examples/l3fwd-power/main.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2010-2016 Intel Corporation
   3  */
   4
   5 #include <stdio.h>
   6 #include <stdlib.h>
   7 #include <stdint.h>
   8 #include <inttypes.h>
   9 #include <sys/types.h>
  10 #include <string.h>
  11 #include <sys/queue.h>
  12 #include <stdarg.h>
  13 #include <errno.h>
  14 #include <getopt.h>
  15 #include <unistd.h>
  16 #include <signal.h>
  17
  18 #include <rte_common.h>
  19 #include <rte_byteorder.h>
  20 #include <rte_log.h>
  21 #include <rte_malloc.h>
  22 #include <rte_memory.h>
  23 #include <rte_memcpy.h>
  24 #include <rte_eal.h>
  25 #include <rte_launch.h>
  26 #include <rte_atomic.h>
  27 #include <rte_cycles.h>
  28 #include <rte_prefetch.h>
  29 #include <rte_lcore.h>
  30 #include <rte_per_lcore.h>
  31 #include <rte_branch_prediction.h>
  32 #include <rte_interrupts.h>
  33 #include <rte_random.h>
  34 #include <rte_debug.h>
  35 #include <rte_ether.h>
  36 #include <rte_ethdev.h>
  37 #include <rte_mempool.h>
  38 #include <rte_mbuf.h>
  39 #include <rte_ip.h>
  40 #include <rte_tcp.h>
  41 #include <rte_udp.h>
  42 #include <rte_string_fns.h>
  43 #include <rte_timer.h>
  44 #include <rte_power.h>
  45 #include <rte_spinlock.h>
  46
  47 #define RTE_LOGTYPE_L3FWD_POWER RTE_LOGTYPE_USER1
  48
  49 #define MAX_PKT_BURST 32
  50
  51 #define MIN_ZERO_POLL_COUNT 10
  52
  53 /* 100 ms interval */
  54 #define TIMER_NUMBER_PER_SECOND           10
  55 /* 100000 us */
  56 #define SCALING_PERIOD                    (1000000/TIMER_NUMBER_PER_SECOND)
  57 #define SCALING_DOWN_TIME_RATIO_THRESHOLD 0.25
  58
  59 #define APP_LOOKUP_EXACT_MATCH          0
  60 #define APP_LOOKUP_LPM                  1
  61 #define DO_RFC_1812_CHECKS
  62
  63 #ifndef APP_LOOKUP_METHOD
  64 #define APP_LOOKUP_METHOD             APP_LOOKUP_LPM
  65 #endif
  66
  67 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
  68 #include <rte_hash.h>
  69 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
  70 #include <rte_lpm.h>
  71 #else
  72 #error "APP_LOOKUP_METHOD set to incorrect value"
  73 #endif
  74
  75 #ifndef IPv6_BYTES
  76 #define IPv6_BYTES_FMT "%02x%02x:%02x%02x:%02x%02x:%02x%02x:"\
  77                        "%02x%02x:%02x%02x:%02x%02x:%02x%02x"
  78 #define IPv6_BYTES(addr) \
  79         addr[0],  addr[1], addr[2],  addr[3], \
  80         addr[4],  addr[5], addr[6],  addr[7], \
  81         addr[8],  addr[9], addr[10], addr[11],\
  82         addr[12], addr[13],addr[14], addr[15]
  83 #endif
  84
  85 #define MAX_JUMBO_PKT_LEN  9600
  86
  87 #define IPV6_ADDR_LEN 16
  88
  89 #define MEMPOOL_CACHE_SIZE 256
  90
  91 /*
  92  * This expression is used to calculate the number of mbufs needed depending on
  93  * user input, taking into account memory for rx and tx hardware rings, cache
  94  * per lcore and mtable per port per lcore. RTE_MAX is used to ensure that
  95  * NB_MBUF never goes below a minimum value of 8192.
  96  */
  97
  98 #define NB_MBUF RTE_MAX ( \
  99         (nb_ports*nb_rx_queue*nb_rxd + \
 100         nb_ports*nb_lcores*MAX_PKT_BURST + \
 101         nb_ports*n_tx_queue*nb_txd + \
 102         nb_lcores*MEMPOOL_CACHE_SIZE), \
 103         (unsigned)8192)
 104
 105 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
 106
 107 #define NB_SOCKETS 8
 108
 109 /* Configure how many packets ahead to prefetch, when reading packets */
 110 #define PREFETCH_OFFSET 3
 111
 112 /*
 113  * Configurable number of RX/TX ring descriptors
 114  */
 115 #define RTE_TEST_RX_DESC_DEFAULT 512
 116 #define RTE_TEST_TX_DESC_DEFAULT 512
 117 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
 118 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
 119
 120 /* ethernet addresses of ports */
 121 static struct ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
 122
 123 /* ethernet addresses of ports */
 124 static rte_spinlock_t locks[RTE_MAX_ETHPORTS];
 125
 126 /* mask of enabled ports */
 127 static uint32_t enabled_port_mask = 0;
 128 /* Ports set in promiscuous mode off by default. */
 129 static int promiscuous_on = 0;
 130 /* NUMA is enabled by default. */
 131 static int numa_on = 1;
 132 static int parse_ptype; /**< Parse packet type using rx callback, and */
 133                         /**< disabled by default */
 134
 135 enum freq_scale_hint_t
 136 {
 137         FREQ_LOWER    =      -1,
 138         FREQ_CURRENT  =       0,
 139         FREQ_HIGHER   =       1,
 140         FREQ_HIGHEST  =       2
 141 };
 142
 143 struct lcore_rx_queue {
 144         uint16_t port_id;
 145         uint8_t queue_id;
 146         enum freq_scale_hint_t freq_up_hint;
 147         uint32_t zero_rx_packet_count;
 148         uint32_t idle_hint;
 149 } __rte_cache_aligned;
 150
 151 #define MAX_RX_QUEUE_PER_LCORE 16
 152 #define MAX_TX_QUEUE_PER_PORT RTE_MAX_ETHPORTS
 153 #define MAX_RX_QUEUE_PER_PORT 128
 154
 155 #define MAX_RX_QUEUE_INTERRUPT_PER_PORT 16
 156
 157
 158 #define MAX_LCORE_PARAMS 1024
 159 struct lcore_params {
 160         uint16_t port_id;
 161         uint8_t queue_id;
 162         uint8_t lcore_id;
 163 } __rte_cache_aligned;
 164
 165 static struct lcore_params lcore_params_array[MAX_LCORE_PARAMS];
 166 static struct lcore_params lcore_params_array_default[] = {
 167         {0, 0, 2},
 168         {0, 1, 2},
 169         {0, 2, 2},
 170         {1, 0, 2},
 171         {1, 1, 2},
 172         {1, 2, 2},
 173         {2, 0, 2},
 174         {3, 0, 3},
 175         {3, 1, 3},
 176 };
 177
 178 static struct lcore_params * lcore_params = lcore_params_array_default;
 179 static uint16_t nb_lcore_params = sizeof(lcore_params_array_default) /
 180                                 sizeof(lcore_params_array_default[0]);
 181
 182 static struct rte_eth_conf port_conf = {
 183         .rxmode = {
 184                 .mq_mode        = ETH_MQ_RX_RSS,
 185                 .max_rx_pkt_len = ETHER_MAX_LEN,
 186                 .split_hdr_size = 0,
 187                 .header_split   = 0, /**< Header Split disabled */
 188                 .hw_ip_checksum = 1, /**< IP checksum offload enabled */
 189                 .hw_vlan_filter = 0, /**< VLAN filtering disabled */
 190                 .jumbo_frame    = 0, /**< Jumbo Frame Support disabled */
 191                 .hw_strip_crc   = 1, /**< CRC stripped by hardware */
 192         },
 193         .rx_adv_conf = {
 194                 .rss_conf = {
 195                         .rss_key = NULL,
 196                         .rss_hf = ETH_RSS_UDP,
 197                 },
 198         },
 199         .txmode = {
 200                 .mq_mode = ETH_MQ_TX_NONE,
 201         },
 202         .intr_conf = {
 203                 .lsc = 1,
 204                 .rxq = 1,
 205         },
 206 };
 207
 208 static struct rte_mempool * pktmbuf_pool[NB_SOCKETS];
 209
 210
 211 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 212
 213 #ifdef RTE_ARCH_X86
 214 #include <rte_hash_crc.h>
 215 #define DEFAULT_HASH_FUNC       rte_hash_crc
 216 #else
 217 #include <rte_jhash.h>
 218 #define DEFAULT_HASH_FUNC       rte_jhash
 219 #endif
 220
 221 struct ipv4_5tuple {
 222         uint32_t ip_dst;
 223         uint32_t ip_src;
 224         uint16_t port_dst;
 225         uint16_t port_src;
 226         uint8_t  proto;
 227 } __attribute__((__packed__));
 228
 229 struct ipv6_5tuple {
 230         uint8_t  ip_dst[IPV6_ADDR_LEN];
 231         uint8_t  ip_src[IPV6_ADDR_LEN];
 232         uint16_t port_dst;
 233         uint16_t port_src;
 234         uint8_t  proto;
 235 } __attribute__((__packed__));
 236
 237 struct ipv4_l3fwd_route {
 238         struct ipv4_5tuple key;
 239         uint8_t if_out;
 240 };
 241
 242 struct ipv6_l3fwd_route {
 243         struct ipv6_5tuple key;
 244         uint8_t if_out;
 245 };
 246
 247 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
 248         {{IPv4(100,10,0,1), IPv4(200,10,0,1), 101, 11, IPPROTO_TCP}, 0},
 249         {{IPv4(100,20,0,2), IPv4(200,20,0,2), 102, 12, IPPROTO_TCP}, 1},
 250         {{IPv4(100,30,0,3), IPv4(200,30,0,3), 103, 13, IPPROTO_TCP}, 2},
 251         {{IPv4(100,40,0,4), IPv4(200,40,0,4), 104, 14, IPPROTO_TCP}, 3},
 252 };
 253
 254 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
 255         {
 256                 {
 257                         {0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
