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 /*
 125  * This expression is used to calculate the number of mbufs needed depending on
 126  * user input, taking into account memory for rx and tx hardware rings, cache
 127  * per lcore and mtable per port per lcore. RTE_MAX is used to ensure that
 128  * NB_MBUF never goes below a minimum value of 8192.
 129  */
 130
 131 #define NB_MBUF RTE_MAX ( \
 132         (nb_ports*nb_rx_queue*RTE_TEST_RX_DESC_DEFAULT + \
 133         nb_ports*nb_lcores*MAX_PKT_BURST + \
 134         nb_ports*n_tx_queue*RTE_TEST_TX_DESC_DEFAULT + \
 135         nb_lcores*MEMPOOL_CACHE_SIZE), \
 136         (unsigned)8192)
 137
 138 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
 139
 140 #define NB_SOCKETS 8
 141
 142 /* Configure how many packets ahead to prefetch, when reading packets */
 143 #define PREFETCH_OFFSET 3
 144
 145 /*
 146  * Configurable number of RX/TX ring descriptors
 147  */
 148 #define RTE_TEST_RX_DESC_DEFAULT 128
 149 #define RTE_TEST_TX_DESC_DEFAULT 512
 150 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
 151 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
 152
 153 /* ethernet addresses of ports */
 154 static struct ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
 155
 156 /* mask of enabled ports */
 157 static uint32_t enabled_port_mask = 0;
 158 /* Ports set in promiscuous mode off by default. */
 159 static int promiscuous_on = 0;
 160 /* NUMA is enabled by default. */
 161 static int numa_on = 1;
 162
 163 enum freq_scale_hint_t
 164 {
 165         FREQ_LOWER    =      -1,
 166         FREQ_CURRENT  =       0,
 167         FREQ_HIGHER   =       1,
 168         FREQ_HIGHEST  =       2
 169 };
 170
 171 struct mbuf_table {
 172         uint16_t len;
 173         struct rte_mbuf *m_table[MAX_PKT_BURST];
 174 };
 175
 176 struct lcore_rx_queue {
 177         uint8_t port_id;
 178         uint8_t queue_id;
 179         enum freq_scale_hint_t freq_up_hint;
 180         uint32_t zero_rx_packet_count;
 181         uint32_t idle_hint;
 182 } __rte_cache_aligned;
 183
 184 #define MAX_RX_QUEUE_PER_LCORE 16
 185 #define MAX_TX_QUEUE_PER_PORT RTE_MAX_ETHPORTS
 186 #define MAX_RX_QUEUE_PER_PORT 128
 187
 188 #define MAX_LCORE_PARAMS 1024
 189 struct lcore_params {
 190         uint8_t port_id;
 191         uint8_t queue_id;
 192         uint8_t lcore_id;
 193 } __rte_cache_aligned;
 194
 195 static struct lcore_params lcore_params_array[MAX_LCORE_PARAMS];
 196 static struct lcore_params lcore_params_array_default[] = {
 197         {0, 0, 2},
 198         {0, 1, 2},
 199         {0, 2, 2},
 200         {1, 0, 2},
 201         {1, 1, 2},
 202         {1, 2, 2},
 203         {2, 0, 2},
 204         {3, 0, 3},
 205         {3, 1, 3},
 206 };
 207
 208 static struct lcore_params * lcore_params = lcore_params_array_default;
 209 static uint16_t nb_lcore_params = sizeof(lcore_params_array_default) /
 210                                 sizeof(lcore_params_array_default[0]);
 211
 212 static struct rte_eth_conf port_conf = {
 213         .rxmode = {
 214                 .mq_mode        = ETH_MQ_RX_RSS,
 215                 .max_rx_pkt_len = ETHER_MAX_LEN,
 216                 .split_hdr_size = 0,
 217                 .header_split   = 0, /**< Header Split disabled */
 218                 .hw_ip_checksum = 1, /**< IP checksum offload enabled */
 219                 .hw_vlan_filter = 0, /**< VLAN filtering disabled */
 220                 .jumbo_frame    = 0, /**< Jumbo Frame Support disabled */
 221                 .hw_strip_crc   = 0, /**< CRC stripped by hardware */
 222         },
 223         .rx_adv_conf = {
 224                 .rss_conf = {
 225                         .rss_key = NULL,
 226                         .rss_hf = ETH_RSS_IP,
 227                 },
 228         },
 229         .txmode = {
 230                 .mq_mode = ETH_DCB_NONE,
 231         },
 232 };
 233
 234 static struct rte_mempool * pktmbuf_pool[NB_SOCKETS];
 235
 236
 237 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 238
 239 #ifdef RTE_MACHINE_CPUFLAG_SSE4_2
 240 #include <rte_hash_crc.h>
 241 #define DEFAULT_HASH_FUNC       rte_hash_crc
 242 #else
 243 #include <rte_jhash.h>
 244 #define DEFAULT_HASH_FUNC       rte_jhash
 245 #endif
 246
 247 struct ipv4_5tuple {
 248         uint32_t ip_dst;
 249         uint32_t ip_src;
 250         uint16_t port_dst;
