examples/l3fwd-power/main.c

   1 /*-
   2  *   BSD LICENSE
   3  *
   4  *   Copyright(c) 2010-2014 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_memory.h>
  51 #include <rte_memcpy.h>
  52 #include <rte_memzone.h>
  53 #include <rte_eal.h>
  54 #include <rte_per_lcore.h>
  55 #include <rte_launch.h>
  56 #include <rte_atomic.h>
  57 #include <rte_cycles.h>
  58 #include <rte_prefetch.h>
  59 #include <rte_lcore.h>
  60 #include <rte_per_lcore.h>
  61 #include <rte_branch_prediction.h>
  62 #include <rte_interrupts.h>
  63 #include <rte_pci.h>
  64 #include <rte_random.h>
  65 #include <rte_debug.h>
  66 #include <rte_ether.h>
  67 #include <rte_ethdev.h>
  68 #include <rte_ring.h>
  69 #include <rte_mempool.h>
  70 #include <rte_mbuf.h>
  71 #include <rte_ip.h>
  72 #include <rte_tcp.h>
  73 #include <rte_udp.h>
  74 #include <rte_string_fns.h>
  75 #include <rte_timer.h>
  76 #include <rte_power.h>
  77
  78 #define RTE_LOGTYPE_L3FWD_POWER RTE_LOGTYPE_USER1
  79
  80 #define MAX_PKT_BURST 32
  81
  82 #define MIN_ZERO_POLL_COUNT 5
  83
  84 /* around 100ms at 2 Ghz */
  85 #define TIMER_RESOLUTION_CYCLES           200000000ULL
  86 /* 100 ms interval */
  87 #define TIMER_NUMBER_PER_SECOND           10
  88 /* 100000 us */
  89 #define SCALING_PERIOD                    (1000000/TIMER_NUMBER_PER_SECOND)
  90 #define SCALING_DOWN_TIME_RATIO_THRESHOLD 0.25
  91
  92 #define APP_LOOKUP_EXACT_MATCH          0
  93 #define APP_LOOKUP_LPM                  1
  94 #define DO_RFC_1812_CHECKS
  95
  96 #ifndef APP_LOOKUP_METHOD
  97 #define APP_LOOKUP_METHOD             APP_LOOKUP_LPM
  98 #endif
  99
 100 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 101 #include <rte_hash.h>
 102 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 103 #include <rte_lpm.h>
 104 #else
 105 #error "APP_LOOKUP_METHOD set to incorrect value"
 106 #endif
 107
 108 #ifndef IPv6_BYTES
 109 #define IPv6_BYTES_FMT "%02x%02x:%02x%02x:%02x%02x:%02x%02x:"\
 110                        "%02x%02x:%02x%02x:%02x%02x:%02x%02x"
 111 #define IPv6_BYTES(addr) \
 112         addr[0],  addr[1], addr[2],  addr[3], \
 113         addr[4],  addr[5], addr[6],  addr[7], \
 114         addr[8],  addr[9], addr[10], addr[11],\
 115         addr[12], addr[13],addr[14], addr[15]
 116 #endif
 117
 118 #define MAX_JUMBO_PKT_LEN  9600
 119
 120 #define IPV6_ADDR_LEN 16
 121
 122 #define MEMPOOL_CACHE_SIZE 256
 123
 124 #define MBUF_DATA_SIZE (2048 + RTE_PKTMBUF_HEADROOM)
 125
 126 /*
 127  * This expression is used to calculate the number of mbufs needed depending on
 128  * user input, taking into account memory for rx and tx hardware rings, cache
 129  * per lcore and mtable per port per lcore. RTE_MAX is used to ensure that
 130  * NB_MBUF never goes below a minimum value of 8192.
 131  */
 132
 133 #define NB_MBUF RTE_MAX ( \
 134         (nb_ports*nb_rx_queue*RTE_TEST_RX_DESC_DEFAULT + \
 135         nb_ports*nb_lcores*MAX_PKT_BURST + \
 136         nb_ports*n_tx_queue*RTE_TEST_TX_DESC_DEFAULT + \
 137         nb_lcores*MEMPOOL_CACHE_SIZE), \
 138         (unsigned)8192)
 139
 140 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
 141
 142 #define NB_SOCKETS 8
 143
 144 /* Configure how many packets ahead to prefetch, when reading packets */
 145 #define PREFETCH_OFFSET 3
 146
 147 /*
 148  * Configurable number of RX/TX ring descriptors
 149  */
 150 #define RTE_TEST_RX_DESC_DEFAULT 128
 151 #define RTE_TEST_TX_DESC_DEFAULT 512
 152 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
 153 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
 154
 155 /* ethernet addresses of ports */
 156 static struct ether_addr ports_eth_addr[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
 165 enum freq_scale_hint_t
 166 {
 167         FREQ_LOWER    =      -1,
 168         FREQ_CURRENT  =       0,
 169         FREQ_HIGHER   =       1,
 170         FREQ_HIGHEST  =       2
 171 };
 172
 173 struct mbuf_table {
 174         uint16_t len;
 175         struct rte_mbuf *m_table[MAX_PKT_BURST];
 176 };
 177
 178 struct lcore_rx_queue {
 179         uint8_t port_id;
 180         uint8_t queue_id;
 181         enum freq_scale_hint_t freq_up_hint;
 182         uint32_t zero_rx_packet_count;
 183         uint32_t idle_hint;
 184 } __rte_cache_aligned;
 185
 186 #define MAX_RX_QUEUE_PER_LCORE 16
 187 #define MAX_TX_QUEUE_PER_PORT RTE_MAX_ETHPORTS
 188 #define MAX_RX_QUEUE_PER_PORT 128
 189
 190 #define MAX_LCORE_PARAMS 1024
 191 struct lcore_params {
 192         uint8_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   = 0, /**< CRC stripped by hardware */
 224         },
 225         .rx_adv_conf = {
 226                 .rss_conf = {
 227                         .rss_key = NULL,
 228                         .rss_hf = ETH_RSS_IP,
 229                 },
 230         },
 231         .txmode = {
 232                 .mq_mode = ETH_DCB_NONE,
 233         },
 234 };
 235
 236 static struct rte_mempool * pktmbuf_pool[NB_SOCKETS];
 237
 238
 239 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 240
 241 #ifdef RTE_MACHINE_CPUFLAG_SSE4_2
 242 #include <rte_hash_crc.h>
 243 #define DEFAULT_HASH_FUNC       rte_hash_crc
 244 #else
 245 #include <rte_jhash.h>
 246 #define DEFAULT_HASH_FUNC       rte_jhash
 247 #endif
 248
 249 struct ipv4_5tuple {
 250         uint32_t ip_dst;
 251         uint32_t ip_src;
 252         uint16_t port_dst;
 253         uint16_t port_src;
 254         uint8_t  proto;
 255 } __attribute__((__packed__));
 256
 257 struct ipv6_5tuple {
 258         uint8_t  ip_dst[IPV6_ADDR_LEN];
 259         uint8_t  ip_src[IPV6_ADDR_LEN];
 260         uint16_t port_dst;
 261         uint16_t port_src;
 262         uint8_t  proto;
 263 } __attribute__((__packed__));
 264
 265 struct ipv4_l3fwd_route {
 266         struct ipv4_5tuple key;
 267         uint8_t if_out;
 268 };
 269
 270 struct ipv6_l3fwd_route {
 271         struct ipv6_5tuple key;
 272         uint8_t if_out;
 273 };
 274
 275 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
 276         {{IPv4(100,10,0,1), IPv4(200,10,0,1), 101, 11, IPPROTO_TCP}, 0},
 277         {{IPv4(100,20,0,2), IPv4(200,20,0,2), 102, 12, IPPROTO_TCP}, 1},
 278         {{IPv4(100,30,0,3), IPv4(200,30,0,3), 103, 13, IPPROTO_TCP}, 2},
 279         {{IPv4(100,40,0,4), IPv4(200,40,0,4), 104, 14, IPPROTO_TCP}, 3},
 280 };
 281
 282 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
 283         {
 284                 {
 285                         {0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
 286                          0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
 287                         {0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
 288                          0x02, 0x1e, 0x67, 0xff, 0xfe, 0x0d, 0xb6, 0x0a},
 289                          1, 10, IPPROTO_UDP
 290                 }, 4
 291         },
 292 };
 293
 294 typedef struct rte_hash lookup_struct_t;
 295 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
 296 static lookup_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
 297