 258                          0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
 259                         {0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
 260                          0x02, 0x1e, 0x67, 0xff, 0xfe, 0x0d, 0xb6, 0x0a},
 261                          1, 10, IPPROTO_UDP
 262                 }, 4
 263         },
 264 };
 265
 266 typedef struct rte_hash lookup_struct_t;
 267 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
 268 static lookup_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
 269
 270 #define L3FWD_HASH_ENTRIES      1024
 271
 272 #define IPV4_L3FWD_NUM_ROUTES \
 273         (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
 274
 275 #define IPV6_L3FWD_NUM_ROUTES \
 276         (sizeof(ipv6_l3fwd_route_array) / sizeof(ipv6_l3fwd_route_array[0]))
 277
 278 static uint16_t ipv4_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 279 static uint16_t ipv6_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 280 #endif
 281
 282 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 283 struct ipv4_l3fwd_route {
 284         uint32_t ip;
 285         uint8_t  depth;
 286         uint8_t  if_out;
 287 };
 288
 289 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
 290         {IPv4(1,1,1,0), 24, 0},
 291         {IPv4(2,1,1,0), 24, 1},
 292         {IPv4(3,1,1,0), 24, 2},
 293         {IPv4(4,1,1,0), 24, 3},
 294         {IPv4(5,1,1,0), 24, 4},
 295         {IPv4(6,1,1,0), 24, 5},
 296         {IPv4(7,1,1,0), 24, 6},
 297         {IPv4(8,1,1,0), 24, 7},
 298 };
 299
 300 #define IPV4_L3FWD_NUM_ROUTES \
 301         (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
 302
 303 #define IPV4_L3FWD_LPM_MAX_RULES     1024
 304
 305 typedef struct rte_lpm lookup_struct_t;
 306 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
 307 #endif
 308
 309 struct lcore_conf {
 310         uint16_t n_rx_queue;
 311         struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
 312         uint16_t n_tx_port;
 313         uint16_t tx_port_id[RTE_MAX_ETHPORTS];
 314         uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
 315         struct rte_eth_dev_tx_buffer *tx_buffer[RTE_MAX_ETHPORTS];
 316         lookup_struct_t * ipv4_lookup_struct;
 317         lookup_struct_t * ipv6_lookup_struct;
 318 } __rte_cache_aligned;
 319
 320 struct lcore_stats {
 321         /* total sleep time in ms since last frequency scaling down */
 322         uint32_t sleep_time;
 323         /* number of long sleep recently */
 324         uint32_t nb_long_sleep;
 325         /* freq. scaling up trend */
 326         uint32_t trend;
 327         /* total packet processed recently */
 328         uint64_t nb_rx_processed;
 329         /* total iterations looped recently */
 330         uint64_t nb_iteration_looped;
 331         uint32_t padding[9];
 332 } __rte_cache_aligned;
 333
 334 static struct lcore_conf lcore_conf[RTE_MAX_LCORE] __rte_cache_aligned;
 335 static struct lcore_stats stats[RTE_MAX_LCORE] __rte_cache_aligned;
 336 static struct rte_timer power_timers[RTE_MAX_LCORE];
 337
 338 static inline uint32_t power_idle_heuristic(uint32_t zero_rx_packet_count);
 339 static inline enum freq_scale_hint_t power_freq_scaleup_heuristic( \
 340                 unsigned int lcore_id, uint16_t port_id, uint16_t queue_id);
 341
 342 /* exit signal handler */
 343 static void
 344 signal_exit_now(int sigtype)
 345 {
 346         unsigned lcore_id;
 347         unsigned int portid, nb_ports;
 348         int ret;
 349
 350         if (sigtype == SIGINT) {
 351                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
 352                         if (rte_lcore_is_enabled(lcore_id) == 0)
 353                                 continue;
 354
 355                         /* init power management library */
 356                         ret = rte_power_exit(lcore_id);
 357                         if (ret)
 358                                 rte_exit(EXIT_FAILURE, "Power management "
 359                                         "library de-initialization failed on "
 360                                                         "core%u\n", lcore_id);
 361                 }
 362
 363                 nb_ports = rte_eth_dev_count();
 364                 for (portid = 0; portid < nb_ports; portid++) {
 365                         if ((enabled_port_mask & (1 << portid)) == 0)
 366                                 continue;
 367
 368                         rte_eth_dev_stop(portid);
 369                         rte_eth_dev_close(portid);
 370                 }
 371         }
 372
 373         rte_exit(EXIT_SUCCESS, "User forced exit\n");
 374 }
 375
 376 /*  Freqency scale down timer callback */
 377 static void
 378 power_timer_cb(__attribute__((unused)) struct rte_timer *tim,
 379                           __attribute__((unused)) void *arg)
 380 {
 381         uint64_t hz;
 382         float sleep_time_ratio;
 383         unsigned lcore_id = rte_lcore_id();
 384
 385         /* accumulate total execution time in us when callback is invoked */
 386         sleep_time_ratio = (float)(stats[lcore_id].sleep_time) /
 387                                         (float)SCALING_PERIOD;
 388         /**
 389          * check whether need to scale down frequency a step if it sleep a lot.
 390          */
 391         if (sleep_time_ratio >= SCALING_DOWN_TIME_RATIO_THRESHOLD) {
 392                 if (rte_power_freq_down)
 393                         rte_power_freq_down(lcore_id);
 394         }
 395         else if ( (unsigned)(stats[lcore_id].nb_rx_processed /
 396                 stats[lcore_id].nb_iteration_looped) < MAX_PKT_BURST) {
 397                 /**
 398                  * scale down a step if average packet per iteration less
 399                  * than expectation.
 400                  */
 401                 if (rte_power_freq_down)
 402                         rte_power_freq_down(lcore_id);
 403         }
 404
 405         /**
 406          * initialize another timer according to current frequency to ensure
 407          * timer interval is relatively fixed.
 408          */
 409         hz = rte_get_timer_hz();
 410         rte_timer_reset(&power_timers[lcore_id], hz/TIMER_NUMBER_PER_SECOND,
 411                                 SINGLE, lcore_id, power_timer_cb, NULL);
 412
 413         stats[lcore_id].nb_rx_processed = 0;
 414         stats[lcore_id].nb_iteration_looped = 0;
 415
 416         stats[lcore_id].sleep_time = 0;
 417 }
 418
 419 /* Enqueue a single packet, and send burst if queue is filled */
 420 static inline int
 421 send_single_packet(struct rte_mbuf *m, uint16_t port)
 422 {
 423         uint32_t lcore_id;
 424         struct lcore_conf *qconf;
 425
 426         lcore_id = rte_lcore_id();
 427         qconf = &lcore_conf[lcore_id];
 428
 429         rte_eth_tx_buffer(port, qconf->tx_queue_id[port],
 430                         qconf->tx_buffer[port], m);
 431
 432         return 0;
 433 }
 434
 435 #ifdef DO_RFC_1812_CHECKS
 436 static inline int
 437 is_valid_ipv4_pkt(struct ipv4_hdr *pkt, uint32_t link_len)
 438 {
 439         /* From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2 */
 440         /*
 441          * 1. The packet length reported by the Link Layer must be large
 442          * enough to hold the minimum length legal IP datagram (20 bytes).
 443          */
 444         if (link_len < sizeof(struct ipv4_hdr))
 445                 return -1;
 446
 447         /* 2. The IP checksum must be correct. */
 448         /* this is checked in H/W */
 449
 450         /*
 451          * 3. The IP version number must be 4. If the version number is not 4
 452          * then the packet may be another version of IP, such as IPng or
 453          * ST-II.
 454          */
 455         if (((pkt->version_ihl) >> 4) != 4)
 456                 return -3;
 457         /*
 458          * 4. The IP header length field must be large enough to hold the
 459          * minimum length legal IP datagram (20 bytes = 5 words).
 460          */
 461         if ((pkt->version_ihl & 0xf) < 5)
 462                 return -4;
 463
 464         /*
 465          * 5. The IP total length field must be large enough to hold the IP
 466          * datagram header, whose length is specified in the IP header length
 467          * field.