 251         uint16_t port_src;
 252         uint8_t  proto;
 253 } __attribute__((__packed__));
 254
 255 struct ipv6_5tuple {
 256         uint8_t  ip_dst[IPV6_ADDR_LEN];
 257         uint8_t  ip_src[IPV6_ADDR_LEN];
 258         uint16_t port_dst;
 259         uint16_t port_src;
 260         uint8_t  proto;
 261 } __attribute__((__packed__));
 262
 263 struct ipv4_l3fwd_route {
 264         struct ipv4_5tuple key;
 265         uint8_t if_out;
 266 };
 267
 268 struct ipv6_l3fwd_route {
 269         struct ipv6_5tuple key;
 270         uint8_t if_out;
 271 };
 272
 273 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
 274         {{IPv4(100,10,0,1), IPv4(200,10,0,1), 101, 11, IPPROTO_TCP}, 0},
 275         {{IPv4(100,20,0,2), IPv4(200,20,0,2), 102, 12, IPPROTO_TCP}, 1},
 276         {{IPv4(100,30,0,3), IPv4(200,30,0,3), 103, 13, IPPROTO_TCP}, 2},
 277         {{IPv4(100,40,0,4), IPv4(200,40,0,4), 104, 14, IPPROTO_TCP}, 3},
 278 };
 279
 280 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
 281         {
 282                 {
 283                         {0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
 284                          0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
 285                         {0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
 286                          0x02, 0x1e, 0x67, 0xff, 0xfe, 0x0d, 0xb6, 0x0a},
 287                          1, 10, IPPROTO_UDP
 288                 }, 4
 289         },
 290 };
 291
 292 typedef struct rte_hash lookup_struct_t;
 293 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
 294 static lookup_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
 295
 296 #define L3FWD_HASH_ENTRIES      1024
 297
 298 #define IPV4_L3FWD_NUM_ROUTES \
 299         (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
 300
 301 #define IPV6_L3FWD_NUM_ROUTES \
 302         (sizeof(ipv6_l3fwd_route_array) / sizeof(ipv6_l3fwd_route_array[0]))
 303
 304 static uint8_t ipv4_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 305 static uint8_t ipv6_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 306 #endif
 307
 308 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 309 struct ipv4_l3fwd_route {
 310         uint32_t ip;
 311         uint8_t  depth;
 312         uint8_t  if_out;
 313 };
 314
 315 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
 316         {IPv4(1,1,1,0), 24, 0},
 317         {IPv4(2,1,1,0), 24, 1},
 318         {IPv4(3,1,1,0), 24, 2},
 319         {IPv4(4,1,1,0), 24, 3},
 320         {IPv4(5,1,1,0), 24, 4},
 321         {IPv4(6,1,1,0), 24, 5},
 322         {IPv4(7,1,1,0), 24, 6},
 323         {IPv4(8,1,1,0), 24, 7},
 324 };
 325
 326 #define IPV4_L3FWD_NUM_ROUTES \
 327         (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
 328
 329 #define IPV4_L3FWD_LPM_MAX_RULES     1024
 330
 331 typedef struct rte_lpm lookup_struct_t;
 332 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
 333 #endif
 334
 335 struct lcore_conf {
 336         uint16_t n_rx_queue;
 337         struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
 338         uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
 339         struct mbuf_table tx_mbufs[RTE_MAX_ETHPORTS];
 340         lookup_struct_t * ipv4_lookup_struct;
 341         lookup_struct_t * ipv6_lookup_struct;
 342 } __rte_cache_aligned;
 343
 344 struct lcore_stats {
 345         /* total sleep time in ms since last frequency scaling down */
 346         uint32_t sleep_time;
 347         /* number of long sleep recently */
 348         uint32_t nb_long_sleep;
 349         /* freq. scaling up trend */
 350         uint32_t trend;
 351         /* total packet processed recently */
 352         uint64_t nb_rx_processed;
 353         /* total iterations looped recently */
 354         uint64_t nb_iteration_looped;
 355         uint32_t padding[9];
 356 } __rte_cache_aligned;
 357
 358 static struct lcore_conf lcore_conf[RTE_MAX_LCORE] __rte_cache_aligned;
 359 static struct lcore_stats stats[RTE_MAX_LCORE] __rte_cache_aligned;
 360 static struct rte_timer power_timers[RTE_MAX_LCORE];
 361
 362 static inline uint32_t power_idle_heuristic(uint32_t zero_rx_packet_count);
 363 static inline enum freq_scale_hint_t power_freq_scaleup_heuristic( \
 364                         unsigned lcore_id, uint8_t port_id, uint16_t queue_id);
 365
 366 /* exit signal handler */