 298 #define L3FWD_HASH_ENTRIES      1024
 299
 300 #define IPV4_L3FWD_NUM_ROUTES \
 301         (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
 302
 303 #define IPV6_L3FWD_NUM_ROUTES \
 304         (sizeof(ipv6_l3fwd_route_array) / sizeof(ipv6_l3fwd_route_array[0]))
 305
 306 static uint8_t ipv4_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 307 static uint8_t ipv6_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 308 #endif
 309
 310 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 311 struct ipv4_l3fwd_route {
 312         uint32_t ip;
 313         uint8_t  depth;
 314         uint8_t  if_out;
 315 };
 316
 317 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
 318         {IPv4(1,1,1,0), 24, 0},
 319         {IPv4(2,1,1,0), 24, 1},
 320         {IPv4(3,1,1,0), 24, 2},
 321         {IPv4(4,1,1,0), 24, 3},
 322         {IPv4(5,1,1,0), 24, 4},
 323         {IPv4(6,1,1,0), 24, 5},
 324         {IPv4(7,1,1,0), 24, 6},
 325         {IPv4(8,1,1,0), 24, 7},
 326 };
 327
 328 #define IPV4_L3FWD_NUM_ROUTES \
 329         (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
 330
 331 #define IPV4_L3FWD_LPM_MAX_RULES     1024
 332
 333 typedef struct rte_lpm lookup_struct_t;
 334 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
 335 #endif
 336
 337 struct lcore_conf {
 338         uint16_t n_rx_queue;
 339         struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
 340         uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
 341         struct mbuf_table tx_mbufs[RTE_MAX_ETHPORTS];
 342         lookup_struct_t * ipv4_lookup_struct;
 343         lookup_struct_t * ipv6_lookup_struct;
 344 } __rte_cache_aligned;
 345
 346 struct lcore_stats {
 347         /* total sleep time in ms since last frequency scaling down */
 348         uint32_t sleep_time;
 349         /* number of long sleep recently */
 350         uint32_t nb_long_sleep;
 351         /* freq. scaling up trend */
 352         uint32_t trend;
 353         /* total packet processed recently */
 354         uint64_t nb_rx_processed;
 355         /* total iterations looped recently */
 356         uint64_t nb_iteration_looped;
 357         uint32_t padding[9];
 358 } __rte_cache_aligned;
 359
 360 static struct lcore_conf lcore_conf[RTE_MAX_LCORE] __rte_cache_aligned;
 361 static struct lcore_stats stats[RTE_MAX_LCORE] __rte_cache_aligned;
 362 static struct rte_timer power_timers[RTE_MAX_LCORE];
 363
 364 static inline uint32_t power_idle_heuristic(uint32_t zero_rx_packet_count);
 365 static inline enum freq_scale_hint_t power_freq_scaleup_heuristic( \
 366                         unsigned lcore_id, uint8_t port_id, uint16_t queue_id);
 367
 368 /* exit signal handler */
 369 static void
 370 signal_exit_now(int sigtype)
 371 {
 372         unsigned lcore_id;
 373         int ret;
 374
 375         if (sigtype == SIGINT) {
 376                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
 377                         if (rte_lcore_is_enabled(lcore_id) == 0)
 378                                 continue;
 379
 380                         /* init power management library */
 381                         ret = rte_power_exit(lcore_id);
 382                         if (ret)
 383                                 rte_exit(EXIT_FAILURE, "Power management "
 384                                         "library de-initialization failed on "
 385                                                         "core%u\n", lcore_id);
 386                 }
 387         }
 388
 389         rte_exit(EXIT_SUCCESS, "User forced exit\n");
 390 }
 391
 392 /*  Freqency scale down timer callback */
 393 static void
 394 power_timer_cb(__attribute__((unused)) struct rte_timer *tim,
 395                           __attribute__((unused)) void *arg)
 396 {
 397         uint64_t hz;
 398         float sleep_time_ratio;
 399         unsigned lcore_id = rte_lcore_id();
 400
 401         /* accumulate total execution time in us when callback is invoked */
 402         sleep_time_ratio = (float)(stats[lcore_id].sleep_time) /
 403                                         (float)SCALING_PERIOD;
 404
 405         /**
 406          * check whether need to scale down frequency a step if it sleep a lot.
 407          */
 408         if (sleep_time_ratio >= SCALING_DOWN_TIME_RATIO_THRESHOLD)
 409                 rte_power_freq_down(lcore_id);
 410         else if ( (unsigned)(stats[lcore_id].nb_rx_processed /
 411                 stats[lcore_id].nb_iteration_looped) < MAX_PKT_BURST)
 412                 /**
 413                  * scale down a step if average packet per iteration less
 414                  * than expectation.
 415                  */
 416                 rte_power_freq_down(lcore_id);
 417
 418         /**
 419          * initialize another timer according to current frequency to ensure
 420          * timer interval is relatively fixed.
 421          */
 422         hz = rte_get_timer_hz();
 423         rte_timer_reset(&power_timers[lcore_id], hz/TIMER_NUMBER_PER_SECOND,
 424                                 SINGLE, lcore_id, power_timer_cb, NULL);
 425
 426         stats[lcore_id].nb_rx_processed = 0;
 427         stats[lcore_id].nb_iteration_looped = 0;
 428
 429         stats[lcore_id].sleep_time = 0;
 430 }
 431
 432 /* Send burst of packets on an output interface */
 433 static inline int
 434 send_burst(struct lcore_conf *qconf, uint16_t n, uint8_t port)
 435 {
 436         struct rte_mbuf **m_table;
 437         int ret;
 438         uint16_t queueid;
 439
 440         queueid = qconf->tx_queue_id[port];
 441         m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
 442
 443         ret = rte_eth_tx_burst(port, queueid, m_table, n);
 444         if (unlikely(ret < n)) {
 445                 do {
 446                         rte_pktmbuf_free(m_table[ret]);
 447                 } while (++ret < n);
 448         }
 449
 450         return 0;
 451 }
 452
 453 /* Enqueue a single packet, and send burst if queue is filled */
 454 static inline int
 455 send_single_packet(struct rte_mbuf *m, uint8_t port)
 456 {
 457         uint32_t lcore_id;
 458         uint16_t len;
 459         struct lcore_conf *qconf;
 460
 461         lcore_id = rte_lcore_id();
 462
 463         qconf = &lcore_conf[lcore_id];
 464         len = qconf->tx_mbufs[port].len;
 465         qconf->tx_mbufs[port].m_table[len] = m;
 466         len++;
 467
 468         /* enough pkts to be sent */
 469         if (unlikely(len == MAX_PKT_BURST)) {
 470                 send_burst(qconf, MAX_PKT_BURST, port);
 471                 len = 0;
 472         }
 473
 474         qconf->tx_mbufs[port].len = len;
 475         return 0;
 476 }
 477
 478 #ifdef DO_RFC_1812_CHECKS
 479 static inline int
 480 is_valid_ipv4_pkt(struct ipv4_hdr *pkt, uint32_t link_len)
 481 {
 482         /* From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2 */
 483         /*
 484          * 1. The packet length reported by the Link Layer must be large
 485          * enough to hold the minimum length legal IP datagram (20 bytes).
