid,
 527                 lookup_struct_t * ipv4_l3fwd_lookup_struct)
 528 {
 529         struct ipv4_5tuple key;
 530         struct tcp_hdr *tcp;
 531         struct udp_hdr *udp;
 532         int ret = 0;
 533
 534         key.ip_dst = rte_be_to_cpu_32(ipv4_hdr->dst_addr);
 535         key.ip_src = rte_be_to_cpu_32(ipv4_hdr->src_addr);
 536         key.proto = ipv4_hdr->next_proto_id;
 537
 538         switch (ipv4_hdr->next_proto_id) {
 539         case IPPROTO_TCP:
 540                 tcp = (struct tcp_hdr *)((unsigned char *)ipv4_hdr +
 541                                         sizeof(struct ipv4_hdr));
 542                 key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
 543                 key.port_src = rte_be_to_cpu_16(tcp->src_port);
 544                 break;
 545
 546         case IPPROTO_UDP:
 547                 udp = (struct udp_hdr *)((unsigned char *)ipv4_hdr +
 548                                         sizeof(struct ipv4_hdr));
 549                 key.port_dst = rte_be_to_cpu_16(udp->dst_port);
 550                 key.port_src = rte_be_to_cpu_16(udp->src_port);
 551                 break;
 552
 553         default:
 554                 key.port_dst = 0;
 555                 key.port_src = 0;
 556                 break;
 557         }
 558
 559         /* Find destination port */
 560         ret = rte_hash_lookup(ipv4_l3fwd_lookup_struct, (const void *)&key);
 561         return ((ret < 0) ? portid : ipv4_l3fwd_out_if[ret]);
 562 }
 563
 564 static inline uint16_t
 565 get_ipv6_dst_port(struct ipv6_hdr *ipv6_hdr, uint16_t portid,
 566                         lookup_struct_t *ipv6_l3fwd_lookup_struct)
 567 {
 568         struct ipv6_5tuple key;
 569         struct tcp_hdr *tcp;
 570         struct udp_hdr *udp;
 571         int ret = 0;
 572
 573         memcpy(key.ip_dst, ipv6_hdr->dst_addr, IPV6_ADDR_LEN);
 574         memcpy(key.ip_src, ipv6_hdr->src_addr, IPV6_ADDR_LEN);
 575
 576         key.proto = ipv6_hdr->proto;
 577
 578         switch (ipv6_hdr->proto) {
 579         case IPPROTO_TCP:
 580                 tcp = (struct tcp_hdr *)((unsigned char *) ipv6_hdr +
 581                                         sizeof(struct ipv6_hdr));
 582                 key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
 583                 key.port_src = rte_be_to_cpu_16(tcp->src_port);
 584                 break;
 585
 586         case IPPROTO_UDP:
 587                 udp = (struct udp_hdr *)((unsigned char *) ipv6_hdr +
 588                                         sizeof(struct ipv6_hdr));
 589                 key.port_dst = rte_be_to_cpu_16(udp->dst_port);
 590                 key.port_src = rte_be_to_cpu_16(udp->src_port);
 591                 break;
 592
 593         default:
 594                 key.port_dst = 0;
 595                 key.port_src = 0;
 596                 break;
 597         }
 598
 599         /* Find destination port */
 600         ret = rte_hash_lookup(ipv6_l3fwd_lookup_struct, (const void *)&key);
 601         return ((ret < 0) ? portid : ipv6_l3fwd_out_if[ret]);
 602 }
 603 #endif
 604
 605 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 606 static inline uint16_t
 607 get_ipv4_dst_port(struct ipv4_hdr *ipv4_hdr, uint16_t portid,
 608                 lookup_struct_t *ipv4_l3fwd_lookup_struct)
 609 {
 610         uint32_t next_hop;
 611
 612         return ((rte_lpm_lookup(ipv4_l3fwd_lookup_struct,
 613                         rte_be_to_cpu_32(ipv4_hdr->dst_addr), &next_hop) == 0)?
 614                         next_hop : portid);
 615 }
 616 #endif
 617
 618 static inline void
 619 parse_ptype_one(struct rte_mbuf *m)
 620 {
 621         struct ether_hdr *eth_hdr;
 622         uint32_t packet_type = RTE_PTYPE_UNKNOWN;
 623         uint16_t ether_type;
 624
 625         eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
 626         ether_type = eth_hdr->ether_type;
 627         if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv4))
 628                 packet_type |= RTE_PTYPE_L3_IPV4_EXT_UNKNOWN;
 629         else if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv6))
 630                 packet_type |= RTE_PTYPE_L3_IPV6_EXT_UNKNOWN;
 631
 632         m->packet_type = packet_type;
 633 }
 634
 635 static uint16_t
 636 cb_parse_ptype(uint16_t port __rte_unused, uint16_t queue __rte_unused,
 637                struct rte_mbuf *pkts[], uint16_t nb_pkts,
 638                uint16_t max_pkts __rte_unused,
 639                void *user_param __rte_unused)
 640 {
 641         unsigned int i;
 642
 643         for (i = 0; i < nb_pkts; ++i)
 644                 parse_ptype_one(pkts[i]);
 645
 646         return nb_pkts;
 647 }
 648
 649 static int
 650 add_cb_parse_ptype(uint16_t portid, uint16_t queueid)
 651 {
 652         printf("Port %d: softly parse packet type info\n", portid);
 653         if (rte_eth_add_rx_callback(portid, queueid, cb_parse_ptype, NULL))
 654                 return 0;
 655
 656         printf("Failed to add rx callback: port=%d\n", portid);
 657         return -1;
 658 }
 659
 660 static inline void
 661 l3fwd_simple_forward(struct rte_mbuf *m, uint16_t portid,
 662                                 struct lcore_conf *qconf)
 663 {
 664         struct ether_hdr *eth_hdr;
 665         struct ipv4_hdr *ipv4_hdr;
 666         void *d_addr_bytes;
 667         uint16_t dst_port;