 468          */
 469         if (rte_cpu_to_be_16(pkt->total_length) < sizeof(struct ipv4_hdr))
 470                 return -5;
 471
 472         return 0;
 473 }
 474 #endif
 475
 476 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 477 static void
 478 print_ipv4_key(struct ipv4_5tuple key)
 479 {
 480         printf("IP dst = %08x, IP src = %08x, port dst = %d, port src = %d, "
 481                 "proto = %d\n", (unsigned)key.ip_dst, (unsigned)key.ip_src,
 482                                 key.port_dst, key.port_src, key.proto);
 483 }
 484 static void
 485 print_ipv6_key(struct ipv6_5tuple key)
 486 {
 487         printf( "IP dst = " IPv6_BYTES_FMT ", IP src = " IPv6_BYTES_FMT ", "
 488                 "port dst = %d, port src = %d, proto = %d\n",
 489                 IPv6_BYTES(key.ip_dst), IPv6_BYTES(key.ip_src),
 490                 key.port_dst, key.port_src, key.proto);
 491 }
 492
 493 static inline uint16_t
 494 get_ipv4_dst_port(struct ipv4_hdr *ipv4_hdr, uint16_t portid,
 495                 lookup_struct_t * ipv4_l3fwd_lookup_struct)
 496 {
 497         struct ipv4_5tuple key;
 498         struct tcp_hdr *tcp;
 499         struct udp_hdr *udp;
 500         int ret = 0;
 501
 502         key.ip_dst = rte_be_to_cpu_32(ipv4_hdr->dst_addr);
 503         key.ip_src = rte_be_to_cpu_32(ipv4_hdr->src_addr);
 504         key.proto = ipv4_hdr->next_proto_id;
 505
 506         switch (ipv4_hdr->next_proto_id) {
 507         case IPPROTO_TCP:
 508                 tcp = (struct tcp_hdr *)((unsigned char *)ipv4_hdr +
 509                                         sizeof(struct ipv4_hdr));
 510                 key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
 511                 key.port_src = rte_be_to_cpu_16(tcp->src_port);
 512                 break;
 513
 514         case IPPROTO_UDP:
 515                 udp = (struct udp_hdr *)((unsigned char *)ipv4_hdr +
 516                                         sizeof(struct ipv4_hdr));
 517                 key.port_dst = rte_be_to_cpu_16(udp->dst_port);
 518                 key.port_src = rte_be_to_cpu_16(udp->src_port);
 519                 break;
 520
 521         default:
 522                 key.port_dst = 0;
 523                 key.port_src = 0;
 524                 break;
 525         }
 526
 527         /* Find destination port */
 528         ret = rte_hash_lookup(ipv4_l3fwd_lookup_struct, (const void *)&key);
 529         return ((ret < 0) ? portid : ipv4_l3fwd_out_if[ret]);
 530 }
 531
 532 static inline uint16_t
 533 get_ipv6_dst_port(struct ipv6_hdr *ipv6_hdr, uint16_t portid,
 534                         lookup_struct_t *ipv6_l3fwd_lookup_struct)
 535 {
 536         struct ipv6_5tuple key;
 537         struct tcp_hdr *tcp;
 538         struct udp_hdr *udp;
 539         int ret = 0;
 540
 541         memcpy(key.ip_dst, ipv6_hdr->dst_addr, IPV6_ADDR_LEN);
 542         memcpy(key.ip_src, ipv6_hdr->src_addr, IPV6_ADDR_LEN);
 543
 544         key.proto = ipv6_hdr->proto;
 545
 546         switch (ipv6_hdr->proto) {
 547         case IPPROTO_TCP:
 548                 tcp = (struct tcp_hdr *)((unsigned char *) ipv6_hdr +
 549                                         sizeof(struct ipv6_hdr));
 550                 key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
 551                 key.port_src = rte_be_to_cpu_16(tcp->src_port);
 552                 break;
 553
 554         case IPPROTO_UDP:
 555                 udp = (struct udp_hdr *)((unsigned char *) ipv6_hdr +
 556                                         sizeof(struct ipv6_hdr));
 557                 key.port_dst = rte_be_to_cpu_16(udp->dst_port);
 558                 key.port_src = rte_be_to_cpu_16(udp->src_port);
 559                 break;
 560
 561         default:
 562                 key.port_dst = 0;
 563                 key.port_src = 0;
 564                 break;
 565         }
 566
 567         /* Find destination port */
 568         ret = rte_hash_lookup(ipv6_l3fwd_lookup_struct, (const void *)&key);
 569         return ((ret < 0) ? portid : ipv6_l3fwd_out_if[ret]);
 570 }
 571 #endif
 572
 573 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 574 static inline uint16_t
 575 get_ipv4_dst_port(struct ipv4_hdr *ipv4_hdr, uint16_t portid,
 576                 lookup_struct_t *ipv4_l3fwd_lookup_struct)
 577 {
 578         uint32_t next_hop;
 579
 580         return ((rte_lpm_lookup(ipv4_l3fwd_lookup_struct,
 581                         rte_be_to_cpu_32(ipv4_hdr->dst_addr), &next_hop) == 0)?
 582                         next_hop : portid);
 583 }
 584 #endif
 585
 586 static inline void
 587 parse_ptype_one(struct rte_mbuf *m)
 588 {
 589         struct ether_hdr *eth_hdr;
 590         uint32_t packet_type = RTE_PTYPE_UNKNOWN;
 591         uint16_t ether_type;
 592
 593         eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
 594         ether_type = eth_hdr->ether_type;
 595         if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv4))
 596                 packet_type |= RTE_PTYPE_L3_IPV4_EXT_UNKNOWN;
 597         else if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv6))
 598                 packet_type |= RTE_PTYPE_L3_IPV6_EXT_UNKNOWN;
 599
 600         m->packet_type = packet_type;
 601 }
 602
 603 static uint16_t
 604 cb_parse_ptype(uint16_t port __rte_unused, uint16_t queue __rte_unused,
 605                struct rte_mbuf *pkts[], uint16_t nb_pkts,
 606                uint16_t max_pkts __rte_unused,
 607                void *user_param __rte_unused)
 608 {
 609         unsigned int i;
 610
 611         for (i = 0; i < nb_pkts; ++i)
 612                 parse_ptype_one(pkts[i]);
 613
 614         return nb_pkts;
 615 }
 616
 617 static int
 618 add_cb_parse_ptype(uint16_t portid, uint16_t queueid)
 619 {
 620         printf("Port %d: softly parse packet type info\n", portid);
 621         if (rte_eth_add_rx_callback(portid, queueid, cb_parse_ptype, NULL))
 622                 return 0;
 623
 624         printf("Failed to add rx callback: port=%d\n", portid);
 625         return -1;
 626 }
 627
 628 static inline void
 629 l3fwd_simple_forward(struct rte_mbuf *m, uint16_t portid,
 630                                 struct lcore_conf *qconf)
 631 {
 632         struct ether_hdr *eth_hdr;
 633         struct ipv4_hdr *ipv4_hdr;
 634         void *d_addr_bytes;
 635         uint16_t dst_port;
 636
 637         eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
 638
 639         if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) {
 640                 /* Handle IPv4 headers.*/
 641                 ipv4_hdr =
 642                         rte_pktmbuf_mtod_offset(m, struct ipv4_hdr *,
 643                                                 sizeof(struct ether_hdr));
 644
 645 #ifdef DO_RFC_1812_CHECKS
 646                 /* Check to make sure the packet is valid (RFC1812) */
 647                 if (is_valid_ipv4_pkt(ipv4_hdr, m->pkt_len) < 0) {
 648                         rte_pktmbuf_free(m);
 649                         return;
 650                 }
 651 #endif
 652
 653                 dst_port = get_ipv4_dst_port(ipv4_hdr, portid,
 654                                         qconf->ipv4_lookup_struct);
 655                 if (dst_port >= RTE_MAX_ETHPORTS ||
 656                                 (enabled_port_mask & 1 << dst_port) == 0)
 657                         dst_port = portid;
 658
 659                 /* 02:00:00:00:00:xx */
 660                 d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
 661                 *((uint64_t *)d_addr_bytes) =
 662                         0x000000000002 + ((uint64_t)dst_port << 40);
 663
 664 #ifdef DO_RFC_1812_CHECKS
 665                 /* Update time to live and header checksum */
 666                 --(ipv4_hdr->time_to_live);
 667                 ++(ipv4_hdr->hdr_checksum);
 668 #endif
 669
 670                 /* src addr */
 671                 ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
 672
 673                 send_single_packet(m, dst_port);
 674         } else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) {
 675                 /* Handle IPv6 headers.*/
 676 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 677                 struct ipv6_hdr *ipv6_hdr;
 678
 679                 ipv6_hdr =
 680                         rte_pktmbuf_mtod_offset(m, struct ipv6_hdr *,
 681                                                 sizeof(struct ether_hdr));
 682
 683                 dst_port = get_ipv6_dst_port(ipv6_hdr, portid,
 684                                         qconf->ipv6_lookup_struct);
 685
 686                 if (dst_port >= RTE_MAX_ETHPORTS ||
 687                                 (enabled_port_mask & 1 << dst_port) == 0)
 688                         dst_port = portid;
 689
 690                 /* 02:00:00:00:00:xx */
 691                 d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
 692                 *((uint64_t *)d_addr_bytes) =
 693                         0x000000000002 + ((uint64_t)dst_port << 40);
 694
 695                 /* src addr */
 696                 ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
 697