 486          */
 487         if (link_len < sizeof(struct ipv4_hdr))
 488                 return -1;
 489
 490         /* 2. The IP checksum must be correct. */
 491         /* this is checked in H/W */
 492
 493         /*
 494          * 3. The IP version number must be 4. If the version number is not 4
 495          * then the packet may be another version of IP, such as IPng or
 496          * ST-II.
 497          */
 498         if (((pkt->version_ihl) >> 4) != 4)
 499                 return -3;
 500         /*
 501          * 4. The IP header length field must be large enough to hold the
 502          * minimum length legal IP datagram (20 bytes = 5 words).
 503          */
 504         if ((pkt->version_ihl & 0xf) < 5)
 505                 return -4;
 506
 507         /*
 508          * 5. The IP total length field must be large enough to hold the IP
 509          * datagram header, whose length is specified in the IP header length
 510          * field.
 511          */
 512         if (rte_cpu_to_be_16(pkt->total_length) < sizeof(struct ipv4_hdr))
 513                 return -5;
 514
 515         return 0;
 516 }
 517 #endif
 518
 519 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 520 static void
 521 print_ipv4_key(struct ipv4_5tuple key)
 522 {
 523         printf("IP dst = %08x, IP src = %08x, port dst = %d, port src = %d, "
 524                 "proto = %d\n", (unsigned)key.ip_dst, (unsigned)key.ip_src,
 525                                 key.port_dst, key.port_src, key.proto);
 526 }
 527 static void
 528 print_ipv6_key(struct ipv6_5tuple key)
 529 {
 530         printf( "IP dst = " IPv6_BYTES_FMT ", IP src = " IPv6_BYTES_FMT ", "
 531                 "port dst = %d, port src = %d, proto = %d\n",
 532                 IPv6_BYTES(key.ip_dst), IPv6_BYTES(key.ip_src),
 533                 key.port_dst, key.port_src, key.proto);
 534 }
 535
 536 static inline uint8_t
 537 get_ipv4_dst_port(struct ipv4_hdr *ipv4_hdr, uint8_t portid,
 538                 lookup_struct_t * ipv4_l3fwd_lookup_struct)
 539 {
 540         struct ipv4_5tuple key;
 541         struct tcp_hdr *tcp;
 542         struct udp_hdr *udp;
 543         int ret = 0;
 544
 545         key.ip_dst = rte_be_to_cpu_32(ipv4_hdr->dst_addr);
 546         key.ip_src = rte_be_to_cpu_32(ipv4_hdr->src_addr);
 547         key.proto = ipv4_hdr->next_proto_id;
 548
 549         switch (ipv4_hdr->next_proto_id) {
 550         case IPPROTO_TCP:
 551                 tcp = (struct tcp_hdr *)((unsigned char *)ipv4_hdr +
 552                                         sizeof(struct ipv4_hdr));
 553                 key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
 554                 key.port_src = rte_be_to_cpu_16(tcp->src_port);
 555                 break;
 556
 557         case IPPROTO_UDP:
 558                 udp = (struct udp_hdr *)((unsigned char *)ipv4_hdr +
 559                                         sizeof(struct ipv4_hdr));
 560                 key.port_dst = rte_be_to_cpu_16(udp->dst_port);
 561                 key.port_src = rte_be_to_cpu_16(udp->src_port);
 562                 break;
 563
 564         default:
 565                 key.port_dst = 0;
 566                 key.port_src = 0;
 567                 break;
 568         }
 569
 570         /* Find destination port */
 571         ret = rte_hash_lookup(ipv4_l3fwd_lookup_struct, (const void *)&key);
 572         return (uint8_t)((ret < 0)? portid : ipv4_l3fwd_out_if[ret]);
 573 }
 574
 575 static inline uint8_t
 576 get_ipv6_dst_port(struct ipv6_hdr *ipv6_hdr,  uint8_t portid,
 577                         lookup_struct_t *ipv6_l3fwd_lookup_struct)
 578 {
 579         struct ipv6_5tuple key;
 580         struct tcp_hdr *tcp;
 581         struct udp_hdr *udp;
 582         int ret = 0;
 583
 584         memcpy(key.ip_dst, ipv6_hdr->dst_addr, IPV6_ADDR_LEN);
 585         memcpy(key.ip_src, ipv6_hdr->src_addr, IPV6_ADDR_LEN);
 586
 587         key.proto = ipv6_hdr->proto;
 588
 589         switch (ipv6_hdr->proto) {
 590         case IPPROTO_TCP:
 591                 tcp = (struct tcp_hdr *)((unsigned char *) ipv6_hdr +
 592                                         sizeof(struct ipv6_hdr));
 593                 key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
 594                 key.port_src = rte_be_to_cpu_16(tcp->src_port);
 595                 break;
 596
 597         case IPPROTO_UDP:
 598                 udp = (struct udp_hdr *)((unsigned char *) ipv6_hdr +
 599                                         sizeof(struct ipv6_hdr));
 600                 key.port_dst = rte_be_to_cpu_16(udp->dst_port);
 601                 key.port_src = rte_be_to_cpu_16(udp->src_port);
 602                 break;
 603
 604         default:
 605                 key.port_dst = 0;
 606                 key.port_src = 0;
 607                 break;
 608         }
 609
 610         /* Find destination port */
 611         ret = rte_hash_lookup(ipv6_l3fwd_lookup_struct, (const void *)&key);
 612         return (uint8_t)((ret < 0)? portid : ipv6_l3fwd_out_if[ret]);
 613 }
 614 #endif
 615
 616 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 617 static inline uint8_t
 618 get_ipv4_dst_port(struct ipv4_hdr *ipv4_hdr, uint8_t portid,
 619                 lookup_struct_t *ipv4_l3fwd_lookup_struct)
 620 {
 621         uint8_t next_hop;
 622
 623         return (uint8_t) ((rte_lpm_lookup(ipv4_l3fwd_lookup_struct,
 624                         rte_be_to_cpu_32(ipv4_hdr->dst_addr), &next_hop) == 0)?