 668
 669         eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
 670
 671         if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) {
 672                 /* Handle IPv4 headers.*/
 673                 ipv4_hdr =
 674                         rte_pktmbuf_mtod_offset(m, struct ipv4_hdr *,
 675                                                 sizeof(struct ether_hdr));
 676
 677 #ifdef DO_RFC_1812_CHECKS
 678                 /* Check to make sure the packet is valid (RFC1812) */
 679                 if (is_valid_ipv4_pkt(ipv4_hdr, m->pkt_len) < 0) {
 680                         rte_pktmbuf_free(m);
 681                         return;
 682                 }
 683 #endif
 684
 685                 dst_port = get_ipv4_dst_port(ipv4_hdr, portid,
 686                                         qconf->ipv4_lookup_struct);
 687                 if (dst_port >= RTE_MAX_ETHPORTS ||
 688                                 (enabled_port_mask & 1 << dst_port) == 0)
 689                         dst_port = portid;
 690
 691                 /* 02:00:00:00:00:xx */
 692                 d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
 693                 *((uint64_t *)d_addr_bytes) =
 694                         0x000000000002 + ((uint64_t)dst_port << 40);
 695
 696 #ifdef DO_RFC_1812_CHECKS
 697                 /* Update time to live and header checksum */
 698                 --(ipv4_hdr->time_to_live);
 699                 ++(ipv4_hdr->hdr_checksum);
 700 #endif
 701
 702                 /* src addr */
 703                 ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
 704
 705                 send_single_packet(m, dst_port);
 706         } else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) {
 707                 /* Handle IPv6 headers.*/
 708 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 709                 struct ipv6_hdr *ipv6_hdr;
 710
 711                 ipv6_hdr =
 712                         rte_pktmbuf_mtod_offset(m, struct ipv6_hdr *,
 713                                                 sizeof(struct ether_hdr));
 714
 715                 dst_port = get_ipv6_dst_port(ipv6_hdr, portid,
 716                                         qconf->ipv6_lookup_struct);
 717
 718                 if (dst_port >= RTE_MAX_ETHPORTS ||
 719                                 (enabled_port_mask & 1 << dst_port) == 0)
 720                         dst_port = portid;
 721
 722                 /* 02:00:00:00:00:xx */
 723                 d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
 724                 *((uint64_t *)d_addr_bytes) =
 725                         0x000000000002 + ((uint64_t)dst_port << 40);
 726
 727                 /* src addr */
 728                 ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
 729
 730                 send_single_packet(m, dst_port);
 731 #else
 732                 /* We don't currently handle IPv6 packets in LPM mode. */
 733                 rte_pktmbuf_free(m);
 734 #endif
 735         } else
 736                 rte_pktmbuf_free(m);
 737
 738 }
 739
 740 #define MINIMUM_SLEEP_TIME         1
 741 #define SUSPEND_THRESHOLD          300
 742
 743 static inline uint32_t
 744 power_idle_heuristic(uint32_t zero_rx_packet_count)
 745 {
 746         /* If zero count is less than 100,  sleep 1us */
 747         if (zero_rx_packet_count < SUSPEND_THRESHOLD)
 748                 return MINIMUM_SLEEP_TIME;
 749         /* If zero count is less than 1000, sleep 100 us which is the
 750                 minimum latency switching from C3/C6 to C0
 751         */
 752         else
 753                 return SUSPEND_THRESHOLD;
 754 }
 755
 756 static inline enum freq_scale_hint_t
 757 power_freq_scaleup_heuristic(unsigned lcore_id,
 758                              uint16_t port_id,
 759                              uint16_t queue_id)
 760 {
 761 /**
 762  * HW Rx queue size is 128 by default, Rx burst read at maximum 32 entries
 763  * per iteration
 764  */
 765 #define FREQ_GEAR1_RX_PACKET_THRESHOLD             MAX_PKT_BURST
 766 #define FREQ_GEAR2_RX_PACKET_THRESHOLD             (MAX_PKT_BURST*2)
 767 #define FREQ_GEAR3_RX_PACKET_THRESHOLD             (MAX_PKT_BURST*3)
 768 #define FREQ_UP_TREND1_ACC   1
 769 #define FREQ_UP_TREND2_ACC   100
 770 #define FREQ_UP_THRESHOLD    10000
 771
 772         if (likely(rte_eth_rx_descriptor_done(port_id, queue_id,
 773                         FREQ_GEAR3_RX_PACKET_THRESHOLD) > 0)) {
 774                 stats[lcore_id].trend = 0;
 775                 return FREQ_HIGHEST;
 776         } else if (likely(rte_eth_rx_descriptor_done(port_id, queue_id,
 777                         FREQ_GEAR2_RX_PACKET_THRESHOLD) > 0))
 778                 stats[lcore_id].trend += FREQ_UP_TREND2_ACC;
 779         else if (likely(rte_eth_rx_descriptor_done(port_id, queue_id,
 780                         FREQ_GEAR1_RX_PACKET_THRESHOLD) > 0))
 781                 stats[lcore_id].trend += FREQ_UP_TREND1_ACC;
 782
 783         if (likely(stats[lcore_id].trend > FREQ_UP_THRESHOLD)) {
 784                 stats[lcore_id].trend = 0;
 785                 return FREQ_HIGHER;
 786         }
 787
 788         return FREQ_CURRENT;
 789 }
 790
 791 /**
 792  * force polling thread sleep until one-shot rx interrupt triggers
 793  * @param port_id
 794  *  Port id.
 795  * @param queue_id
 796  *  Rx queue id.