 698                 send_single_packet(m, dst_port);
 699 #else
 700                 /* We don't currently handle IPv6 packets in LPM mode. */
 701                 rte_pktmbuf_free(m);
 702 #endif
 703         } else
 704                 rte_pktmbuf_free(m);
 705
 706 }
 707
 708 #define MINIMUM_SLEEP_TIME         1
 709 #define SUSPEND_THRESHOLD          300
 710
 711 static inline uint32_t
 712 power_idle_heuristic(uint32_t zero_rx_packet_count)
 713 {
 714         /* If zero count is less than 100,  sleep 1us */
 715         if (zero_rx_packet_count < SUSPEND_THRESHOLD)
 716                 return MINIMUM_SLEEP_TIME;
 717         /* If zero count is less than 1000, sleep 100 us which is the
 718                 minimum latency switching from C3/C6 to C0
 719         */
 720         else
 721                 return SUSPEND_THRESHOLD;
 722 }
 723
 724 static inline enum freq_scale_hint_t
 725 power_freq_scaleup_heuristic(unsigned lcore_id,
 726                              uint16_t port_id,
 727                              uint16_t queue_id)
 728 {
 729         uint32_t rxq_count = rte_eth_rx_queue_count(port_id, queue_id);
 730 /**
 731  * HW Rx queue size is 128 by default, Rx burst read at maximum 32 entries
 732  * per iteration
 733  */
 734 #define FREQ_GEAR1_RX_PACKET_THRESHOLD             MAX_PKT_BURST
 735 #define FREQ_GEAR2_RX_PACKET_THRESHOLD             (MAX_PKT_BURST*2)
 736 #define FREQ_GEAR3_RX_PACKET_THRESHOLD             (MAX_PKT_BURST*3)
 737 #define FREQ_UP_TREND1_ACC   1
 738 #define FREQ_UP_TREND2_ACC   100
 739 #define FREQ_UP_THRESHOLD    10000
 740
 741         if (likely(rxq_count > FREQ_GEAR3_RX_PACKET_THRESHOLD)) {
 742                 stats[lcore_id].trend = 0;
 743                 return FREQ_HIGHEST;
 744         } else if (likely(rxq_count > FREQ_GEAR2_RX_PACKET_THRESHOLD))
 745                 stats[lcore_id].trend += FREQ_UP_TREND2_ACC;
 746         else if (likely(rxq_count > FREQ_GEAR1_RX_PACKET_THRESHOLD))
 747                 stats[lcore_id].trend += FREQ_UP_TREND1_ACC;
 748
 749         if (likely(stats[lcore_id].trend > FREQ_UP_THRESHOLD)) {
 750                 stats[lcore_id].trend = 0;
 751                 return FREQ_HIGHER;
 752         }
 753
 754         return FREQ_CURRENT;
 755 }
 756
 757 /**
 758  * force polling thread sleep until one-shot rx interrupt triggers
 759  * @param port_id
 760  *  Port id.
 761  * @param queue_id
 762  *  Rx queue id.
 763  * @return
 764  *  0 on success
 765  */
 766 static int
 767 sleep_until_rx_interrupt(int num)
 768 {
 769         struct rte_epoll_event event[num];
 770         int n, i;
 771         uint16_t port_id;
 772         uint8_t queue_id;
 773         void *data;
 774
 775         RTE_LOG(INFO, L3FWD_POWER,
 776                 "lcore %u sleeps until interrupt triggers\n",
 777                 rte_lcore_id());
 778
 779         n = rte_epoll_wait(RTE_EPOLL_PER_THREAD, event, num, -1);
 780         for (i = 0; i < n; i++) {
 781                 data = event[i].epdata.data;
 782                 port_id = ((uintptr_t)data) >> CHAR_BIT;
 783                 queue_id = ((uintptr_t)data) &
 784                         RTE_LEN2MASK(CHAR_BIT, uint8_t);
 785                 rte_eth_dev_rx_intr_disable(port_id, queue_id);
 786                 RTE_LOG(INFO, L3FWD_POWER,
 787                         "lcore %u is waked up from rx interrupt on"
 788                         " port %d queue %d\n",
 789                         rte_lcore_id(), port_id, queue_id);
 790         }
 791
 792         return 0;
 793 }
 794
 795 static void turn_on_intr(struct lcore_conf *qconf)
 796 {
 797         int i;
 798         struct lcore_rx_queue *rx_queue;
 799         uint8_t queue_id;
 800         uint16_t port_id;
 801
 802         for (i = 0; i < qconf->n_rx_queue; ++i) {
 803                 rx_queue = &(qconf->rx_queue_list[i]);
 804                 port_id = rx_queue->port_id;
 805                 queue_id = rx_queue->queue_id;
 806
 807                 rte_spinlock_lock(&(locks[port_id]));
 808                 rte_eth_dev_rx_intr_enable(port_id, queue_id);
 809                 rte_spinlock_unlock(&(locks[port_id]));
 810         }
 811 }
 812
 813 static int event_register(struct lcore_conf *qconf)
 814 {
 815         struct lcore_rx_queue *rx_queue;
 816         uint8_t queueid;
 817         uint16_t portid;
 818         uint32_t data;
 819         int ret;
 820         int i;
 821
 822         for (i = 0; i < qconf->n_rx_queue; ++i) {
 823                 rx_queue = &(qconf->rx_queue_list[i]);
 824                 portid = rx_queue->port_id;
 825                 queueid = rx_queue->queue_id;
 826                 data = portid << CHAR_BIT | queueid;
 827
 828                 ret = rte_eth_dev_rx_intr_ctl_q(portid, queueid,
 829                                                 RTE_EPOLL_PER_THREAD,
 830                                                 RTE_INTR_EVENT_ADD,
 831                                                 (void *)((uintptr_t)data));
 832                 if (ret)
 833                         return ret;
 834         }
 835
 836         return 0;
 837 }
 838
 839 /* main processing loop */
 840 static int
 841 main_loop(__attribute__((unused)) void *dummy)
 842 {
 843         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
 844         unsigned lcore_id;
 845         uint64_t prev_tsc, diff_tsc, cur_tsc, tim_res_tsc, hz;
 846         uint64_t prev_tsc_power = 0, cur_tsc_power, diff_tsc_power;
 847         int i, j, nb_rx;
 848         uint8_t queueid;
 849         uint16_t portid;
 850         struct lcore_conf *qconf;
 851         struct lcore_rx_queue *rx_queue;
 852         enum freq_scale_hint_t lcore_scaleup_hint;
 853         uint32_t lcore_rx_idle_count = 0;
 854         uint32_t lcore_idle_hint = 0;
 855         int intr_en = 0;
 856
 857         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
 858
 859         prev_tsc = 0;
 860         hz = rte_get_timer_hz();
 861         tim_res_tsc = hz/TIMER_NUMBER_PER_SECOND;
 862
 863         lcore_id = rte_lcore_id();
 864         qconf = &lcore_conf[lcore_id];
 865
 866         if (qconf->n_rx_queue == 0) {
 867                 RTE_LOG(INFO, L3FWD_POWER, "lcore %u has nothing to do\n", lcore_id);
 868                 return 0;
 869         }
 870
 871         RTE_LOG(INFO, L3FWD_POWER, "entering main loop on lcore %u\n", lcore_id);
 872
 873         for (i = 0; i < qconf->n_rx_queue; i++) {
 874                 portid = qconf->rx_queue_list[i].port_id;
 875                 queueid = qconf->rx_queue_list[i].queue_id;
 876                 RTE_LOG(INFO, L3FWD_POWER, " -- lcoreid=%u portid=%u "
 877                         "rxqueueid=%hhu\n", lcore_id, portid, queueid);
 878         }
 879
 880         /* add into event wait list */
 881         if (event_register(qconf) == 0)
 882                 intr_en = 1;
 883         else
 884                 RTE_LOG(INFO, L3FWD_POWER, "RX interrupt won't enable.\n");
 885
 886         while (1) {
 887                 stats[lcore_id].nb_iteration_looped++;
 888
 889                 cur_tsc = rte_rdtsc();
 890                 cur_tsc_power = cur_tsc;
 891
 892                 /*
 893                  * TX burst queue drain
 894                  */
 895                 diff_tsc = cur_tsc - prev_tsc;
 896                 if (unlikely(diff_tsc > drain_tsc)) {
 897                         for (i = 0; i < qconf->n_tx_port; ++i) {
 898                                 portid = qconf->tx_port_id[i];
 899                                 rte_eth_tx_buffer_flush(portid,
 900                                                 qconf->tx_queue_id[portid],
 901                                                 qconf->tx_buffer[portid]);
 902                         }
 903                         prev_tsc = cur_tsc;
 904                 }
 905
 906                 diff_tsc_power = cur_tsc_power - prev_tsc_power;
 907                 if (diff_tsc_power > tim_res_tsc) {
 908                         rte_timer_manage();
 909                         prev_tsc_power = cur_tsc_power;
 910                 }
 911
 912 start_rx:
 913                 /*
 914                  * Read packet from RX queues
 915                  */
 916                 lcore_scaleup_hint = FREQ_CURRENT;
 917                 lcore_rx_idle_count = 0;
 918                 for (i = 0; i < qconf->n_rx_queue; ++i) {
 919                         rx_queue = &(qconf->rx_queue_list[i]);
 920                         rx_queue->idle_hint = 0;
 921                         portid = rx_queue->port_id;
 922                         queueid = rx_queue->queue_id;
 923
 924                         nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
 925                                                                 MAX_PKT_BURST);
 926
 927                         stats[lcore_id].nb_rx_processed += nb_rx;
 928                         if (unlikely(nb_rx == 0)) {
 929                                 /**
 930                                  * no packet received from rx queue, try to
 931                                  * sleep for a while forcing CPU enter deeper
 932                                  * C states.