 625                         next_hop : portid);
 626 }
 627 #endif
 628
 629 static inline void
 630 l3fwd_simple_forward(struct rte_mbuf *m, uint8_t portid,
 631                                 struct lcore_conf *qconf)
 632 {
 633         struct ether_hdr *eth_hdr;
 634         struct ipv4_hdr *ipv4_hdr;
 635         void *d_addr_bytes;
 636         uint8_t dst_port;
 637
 638         eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
 639
 640         if (m->ol_flags & PKT_RX_IPV4_HDR) {
 641                 /* Handle IPv4 headers.*/
 642                 ipv4_hdr =
 643                         (struct ipv4_hdr *)(rte_pktmbuf_mtod(m, unsigned char*)
 644                                                 + sizeof(struct ether_hdr));
 645
 646 #ifdef DO_RFC_1812_CHECKS
 647                 /* Check to make sure the packet is valid (RFC1812) */
 648                 if (is_valid_ipv4_pkt(ipv4_hdr, m->pkt_len) < 0) {
 649                         rte_pktmbuf_free(m);
 650                         return;
 651                 }
 652 #endif
 653
 654                 dst_port = get_ipv4_dst_port(ipv4_hdr, portid,
 655                                         qconf->ipv4_lookup_struct);
 656                 if (dst_port >= RTE_MAX_ETHPORTS ||
 657                                 (enabled_port_mask & 1 << dst_port) == 0)
 658                         dst_port = portid;
 659
 660                 /* 02:00:00:00:00:xx */
 661                 d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
 662                 *((uint64_t *)d_addr_bytes) =
 663                         0x000000000002 + ((uint64_t)dst_port << 40);
 664
 665 #ifdef DO_RFC_1812_CHECKS
 666                 /* Update time to live and header checksum */
 667                 --(ipv4_hdr->time_to_live);
 668                 ++(ipv4_hdr->hdr_checksum);
 669 #endif
 670
 671                 /* src addr */
 672                 ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
 673
 674                 send_single_packet(m, dst_port);
 675         }
 676         else {
 677                 /* Handle IPv6 headers.*/
 678 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 679                 struct ipv6_hdr *ipv6_hdr;
 680
 681                 ipv6_hdr =
 682                         (struct ipv6_hdr *)(rte_pktmbuf_mtod(m, unsigned char*)
 683                                                 + sizeof(struct ether_hdr));
 684
 685                 dst_port = get_ipv6_dst_port(ipv6_hdr, portid,
 686                                         qconf->ipv6_lookup_struct);
 687
 688                 if (dst_port >= RTE_MAX_ETHPORTS ||
 689                                 (enabled_port_mask & 1 << dst_port) == 0)
 690                         dst_port = portid;
 691
 692                 /* 02:00:00:00:00:xx */
 693                 d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
 694                 *((uint64_t *)d_addr_bytes) =
 695                         0x000000000002 + ((uint64_t)dst_port << 40);
 696
 697                 /* src addr */
 698                 ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
 699
 700                 send_single_packet(m, dst_port);
 701 #else
 702                 /* We don't currently handle IPv6 packets in LPM mode. */
 703                 rte_pktmbuf_free(m);
 704 #endif
 705         }
 706
 707 }
 708
 709 #define SLEEP_GEAR1_THRESHOLD            100
 710 #define SLEEP_GEAR2_THRESHOLD            1000
 711
 712 static inline uint32_t
 713 power_idle_heuristic(uint32_t zero_rx_packet_count)
 714 {
 715         /* If zero count is less than 100, use it as the sleep time in us */
 716         if (zero_rx_packet_count < SLEEP_GEAR1_THRESHOLD)
 717                 return zero_rx_packet_count;
 718         /* If zero count is less than 1000, sleep time should be 100 us */
 719         else if ((zero_rx_packet_count >= SLEEP_GEAR1_THRESHOLD) &&
 720                         (zero_rx_packet_count < SLEEP_GEAR2_THRESHOLD))
 721                 return SLEEP_GEAR1_THRESHOLD;
 722         /* If zero count is greater than 1000, sleep time should be 1000 us */
 723         else if (zero_rx_packet_count >= SLEEP_GEAR2_THRESHOLD)
 724                 return SLEEP_GEAR2_THRESHOLD;
 725
 726         return 0;
 727 }
 728
 729 static inline enum freq_scale_hint_t
 730 power_freq_scaleup_heuristic(unsigned lcore_id,
 731                              uint8_t port_id,
 732                              uint16_t queue_id)
 733 {
 734 /**
 735  * HW Rx queue size is 128 by default, Rx burst read at maximum 32 entries
 736  * per iteration
 737  */
 738 #define FREQ_GEAR1_RX_PACKET_THRESHOLD             MAX_PKT_BURST
 739 #define FREQ_GEAR2_RX_PACKET_THRESHOLD             (MAX_PKT_BURST*2)
 740 #define FREQ_GEAR3_RX_PACKET_THRESHOLD             (MAX_PKT_BURST*3)
 741 #define FREQ_UP_TREND1_ACC   1
 742 #define FREQ_UP_TREND2_ACC   100
 743 #define FREQ_UP_THRESHOLD    10000
 744
 745         if (likely(rte_eth_rx_descriptor_done(port_id, queue_id,
 746                         FREQ_GEAR3_RX_PACKET_THRESHOLD) > 0)) {
 747                 stats[lcore_id].trend = 0;
 748                 return FREQ_HIGHEST;
 749         } else if (likely(rte_eth_rx_descriptor_done(port_id, queue_id,
 750                         FREQ_GEAR2_RX_PACKET_THRESHOLD) > 0))
 751                 stats[lcore_id].trend += FREQ_UP_TREND2_ACC;
 752         else if (likely(rte_eth_rx_descriptor_done(port_id, queue_id,
 753                         FREQ_GEAR1_RX_PACKET_THRESHOLD) > 0))
 754                 stats[lcore_id].trend += FREQ_UP_TREND1_ACC;
 755
 756         if (likely(stats[lcore_id].trend > FREQ_UP_THRESHOLD)) {
 757                 stats[lcore_id].trend = 0;
 758                 return FREQ_HIGHER;
 759         }
 760
 761         return FREQ_CURRENT;
 762 }
 763
 764 /* main processing loop */
 765 static int
 766 main_loop(__attribute__((unused)) void *dummy)
 767 {
 768         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
 769         unsigned lcore_id;
 770         uint64_t prev_tsc, diff_tsc, cur_tsc;
 771         uint64_t prev_tsc_power = 0, cur_tsc_power, diff_tsc_power;
 772         int i, j, nb_rx;
 773         uint8_t portid, queueid;
 774         struct lcore_conf *qconf;
 775         struct lcore_rx_queue *rx_queue;
 776         enum freq_scale_hint_t lcore_scaleup_hint;
 777
 778         uint32_t lcore_rx_idle_count = 0;
 779         uint32_t lcore_idle_hint = 0;
 780
 781         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
 782
 783         prev_tsc = 0;
 784
 785         lcore_id = rte_lcore_id();
 786         qconf = &lcore_conf[lcore_id];
 787
 788         if (qconf->n_rx_queue == 0) {
 789                 RTE_LOG(INFO, L3FWD_POWER, "lcore %u has nothing to do\n", lcore_id);
 790                 return 0;
 791         }
 792
 793         RTE_LOG(INFO, L3FWD_POWER, "entering main loop on lcore %u\n", lcore_id);
 794
 795         for (i = 0; i < qconf->n_rx_queue; i++) {
 796
 797                 portid = qconf->rx_queue_list[i].port_id;
 798                 queueid = qconf->rx_queue_list[i].queue_id;
 799                 RTE_LOG(INFO, L3FWD_POWER, " -- lcoreid=%u portid=%hhu "
 800                         "rxqueueid=%hhu\n", lcore_id, portid, queueid);
 801         }
 802
 803         while (1) {
 804                 stats[lcore_id].nb_iteration_looped++;
 805
 806                 cur_tsc = rte_rdtsc();
 807                 cur_tsc_power = cur_tsc;
 808
 809                 /*
 810                  * TX burst queue drain
 811                  */
 812                 diff_tsc = cur_tsc - prev_tsc;
 813                 if (unlikely(diff_tsc > drain_tsc)) {
 814
 815                         /*
 816                          * This could be optimized (use queueid instead of
 817                          * portid), but it is not called so often
 818                          */
 819                         for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
 820                                 if (qconf->tx_mbufs[portid].len == 0)
 821                                         continue;
 822                                 send_burst(&lcore_conf[lcore_id],
 823                                         qconf->tx_mbufs[portid].len,
 824                                         portid);
 825                                 qconf->tx_mbufs[portid].len = 0;
 826                         }
 827
 828                         prev_tsc = cur_tsc;
 829                 }
 830
 831                 diff_tsc_power = cur_tsc_power - prev_tsc_power;
 832                 if (diff_tsc_power > TIMER_RESOLUTION_CYCLES) {
 833                         rte_timer_manage();
 834                         prev_tsc_power = cur_tsc_power;
 835                 }
 836
 837                 /*
 838                  * Read packet from RX queues
 839                  */
 840                 lcore_scaleup_hint = FREQ_CURRENT;
 841                 lcore_rx_idle_count = 0;
 842                 for (i = 0; i < qconf->n_rx_queue; ++i) {
 843                         rx_queue = &(qconf->rx_queue_list[i]);
 844                         rx_queue->idle_hint = 0;
 845                         portid = rx_queue->port_id;
 846                         queueid = rx_queue->queue_id;
 847
 848                         nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
 849                                                                 MAX_PKT_BURST);
 850                         stats[lcore_id].nb_rx_processed += nb_rx;
 851                         if (unlikely(nb_rx == 0)) {
 852                                 /**
 853                                  * no packet received from rx queue, try to
 854                                  * sleep for a while forcing CPU enter deeper
 855                                  * C states.