 797  * @return
 798  *  0 on success
 799  */
 800 static int
 801 sleep_until_rx_interrupt(int num)
 802 {
 803         struct rte_epoll_event event[num];
 804         int n, i;
 805         uint16_t port_id;
 806         uint8_t queue_id;
 807         void *data;
 808
 809         RTE_LOG(INFO, L3FWD_POWER,
 810                 "lcore %u sleeps until interrupt triggers\n",
 811                 rte_lcore_id());
 812
 813         n = rte_epoll_wait(RTE_EPOLL_PER_THREAD, event, num, -1);
 814         for (i = 0; i < n; i++) {
 815                 data = event[i].epdata.data;
 816                 port_id = ((uintptr_t)data) >> CHAR_BIT;
 817                 queue_id = ((uintptr_t)data) &
 818                         RTE_LEN2MASK(CHAR_BIT, uint8_t);
 819                 rte_eth_dev_rx_intr_disable(port_id, queue_id);
 820                 RTE_LOG(INFO, L3FWD_POWER,
 821                         "lcore %u is waked up from rx interrupt on"
 822                         " port %d queue %d\n",
 823                         rte_lcore_id(), port_id, queue_id);
 824         }
 825
 826         return 0;
 827 }
 828
 829 static void turn_on_intr(struct lcore_conf *qconf)
 830 {
 831         int i;
 832         struct lcore_rx_queue *rx_queue;
 833         uint8_t queue_id;
 834         uint16_t port_id;
 835
 836         for (i = 0; i < qconf->n_rx_queue; ++i) {
 837                 rx_queue = &(qconf->rx_queue_list[i]);
 838                 port_id = rx_queue->port_id;
 839                 queue_id = rx_queue->queue_id;
 840
 841                 rte_spinlock_lock(&(locks[port_id]));
 842                 rte_eth_dev_rx_intr_enable(port_id, queue_id);
 843                 rte_spinlock_unlock(&(locks[port_id]));
 844         }
 845 }
 846
 847 static int event_register(struct lcore_conf *qconf)
 848 {
 849         struct lcore_rx_queue *rx_queue;
 850         uint8_t queueid;
 851         uint16_t portid;
 852         uint32_t data;
 853         int ret;
 854         int i;
 855
 856         for (i = 0; i < qconf->n_rx_queue; ++i) {
 857                 rx_queue = &(qconf->rx_queue_list[i]);
 858                 portid = rx_queue->port_id;
 859                 queueid = rx_queue->queue_id;
 860                 data = portid << CHAR_BIT | queueid;
 861
 862                 ret = rte_eth_dev_rx_intr_ctl_q(portid, queueid,
 863                                                 RTE_EPOLL_PER_THREAD,
 864                                                 RTE_INTR_EVENT_ADD,
 865                                                 (void *)((uintptr_t)data));
 866                 if (ret)
 867                         return ret;
 868         }
 869
 870         return 0;
 871 }
 872
 873 /* main processing loop */
 874 static int
 875 main_loop(__attribute__((unused)) void *dummy)
 876 {
 877         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
 878         unsigned lcore_id;
 879         uint64_t prev_tsc, diff_tsc, cur_tsc;
 880         uint64_t prev_tsc_power = 0, cur_tsc_power, diff_tsc_power;
 881         int i, j, nb_rx;
 882         uint8_t queueid;
 883         uint16_t portid;
 884         struct lcore_conf *qconf;
 885         struct lcore_rx_queue *rx_queue;
 886         enum freq_scale_hint_t lcore_scaleup_hint;
 887         uint32_t lcore_rx_idle_count = 0;
 888         uint32_t lcore_idle_hint = 0;
 889         int intr_en = 0;
 890
 891         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
 892
 893         prev_tsc = 0;
 894
 895         lcore_id = rte_lcore_id();
 896         qconf = &lcore_conf[lcore_id];
 897
 898         if (qconf->n_rx_queue == 0) {
 899                 RTE_LOG(INFO, L3FWD_POWER, "lcore %u has nothing to do\n", lcore_id);
 900                 return 0;
 901         }
 902
 903         RTE_LOG(INFO, L3FWD_POWER, "entering main loop on lcore %u\n", lcore_id);
 904
 905         for (i = 0; i < qconf->n_rx_queue; i++) {
 906                 portid = qconf->rx_queue_list[i].port_id;
 907                 queueid = qconf->rx_queue_list[i].queue_id;
 908                 RTE_LOG(INFO, L3FWD_POWER, " -- lcoreid=%u portid=%u "
 909                         "rxqueueid=%hhu\n", lcore_id, portid, queueid);
 910         }
 911
 912         /* add into event wait list */
 913         if (event_register(qconf) == 0)
 914                 intr_en = 1;
 915         else
 916                 RTE_LOG(INFO, L3FWD_POWER, "RX interrupt won't enable.\n");
 917
 918         while (1) {
 919                 stats[lcore_id].nb_iteration_looped++;
 920
 921                 cur_tsc = rte_rdtsc();
 922                 cur_tsc_power = cur_tsc;
 923
 924                 /*
 925                  * TX burst queue drain
 926                  */
 927                 diff_tsc = cur_tsc - prev_tsc;
 928                 if (unlikely(diff_tsc > drain_tsc)) {
 929                         for (i = 0; i < qconf->n_tx_port; ++i) {
 930                                 portid = qconf->tx_port_id[i];
 931                                 rte_eth_tx_buffer_flush(portid,
 932                                                 qconf->tx_queue_id[portid],
 933                                                 qconf->tx_buffer[portid]);
 934                         }
 935                         prev_tsc = cur_tsc;
 936                 }
 937
 938                 diff_tsc_power = cur_tsc_power - prev_tsc_power;
 939                 if (diff_tsc_power > TIMER_RESOLUTION_CYCLES) {
 940                         rte_timer_manage();
 941                         prev_tsc_power = cur_tsc_power;
 942                 }
 943
 944 start_rx:
 945                 /*
 946                  * Read packet from RX queues
 947                  */
 948                 lcore_scaleup_hint = FREQ_CURRENT;
 949                 lcore_rx_idle_count = 0;
 950                 for (i = 0; i < qconf->n_rx_queue; ++i) {
 951                         rx_queue = &(qconf->rx_queue_list[i]);
 952                         rx_queue->idle_hint = 0;
 953                         portid = rx_queue->port_id;
 954                         queueid = rx_queue->queue_id;
 955
 956                         nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
 957                                                                 MAX_PKT_BURST);
 958
 959                         stats[lcore_id].nb_rx_processed += nb_rx;
 960                         if (unlikely(nb_rx == 0)) {
 961                                 /**
 962                                  * no packet received from rx queue, try to
 963                                  * sleep for a while forcing CPU enter deeper
 964                                  * C states.
 965                                  */
 966                                 rx_queue->zero_rx_packet_count++;
 967
 968                                 if (rx_queue->zero_rx_packet_count <=
 969                                                         MIN_ZERO_POLL_COUNT)
 970                                         continue;
 971
 972                                 rx_queue->idle_hint = power_idle_heuristic(\
 973                                         rx_queue->zero_rx_packet_count);
 974                                 lcore_rx_idle_count++;
 975                         } else {
 976                                 rx_queue->zero_rx_packet_count = 0;
 977
 978                                 /**
 979                                  * do not scale up frequency immediately as
 980                                  * user to kernel space communication is costly
 981                                  * which might impact packet I/O for received
 982                                  * packets.