 367 static void
 368 signal_exit_now(int sigtype)
 369 {
 370         unsigned lcore_id;
 371         int ret;
 372
 373         if (sigtype == SIGINT) {
 374                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
 375                         if (rte_lcore_is_enabled(lcore_id) == 0)
 376                                 continue;
 377
 378                         /* init power management library */
 379                         ret = rte_power_exit(lcore_id);
 380                         if (ret)
 381                                 rte_exit(EXIT_FAILURE, "Power management "
 382                                         "library de-initialization failed on "
 383                                                         "core%u\n", lcore_id);
 384                 }
 385         }
 386
 387         rte_exit(EXIT_SUCCESS, "User forced exit\n");
 388 }
 389
 390 /*  Freqency scale down timer callback */
 391 static void
 392 power_timer_cb(__attribute__((unused)) struct rte_timer *tim,
 393                           __attribute__((unused)) void *arg)
 394 {
 395         uint64_t hz;
 396         float sleep_time_ratio;
 397         unsigned lcore_id = rte_lcore_id();
 398
 399         /* accumulate total execution time in us when callback is invoked */
 400         sleep_time_ratio = (float)(stats[lcore_id].sleep_time) /
 401                                         (float)SCALING_PERIOD;
 402
 403         /**
 404          * check whether need to scale down frequency a step if it sleep a lot.
 405          */
 406         if (sleep_time_ratio >= SCALING_DOWN_TIME_RATIO_THRESHOLD)
 407                 rte_power_freq_down(lcore_id);
 408         else if ( (unsigned)(stats[lcore_id].nb_rx_processed /
 409                 stats[lcore_id].nb_iteration_looped) < MAX_PKT_BURST)
 410                 /**
 411                  * scale down a step if average packet per iteration less
 412                  * than expectation.
 413                  */
 414                 rte_power_freq_down(lcore_id);
 415
 416         /**
 417          * initialize another timer according to current frequency to ensure
 418          * timer interval is relatively fixed.
 419          */
 420         hz = rte_get_timer_hz();
 421         rte_timer_reset(&power_timers[lcore_id], hz/TIMER_NUMBER_PER_SECOND,
 422                                 SINGLE, lcore_id, power_timer_cb, NULL);
 423
 424         stats[lcore_id].nb_rx_processed = 0;
 425         stats[lcore_id].nb_iteration_looped = 0;
 426
 427         stats[lcore_id].sleep_time = 0;
 428 }
 429
 430 /* Send burst of packets on an output interface */
 431 static inline int
 432 send_burst(struct lcore_conf *qconf, uint16_t n, uint8_t port)
 433 {
 434         struct rte_mbuf **m_table;
 435         int ret;
 436         uint16_t queueid;
 437
 438         queueid = qconf->tx_queue_id[port];
 439         m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
 440
 441         ret = rte_eth_tx_burst(port, queueid, m_table, n);
 442         if (unlikely(ret < n)) {
 443                 do {
 444                         rte_pktmbuf_free(m_table[ret]);
 445                 } while (++ret < n);
 446         }
 447
 448         return 0;
 449 }
 450
 451 /* Enqueue a single packet, and send burst if queue is filled */
 452 static inline int
 453 send_single_packet(struct rte_mbuf *m, uint8_t port)
 454 {
 455         uint32_t lcore_id;
 456         uint16_t len;
 457         struct lcore_conf *qconf;
 458
 459         lcore_id = rte_lcore_id();
 460
 461         qconf = &lcore_conf[lcore_id];
 462         len = qconf->tx_mbufs[port].len;
 463         qconf->tx_mbufs[port].m_table[len] = m;
 464         len++;
 465
 466         /* enough pkts to be sent */
 467         if (unlikely(len == MAX_PKT_BURST)) {
 468                 send_burst(qconf, MAX_PKT_BURST, port);
 469                 len = 0;
 470         }
 471
 472         qconf->tx_mbufs[port].len = len;
 473         return 0;
 474 }
 475
 476 #ifdef DO_RFC_1812_CHECKS
 477 static inline int
 478 is_valid_ipv4_pkt(struct ipv4_hdr *pkt, uint32_t link_len)
 479 {
 480         /* From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2 */
 481         /*
 482          * 1. The packet length reported by the Link Layer must be large
 483          * enough to hold the minimum length legal IP datagram (20 bytes).