 856                                  */
 857                                 rx_queue->zero_rx_packet_count++;
 858
 859                                 if (rx_queue->zero_rx_packet_count <=
 860                                                         MIN_ZERO_POLL_COUNT)
 861                                         continue;
 862
 863                                 rx_queue->idle_hint = power_idle_heuristic(\
 864                                         rx_queue->zero_rx_packet_count);
 865                                 lcore_rx_idle_count++;
 866                         } else {
 867                                 rx_queue->zero_rx_packet_count = 0;
 868
 869                                 /**
 870                                  * do not scale up frequency immediately as
 871                                  * user to kernel space communication is costly
 872                                  * which might impact packet I/O for received
 873                                  * packets.
 874                                  */
 875                                 rx_queue->freq_up_hint =
 876                                         power_freq_scaleup_heuristic(lcore_id,
 877                                                         portid, queueid);
 878                         }
 879
 880                         /* Prefetch first packets */
 881                         for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
 882                                 rte_prefetch0(rte_pktmbuf_mtod(
 883                                                 pkts_burst[j], void *));
 884                         }
 885
 886                         /* Prefetch and forward already prefetched packets */
 887                         for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
 888                                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
 889                                                 j + PREFETCH_OFFSET], void *));
 890                                 l3fwd_simple_forward(pkts_burst[j], portid,
 891                                                                 qconf);
 892                         }
 893
 894                         /* Forward remaining prefetched packets */
 895                         for (; j < nb_rx; j++) {
 896                                 l3fwd_simple_forward(pkts_burst[j], portid,
 897                                                                 qconf);
 898                         }
 899                 }
 900
 901                 if (likely(lcore_rx_idle_count != qconf->n_rx_queue)) {
 902                         for (i = 1, lcore_scaleup_hint =
 903                                 qconf->rx_queue_list[0].freq_up_hint;
 904                                         i < qconf->n_rx_queue; ++i) {
 905                                 rx_queue = &(qconf->rx_queue_list[i]);
 906                                 if (rx_queue->freq_up_hint >
 907                                                 lcore_scaleup_hint)
 908                                         lcore_scaleup_hint =
 909                                                 rx_queue->freq_up_hint;
 910                         }
 911
 912                         if (lcore_scaleup_hint == FREQ_HIGHEST)
 913                                 rte_power_freq_max(lcore_id);
 914                         else if (lcore_scaleup_hint == FREQ_HIGHER)
 915                                 rte_power_freq_up(lcore_id);
 916                 } else {
 917                         /**
 918                          * All Rx queues empty in recent consecutive polls,
 919                          * sleep in a conservative manner, meaning sleep as
 920                          * less as possible.
 921                          */
 922                         for (i = 1, lcore_idle_hint =
 923                                 qconf->rx_queue_list[0].idle_hint;
 924                                         i < qconf->n_rx_queue; ++i) {
 925                                 rx_queue = &(qconf->rx_queue_list[i]);
 926                                 if (rx_queue->idle_hint < lcore_idle_hint)
 927                                         lcore_idle_hint = rx_queue->idle_hint;
 928                         }
 929
 930                         if ( lcore_idle_hint < SLEEP_GEAR1_THRESHOLD)
 931                                 /**
 932                                  * execute "pause" instruction to avoid context
 933                                  * switch for short sleep.
 934                                  */
 935                                 rte_delay_us(lcore_idle_hint);
 936                         else
 937                                 /* long sleep force runing thread to suspend */
 938                                 usleep(lcore_idle_hint);
 939
 940                         stats[lcore_id].sleep_time += lcore_idle_hint;
 941                 }
 942         }
 943 }
 944
 945 static int
 946 check_lcore_params(void)
 947 {
 948         uint8_t queue, lcore;
 949         uint16_t i;
 950         int socketid;
 951
 952         for (i = 0; i < nb_lcore_params; ++i) {
 953                 queue = lcore_params[i].queue_id;
 954                 if (queue >= MAX_RX_QUEUE_PER_PORT) {
 955                         printf("invalid queue number: %hhu\n", queue);
 956                         return -1;
 957                 }
 958                 lcore = lcore_params[i].lcore_id;
 959                 if (!rte_lcore_is_enabled(lcore)) {
 960                         printf("error: lcore %hhu is not enabled in lcore "
 961                                                         "mask\n", lcore);
 962                         return -1;
 963                 }
 964                 if ((socketid = rte_lcore_to_socket_id(lcore) != 0) &&
 965                                                         (numa_on == 0)) {
 966                         printf("warning: lcore %hhu is on socket %d with numa "
 967                                                 "off\n", lcore, socketid);
 968                 }
 969         }
 970         return 0;
 971 }
 972
 973 static int
 974 check_port_config(const unsigned nb_ports)
 975 {
 976         unsigned portid;
 977         uint16_t i;
 978
 979         for (i = 0; i < nb_lcore_params; ++i) {
 980                 portid = lcore_params[i].port_id;
 981                 if ((enabled_port_mask & (1 << portid)) == 0) {
 982                         printf("port %u is not enabled in port mask\n",
 983                                                                 portid);
 984                         return -1;
 985                 }
 986                 if (portid >= nb_ports) {
 987                         printf("port %u is not present on the board\n",
 988                                                                 portid);
 989                         return -1;
 990                 }
 991         }
 992         return 0;
 993 }
 994
 995 static uint8_t
 996 get_port_n_rx_queues(const uint8_t port)
 997 {
 998         int queue = -1;
 999         uint16_t i;
1000
1001         for (i = 0; i < nb_lcore_params; ++i) {
1002                 if (lcore_params[i].port_id == port &&
1003                                 lcore_params[i].queue_id > queue)
1004                         queue = lcore_params[i].queue_id;
1005         }
1006         return (uint8_t)(++queue);
1007 }
1008
1009 static int
1010 init_lcore_rx_queues(void)
1011 {
1012         uint16_t i, nb_rx_queue;
1013         uint8_t lcore;
1014
1015         for (i = 0; i < nb_lcore_params; ++i) {
1016                 lcore = lcore_params[i].lcore_id;
1017                 nb_rx_queue = lcore_conf[lcore].n_rx_queue;
1018                 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
1019                         printf("error: too many queues (%u) for lcore: %u\n",
1020                                 (unsigned)nb_rx_queue + 1, (unsigned)lcore);
1021                         return -1;
1022                 } else {
1023                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
1024                                 lcore_params[i].port_id;
1025                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
1026                                 lcore_params[i].queue_id;
1027                         lcore_conf[lcore].n_rx_queue++;
1028                 }
1029         }
1030         return 0;
1031 }
1032
1033 /* display usage */
1034 static void
1035 print_usage(const char *prgname)
1036 {
1037         printf ("%s [EAL options] -- -p PORTMASK -P"
1038                 "  [--config (port,queue,lcore)[,(port,queue,lcore]]"
1039                 "  [--enable-jumbo [--max-pkt-len PKTLEN]]\n"
1040                 "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
1041                 "  -P : enable promiscuous mode\n"
1042                 "  --config (port,queue,lcore): rx queues configuration\n"