 933                                  */
 934                                 rx_queue->zero_rx_packet_count++;
 935
 936                                 if (rx_queue->zero_rx_packet_count <=
 937                                                         MIN_ZERO_POLL_COUNT)
 938                                         continue;
 939
 940                                 rx_queue->idle_hint = power_idle_heuristic(\
 941                                         rx_queue->zero_rx_packet_count);
 942                                 lcore_rx_idle_count++;
 943                         } else {
 944                                 rx_queue->zero_rx_packet_count = 0;
 945
 946                                 /**
 947                                  * do not scale up frequency immediately as
 948                                  * user to kernel space communication is costly
 949                                  * which might impact packet I/O for received
 950                                  * packets.
 951                                  */
 952                                 rx_queue->freq_up_hint =
 953                                         power_freq_scaleup_heuristic(lcore_id,
 954                                                         portid, queueid);
 955                         }
 956
 957                         /* Prefetch first packets */
 958                         for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
 959                                 rte_prefetch0(rte_pktmbuf_mtod(
 960                                                 pkts_burst[j], void *));
 961                         }
 962
 963                         /* Prefetch and forward already prefetched packets */
 964                         for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
 965                                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
 966                                                 j + PREFETCH_OFFSET], void *));
 967                                 l3fwd_simple_forward(pkts_burst[j], portid,
 968                                                                 qconf);
 969                         }
 970
 971                         /* Forward remaining prefetched packets */
 972                         for (; j < nb_rx; j++) {
 973                                 l3fwd_simple_forward(pkts_burst[j], portid,
 974                                                                 qconf);
 975                         }
 976                 }
 977
 978                 if (likely(lcore_rx_idle_count != qconf->n_rx_queue)) {
 979                         for (i = 1, lcore_scaleup_hint =
 980                                 qconf->rx_queue_list[0].freq_up_hint;
 981                                         i < qconf->n_rx_queue; ++i) {
 982                                 rx_queue = &(qconf->rx_queue_list[i]);
 983                                 if (rx_queue->freq_up_hint >
 984                                                 lcore_scaleup_hint)
 985                                         lcore_scaleup_hint =
 986                                                 rx_queue->freq_up_hint;
 987                         }
 988
 989                         if (lcore_scaleup_hint == FREQ_HIGHEST) {
 990                                 if (rte_power_freq_max)
 991                                         rte_power_freq_max(lcore_id);
 992                         } else if (lcore_scaleup_hint == FREQ_HIGHER) {
 993                                 if (rte_power_freq_up)
 994                                         rte_power_freq_up(lcore_id);
 995                         }
 996                 } else {
 997                         /**
 998                          * All Rx queues empty in recent consecutive polls,
 999                          * sleep in a conservative manner, meaning sleep as
1000                          * less as possible.
1001                          */
1002                         for (i = 1, lcore_idle_hint =
1003                                 qconf->rx_queue_list[0].idle_hint;
1004                                         i < qconf->n_rx_queue; ++i) {
1005                                 rx_queue = &(qconf->rx_queue_list[i]);
1006                                 if (rx_queue->idle_hint < lcore_idle_hint)
1007                                         lcore_idle_hint = rx_queue->idle_hint;
1008                         }
1009
1010                         if (lcore_idle_hint < SUSPEND_THRESHOLD)
1011                                 /**
1012                                  * execute "pause" instruction to avoid context
1013                                  * switch which generally take hundred of
1014                                  * microseconds for short sleep.
1015                                  */
1016                                 rte_delay_us(lcore_idle_hint);
1017                         else {
1018                                 /* suspend until rx interrupt trigges */
1019                                 if (intr_en) {
1020                                         turn_on_intr(qconf);
1021                                         sleep_until_rx_interrupt(
1022                                                 qconf->n_rx_queue);
1023                                         /**
1024                                          * start receiving packets immediately
1025                                          */
1026                                         goto start_rx;
1027                                 }
1028                         }
1029                         stats[lcore_id].sleep_time += lcore_idle_hint;
1030                 }
1031         }
1032 }
1033
1034 static int
1035 check_lcore_params(void)
1036 {
1037         uint8_t queue, lcore;
1038         uint16_t i;
1039         int socketid;
1040
1041         for (i = 0; i < nb_lcore_params; ++i) {
1042                 queue = lcore_params[i].queue_id;
1043                 if (queue >= MAX_RX_QUEUE_PER_PORT) {
1044                         printf("invalid queue number: %hhu\n", queue);
1045                         return -1;
1046                 }
1047                 lcore = lcore_params[i].lcore_id;
1048                 if (!rte_lcore_is_enabled(lcore)) {
1049                         printf("error: lcore %hhu is not enabled in lcore "
1050                                                         "mask\n", lcore);
1051                         return -1;
1052                 }
1053                 if ((socketid = rte_lcore_to_socket_id(lcore) != 0) &&
1054                                                         (numa_on == 0)) {
1055                         printf("warning: lcore %hhu is on socket %d with numa "
1056                                                 "off\n", lcore, socketid);
1057                 }
1058         }
1059         return 0;
1060 }
1061
1062 static int
1063 check_port_config(const unsigned nb_ports)
1064 {
1065         unsigned portid;
1066         uint16_t i;
1067
1068         for (i = 0; i < nb_lcore_params; ++i) {
1069                 portid = lcore_params[i].port_id;
1070                 if ((enabled_port_mask & (1 << portid)) == 0) {
1071                         printf("port %u is not enabled in port mask\n",
1072                                                                 portid);
1073                         return -1;
1074                 }
1075                 if (portid >= nb_ports) {
1076                         printf("port %u is not present on the board\n",
1077                                                                 portid);
1078                         return -1;
1079                 }
1080         }
1081         return 0;
1082 }
1083
1084 static uint8_t
1085 get_port_n_rx_queues(const uint16_t port)
1086 {
1087         int queue = -1;
1088         uint16_t i;
1089
1090         for (i = 0; i < nb_lcore_params; ++i) {
1091                 if (lcore_params[i].port_id == port &&
1092                                 lcore_params[i].queue_id > queue)
1093                         queue = lcore_params[i].queue_id;
1094         }
1095         return (uint8_t)(++queue);
1096 }
1097
1098 static int
1099 init_lcore_rx_queues(void)
1100 {
1101         uint16_t i, nb_rx_queue;
1102         uint8_t lcore;
1103
1104         for (i = 0; i < nb_lcore_params; ++i) {
1105                 lcore = lcore_params[i].lcore_id;
1106                 nb_rx_queue = lcore_conf[lcore].n_rx_queue;
1107                 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
1108                         printf("error: too many queues (%u) for lcore: %u\n",
1109                                 (unsigned)nb_rx_queue + 1, (unsigned)lcore);
1110                         return -1;
1111                 } else {
1112                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
1113                                 lcore_params[i].port_id;
1114                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
1115                                 lcore_params[i].queue_id;
1116                         lcore_conf[lcore].n_rx_queue++;
1117                 }
1118         }
1119         return 0;
1120 }
1121
1122 /* display usage */
1123 static void
1124 print_usage(const char *prgname)
1125 {
1126         printf ("%s [EAL options] -- -p PORTMASK -P"
1127                 "  [--config (port,queue,lcore)[,(port,queue,lcore]]"
1128                 "  [--enable-jumbo [--max-pkt-len PKTLEN]]\n"
1129                 "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
1130                 "  -P : enable promiscuous mode\n"
1131                 "  --config (port,queue,lcore): rx queues configuration\n"
1132                 "  --no-numa: optional, disable numa awareness\n"
1133                 "  --enable-jumbo: enable jumbo frame"
1134                 " which max packet len is PKTLEN in decimal (64-9600)\n"
1135                 "  --parse-ptype: parse packet type by software\n",
1136                 prgname);
1137 }
1138
1139 static int parse_max_pkt_len(const char *pktlen)
1140 {
1141         char *end = NULL;
1142         unsigned long len;
1143
1144         /* parse decimal string */
1145         len = strtoul(pktlen, &end, 10);
1146         if ((pktlen[0] == '\0') || (end == NULL) || (*end != '\0'))
1147                 return -1;
1148
1149         if (len == 0)