 983                                  */
 984                                 rx_queue->freq_up_hint =
 985                                         power_freq_scaleup_heuristic(lcore_id,
 986                                                         portid, queueid);
 987                         }
 988
 989                         /* Prefetch first packets */
 990                         for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
 991                                 rte_prefetch0(rte_pktmbuf_mtod(
 992                                                 pkts_burst[j], void *));
 993                         }
 994
 995                         /* Prefetch and forward already prefetched packets */
 996                         for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
 997                                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
 998                                                 j + PREFETCH_OFFSET], void *));
 999                                 l3fwd_simple_forward(pkts_burst[j], portid,
1000                                                                 qconf);
1001                         }
1002
1003                         /* Forward remaining prefetched packets */
1004                         for (; j < nb_rx; j++) {
1005                                 l3fwd_simple_forward(pkts_burst[j], portid,
1006                                                                 qconf);
1007                         }
1008                 }
1009
1010                 if (likely(lcore_rx_idle_count != qconf->n_rx_queue)) {
1011                         for (i = 1, lcore_scaleup_hint =
1012                                 qconf->rx_queue_list[0].freq_up_hint;
1013                                         i < qconf->n_rx_queue; ++i) {
1014                                 rx_queue = &(qconf->rx_queue_list[i]);
1015                                 if (rx_queue->freq_up_hint >
1016                                                 lcore_scaleup_hint)
1017                                         lcore_scaleup_hint =
1018                                                 rx_queue->freq_up_hint;
1019                         }
1020
1021                         if (lcore_scaleup_hint == FREQ_HIGHEST) {
1022                                 if (rte_power_freq_max)
1023                                         rte_power_freq_max(lcore_id);
1024                         } else if (lcore_scaleup_hint == FREQ_HIGHER) {
1025                                 if (rte_power_freq_up)
1026                                         rte_power_freq_up(lcore_id);
1027                         }
1028                 } else {
1029                         /**
1030                          * All Rx queues empty in recent consecutive polls,
1031                          * sleep in a conservative manner, meaning sleep as
1032                          * less as possible.
1033                          */
1034                         for (i = 1, lcore_idle_hint =
1035                                 qconf->rx_queue_list[0].idle_hint;
1036                                         i < qconf->n_rx_queue; ++i) {
1037                                 rx_queue = &(qconf->rx_queue_list[i]);
1038                                 if (rx_queue->idle_hint < lcore_idle_hint)
1039                                         lcore_idle_hint = rx_queue->idle_hint;
1040                         }
1041
1042                         if (lcore_idle_hint < SUSPEND_THRESHOLD)
1043                                 /**
1044                                  * execute "pause" instruction to avoid context
1045                                  * switch which generally take hundred of
1046                                  * microseconds for short sleep.
1047                                  */
1048                                 rte_delay_us(lcore_idle_hint);
1049                         else {
1050                                 /* suspend until rx interrupt trigges */
1051                                 if (intr_en) {
1052                                         turn_on_intr(qconf);
1053                                         sleep_until_rx_interrupt(
1054                                                 qconf->n_rx_queue);
1055                                 }
1056                                 /* start receiving packets immediately */
1057                                 goto start_rx;
1058                         }
1059                         stats[lcore_id].sleep_time += lcore_idle_hint;
1060                 }
1061         }
1062 }
1063
1064 static int
1065 check_lcore_params(void)
1066 {
1067         uint8_t queue, lcore;
1068         uint16_t i;
1069         int socketid;
1070
1071         for (i = 0; i < nb_lcore_params; ++i) {
1072                 queue = lcore_params[i].queue_id;
1073                 if (queue >= MAX_RX_QUEUE_PER_PORT) {
1074                         printf("invalid queue number: %hhu\n", queue);
1075                         return -1;
1076                 }
1077                 lcore = lcore_params[i].lcore_id;
1078                 if (!rte_lcore_is_enabled(lcore)) {
1079                         printf("error: lcore %hhu is not enabled in lcore "
1080                                                         "mask\n", lcore);
1081                         return -1;
1082                 }
1083                 if ((socketid = rte_lcore_to_socket_id(lcore) != 0) &&
1084                                                         (numa_on == 0)) {
1085                         printf("warning: lcore %hhu is on socket %d with numa "
1086                                                 "off\n", lcore, socketid);
1087                 }
1088         }
1089         return 0;
1090 }
1091
1092 static int
1093 check_port_config(const unsigned nb_ports)
1094 {
1095         unsigned portid;
1096         uint16_t i;
1097
1098         for (i = 0; i < nb_lcore_params; ++i) {
1099                 portid = lcore_params[i].port_id;
1100                 if ((enabled_port_mask & (1 << portid)) == 0) {
1101                         printf("port %u is not enabled in port mask\n",
1102                                                                 portid);
1103                         return -1;
1104                 }
1105                 if (portid >= nb_ports) {
1106                         printf("port %u is not present on the board\n",
1107                                                                 portid);
1108                         return -1;
1109                 }
1110         }
1111         return 0;
1112 }
1113
1114 static uint8_t
1115 get_port_n_rx_queues(const uint16_t port)
1116 {
1117         int queue = -1;
1118         uint16_t i;
1119
1120         for (i = 0; i < nb_lcore_params; ++i) {
1121                 if (lcore_params[i].port_id == port &&
1122                                 lcore_params[i].queue_id > queue)
1123                         queue = lcore_params[i].queue_id;
1124         }
1125         return (uint8_t)(++queue);
1126 }
1127
1128 static int
1129 init_lcore_rx_queues(void)
1130 {
1131         uint16_t i, nb_rx_queue;
1132         uint8_t lcore;
1133
1134         for (i = 0; i < nb_lcore_params; ++i) {
1135                 lcore = lcore_params[i].lcore_id;
1136                 nb_rx_queue = lcore_conf[lcore].n_rx_queue;
1137                 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
1138                         printf("error: too many queues (%u) for lcore: %u\n",
1139                                 (unsigned)nb_rx_queue + 1, (unsigned)lcore);
1140                         return -1;
1141                 } else {
1142                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
1143                                 lcore_params[i].port_id;
1144                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
1145                                 lcore_params[i].queue_id;
1146                         lcore_conf[lcore].n_rx_queue++;
1147                 }
1148         }
1149         return 0;
1150 }