1043                 "  --no-numa: optional, disable numa awareness\n"
1044                 "  --enable-jumbo: enable jumbo frame"
1045                 " which max packet len is PKTLEN in decimal (64-9600)\n",
1046                 prgname);
1047 }
1048
1049 static int parse_max_pkt_len(const char *pktlen)
1050 {
1051         char *end = NULL;
1052         unsigned long len;
1053
1054         /* parse decimal string */
1055         len = strtoul(pktlen, &end, 10);
1056         if ((pktlen[0] == '\0') || (end == NULL) || (*end != '\0'))
1057                 return -1;
1058
1059         if (len == 0)
1060                 return -1;
1061
1062         return len;
1063 }
1064
1065 static int
1066 parse_portmask(const char *portmask)
1067 {
1068         char *end = NULL;
1069         unsigned long pm;
1070
1071         /* parse hexadecimal string */
1072         pm = strtoul(portmask, &end, 16);
1073         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
1074                 return -1;
1075
1076         if (pm == 0)
1077                 return -1;
1078
1079         return pm;
1080 }
1081
1082 static int
1083 parse_config(const char *q_arg)
1084 {
1085         char s[256];
1086         const char *p, *p0 = q_arg;
1087         char *end;
1088         enum fieldnames {
1089                 FLD_PORT = 0,
1090                 FLD_QUEUE,
1091                 FLD_LCORE,
1092                 _NUM_FLD
1093         };
1094         unsigned long int_fld[_NUM_FLD];
1095         char *str_fld[_NUM_FLD];
1096         int i;
1097         unsigned size;
1098
1099         nb_lcore_params = 0;
1100
1101         while ((p = strchr(p0,'(')) != NULL) {
1102                 ++p;
1103                 if((p0 = strchr(p,')')) == NULL)
1104                         return -1;
1105
1106                 size = p0 - p;
1107                 if(size >= sizeof(s))
1108                         return -1;
1109
1110                 snprintf(s, sizeof(s), "%.*s", size, p);
1111                 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') !=
1112                                                                 _NUM_FLD)
1113                         return -1;
1114                 for (i = 0; i < _NUM_FLD; i++){
1115                         errno = 0;
1116                         int_fld[i] = strtoul(str_fld[i], &end, 0);
1117                         if (errno != 0 || end == str_fld[i] || int_fld[i] >
1118                                                                         255)
1119                                 return -1;
1120                 }
1121                 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
1122                         printf("exceeded max number of lcore params: %hu\n",
1123                                 nb_lcore_params);
1124                         return -1;
1125                 }
1126                 lcore_params_array[nb_lcore_params].port_id =
1127                                 (uint8_t)int_fld[FLD_PORT];
1128                 lcore_params_array[nb_lcore_params].queue_id =
1129                                 (uint8_t)int_fld[FLD_QUEUE];
1130                 lcore_params_array[nb_lcore_params].lcore_id =
1131                                 (uint8_t)int_fld[FLD_LCORE];
1132                 ++nb_lcore_params;
1133         }
1134         lcore_params = lcore_params_array;
1135
1136         return 0;
1137 }
1138
1139 /* Parse the argument given in the command line of the application */
1140 static int
1141 parse_args(int argc, char **argv)
1142 {
1143         int opt, ret;
1144         char **argvopt;
1145         int option_index;
1146         char *prgname = argv[0];
1147         static struct option lgopts[] = {
1148                 {"config", 1, 0, 0},
1149                 {"no-numa", 0, 0, 0},
1150                 {"enable-jumbo", 0, 0, 0},
1151                 {NULL, 0, 0, 0}
1152         };
1153
1154         argvopt = argv;
1155
1156         while ((opt = getopt_long(argc, argvopt, "p:P",
1157                                 lgopts, &option_index)) != EOF) {
1158
1159                 switch (opt) {
1160                 /* portmask */
1161                 case 'p':
1162                         enabled_port_mask = parse_portmask(optarg);
1163                         if (enabled_port_mask == 0) {
1164                                 printf("invalid portmask\n");
1165                                 print_usage(prgname);
1166                                 return -1;
1167                         }
1168                         break;
1169                 case 'P':
1170                         printf("Promiscuous mode selected\n");
1171                         promiscuous_on = 1;
1172                         break;
1173
1174                 /* long options */
1175                 case 0:
1176                         if (!strncmp(lgopts[option_index].name, "config", 6)) {
1177                                 ret = parse_config(optarg);
1178                                 if (ret) {
1179                                         printf("invalid config\n");
1180                                         print_usage(prgname);
1181                                         return -1;
1182                                 }
1183                         }
1184
1185                         if (!strncmp(lgopts[option_index].name,
1186                                                 "no-numa", 7)) {
1187                                 printf("numa is disabled \n");
1188                                 numa_on = 0;
1189                         }
1190
1191                         if (!strncmp(lgopts[option_index].name,
1192                                         "enable-jumbo", 12)) {
1193                                 struct option lenopts =
1194                                         {"max-pkt-len", required_argument, \
1195                                                                         0, 0};
1196
1197                                 printf("jumbo frame is enabled \n");
1198                                 port_conf.rxmode.jumbo_frame = 1;
1199
1200                                 /**
1201                                  * if no max-pkt-len set, use the default value
1202                                  * ETHER_MAX_LEN
1203                                  */
1204                                 if (0 == getopt_long(argc, argvopt, "",
1205                                                 &lenopts, &option_index)) {
1206                                         ret = parse_max_pkt_len(optarg);
1207                                         if ((ret < 64) ||
1208                                                 (ret > MAX_JUMBO_PKT_LEN)){
1209                                                 printf("invalid packet "
1210                                                                 "length\n");
1211                                                 print_usage(prgname);
1212                                                 return -1;
1213                                         }
1214                                         port_conf.rxmode.max_rx_pkt_len = ret;
1215                                 }
1216                                 printf("set jumbo frame "
1217                                         "max packet length to %u\n",
1218                                 (unsigned int)port_conf.rxmode.max_rx_pkt_len);
1219                         }
1220
1221                         break;
1222
1223                 default:
1224                         print_usage(prgname);
1225                         return -1;
1226                 }
1227         }
1228
1229         if (optind >= 0)
1230                 argv[optind-1] = prgname;
1231
1232         ret = optind-1;
1233         optind = 0; /* reset getopt lib */
1234         return ret;
1235 }
1236
1237 static void
1238 print_ethaddr(const char *name, const struct ether_addr *eth_addr)
1239 {
1240         char buf[ETHER_ADDR_FMT_SIZE];
1241         ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr);
1242         printf("%s%s", name, buf);
1243 }
1244
1245 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1246 static void
1247 setup_hash(int socketid)
1248 {
1249         struct rte_hash_parameters ipv4_l3fwd_hash_params = {
1250                 .name = NULL,
1251                 .entries = L3FWD_HASH_ENTRIES,
1252                 .bucket_entries = 4,
1253                 .key_len = sizeof(struct ipv4_5tuple),
1254                 .hash_func = DEFAULT_HASH_FUNC,
1255                 .hash_func_init_val = 0,
1256         };
1257
1258         struct rte_hash_parameters ipv6_l3fwd_hash_params = {
1259                 .name = NULL,
1260                 .entries = L3FWD_HASH_ENTRIES,
1261                 .bucket_entries = 4,
1262                 .key_len = sizeof(struct ipv6_5tuple),
1263                 .hash_func = DEFAULT_HASH_FUNC,