 484          */
 485         if (link_len < sizeof(struct ipv4_hdr))
 486                 return -1;
 487
 488         /* 2. The IP checksum must be correct. */
 489         /* this is checked in H/W */
 490
 491         /*
 492          * 3. The IP version number must be 4. If the version number is not 4
 493          * then the packet may be another version of IP, such as IPng or
 494          * ST-II.
 495          */
 496         if (((pkt->version_ihl) >> 4) != 4)
 497                 return -3;
 498         /*
 499          * 4. The IP header length field must be large enough to hold the
 500          * minimum length legal IP datagram (20 bytes = 5 words).
 501          */
 502         if ((pkt->version_ihl & 0xf) < 5)
 503                 return -4;
 504
 505         /*
 506          * 5. The IP total length field must be large enough to hold the IP
 507          * datagram header, whose length is specified in the IP header length
 508          * field.
 509          */
 510         if (rte_cpu_to_be_16(pkt->total_length) < sizeof(struct ipv4_hdr))
 511                 return -5;
 512
 513         return 0;
 514 }
 515 #endif
 516
 517 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 518 static void
 519 print_ipv4_key(struct ipv4_5tuple key)
 520 {
 521         printf("IP dst = %08x, IP src = %08x, port dst = %d, port src = %d, "
 522                 "proto = %d\n", (unsigned)key.ip_dst, (unsigned)key.ip_src,
 523                                 key.port_dst, key.port_src, key.proto);
 524 }
 525 static void
 526 print_ipv6_key(struct ipv6_5tuple key)
 527 {
 528         printf( "IP dst = " IPv6_BYTES_FMT ", IP src = " IPv6_BYTES_FMT ", "
 529                 "port dst = %d, port src = %d, proto = %d\n",
 530                 IPv6_BYTES(key.ip_dst), IPv6_BYTES(key.ip_src),
 531                 key.port_dst, key.port_src, key.proto);
 532 }
 533
 534 static inline uint8_t
 535 get_ipv4_dst_port(struct ipv4_hdr *ipv4_hdr, uint8_t portid,
 536                 lookup_struct_t * ipv4_l3fwd_lookup_struct)
 537 {
 538         struct ipv4_5tuple key;
 539         struct tcp_hdr *tcp;
 540         struct udp_hdr *udp;
 541         int ret = 0;
 542
 543         key.ip_dst = rte_be_to_cpu_32(ipv4_hdr->dst_addr);
 544         key.ip_src = rte_be_to_cpu_32(ipv4_hdr->src_addr);
 545         key.proto = ipv4_hdr->next_proto_id;
 546
 547         switch (ipv4_hdr->next_proto_id) {
 548         case IPPROTO_TCP:
 549                 tcp = (struct tcp_hdr *)((unsigned char *)ipv4_hdr +
 550                                         sizeof(struct ipv4_hdr));
 551                 key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
 552                 key.port_src = rte_be_to_cpu_16(tcp->src_port);
 553                 break;
 554
 555         case IPPROTO_UDP:
 556                 udp = (struct udp_hdr *)((unsigned char *)ipv4_hdr +
 557                                         sizeof(struct ipv4_hdr));
 558                 key.port_dst = rte_be_to_cpu_16(udp->dst_port);
 559                 key.port_src = rte_be_to_cpu_16(udp->src_port);
 560                 break;
 561
 562         default:
 563                 key.port_dst = 0;
 564                 key.port_src = 0;
 565                 break;
 566         }
 567
 568         /* Find destination port */
 569         ret = rte_hash_lookup(ipv4_l3fwd_lookup_struct, (const void *)&key);
 570         return (uint8_t)((ret < 0)? portid : ipv4_l3fwd_out_if[ret]);
 571 }
 572
 573 static inline uint8_t
 574 get_ipv6_dst_port(struct ipv6_hdr *ipv6_hdr,  uint8_t portid,
 575                         lookup_struct_t *ipv6_l3fwd_lookup_struct)
 576 {
 577         struct ipv6_5tuple key;
 578         struct tcp_hdr *tcp;
 579         struct udp_hdr *udp;
 580         int ret = 0;
 581
 582         memcpy(key.ip_dst, ipv6_hdr->dst_addr, IPV6_ADDR_LEN);
 583         memcpy(key.ip_src, ipv6_hdr->src_addr, IPV6_ADDR_LEN);
 584
 585         key.proto = ipv6_hdr->proto;
 586
 587         switch (ipv6_hdr->proto) {
 588         case IPPROTO_TCP:
 589                 tcp = (struct tcp_hdr *)((unsigned char *) ipv6_hdr +