1151
1152 /* display usage */
1153 static void
1154 print_usage(const char *prgname)
1155 {
1156         printf ("%s [EAL options] -- -p PORTMASK -P"
1157                 "  [--config (port,queue,lcore)[,(port,queue,lcore]]"
1158                 "  [--enable-jumbo [--max-pkt-len PKTLEN]]\n"
1159                 "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
1160                 "  -P : enable promiscuous mode\n"
1161                 "  --config (port,queue,lcore): rx queues configuration\n"
1162                 "  --no-numa: optional, disable numa awareness\n"
1163                 "  --enable-jumbo: enable jumbo frame"
1164                 " which max packet len is PKTLEN in decimal (64-9600)\n"
1165                 "  --parse-ptype: parse packet type by software\n",
1166                 prgname);
1167 }
1168
1169 static int parse_max_pkt_len(const char *pktlen)
1170 {
1171         char *end = NULL;
1172         unsigned long len;
1173
1174         /* parse decimal string */
1175         len = strtoul(pktlen, &end, 10);
1176         if ((pktlen[0] == '\0') || (end == NULL) || (*end != '\0'))
1177                 return -1;
1178
1179         if (len == 0)
1180                 return -1;
1181
1182         return len;
1183 }
1184
1185 static int
1186 parse_portmask(const char *portmask)
1187 {
1188         char *end = NULL;
1189         unsigned long pm;
1190
1191         /* parse hexadecimal string */
1192         pm = strtoul(portmask, &end, 16);
1193         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
1194                 return -1;
1195
1196         if (pm == 0)
1197                 return -1;
1198
1199         return pm;
1200 }
1201
1202 static int
1203 parse_config(const char *q_arg)
1204 {
1205         char s[256];
1206         const char *p, *p0 = q_arg;
1207         char *end;
1208         enum fieldnames {
1209                 FLD_PORT = 0,
1210                 FLD_QUEUE,
1211                 FLD_LCORE,
1212                 _NUM_FLD
1213         };
1214         unsigned long int_fld[_NUM_FLD];
1215         char *str_fld[_NUM_FLD];
1216         int i;
1217         unsigned size;
1218
1219         nb_lcore_params = 0;
1220
1221         while ((p = strchr(p0,'(')) != NULL) {
1222                 ++p;
1223                 if((p0 = strchr(p,')')) == NULL)
1224                         return -1;
1225
1226                 size = p0 - p;
1227                 if(size >= sizeof(s))
1228                         return -1;
1229
1230                 snprintf(s, sizeof(s), "%.*s", size, p);
1231                 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') !=
1232                                                                 _NUM_FLD)
1233                         return -1;
1234                 for (i = 0; i < _NUM_FLD; i++){
1235                         errno = 0;
1236                         int_fld[i] = strtoul(str_fld[i], &end, 0);
1237                         if (errno != 0 || end == str_fld[i] || int_fld[i] >
1238                                                                         255)
1239                                 return -1;
1240                 }
1241                 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
1242                         printf("exceeded max number of lcore params: %hu\n",
1243                                 nb_lcore_params);
1244                         return -1;
1245                 }
1246                 lcore_params_array[nb_lcore_params].port_id =
1247                                 (uint8_t)int_fld[FLD_PORT];
1248                 lcore_params_array[nb_lcore_params].queue_id =
1249                                 (uint8_t)int_fld[FLD_QUEUE];
1250                 lcore_params_array[nb_lcore_params].lcore_id =
1251                                 (uint8_t)int_fld[FLD_LCORE];
1252                 ++nb_lcore_params;
1253         }
1254         lcore_params = lcore_params_array;
1255
1256         return 0;
1257 }
1258
1259 #define CMD_LINE_OPT_PARSE_PTYPE "parse-ptype"
1260
1261 /* Parse the argument given in the command line of the application */
1262 static int
1263 parse_args(int argc, char **argv)
1264 {
1265         int opt, ret;
1266         char **argvopt;
1267         int option_index;
1268         char *prgname = argv[0];
1269         static struct option lgopts[] = {
1270                 {"config", 1, 0, 0},
1271                 {"no-numa", 0, 0, 0},
1272                 {"enable-jumbo", 0, 0, 0},
1273                 {CMD_LINE_OPT_PARSE_PTYPE, 0, 0, 0},
1274                 {NULL, 0, 0, 0}
1275         };
1276
1277         argvopt = argv;
1278
1279         while ((opt = getopt_long(argc, argvopt, "p:P",
1280                                 lgopts, &option_index)) != EOF) {
1281
1282                 switch (opt) {
1283                 /* portmask */
1284                 case 'p':
1285                         enabled_port_mask = parse_portmask(optarg);
1286                         if (enabled_port_mask == 0) {
1287                                 printf("invalid portmask\n");
1288                                 print_usage(prgname);
1289                                 return -1;
1290                         }
1291                         break;
1292                 case 'P':
1293                         printf("Promiscuous mode selected\n");
1294                         promiscuous_on = 1;
1295                         break;
1296
1297                 /* long options */
1298                 case 0:
1299                         if (!strncmp(lgopts[option_index].name, "config", 6)) {
1300                                 ret = parse_config(optarg);
1301                                 if (ret) {
1302                                         printf("invalid config\n");
1303                                         print_usage(prgname);
1304                                         return -1;
1305                                 }
1306                         }
1307
1308                         if (!strncmp(lgopts[option_index].name,
1309                                                 "no-numa", 7)) {
1310                                 printf("numa is disabled \n");
1311                                 numa_on = 0;
1312                         }
1313
1314                         if (!strncmp(lgopts[option_index].name,
1315                                         "enable-jumbo", 12)) {
1316                                 struct option lenopts =
1317                                         {"max-pkt-len", required_argument, \
1318                                                                         0, 0};
1319
1320                                 printf("jumbo frame is enabled \n");
1321                                 port_conf.rxmode.jumbo_frame = 1;
1322
1323                                 /**
1324                                  * if no max-pkt-len set, use the default value
1325                                  * ETHER_MAX_LEN
1326                                  */
1327                                 if (0 == getopt_long(argc, argvopt, "",
1328                                                 &lenopts, &option_index)) {
1329                                         ret = parse_max_pkt_len(optarg);
1330                                         if ((ret < 64) ||
1331                                                 (ret > MAX_JUMBO_PKT_LEN)){
1332                                                 printf("invalid packet "
1333                                                                 "length\n");
1334                                                 print_usage(prgname);
1335                                                 return -1;
1336                                         }
1337                                         port_conf.rxmode.max_rx_pkt_len = ret;
1338                                 }
1339                                 printf("set jumbo frame "
1340                                         "max packet length to %u\n",
1341                                 (unsigned int)port_conf.rxmode.max_rx_pkt_len);