1150                 return -1;
1151
1152         return len;
1153 }
1154
1155 static int
1156 parse_portmask(const char *portmask)
1157 {
1158         char *end = NULL;
1159         unsigned long pm;
1160
1161         /* parse hexadecimal string */
1162         pm = strtoul(portmask, &end, 16);
1163         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
1164                 return -1;
1165
1166         if (pm == 0)
1167                 return -1;
1168
1169         return pm;
1170 }
1171
1172 static int
1173 parse_config(const char *q_arg)
1174 {
1175         char s[256];
1176         const char *p, *p0 = q_arg;
1177         char *end;
1178         enum fieldnames {
1179                 FLD_PORT = 0,
1180                 FLD_QUEUE,
1181                 FLD_LCORE,
1182                 _NUM_FLD
1183         };
1184         unsigned long int_fld[_NUM_FLD];
1185         char *str_fld[_NUM_FLD];
1186         int i;
1187         unsigned size;
1188
1189         nb_lcore_params = 0;
1190
1191         while ((p = strchr(p0,'(')) != NULL) {
1192                 ++p;
1193                 if((p0 = strchr(p,')')) == NULL)
1194                         return -1;
1195
1196                 size = p0 - p;
1197                 if(size >= sizeof(s))
1198                         return -1;
1199
1200                 snprintf(s, sizeof(s), "%.*s", size, p);
1201                 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') !=
1202                                                                 _NUM_FLD)
1203                         return -1;
1204                 for (i = 0; i < _NUM_FLD; i++){
1205                         errno = 0;
1206                         int_fld[i] = strtoul(str_fld[i], &end, 0);
1207                         if (errno != 0 || end == str_fld[i] || int_fld[i] >
1208                                                                         255)
1209                                 return -1;
1210                 }
1211                 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
1212                         printf("exceeded max number of lcore params: %hu\n",
1213                                 nb_lcore_params);
1214                         return -1;
1215                 }
1216                 lcore_params_array[nb_lcore_params].port_id =
1217                                 (uint8_t)int_fld[FLD_PORT];
1218                 lcore_params_array[nb_lcore_params].queue_id =
1219                                 (uint8_t)int_fld[FLD_QUEUE];
1220                 lcore_params_array[nb_lcore_params].lcore_id =
1221                                 (uint8_t)int_fld[FLD_LCORE];
1222                 ++nb_lcore_params;
1223         }
1224         lcore_params = lcore_params_array;
1225
1226         return 0;
1227 }
1228
1229 #define CMD_LINE_OPT_PARSE_PTYPE "parse-ptype"
1230
1231 /* Parse the argument given in the command line of the application */
1232 static int
1233 parse_args(int argc, char **argv)
1234 {
1235         int opt, ret;
1236         char **argvopt;
1237         int option_index;
1238         char *prgname = argv[0];
1239         static struct option lgopts[] = {
1240                 {"config", 1, 0, 0},
1241                 {"no-numa", 0, 0, 0},
1242                 {"enable-jumbo", 0, 0, 0},
1243                 {CMD_LINE_OPT_PARSE_PTYPE, 0, 0, 0},
1244                 {NULL, 0, 0, 0}
1245         };
1246
1247         argvopt = argv;
1248
1249         while ((opt = getopt_long(argc, argvopt, "p:P",
1250                                 lgopts, &option_index)) != EOF) {
1251
1252                 switch (opt) {
1253                 /* portmask */
1254                 case 'p':
1255                         enabled_port_mask = parse_portmask(optarg);
1256                         if (enabled_port_mask == 0) {
1257                                 printf("invalid portmask\n");
1258                                 print_usage(prgname);
1259                                 return -1;
1260                         }
1261                         break;
1262                 case 'P':
1263                         printf("Promiscuous mode selected\n");
1264                         promiscuous_on = 1;
1265                         break;
1266
1267                 /* long options */
1268                 case 0:
1269                         if (!strncmp(lgopts[option_index].name, "config", 6)) {
1270                                 ret = parse_config(optarg);
1271                                 if (ret) {
1272                                         printf("invalid config\n");
1273                                         print_usage(prgname);
1274                                         return -1;
1275                                 }
1276                         }
1277
1278                         if (!strncmp(lgopts[option_index].name,
1279                                                 "no-numa", 7)) {
1280                                 printf("numa is disabled \n");
1281                                 numa_on = 0;
1282                         }
1283
1284                         if (!strncmp(lgopts[option_index].name,
1285                                         "enable-jumbo", 12)) {
1286                                 struct option lenopts =
1287                                         {"max-pkt-len", required_argument, \
1288                                                                         0, 0};
1289
1290                                 printf("jumbo frame is enabled \n");
1291                                 port_conf.rxmode.jumbo_frame = 1;
1292
1293                                 /**
1294                                  * if no max-pkt-len set, use the default value
1295                                  * ETHER_MAX_LEN
1296                                  */
1297                                 if (0 == getopt_long(argc, argvopt, "",
1298                                                 &lenopts, &option_index)) {
1299                                         ret = parse_max_pkt_len(optarg);
1300                                         if ((ret < 64) ||
1301                                                 (ret > MAX_JUMBO_PKT_LEN)){
1302                                                 printf("invalid packet "
1303                                                                 "length\n");
1304                                                 print_usage(prgname);
1305                                                 return -1;
1306                                         }
1307                                         port_conf.rxmode.max_rx_pkt_len = ret;
1308                                 }
1309                                 printf("set jumbo frame "
1310                                         "max packet length to %u\n",
1311                                 (unsigned int)port_conf.rxmode.max_rx_pkt_len);
1312                         }
1313
1314                         if (!strncmp(lgopts[option_index].name,
1315                                      CMD_LINE_OPT_PARSE_PTYPE,
1316                                      sizeof(CMD_LINE_OPT_PARSE_PTYPE))) {
1317                                 printf("soft parse-ptype is enabled\n");
1318                                 parse_ptype = 1;
1319                         }
1320
1321                         break;
1322
1323                 default:
1324                         print_usage(prgname);
1325                         return -1;
1326                 }
1327         }
1328
1329         if (optind >= 0)
1330                 argv[optind-1] = prgname;
1331
1332         ret = optind-1;
1333         optind = 1; /* reset getopt lib */
1334         return ret;
1335 }
1336
1337 static void
1338 print_ethaddr(const char *name, const struct ether_addr *eth_addr)
1339 {
1340         char buf[ETHER_ADDR_FMT_SIZE];
1341         ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr);
1342         printf("%s%s", name, buf);
1343 }
1344
1345 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1346 static void
1347 setup_hash(int socketid)
1348 {
1349         struct rte_hash_parameters ipv4_l3fwd_hash_params = {
1350                 .name = NULL,
1351                 .entries = L3FWD_HASH_ENTRIES,
1352                 .key_len = sizeof(struct ipv4_5tuple),
1353                 .hash_func = DEFAULT_HASH_FUNC,
1354                 .hash_func_init_val = 0,
1355         };
1356
1357         struct rte_hash_parameters ipv6_l3fwd_hash_params = {
1358                 .name = NULL,
1359                 .entries = L3FWD_HASH_ENTRIES,
1360                 .key_len = sizeof(struct ipv6_5tuple),
1361                 .hash_func = DEFAULT_HASH_FUNC,
1362                 .hash_func_init_val = 0,
1363         };
1364
1365         unsigned i;
1366         int ret;
1367         char s[64];
1368
1369         /* create ipv4 hash */
1370         snprintf(s, sizeof(s), "ipv4_l3fwd_hash_%d", socketid);
1371         ipv4_l3fwd_hash_params.name = s;
1372         ipv4_l3fwd_hash_params.socket_id = socketid;
1373         ipv4_l3fwd_lookup_struct[socketid] =
1374                 rte_hash_create(&ipv4_l3fwd_hash_params);
1375         if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
1376                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
1377                                 "socket %d\n", socketid);
1378
1379         /* create ipv6 hash */
1380         snprintf(s, sizeof(s), "ipv6_l3fwd_hash_%d", socketid);
1381         ipv6_l3fwd_hash_params.name = s;
1382         ipv6_l3fwd_hash_params.socket_id = socketid;
1383         ipv6_l3fwd_lookup_struct[socketid] =
1384                 rte_hash_create(&ipv6_l3fwd_hash_params);
1385         if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
1386                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
1387                                 "socket %d\n", socketid);
1388
1389
1390         /* populate the ipv4 hash */
1391         for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
1392                 ret = rte_hash_add_key (ipv4_l3fwd_lookup_struct[socketid],
1393                                 (void *) &ipv4_l3fwd_route_array[i].key);
1394                 if (ret < 0) {
1395                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
1396                                 "l3fwd hash on socket %d\n", i, socketid);