1264                 .hash_func_init_val = 0,
1265         };
1266
1267         unsigned i;
1268         int ret;
1269         char s[64];
1270
1271         /* create ipv4 hash */
1272         snprintf(s, sizeof(s), "ipv4_l3fwd_hash_%d", socketid);
1273         ipv4_l3fwd_hash_params.name = s;
1274         ipv4_l3fwd_hash_params.socket_id = socketid;
1275         ipv4_l3fwd_lookup_struct[socketid] =
1276                 rte_hash_create(&ipv4_l3fwd_hash_params);
1277         if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
1278                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
1279                                 "socket %d\n", socketid);
1280
1281         /* create ipv6 hash */
1282         snprintf(s, sizeof(s), "ipv6_l3fwd_hash_%d", socketid);
1283         ipv6_l3fwd_hash_params.name = s;
1284         ipv6_l3fwd_hash_params.socket_id = socketid;
1285         ipv6_l3fwd_lookup_struct[socketid] =
1286                 rte_hash_create(&ipv6_l3fwd_hash_params);
1287         if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
1288                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
1289                                 "socket %d\n", socketid);
1290
1291
1292         /* populate the ipv4 hash */
1293         for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
1294                 ret = rte_hash_add_key (ipv4_l3fwd_lookup_struct[socketid],
1295                                 (void *) &ipv4_l3fwd_route_array[i].key);
1296                 if (ret < 0) {
1297                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
1298                                 "l3fwd hash on socket %d\n", i, socketid);
1299                 }
1300                 ipv4_l3fwd_out_if[ret] = ipv4_l3fwd_route_array[i].if_out;
1301                 printf("Hash: Adding key\n");
1302                 print_ipv4_key(ipv4_l3fwd_route_array[i].key);
1303         }
1304
1305         /* populate the ipv6 hash */
1306         for (i = 0; i < IPV6_L3FWD_NUM_ROUTES; i++) {
1307                 ret = rte_hash_add_key (ipv6_l3fwd_lookup_struct[socketid],
1308                                 (void *) &ipv6_l3fwd_route_array[i].key);
1309                 if (ret < 0) {
1310                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
1311                                 "l3fwd hash on socket %d\n", i, socketid);
1312                 }
1313                 ipv6_l3fwd_out_if[ret] = ipv6_l3fwd_route_array[i].if_out;
1314                 printf("Hash: Adding key\n");
1315                 print_ipv6_key(ipv6_l3fwd_route_array[i].key);
1316         }
1317 }
1318 #endif
1319
1320 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1321 static void
1322 setup_lpm(int socketid)
1323 {
1324         unsigned i;
1325         int ret;
1326         char s[64];
1327
1328         /* create the LPM table */
1329         snprintf(s, sizeof(s), "IPV4_L3FWD_LPM_%d", socketid);
1330         ipv4_l3fwd_lookup_struct[socketid] = rte_lpm_create(s, socketid,
1331                                 IPV4_L3FWD_LPM_MAX_RULES, 0);
1332         if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
1333                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
1334                                 " on socket %d\n", socketid);
1335
1336         /* populate the LPM table */
1337         for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
1338                 ret = rte_lpm_add(ipv4_l3fwd_lookup_struct[socketid],
1339                         ipv4_l3fwd_route_array[i].ip,
1340                         ipv4_l3fwd_route_array[i].depth,
1341                         ipv4_l3fwd_route_array[i].if_out);
1342
1343                 if (ret < 0) {
1344                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
1345                                 "l3fwd LPM table on socket %d\n",
1346                                 i, socketid);
1347                 }
1348
1349                 printf("LPM: Adding route 0x%08x / %d (%d)\n",
1350                         (unsigned)ipv4_l3fwd_route_array[i].ip,
1351                         ipv4_l3fwd_route_array[i].depth,
1352                         ipv4_l3fwd_route_array[i].if_out);
1353         }
1354 }
1355 #endif
1356
1357 static int
1358 init_mem(unsigned nb_mbuf)
1359 {
1360         struct lcore_conf *qconf;
1361         int socketid;
1362         unsigned lcore_id;
1363         char s[64];
1364
1365         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1366                 if (rte_lcore_is_enabled(lcore_id) == 0)
1367                         continue;
1368
1369                 if (numa_on)
1370                         socketid = rte_lcore_to_socket_id(lcore_id);
1371                 else
1372                         socketid = 0;
1373
1374                 if (socketid >= NB_SOCKETS) {
1375                         rte_exit(EXIT_FAILURE, "Socket %d of lcore %u is "
1376                                         "out of range %d\n", socketid,
1377                                                 lcore_id, NB_SOCKETS);
1378                 }
1379                 if (pktmbuf_pool[socketid] == NULL) {
1380                         snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
1381                         pktmbuf_pool[socketid] =
1382                                 rte_pktmbuf_pool_create(s, nb_mbuf,
1383                                         MEMPOOL_CACHE_SIZE, 0, MBUF_DATA_SIZE,
1384                                         socketid);
1385                         if (pktmbuf_pool[socketid] == NULL)
1386                                 rte_exit(EXIT_FAILURE,
1387                                         "Cannot init mbuf pool on socket %d\n",
1388                                                                 socketid);
1389                         else
1390                                 printf("Allocated mbuf pool on socket %d\n",
1391                                                                 socketid);
1392
1393 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1394                         setup_lpm(socketid);
1395 #else
1396                         setup_hash(socketid);
1397 #endif
1398                 }
1399                 qconf = &lcore_conf[lcore_id];
1400                 qconf->ipv4_lookup_struct = ipv4_l3fwd_lookup_struct[socketid];
1401 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1402                 qconf->ipv6_lookup_struct = ipv6_l3fwd_lookup_struct[socketid];
1403 #endif
1404         }
1405         return 0;
1406 }
1407
1408 /* Check the link status of all ports in up to 9s, and print them finally */
1409 static void
1410 check_all_ports_link_status(uint8_t port_num, uint32_t port_mask)
1411 {
1412 #define CHECK_INTERVAL 100 /* 100ms */
1413 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
1414         uint8_t portid, count, all_ports_up, print_flag = 0;
1415         struct rte_eth_link link;
1416
1417         printf("\nChecking link status");
1418         fflush(stdout);
1419         for (count = 0; count <= MAX_CHECK_TIME; count++) {
1420                 all_ports_up = 1;
1421                 for (portid = 0; portid < port_num; portid++) {
1422                         if ((port_mask & (1 << portid)) == 0)
1423                                 continue;
1424                         memset(&link, 0, sizeof(link));
1425                         rte_eth_link_get_nowait(portid, &link);
1426                         /* print link status if flag set */
1427                         if (print_flag == 1) {
1428                                 if (link.link_status)
1429                                         printf("Port %d Link Up - speed %u "
1430                                                 "Mbps - %s\n", (uint8_t)portid,
1431                                                 (unsigned)link.link_speed,
1432                                 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
1433                                         ("full-duplex") : ("half-duplex\n"));
1434                                 else
1435                                         printf("Port %d Link Down\n",
1436                                                 (uint8_t)portid);
1437                                 continue;
1438                         }
1439                         /* clear all_ports_up flag if any link down */
1440                         if (link.link_status == 0) {
1441                                 all_ports_up = 0;
1442                                 break;
1443                         }
1444                 }
1445                 /* after finally printing all link status, get out */
1446                 if (print_flag == 1)
1447                         break;
1448
1449                 if (all_ports_up == 0) {
1450                         printf(".");
1451                         fflush(stdout);
1452                         rte_delay_ms(CHECK_INTERVAL);
1453                 }