1342                         }
1343
1344                         if (!strncmp(lgopts[option_index].name,
1345                                      CMD_LINE_OPT_PARSE_PTYPE,
1346                                      sizeof(CMD_LINE_OPT_PARSE_PTYPE))) {
1347                                 printf("soft parse-ptype is enabled\n");
1348                                 parse_ptype = 1;
1349                         }
1350
1351                         break;
1352
1353                 default:
1354                         print_usage(prgname);
1355                         return -1;
1356                 }
1357         }
1358
1359         if (optind >= 0)
1360                 argv[optind-1] = prgname;
1361
1362         ret = optind-1;
1363         optind = 1; /* reset getopt lib */
1364         return ret;
1365 }
1366
1367 static void
1368 print_ethaddr(const char *name, const struct ether_addr *eth_addr)
1369 {
1370         char buf[ETHER_ADDR_FMT_SIZE];
1371         ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr);
1372         printf("%s%s", name, buf);
1373 }
1374
1375 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1376 static void
1377 setup_hash(int socketid)
1378 {
1379         struct rte_hash_parameters ipv4_l3fwd_hash_params = {
1380                 .name = NULL,
1381                 .entries = L3FWD_HASH_ENTRIES,
1382                 .key_len = sizeof(struct ipv4_5tuple),
1383                 .hash_func = DEFAULT_HASH_FUNC,
1384                 .hash_func_init_val = 0,
1385         };
1386
1387         struct rte_hash_parameters ipv6_l3fwd_hash_params = {
1388                 .name = NULL,
1389                 .entries = L3FWD_HASH_ENTRIES,
1390                 .key_len = sizeof(struct ipv6_5tuple),
1391                 .hash_func = DEFAULT_HASH_FUNC,
1392                 .hash_func_init_val = 0,
1393         };
1394
1395         unsigned i;
1396         int ret;
1397         char s[64];
1398
1399         /* create ipv4 hash */
1400         snprintf(s, sizeof(s), "ipv4_l3fwd_hash_%d", socketid);
1401         ipv4_l3fwd_hash_params.name = s;
1402         ipv4_l3fwd_hash_params.socket_id = socketid;
1403         ipv4_l3fwd_lookup_struct[socketid] =
1404                 rte_hash_create(&ipv4_l3fwd_hash_params);
1405         if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
1406                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
1407                                 "socket %d\n", socketid);
1408
1409         /* create ipv6 hash */
1410         snprintf(s, sizeof(s), "ipv6_l3fwd_hash_%d", socketid);
1411         ipv6_l3fwd_hash_params.name = s;
1412         ipv6_l3fwd_hash_params.socket_id = socketid;
1413         ipv6_l3fwd_lookup_struct[socketid] =
1414                 rte_hash_create(&ipv6_l3fwd_hash_params);
1415         if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
1416                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
1417                                 "socket %d\n", socketid);
1418
1419
1420         /* populate the ipv4 hash */
1421         for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
1422                 ret = rte_hash_add_key (ipv4_l3fwd_lookup_struct[socketid],
1423                                 (void *) &ipv4_l3fwd_route_array[i].key);
1424                 if (ret < 0) {
1425                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
1426                                 "l3fwd hash on socket %d\n", i, socketid);
1427                 }
1428                 ipv4_l3fwd_out_if[ret] = ipv4_l3fwd_route_array[i].if_out;
1429                 printf("Hash: Adding key\n");
1430                 print_ipv4_key(ipv4_l3fwd_route_array[i].key);
1431         }
1432
1433         /* populate the ipv6 hash */
1434         for (i = 0; i < IPV6_L3FWD_NUM_ROUTES; i++) {
1435                 ret = rte_hash_add_key (ipv6_l3fwd_lookup_struct[socketid],
1436                                 (void *) &ipv6_l3fwd_route_array[i].key);
1437                 if (ret < 0) {
1438                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
1439                                 "l3fwd hash on socket %d\n", i, socketid);
1440                 }
1441                 ipv6_l3fwd_out_if[ret] = ipv6_l3fwd_route_array[i].if_out;
1442                 printf("Hash: Adding key\n");
1443                 print_ipv6_key(ipv6_l3fwd_route_array[i].key);
1444         }
1445 }
1446 #endif
1447
1448 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1449 static void
1450 setup_lpm(int socketid)
1451 {
1452         unsigned i;
1453         int ret;
1454         char s[64];
1455
1456         /* create the LPM table */
1457         struct rte_lpm_config lpm_ipv4_config;
1458
1459         lpm_ipv4_config.max_rules = IPV4_L3FWD_LPM_MAX_RULES;
1460         lpm_ipv4_config.number_tbl8s = 256;
1461         lpm_ipv4_config.flags = 0;
1462
1463         snprintf(s, sizeof(s), "IPV4_L3FWD_LPM_%d", socketid);
1464         ipv4_l3fwd_lookup_struct[socketid] =
1465                         rte_lpm_create(s, socketid, &lpm_ipv4_config);
1466         if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
1467                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
1468                                 " on socket %d\n", socketid);
1469
1470         /* populate the LPM table */
1471         for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
1472                 ret = rte_lpm_add(ipv4_l3fwd_lookup_struct[socketid],
1473                         ipv4_l3fwd_route_array[i].ip,
1474                         ipv4_l3fwd_route_array[i].depth,
1475                         ipv4_l3fwd_route_array[i].if_out);
1476
1477                 if (ret < 0) {
1478                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
1479                                 "l3fwd LPM table on socket %d\n",
1480                                 i, socketid);
1481                 }
1482
1483                 printf("LPM: Adding route 0x%08x / %d (%d)\n",
1484                         (unsigned)ipv4_l3fwd_route_array[i].ip,
1485                         ipv4_l3fwd_route_array[i].depth,
1486                         ipv4_l3fwd_route_array[i].if_out);
1487         }
1488 }
1489 #endif
1490
1491 static int
1492 init_mem(unsigned nb_mbuf)
1493 {
1494         struct lcore_conf *qconf;
1495         int socketid;
1496         unsigned lcore_id;
1497         char s[64];
1498
1499         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1500                 if (rte_lcore_is_enabled(lcore_id) == 0)
1501                         continue;
1502
1503                 if (numa_on)
1504                         socketid = rte_lcore_to_socket_id(lcore_id);
1505                 else
1506                         socketid = 0;
1507
1508                 if (socketid >= NB_SOCKETS) {
1509                         rte_exit(EXIT_FAILURE, "Socket %d of lcore %u is "
1510                                         "out of range %d\n", socketid,
1511                                                 lcore_id, NB_SOCKETS);
1512                 }
1513                 if (pktmbuf_pool[socketid] == NULL) {
1514                         snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
1515                         pktmbuf_pool[socketid] =
1516                                 rte_pktmbuf_pool_create(s, nb_mbuf,
1517                                         MEMPOOL_CACHE_SIZE, 0,
1518                                         RTE_MBUF_DEFAULT_BUF_SIZE,
1519                                         socketid);
1520                         if (pktmbuf_pool[socketid] == NULL)
1521                                 rte_exit(EXIT_FAILURE,
1522                                         "Cannot init mbuf pool on socket %d\n",
1523                                                                 socketid);
1524                         else
1525                                 printf("Allocated mbuf pool on socket %d\n",
1526                                                                 socketid);
1527