1397                 }
1398                 ipv4_l3fwd_out_if[ret] = ipv4_l3fwd_route_array[i].if_out;
1399                 printf("Hash: Adding key\n");
1400                 print_ipv4_key(ipv4_l3fwd_route_array[i].key);
1401         }
1402
1403         /* populate the ipv6 hash */
1404         for (i = 0; i < IPV6_L3FWD_NUM_ROUTES; i++) {
1405                 ret = rte_hash_add_key (ipv6_l3fwd_lookup_struct[socketid],
1406                                 (void *) &ipv6_l3fwd_route_array[i].key);
1407                 if (ret < 0) {
1408                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
1409                                 "l3fwd hash on socket %d\n", i, socketid);
1410                 }
1411                 ipv6_l3fwd_out_if[ret] = ipv6_l3fwd_route_array[i].if_out;
1412                 printf("Hash: Adding key\n");
1413                 print_ipv6_key(ipv6_l3fwd_route_array[i].key);
1414         }
1415 }
1416 #endif
1417
1418 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1419 static void
1420 setup_lpm(int socketid)
1421 {
1422         unsigned i;
1423         int ret;
1424         char s[64];
1425
1426         /* create the LPM table */
1427         struct rte_lpm_config lpm_ipv4_config;
1428
1429         lpm_ipv4_config.max_rules = IPV4_L3FWD_LPM_MAX_RULES;
1430         lpm_ipv4_config.number_tbl8s = 256;
1431         lpm_ipv4_config.flags = 0;
1432
1433         snprintf(s, sizeof(s), "IPV4_L3FWD_LPM_%d", socketid);
1434         ipv4_l3fwd_lookup_struct[socketid] =
1435                         rte_lpm_create(s, socketid, &lpm_ipv4_config);
1436         if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
1437                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
1438                                 " on socket %d\n", socketid);
1439
1440         /* populate the LPM table */
1441         for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
1442                 ret = rte_lpm_add(ipv4_l3fwd_lookup_struct[socketid],
1443                         ipv4_l3fwd_route_array[i].ip,
1444                         ipv4_l3fwd_route_array[i].depth,
1445                         ipv4_l3fwd_route_array[i].if_out);
1446
1447                 if (ret < 0) {
1448                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
1449                                 "l3fwd LPM table on socket %d\n",
1450                                 i, socketid);
1451                 }
1452
1453                 printf("LPM: Adding route 0x%08x / %d (%d)\n",
1454                         (unsigned)ipv4_l3fwd_route_array[i].ip,
1455                         ipv4_l3fwd_route_array[i].depth,
1456                         ipv4_l3fwd_route_array[i].if_out);
1457         }
1458 }
1459 #endif
1460
1461 static int
1462 init_mem(unsigned nb_mbuf)
1463 {
1464         struct lcore_conf *qconf;
1465         int socketid;
1466         unsigned lcore_id;
1467         char s[64];
1468
1469         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1470                 if (rte_lcore_is_enabled(lcore_id) == 0)
1471                         continue;
1472
1473                 if (numa_on)
1474                         socketid = rte_lcore_to_socket_id(lcore_id);
1475                 else
1476                         socketid = 0;
1477
1478                 if (socketid >= NB_SOCKETS) {
1479                         rte_exit(EXIT_FAILURE, "Socket %d of lcore %u is "
1480                                         "out of range %d\n", socketid,
1481                                                 lcore_id, NB_SOCKETS);
1482                 }
1483                 if (pktmbuf_pool[socketid] == NULL) {
1484                         snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
1485                         pktmbuf_pool[socketid] =
1486                                 rte_pktmbuf_pool_create(s, nb_mbuf,
1487                                         MEMPOOL_CACHE_SIZE, 0,
1488                                         RTE_MBUF_DEFAULT_BUF_SIZE,
1489                                         socketid);
1490                         if (pktmbuf_pool[socketid] == NULL)
1491                                 rte_exit(EXIT_FAILURE,
1492                                         "Cannot init mbuf pool on socket %d\n",
1493                                                                 socketid);
1494                         else
1495                                 printf("Allocated mbuf pool on socket %d\n",
1496                                                                 socketid);
1497
1498 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1499                         setup_lpm(socketid);
1500 #else
1501                         setup_hash(socketid);
1502 #endif
1503                 }
1504                 qconf = &lcore_conf[lcore_id];
1505                 qconf->ipv4_lookup_struct = ipv4_l3fwd_lookup_struct[socketid];
1506 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1507                 qconf->ipv6_lookup_struct = ipv6_l3fwd_lookup_struct[socketid];
1508 #endif
1509         }
1510         return 0;
1511 }
1512
1513 /* Check the link status of all ports in up to 9s, and print them finally */
1514 static void
1515 check_all_ports_link_status(uint16_t port_num, uint32_t port_mask)
1516 {
1517 #define CHECK_INTERVAL 100 /* 100ms */
1518 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
1519         uint8_t count, all_ports_up, print_flag = 0;
1520         uint16_t portid;
1521         struct rte_eth_link link;
1522
1523         printf("\nChecking link status");
1524         fflush(stdout);
1525         for (count = 0; count <= MAX_CHECK_TIME; count++) {
1526                 all_ports_up = 1;
1527                 for (portid = 0; portid < port_num; portid++) {
1528                         if ((port_mask & (1 << portid)) == 0)
1529                                 continue;
1530                         memset(&link, 0, sizeof(link));
1531                         rte_eth_link_get_nowait(portid, &link);
1532                         /* print link status if flag set */
1533                         if (print_flag == 1) {
1534                                 if (link.link_status)
1535                                         printf("Port %d Link Up - speed %u "
1536                                                 "Mbps - %s\n", (uint8_t)portid,
1537                                                 (unsigned)link.link_speed,
1538                                 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
1539                                         ("full-duplex") : ("half-duplex\n"));
1540                                 else
1541                                         printf("Port %d Link Down\n",
1542                                                 (uint8_t)portid);
1543                                 continue;
1544                         }
1545                         /* clear all_ports_up flag if any link down */
1546                         if (link.link_status == ETH_LINK_DOWN) {
1547                                 all_ports_up = 0;
1548                                 break;
1549                         }
1550                 }
1551                 /* after finally printing all link status, get out */
1552                 if (print_flag == 1)
1553                         break;
1554
1555                 if (all_ports_up == 0) {
1556                         printf(".");
1557                         fflush(stdout);
1558                         rte_delay_ms(CHECK_INTERVAL);
1559                 }
1560
1561                 /* set the print_flag if all ports up or timeout */
1562                 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
1563                         print_flag = 1;
1564                         printf("done\n");
1565                 }
1566         }
1567 }
1568
1569 static int check_ptype(uint16_t portid)
1570 {
1571         int i, ret;
1572         int ptype_l3_ipv4 = 0;
1573 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1574         int ptype_l3_ipv6 = 0;
1575 #endif
1576         uint32_t ptype_mask = RTE_PTYPE_L3_MASK;
1577
1578         ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, NULL, 0);
1579         if (ret <= 0)
1580                 return 0;
1581
1582         uint32_t ptypes[ret];
1583
1584         ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, ptypes, ret);
1585         for (i = 0; i < ret; ++i) {
1586                 if (ptypes[i] & RTE_PTYPE_L3_IPV4)
1587                         ptype_l3_ipv4 = 1;
1588 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1589                 if (ptypes[i] & RTE_PTYPE_L3_IPV6)
1590                         ptype_l3_ipv6 = 1;
1591 #endif
1592         }
1593
1594         if (ptype_l3_ipv4 == 0)
1595                 printf("port %d cannot parse RTE_PTYPE_L3_IPV4\n", portid);
1596
1597 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1598         if (ptype_l3_ipv6 == 0)
1599                 printf("port %d cannot parse RTE_PTYPE_L3_IPV6\n", portid);
1600 #endif
1601
1602 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1603         if (ptype_l3_ipv4)
1604 #else /* APP_LOOKUP_EXACT_MATCH */
1605         if (ptype_l3_ipv4 && ptype_l3_ipv6)
1606 #endif
1607                 return 1;
1608
1609         return 0;
1610
1611 }
1612
1613 int
1614 main(int argc, char **argv)
1615 {
1616         struct lcore_conf *qconf;
1617         struct rte_eth_dev_info dev_info;
1618         struct rte_eth_txconf *txconf;
1619         int ret;
1620         uint16_t nb_ports;
1621         uint16_t queueid;
1622         unsigned lcore_id;
1623         uint64_t hz;
1624         uint32_t n_tx_queue, nb_lcores;
1625         uint32_t dev_rxq_num, dev_txq_num;
1626         uint8_t nb_rx_queue, queue, socketid;
1627         uint16_t portid;
1628         uint16_t org_rxq_intr = port_conf.intr_conf.rxq;
1629
1630         /* catch SIGINT and restore cpufreq governor to ondemand */
1631         signal(SIGINT, signal_exit_now);
1632
1633         /* init EAL */
1634         ret = rte_eal_init(argc, argv);
1635         if (ret < 0)
1636                 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
1637         argc -= ret;
1638         argv += ret;
1639
1640         /* init RTE timer library to be used late */
1641         rte_timer_subsystem_init();
1642
1643         /* parse application arguments (after the EAL ones) */