1454
1455                 /* set the print_flag if all ports up or timeout */
1456                 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
1457                         print_flag = 1;
1458                         printf("done\n");
1459                 }
1460         }
1461 }
1462
1463 int
1464 main(int argc, char **argv)
1465 {
1466         struct lcore_conf *qconf;
1467         struct rte_eth_dev_info dev_info;
1468         struct rte_eth_txconf *txconf;
1469         int ret;
1470         unsigned nb_ports;
1471         uint16_t queueid;
1472         unsigned lcore_id;
1473         uint64_t hz;
1474         uint32_t n_tx_queue, nb_lcores;
1475         uint8_t portid, nb_rx_queue, queue, socketid;
1476
1477         /* catch SIGINT and restore cpufreq governor to ondemand */
1478         signal(SIGINT, signal_exit_now);
1479
1480         /* init EAL */
1481         ret = rte_eal_init(argc, argv);
1482         if (ret < 0)
1483                 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
1484         argc -= ret;
1485         argv += ret;
1486
1487         /* init RTE timer library to be used late */
1488         rte_timer_subsystem_init();
1489
1490         /* parse application arguments (after the EAL ones) */
1491         ret = parse_args(argc, argv);
1492         if (ret < 0)
1493                 rte_exit(EXIT_FAILURE, "Invalid L3FWD parameters\n");
1494
1495         if (check_lcore_params() < 0)
1496                 rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
1497
1498         ret = init_lcore_rx_queues();
1499         if (ret < 0)
1500                 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
1501
1502
1503         nb_ports = rte_eth_dev_count();
1504         if (nb_ports > RTE_MAX_ETHPORTS)
1505                 nb_ports = RTE_MAX_ETHPORTS;
1506
1507         if (check_port_config(nb_ports) < 0)
1508                 rte_exit(EXIT_FAILURE, "check_port_config failed\n");
1509
1510         nb_lcores = rte_lcore_count();
1511
1512         /* initialize all ports */
1513         for (portid = 0; portid < nb_ports; portid++) {
1514                 /* skip ports that are not enabled */
1515                 if ((enabled_port_mask & (1 << portid)) == 0) {
1516                         printf("\nSkipping disabled port %d\n", portid);
1517                         continue;
1518                 }
1519
1520                 /* init port */
1521                 printf("Initializing port %d ... ", portid );
1522                 fflush(stdout);
1523
1524                 nb_rx_queue = get_port_n_rx_queues(portid);
1525                 n_tx_queue = nb_lcores;
1526                 if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
1527                         n_tx_queue = MAX_TX_QUEUE_PER_PORT;
1528                 printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
1529                         nb_rx_queue, (unsigned)n_tx_queue );
1530                 ret = rte_eth_dev_configure(portid, nb_rx_queue,
1531                                         (uint16_t)n_tx_queue, &port_conf);
1532                 if (ret < 0)
1533                         rte_exit(EXIT_FAILURE, "Cannot configure device: "
1534                                         "err=%d, port=%d\n", ret, portid);
1535
1536                 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
1537                 print_ethaddr(" Address:", &ports_eth_addr[portid]);
1538                 printf(", ");
1539
1540                 /* init memory */
1541                 ret = init_mem(NB_MBUF);
1542                 if (ret < 0)
1543                         rte_exit(EXIT_FAILURE, "init_mem failed\n");
1544
1545                 /* init one TX queue per couple (lcore,port) */
1546                 queueid = 0;
1547                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1548                         if (rte_lcore_is_enabled(lcore_id) == 0)
1549                                 continue;
1550
1551                         if (numa_on)
1552                                 socketid = \
1553                                 (uint8_t)rte_lcore_to_socket_id(lcore_id);
1554                         else
1555                                 socketid = 0;
1556
1557                         printf("txq=%u,%d,%d ", lcore_id, queueid, socketid);
1558                         fflush(stdout);
1559
1560                         rte_eth_dev_info_get(portid, &dev_info);
1561                         txconf = &dev_info.default_txconf;
1562                         if (port_conf.rxmode.jumbo_frame)
1563                                 txconf->txq_flags = 0;
1564                         ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
1565                                                      socketid, txconf);
1566                         if (ret < 0)
1567                                 rte_exit(EXIT_FAILURE,
1568                                         "rte_eth_tx_queue_setup: err=%d, "
1569                                                 "port=%d\n", ret, portid);
1570
1571                         qconf = &lcore_conf[lcore_id];
1572                         qconf->tx_queue_id[portid] = queueid;
1573                         queueid++;
1574                 }
1575                 printf("\n");
1576         }
1577
1578         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1579                 if (rte_lcore_is_enabled(lcore_id) == 0)
1580                         continue;
1581
1582                 /* init power management library */
1583                 ret = rte_power_init(lcore_id);
1584                 if (ret)
1585                         rte_exit(EXIT_FAILURE, "Power management library "
1586                                 "initialization failed on core%u\n", lcore_id);
1587
1588                 /* init timer structures for each enabled lcore */
1589                 rte_timer_init(&power_timers[lcore_id]);
1590                 hz = rte_get_timer_hz();
1591                 rte_timer_reset(&power_timers[lcore_id],
1592                         hz/TIMER_NUMBER_PER_SECOND, SINGLE, lcore_id,
1593                                                 power_timer_cb, NULL);
1594
1595                 qconf = &lcore_conf[lcore_id];
1596                 printf("\nInitializing rx queues on lcore %u ... ", lcore_id );
1597                 fflush(stdout);
1598                 /* init RX queues */
1599                 for(queue = 0; queue < qconf->n_rx_queue; ++queue) {
1600                         portid = qconf->rx_queue_list[queue].port_id;
1601                         queueid = qconf->rx_queue_list[queue].queue_id;
1602
1603                         if (numa_on)
1604                                 socketid = \
1605                                 (uint8_t)rte_lcore_to_socket_id(lcore_id);
1606                         else
1607                                 socketid = 0;
1608
1609                         printf("rxq=%d,%d,%d ", portid, queueid, socketid);
1610                         fflush(stdout);
1611
1612                         ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
1613                                 socketid, NULL,
1614                                 pktmbuf_pool[socketid]);
1615                         if (ret < 0)
1616                                 rte_exit(EXIT_FAILURE,
1617                                         "rte_eth_rx_queue_setup: err=%d, "
1618                                                 "port=%d\n", ret, portid);
1619                 }
1620         }
1621
1622         printf("\n");
1623
1624         /* start ports */
1625         for (portid = 0; portid < nb_ports; portid++) {
1626                 if ((enabled_port_mask & (1 << portid)) == 0) {
1627                         continue;
1628                 }
1629                 /* Start device */
1630                 ret = rte_eth_dev_start(portid);
1631                 if (ret < 0)
1632                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, "
1633                                                 "port=%d\n", ret, portid);
1634
1635                 /*
1636                  * If enabled, put device in promiscuous mode.
1637                  * This allows IO forwarding mode to forward packets
1638                  * to itself through 2 cross-connected  ports of the
1639                  * target machine.
1640                  */
1641                 if (promiscuous_on)
1642                         rte_eth_promiscuous_enable(portid);
1643         }
1644
1645         check_all_ports_link_status((uint8_t)nb_ports, enabled_port_mask);
1646
1647         /* launch per-lcore init on every lcore */
1648         rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
1649         RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1650                 if (rte_eal_wait_lcore(lcore_id) < 0)
1651                         return -1;
1652         }
1653
1654         return 0;
1655 }