 590                                         sizeof(struct ipv6_hdr));
 591                 key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
 592                 key.port_src = rte_be_to_cpu_16(tcp->src_port);
 593                 break;
 594
 595         case IPPROTO_UDP:
 596                 udp = (struct udp_hdr *)((unsigned char *) ipv6_hdr +
 597                                         sizeof(struct ipv6_hdr));
 598                 key.port_dst = rte_be_to_cpu_16(udp->dst_port);
 599                 key.port_src = rte_be_to_cpu_16(udp->src_port);
 600                 break;
 601
 602         default:
 603                 key.port_dst = 0;
 604                 key.port_src = 0;
 605                 break;
 606         }
 607
 608         /* Find destination port */
 609         ret = rte_hash_lookup(ipv6_l3fwd_lookup_struct, (const void *)&key);
 610         return (uint8_t)((ret < 0)? portid : ipv6_l3fwd_out_if[ret]);
 611 }
 612 #endif
 613
 614 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 615 static inline uint8_t
 616 get_ipv4_dst_port(struct ipv4_hdr *ipv4_hdr, uint8_t portid,
 617                 lookup_struct_t *ipv4_l3fwd_lookup_struct)
 618 {
 619         uint8_t next_hop;
 620
 621         return (uint8_t) ((rte_lpm_lookup(ipv4_l3fwd_lookup_struct,
 622                         rte_be_to_cpu_32(ipv4_hdr->dst_addr), &next_hop) == 0)?
 623                         next_hop : portid);
 624 }
 625 #endif
 626
 627 static inline void
 628 l3fwd_simple_forward(struct rte_mbuf *m, uint8_t portid,
 629                                 struct lcore_conf *qconf)
 630 {
 631         struct ether_hdr *eth_hdr;
 632         struct ipv4_hdr *ipv4_hdr;
 633         void *d_addr_bytes;
 634         uint8_t dst_port;
 635
 636         eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
 637
 638 #ifdef RTE_NEXT_ABI
 639         if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) {
 640 #else
 641         if (m->ol_flags & PKT_RX_IPV4_HDR) {
 642 #endif
 643                 /* Handle IPv4 headers.*/
 644                 ipv4_hdr =
 645                         rte_pktmbuf_mtod_offset(m, struct ipv4_hdr *,
 646                                                 sizeof(struct ether_hdr));
 647
 648 #ifdef DO_RFC_1812_CHECKS
 649                 /* Check to make sure the packet is valid (RFC1812) */
 650                 if (is_valid_ipv4_pkt(ipv4_hdr, m->pkt_len) < 0) {
 651                         rte_pktmbuf_free(m);
 652                         return;
 653                 }
 654 #endif
 655
 656                 dst_port = get_ipv4_dst_port(ipv4_hdr, portid,
 657                                         qconf->ipv4_lookup_struct);
 658                 if (dst_port >= RTE_MAX_ETHPORTS ||
 659                                 (enabled_port_mask & 1 << dst_port) == 0)
 660                         dst_port = portid;
 661
 662                 /* 02:00:00:00:00:xx */
 663                 d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
 664                 *((uint64_t *)d_addr_bytes) =
 665                         0x000000000002 + ((uint64_t)dst_port << 40);
 666
 667 #ifdef DO_RFC_1812_CHECKS
 668                 /* Update time to live and header checksum */
 669                 --(ipv4_hdr->time_to_live);
 670                 ++(ipv4_hdr->hdr_checksum);
 671 #endif
 672
 673                 /* src addr */
 674                 ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
 675
 676                 send_single_packet(m, dst_port);
 677 #ifdef RTE_NEXT_ABI
 678         } else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) {
 679 #else
 680         }
 681         else {
 682 #endif
 683                 /* Handle IPv6 headers.*/
 684 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 685                 struct ipv6_hdr *ipv6_hdr;
 686
 687                 ipv6_hdr =
 688                         rte_pktmbuf_mtod_offset(m, struct ipv6_hdr *,
 689                                                 sizeof(struct ether_hdr));
 690
 691                 dst_port = get_ipv6_dst_port(ipv6_hdr, portid,
 692                                         qconf->ipv6_lookup_struct);