1644         ret = parse_args(argc, argv);
1645         if (ret < 0)
1646                 rte_exit(EXIT_FAILURE, "Invalid L3FWD parameters\n");
1647
1648         if (check_lcore_params() < 0)
1649                 rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
1650
1651         ret = init_lcore_rx_queues();
1652         if (ret < 0)
1653                 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
1654
1655         nb_ports = rte_eth_dev_count();
1656
1657         if (check_port_config(nb_ports) < 0)
1658                 rte_exit(EXIT_FAILURE, "check_port_config failed\n");
1659
1660         nb_lcores = rte_lcore_count();
1661
1662         /* initialize all ports */
1663         for (portid = 0; portid < nb_ports; portid++) {
1664                 /* skip ports that are not enabled */
1665                 if ((enabled_port_mask & (1 << portid)) == 0) {
1666                         printf("\nSkipping disabled port %d\n", portid);
1667                         continue;
1668                 }
1669
1670                 /* init port */
1671                 printf("Initializing port %d ... ", portid );
1672                 fflush(stdout);
1673
1674                 rte_eth_dev_info_get(portid, &dev_info);
1675                 dev_rxq_num = dev_info.max_rx_queues;
1676                 dev_txq_num = dev_info.max_tx_queues;
1677
1678                 nb_rx_queue = get_port_n_rx_queues(portid);
1679                 if (nb_rx_queue > dev_rxq_num)
1680                         rte_exit(EXIT_FAILURE,
1681                                 "Cannot configure not existed rxq: "
1682                                 "port=%d\n", portid);
1683
1684                 n_tx_queue = nb_lcores;
1685                 if (n_tx_queue > dev_txq_num)
1686                         n_tx_queue = dev_txq_num;
1687                 printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
1688                         nb_rx_queue, (unsigned)n_tx_queue );
1689                 /* If number of Rx queue is 0, no need to enable Rx interrupt */
1690                 if (nb_rx_queue == 0)
1691                         port_conf.intr_conf.rxq = 0;
1692                 ret = rte_eth_dev_configure(portid, nb_rx_queue,
1693                                         (uint16_t)n_tx_queue, &port_conf);
1694                 /* Revert to original value */
1695                 port_conf.intr_conf.rxq = org_rxq_intr;
1696                 if (ret < 0)
1697                         rte_exit(EXIT_FAILURE, "Cannot configure device: "
1698                                         "err=%d, port=%d\n", ret, portid);
1699
1700                 ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
1701                                                        &nb_txd);
1702                 if (ret < 0)
1703                         rte_exit(EXIT_FAILURE,
1704                                  "Cannot adjust number of descriptors: err=%d, port=%d\n",
1705                                  ret, portid);
1706
1707                 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
1708                 print_ethaddr(" Address:", &ports_eth_addr[portid]);
1709                 printf(", ");
1710
1711                 /* init memory */
1712                 ret = init_mem(NB_MBUF);
1713                 if (ret < 0)
1714                         rte_exit(EXIT_FAILURE, "init_mem failed\n");
1715
1716                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1717                         if (rte_lcore_is_enabled(lcore_id) == 0)
1718                                 continue;
1719
1720                         /* Initialize TX buffers */
1721                         qconf = &lcore_conf[lcore_id];
1722                         qconf->tx_buffer[portid] = rte_zmalloc_socket("tx_buffer",
1723                                 RTE_ETH_TX_BUFFER_SIZE(MAX_PKT_BURST), 0,
1724                                 rte_eth_dev_socket_id(portid));
1725                         if (qconf->tx_buffer[portid] == NULL)
1726                                 rte_exit(EXIT_FAILURE, "Can't allocate tx buffer for port %u\n",
1727                                                  portid);
1728
1729                         rte_eth_tx_buffer_init(qconf->tx_buffer[portid], MAX_PKT_BURST);
1730                 }
1731
1732                 /* init one TX queue per couple (lcore,port) */
1733                 queueid = 0;
1734                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1735                         if (rte_lcore_is_enabled(lcore_id) == 0)
1736                                 continue;
1737
1738                         if (queueid >= dev_txq_num)
1739                                 continue;
1740
1741                         if (numa_on)
1742                                 socketid = \
1743                                 (uint8_t)rte_lcore_to_socket_id(lcore_id);
1744                         else
1745                                 socketid = 0;
1746
1747                         printf("txq=%u,%d,%d ", lcore_id, queueid, socketid);
1748                         fflush(stdout);
1749
1750                         rte_eth_dev_info_get(portid, &dev_info);
1751                         txconf = &dev_info.default_txconf;
1752                         if (port_conf.rxmode.jumbo_frame)
1753                                 txconf->txq_flags = 0;
1754                         ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
1755                                                      socketid, txconf);
1756                         if (ret < 0)
1757                                 rte_exit(EXIT_FAILURE,
1758                                         "rte_eth_tx_queue_setup: err=%d, "
1759                                                 "port=%d\n", ret, portid);
1760
1761                         qconf = &lcore_conf[lcore_id];
1762                         qconf->tx_queue_id[portid] = queueid;
1763                         queueid++;
1764
1765                         qconf->tx_port_id[qconf->n_tx_port] = portid;
1766                         qconf->n_tx_port++;
1767                 }
1768                 printf("\n");
1769         }
1770
1771         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1772                 if (rte_lcore_is_enabled(lcore_id) == 0)
1773                         continue;
1774
1775                 /* init power management library */
1776                 ret = rte_power_init(lcore_id);
1777                 if (ret)
1778                         RTE_LOG(ERR, POWER,
1779                                 "Library initialization failed on core %u\n", lcore_id);
1780
1781                 /* init timer structures for each enabled lcore */
1782                 rte_timer_init(&power_timers[lcore_id]);
1783                 hz = rte_get_timer_hz();
1784                 rte_timer_reset(&power_timers[lcore_id],
1785                         hz/TIMER_NUMBER_PER_SECOND, SINGLE, lcore_id,
1786                                                 power_timer_cb, NULL);
1787
1788                 qconf = &lcore_conf[lcore_id];
1789                 printf("\nInitializing rx queues on lcore %u ... ", lcore_id );
1790                 fflush(stdout);
1791                 /* init RX queues */
1792                 for(queue = 0; queue < qconf->n_rx_queue; ++queue) {
1793                         portid = qconf->rx_queue_list[queue].port_id;
1794                         queueid = qconf->rx_queue_list[queue].queue_id;
1795
1796                         if (numa_on)
1797                                 socketid = \
1798                                 (uint8_t)rte_lcore_to_socket_id(lcore_id);
1799                         else
1800                                 socketid = 0;
1801
1802                         printf("rxq=%d,%d,%d ", portid, queueid, socketid);
1803                         fflush(stdout);
1804
1805                         ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
1806                                 socketid, NULL,
1807                                 pktmbuf_pool[socketid]);
1808                         if (ret < 0)
1809                                 rte_exit(EXIT_FAILURE,
1810                                         "rte_eth_rx_queue_setup: err=%d, "
1811                                                 "port=%d\n", ret, portid);
1812
1813                         if (parse_ptype) {
1814                                 if (add_cb_parse_ptype(portid, queueid) < 0)
1815                                         rte_exit(EXIT_FAILURE,
1816                                                  "Fail to add ptype cb\n");
1817                         } else if (!check_ptype(portid))
1818                                 rte_exit(EXIT_FAILURE,
1819                                          "PMD can not provide needed ptypes\n");
1820                 }
1821         }
1822
1823         printf("\n");
1824
1825         /* start ports */
1826         for (portid = 0; portid < nb_ports; portid++) {
1827                 if ((enabled_port_mask & (1 << portid)) == 0) {
1828                         continue;
1829                 }
1830                 /* Start device */
1831                 ret = rte_eth_dev_start(portid);
1832                 if (ret < 0)
1833                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, "
1834                                                 "port=%d\n", ret, portid);
1835                 /*
1836                  * If enabled, put device in promiscuous mode.
1837                  * This allows IO forwarding mode to forward packets
1838                  * to itself through 2 cross-connected  ports of the
1839                  * target machine.
1840                  */
1841                 if (promiscuous_on)
1842                         rte_eth_promiscuous_enable(portid);
1843                 /* initialize spinlock for each port */
1844                 rte_spinlock_init(&(locks[portid]));
1845         }
1846
1847         check_all_ports_link_status(nb_ports, enabled_port_mask);
1848
1849         /* launch per-lcore init on every lcore */
1850         rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
1851         RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1852                 if (rte_eal_wait_lcore(lcore_id) < 0)
1853                         return -1;
1854         }
1855
1856         return 0;
1857 }