1528 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1529                         setup_lpm(socketid);
1530 #else
1531                         setup_hash(socketid);
1532 #endif
1533                 }
1534                 qconf = &lcore_conf[lcore_id];
1535                 qconf->ipv4_lookup_struct = ipv4_l3fwd_lookup_struct[socketid];
1536 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1537                 qconf->ipv6_lookup_struct = ipv6_l3fwd_lookup_struct[socketid];
1538 #endif
1539         }
1540         return 0;
1541 }
1542
1543 /* Check the link status of all ports in up to 9s, and print them finally */
1544 static void
1545 check_all_ports_link_status(uint16_t port_num, uint32_t port_mask)
1546 {
1547 #define CHECK_INTERVAL 100 /* 100ms */
1548 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
1549         uint8_t count, all_ports_up, print_flag = 0;
1550         uint16_t portid;
1551         struct rte_eth_link link;
1552
1553         printf("\nChecking link status");
1554         fflush(stdout);
1555         for (count = 0; count <= MAX_CHECK_TIME; count++) {
1556                 all_ports_up = 1;
1557                 for (portid = 0; portid < port_num; portid++) {
1558                         if ((port_mask & (1 << portid)) == 0)
1559                                 continue;
1560                         memset(&link, 0, sizeof(link));
1561                         rte_eth_link_get_nowait(portid, &link);
1562                         /* print link status if flag set */
1563                         if (print_flag == 1) {
1564                                 if (link.link_status)
1565                                         printf("Port %d Link Up - speed %u "
1566                                                 "Mbps - %s\n", (uint8_t)portid,
1567                                                 (unsigned)link.link_speed,
1568                                 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
1569                                         ("full-duplex") : ("half-duplex\n"));
1570                                 else
1571                                         printf("Port %d Link Down\n",
1572                                                 (uint8_t)portid);
1573                                 continue;
1574                         }
1575                         /* clear all_ports_up flag if any link down */
1576                         if (link.link_status == ETH_LINK_DOWN) {
1577                                 all_ports_up = 0;
1578                                 break;
1579                         }
1580                 }
1581                 /* after finally printing all link status, get out */
1582                 if (print_flag == 1)
1583                         break;
1584
1585                 if (all_ports_up == 0) {
1586                         printf(".");
1587                         fflush(stdout);
1588                         rte_delay_ms(CHECK_INTERVAL);
1589                 }
1590
1591                 /* set the print_flag if all ports up or timeout */
1592                 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
1593                         print_flag = 1;
1594                         printf("done\n");
1595                 }
1596         }
1597 }
1598
1599 static int check_ptype(uint16_t portid)
1600 {
1601         int i, ret;
1602         int ptype_l3_ipv4 = 0;
1603 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1604         int ptype_l3_ipv6 = 0;
1605 #endif
1606         uint32_t ptype_mask = RTE_PTYPE_L3_MASK;
1607
1608         ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, NULL, 0);
1609         if (ret <= 0)
1610                 return 0;
1611
1612         uint32_t ptypes[ret];
1613
1614         ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, ptypes, ret);
1615         for (i = 0; i < ret; ++i) {
1616                 if (ptypes[i] & RTE_PTYPE_L3_IPV4)
1617                         ptype_l3_ipv4 = 1;
1618 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1619                 if (ptypes[i] & RTE_PTYPE_L3_IPV6)
1620                         ptype_l3_ipv6 = 1;
1621 #endif
1622         }
1623
1624         if (ptype_l3_ipv4 == 0)
1625                 printf("port %d cannot parse RTE_PTYPE_L3_IPV4\n", portid);
1626
1627 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1628         if (ptype_l3_ipv6 == 0)
1629                 printf("port %d cannot parse RTE_PTYPE_L3_IPV6\n", portid);
1630 #endif
1631
1632 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1633         if (ptype_l3_ipv4)
1634 #else /* APP_LOOKUP_EXACT_MATCH */
1635         if (ptype_l3_ipv4 && ptype_l3_ipv6)
1636 #endif
1637                 return 1;
1638
1639         return 0;
1640
1641 }
1642
1643 int
1644 main(int argc, char **argv)
1645 {
1646         struct lcore_conf *qconf;
1647         struct rte_eth_dev_info dev_info;
1648         struct rte_eth_txconf *txconf;
1649         int ret;
1650         uint16_t nb_ports;
1651         uint16_t queueid;
1652         unsigned lcore_id;
1653         uint64_t hz;
1654         uint32_t n_tx_queue, nb_lcores;
1655         uint32_t dev_rxq_num, dev_txq_num;
1656         uint8_t nb_rx_queue, queue, socketid;
1657         uint16_t portid;
1658         uint16_t org_rxq_intr = port_conf.intr_conf.rxq;
1659
1660         /* catch SIGINT and restore cpufreq governor to ondemand */
1661         signal(SIGINT, signal_exit_now);
1662
1663         /* init EAL */
1664         ret = rte_eal_init(argc, argv);
1665         if (ret < 0)
1666                 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
1667         argc -= ret;
1668         argv += ret;
1669
1670         /* init RTE timer library to be used late */
1671         rte_timer_subsystem_init();
1672
1673         /* parse application arguments (after the EAL ones) */
1674         ret = parse_args(argc, argv);
1675         if (ret < 0)
1676                 rte_exit(EXIT_FAILURE, "Invalid L3FWD parameters\n");
1677
1678         if (check_lcore_params() < 0)
1679                 rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
1680
1681         ret = init_lcore_rx_queues();
1682         if (ret < 0)
1683                 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
1684
1685         nb_ports = rte_eth_dev_count();
1686
1687         if (check_port_config(nb_ports) < 0)
1688                 rte_exit(EXIT_FAILURE, "check_port_config failed\n");
1689
1690         nb_lcores = rte_lcore_count();
1691
1692         /* initialize all ports */
1693         for (portid = 0; portid < nb_ports; portid++) {
1694                 /* skip ports that are not enabled */
1695                 if ((enabled_port_mask & (1 << portid)) == 0) {
1696                         printf("\nSkipping disabled port %d\n", portid);
1697                         continue;
1698                 }
1699
1700                 /* init port */
1701                 printf("Initializing port %d ... ", portid );
1702                 fflush(stdout);
1703
1704                 rte_eth_dev_info_get(portid, &dev_info);
1705                 dev_rxq_num = dev_info.max_rx_queues;
1706                 dev_txq_num = dev_info.max_tx_queues;
1707
1708                 nb_rx_queue = get_port_n_rx_queues(portid);
1709                 if (nb_rx_queue > dev_rxq_num)
1710                         rte_exit(EXIT_FAILURE,
1711                                 "Cannot configure not existed rxq: "
1712                                 "port=%d\n", portid);
1713
1714                 n_tx_queue = nb_lcores;
1715                 if (n_tx_queue > dev_txq_num)
1716                         n_tx_queue = dev_txq_num;
1717                 printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
1718                         nb_rx_queue, (unsigned)n_tx_queue );
1719                 /* If number of Rx queue is 0, no need to enable Rx interrupt */
1720                 if (nb_rx_queue == 0)
1721                         port_conf.intr_conf.rxq = 0;
1722                 ret = rte_eth_dev_configure(portid, nb_rx_queue,
1723                                         (uint16_t)n_tx_queue, &port_conf);