 693
 694                 if (dst_port >= RTE_MAX_ETHPORTS ||
 695                                 (enabled_port_mask & 1 << dst_port) == 0)
 696                         dst_port = portid;
 697
 698                 /* 02:00:00:00:00:xx */
 699                 d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
 700                 *((uint64_t *)d_addr_bytes) =
 701                         0x000000000002 + ((uint64_t)dst_port << 40);
 702
 703                 /* src addr */
 704                 ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
 705
 706                 send_single_packet(m, dst_port);
 707 #else
 708                 /* We don't currently handle IPv6 packets in LPM mode. */
 709                 rte_pktmbuf_free(m);
 710 #endif
 711         }
 712
 713 }
 714
 715 #define SLEEP_GEAR1_THRESHOLD            100
 716 #define SLEEP_GEAR2_THRESHOLD            1000
 717
 718 static inline uint32_t
 719 power_idle_heuristic(uint32_t zero_rx_packet_count)
 720 {
 721         /* If zero count is less than 100, use it as the sleep time in us */
 722         if (zero_rx_packet_count < SLEEP_GEAR1_THRESHOLD)
 723                 return zero_rx_packet_count;
 724         /* If zero count is less than 1000, sleep time should be 100 us */
 725         else if ((zero_rx_packet_count >= SLEEP_GEAR1_THRESHOLD) &&
 726                         (zero_rx_packet_count < SLEEP_GEAR2_THRESHOLD))
 727                 return SLEEP_GEAR1_THRESHOLD;
 728         /* If zero count is greater than 1000, sleep time should be 1000 us */
 729         else if (zero_rx_packet_count >= SLEEP_GEAR2_THRESHOLD)
 730                 return SLEEP_GEAR2_THRESHOLD;
 731
 732         return 0;
 733 }
 734
 735 static inline enum freq_scale_hint_t
 736 power_freq_scaleup_heuristic(unsigned lcore_id,
 737                              uint8_t port_id,
 738                              uint16_t queue_id)
 739 {
 740 /**
 741  * HW Rx queue size is 128 by default, Rx burst read at maximum 32 entries
 742  * per iteration
 743  */
 744 #define FREQ_GEAR1_RX_PACKET_THRESHOLD             MAX_PKT_BURST
 745 #define FREQ_GEAR2_RX_PACKET_THRESHOLD             (MAX_PKT_BURST*2)
 746 #define FREQ_GEAR3_RX_PACKET_THRESHOLD             (MAX_PKT_BURST*3)
 747 #define FREQ_UP_TREND1_ACC   1
 748 #define FREQ_UP_TREND2_ACC   100
 749 #define FREQ_UP_THRESHOLD    10000
 750
 751         if (likely(rte_eth_rx_descriptor_done(port_id, queue_id,
 752                         FREQ_GEAR3_RX_PACKET_THRESHOLD) > 0)) {
 753                 stats[lcore_id].trend = 0;
 754                 return FREQ_HIGHEST;
 755         } else if (likely(rte_eth_rx_descriptor_done(port_id, queue_id,
 756                         FREQ_GEAR2_RX_PACKET_THRESHOLD) > 0))
 757                 stats[lcore_id].trend += FREQ_UP_TREND2_ACC;
 758         else if (likely(rte_eth_rx_descriptor_done(port_id, queue_id,
 759                         FREQ_GEAR1_RX_PACKET_THRESHOLD) > 0))
 760                 stats[lcore_id].trend += FREQ_UP_TREND1_ACC;
 761
 762         if (likely(stats[lcore_id].trend > FREQ_UP_THRESHOLD)) {
 763                 stats[lcore_id].trend = 0;
 764                 return FREQ_HIGHER;
 765         }
 766
 767         return FREQ_CURRENT;
 768 }
 769
 770 /* main processing loop */
 771 static int
 772 main_loop(__attribute__((unused)) void *dummy)
 773 {
 774         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
 775         unsigned lcore_id;
 776         uint64_t prev_tsc, diff_tsc, cur_tsc;
 777         uint64_t prev_tsc_power = 0, cur_tsc_power, diff_tsc_power;
 778         int i, j, nb_rx;
 779         uint8_t portid, queueid;
 780         struct lcore_conf *qconf;
 781         struct lcore_rx_queue *rx_queue;
 782         enum freq_scale_hint_t lcore_scaleup_hint;
 783
 784         uint32_t lcore_rx_idle_count = 0;
 785         uint32_t lcore_idle_hint = 0;
 786
 787         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
 788
 789         prev_tsc = 0;
 790
 791         lcore_id = rte_lcore_id();
 792         qconf = &lcore_conf[lcore_id];
 793
 794         if (qconf->n_rx_queue == 0) {
 795                 RTE_LOG(INFO, L3FWD_POWER, "lcore %u has nothing to do\n", lcore_id);
 796                 return 0;
 797         }
 798
 799         RTE_LOG(INFO, L3FWD_POWER, "entering main loop on lcore %u\n", lcore_id);
 800
 801         for (i = 0; i < qconf->n_rx_queue; i++) {
 802
 803                 portid = qconf->rx_queue_list[i].port_id;
 804                 queueid = qconf->rx_queue_list[i].queue_id;
 805                 RTE_LOG(INFO, L3FWD_POWER, " -- lcoreid=%u portid=%hhu "
 806                         "rxqueueid=%hhu\n", lcore_id, portid, queueid);
 807         }
 808
 809         while (1) {
 810                 stats[lcore_id].nb_iteration_looped++;
 811
 812                 cur_tsc = rte_rdtsc();
 813                 cur_tsc_power = cur_tsc;
 814
 815                 /*
 816                  * TX burst queue drain
 817                  */
 818                 diff_tsc = cur_tsc - prev_tsc;
 819                 if (unlikely(diff_tsc > drain_tsc)) {
 820
 821                         /*
 822                          * This could be optimized (use queueid instead of
 823                          * portid), but it is not called so often
 824                          */
 825                         for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
 826                                 if (qconf->tx_mbufs[portid].len == 0)
 827                                         continue;
 828                                 send_burst(&lcore_conf[lcore_id],
 829                                         qconf->tx_mbufs[portid].len,
 830                                         portid);
 831                                 qconf->tx_mbufs[portid].len = 0;
 832                         }
 833
 834                         prev_tsc = cur_tsc;
 835                 }
 836
 837                 diff_tsc_power = cur_tsc_power - prev_tsc_power;
 838                 if (diff_tsc_power > TIMER_RESOLUTION_CYCLES) {
 839                         rte_timer_manage();
 840                         prev_tsc_power = cur_tsc_power;
 841                 }
 842
 843                 /*
 844                  * Read packet from RX queues
 845                  */
 846                 lcore_scaleup_hint = FREQ_CURRENT;
 847                 lcore_rx_idle_count = 0;
 848                 for (i = 0; i < qconf->n_rx_queue; ++i) {
 849                         rx_queue = &(qconf->rx_queue_list[i]);
 850                         rx_queue->idle_hint = 0;
 851                         portid = rx_queue->port_id;
 852                         queueid = rx_queue->queue_id;
 853
 854                         nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
 855                                                                 MAX_PKT_BURST);
 856                         stats[lcore_id].nb_rx_processed += nb_rx;
 857                         if (unlikely(nb_rx == 0)) {
 858                                 /**
 859                                  * no packet received from rx queue, try to
 860                                  * sleep for a while forcing CPU enter deeper
 861                                  * C states.
 862                                  */
 863                                 rx_queue->zero_rx_packet_count++;
 864
 865                                 if (rx_queue->zero_rx_packet_count <=
 866                                                         MIN_ZERO_POLL_COUNT)
 867                                         continue;
 868
 869                                 rx_queue->idle_hint = power_idle_heuristic(\
 870                                         rx_queue->zero_rx_packet_count);
 871                                 lcore_rx_idle_count++;
 872                         } else {
 873                                 rx_queue->zero_rx_packet_count = 0;
 874
 875                                 /**
 876                                  * do not scale up frequency immediately as
 877                                  * user to kernel space communication is costly
 878                                  * which might impact packet I/O for received
 879                                  * packets.