1724                 /* Revert to original value */
1725                 port_conf.intr_conf.rxq = org_rxq_intr;
1726                 if (ret < 0)
1727                         rte_exit(EXIT_FAILURE, "Cannot configure device: "
1728                                         "err=%d, port=%d\n", ret, portid);
1729
1730                 ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
1731                                                        &nb_txd);
1732                 if (ret < 0)
1733                         rte_exit(EXIT_FAILURE,
1734                                  "Cannot adjust number of descriptors: err=%d, port=%d\n",
1735                                  ret, portid);
1736
1737                 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
1738                 print_ethaddr(" Address:", &ports_eth_addr[portid]);
1739                 printf(", ");
1740
1741                 /* init memory */
1742                 ret = init_mem(NB_MBUF);
1743                 if (ret < 0)
1744                         rte_exit(EXIT_FAILURE, "init_mem failed\n");
1745
1746                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1747                         if (rte_lcore_is_enabled(lcore_id) == 0)
1748                                 continue;
1749
1750                         /* Initialize TX buffers */
1751                         qconf = &lcore_conf[lcore_id];
1752                         qconf->tx_buffer[portid] = rte_zmalloc_socket("tx_buffer",
1753                                 RTE_ETH_TX_BUFFER_SIZE(MAX_PKT_BURST), 0,
1754                                 rte_eth_dev_socket_id(portid));
1755                         if (qconf->tx_buffer[portid] == NULL)
1756                                 rte_exit(EXIT_FAILURE, "Can't allocate tx buffer for port %u\n",
1757                                                  portid);
1758
1759                         rte_eth_tx_buffer_init(qconf->tx_buffer[portid], MAX_PKT_BURST);
1760                 }
1761
1762                 /* init one TX queue per couple (lcore,port) */
1763                 queueid = 0;
1764                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1765                         if (rte_lcore_is_enabled(lcore_id) == 0)
1766                                 continue;
1767
1768                         if (queueid >= dev_txq_num)
1769                                 continue;
1770
1771                         if (numa_on)
1772                                 socketid = \
1773                                 (uint8_t)rte_lcore_to_socket_id(lcore_id);
1774                         else
1775                                 socketid = 0;
1776
1777                         printf("txq=%u,%d,%d ", lcore_id, queueid, socketid);
1778                         fflush(stdout);
1779
1780                         rte_eth_dev_info_get(portid, &dev_info);
1781                         txconf = &dev_info.default_txconf;
1782                         if (port_conf.rxmode.jumbo_frame)
1783                                 txconf->txq_flags = 0;
1784                         ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
1785                                                      socketid, txconf);
1786                         if (ret < 0)
1787                                 rte_exit(EXIT_FAILURE,
1788                                         "rte_eth_tx_queue_setup: err=%d, "
1789                                                 "port=%d\n", ret, portid);
1790
1791                         qconf = &lcore_conf[lcore_id];
1792                         qconf->tx_queue_id[portid] = queueid;
1793                         queueid++;
1794
1795                         qconf->tx_port_id[qconf->n_tx_port] = portid;
1796                         qconf->n_tx_port++;
1797                 }
1798                 printf("\n");
1799         }
1800
1801         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1802                 if (rte_lcore_is_enabled(lcore_id) == 0)
1803                         continue;
1804
1805                 /* init power management library */
1806                 ret = rte_power_init(lcore_id);
1807                 if (ret)
1808                         RTE_LOG(ERR, POWER,
1809                                 "Library initialization failed on core %u\n", lcore_id);
1810
1811                 /* init timer structures for each enabled lcore */
1812                 rte_timer_init(&power_timers[lcore_id]);
1813                 hz = rte_get_timer_hz();
1814                 rte_timer_reset(&power_timers[lcore_id],
1815                         hz/TIMER_NUMBER_PER_SECOND, SINGLE, lcore_id,
1816                                                 power_timer_cb, NULL);
1817
1818                 qconf = &lcore_conf[lcore_id];
1819                 printf("\nInitializing rx queues on lcore %u ... ", lcore_id );
1820                 fflush(stdout);
1821                 /* init RX queues */
1822                 for(queue = 0; queue < qconf->n_rx_queue; ++queue) {
1823                         portid = qconf->rx_queue_list[queue].port_id;
1824                         queueid = qconf->rx_queue_list[queue].queue_id;
1825
1826                         if (numa_on)
1827                                 socketid = \
1828                                 (uint8_t)rte_lcore_to_socket_id(lcore_id);
1829                         else
1830                                 socketid = 0;
1831
1832                         printf("rxq=%d,%d,%d ", portid, queueid, socketid);
1833                         fflush(stdout);
1834
1835                         ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
1836                                 socketid, NULL,
1837                                 pktmbuf_pool[socketid]);
1838                         if (ret < 0)
1839                                 rte_exit(EXIT_FAILURE,
1840                                         "rte_eth_rx_queue_setup: err=%d, "
1841                                                 "port=%d\n", ret, portid);
1842
1843                         if (parse_ptype) {
1844                                 if (add_cb_parse_ptype(portid, queueid) < 0)
1845                                         rte_exit(EXIT_FAILURE,
1846                                                  "Fail to add ptype cb\n");
1847                         } else if (!check_ptype(portid))
1848                                 rte_exit(EXIT_FAILURE,
1849                                          "PMD can not provide needed ptypes\n");
1850                 }
1851         }
1852
1853         printf("\n");
1854
1855         /* start ports */
1856         for (portid = 0; portid < nb_ports; portid++) {
1857                 if ((enabled_port_mask & (1 << portid)) == 0) {
1858                         continue;
1859                 }
1860                 /* Start device */
1861                 ret = rte_eth_dev_start(portid);
1862                 if (ret < 0)
1863                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, "
1864                                                 "port=%d\n", ret, portid);
1865                 /*
1866                  * If enabled, put device in promiscuous mode.
1867                  * This allows IO forwarding mode to forward packets
1868                  * to itself through 2 cross-connected  ports of the
1869                  * target machine.
1870                  */
1871                 if (promiscuous_on)
1872                         rte_eth_promiscuous_enable(portid);
1873                 /* initialize spinlock for each port */
1874                 rte_spinlock_init(&(locks[portid]));
1875         }
1876
1877         check_all_ports_link_status(nb_ports, enabled_port_mask);
1878
1879         /* launch per-lcore init on every lcore */
1880         rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
1881         RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1882                 if (rte_eal_wait_lcore(lcore_id) < 0)
1883                         return -1;
1884         }
1885
1886         return 0;
1887 }