 880                                  */
 881                                 rx_queue->freq_up_hint =
 882                                         power_freq_scaleup_heuristic(lcore_id,
 883                                                         portid, queueid);
 884                         }
 885
 886                         /* Prefetch first packets */
 887                         for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
 888                                 rte_prefetch0(rte_pktmbuf_mtod(
 889                                                 pkts_burst[j], void *));
 890                         }
 891
 892                         /* Prefetch and forward already prefetched packets */
 893                         for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
 894                                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
 895                                                 j + PREFETCH_OFFSET], void *));
 896                                 l3fwd_simple_forward(pkts_burst[j], portid,
 897                                                                 qconf);
 898                         }
 899
 900                         /* Forward remaining prefetched packets */
 901                         for (; j < nb_rx; j++) {
 902                                 l3fwd_simple_forward(pkts_burst[j], portid,
 903                                                                 qconf);
 904                         }
 905                 }
 906
 907                 if (likely(lcore_rx_idle_count != qconf->n_rx_queue)) {
 908                         for (i = 1, lcore_scaleup_hint =
 909                                 qconf->rx_queue_list[0].freq_up_hint;
 910                                         i < qconf->n_rx_queue; ++i) {
 911                                 rx_queue = &(qconf->rx_queue_list[i]);
 912                                 if (rx_queue->freq_up_hint >
 913                                                 lcore_scaleup_hint)
 914                                         lcore_scaleup_hint =
 915                                                 rx_queue->freq_up_hint;
 916                         }
 917
 918                         if (lcore_scaleup_hint == FREQ_HIGHEST)
 919                                 rte_power_freq_max(lcore_id);
 920                         else if (lcore_scaleup_hint == FREQ_HIGHER)
 921                                 rte_power_freq_up(lcore_id);
 922                 } else {
 923                         /**
 924                          * All Rx queues empty in recent consecutive polls,
 925                          * sleep in a conservative manner, meaning sleep as
 926                          * less as possible.
 927                          */
 928                         for (i = 1, lcore_idle_hint =
 929                                 qconf->rx_queue_list[0].idle_hint;
 930                                         i < qconf->n_rx_queue; ++i) {
 931                                 rx_queue = &(qconf->rx_queue_list[i]);
 932                                 if (rx_queue->idle_hint < lcore_idle_hint)
 933                                         lcore_idle_hint = rx_queue->idle_hint;
 934                         }
 935
 936                         if ( lcore_idle_hint < SLEEP_GEAR1_THRESHOLD)
 937                                 /**
 938                                  * execute "pause" instruction to avoid context
 939                                  * switch for short sleep.
 940                                  */
 941                                 rte_delay_us(lcore_idle_hint);
 942                         else
 943                                 /* long sleep force runing thread to suspend */
 944                                 usleep(lcore_idle_hint);
 945
 946                         stats[lcore_id].sleep_time += lcore_idle_hint;
 947                 }
 948         }
 949 }
 950
 951 static int
 952 check_lcore_params(void)
 953 {
 954         uint8_t queue, lcore;
 955         uint16_t i;
 956         int socketid;
 957
 958         for (i = 0; i < nb_lcore_params; ++i) {
 959                 queue = lcore_params[i].queue_id;
 960                 if (queue >= MAX_RX_QUEUE_PER_PORT) {
 961                         printf("invalid queue number: %hhu\n", queue);
 962                         return -1;
 963                 }
 964                 lcore = lcore_params[i].lcore_id;
 965                 if (!rte_lcore_is_enabled(lcore)) {
 966                         printf("error: lcore %hhu is not enabled in lcore "
 967                                                         "mask\n", lcore);
 968                         return -1;
 969                 }
 970                 if ((socketid = rte_lcore_to_socket_id(lcore) != 0) &&
 971                                                         (numa_on == 0)) {
 972                         printf("warning: lcore %hhu is on socket %d with numa "
 973                                                 "off\n", lcore, socketid);
 974                 }
 975         }
 976         return 0;
 977 }
 978
 979 static int
 980 check_port_config(const unsigned nb_ports)
 981 {
 982         unsigned portid;
 983         uint16_t i;
 984
 985         for (i = 0; i < nb_lcore_params; ++i) {
 986                 portid = lcore_params[i].port_id;
 987                 if ((enabled_port_mask & (1 << portid)) == 0) {
 988                         printf("port %u is not enabled in port mask\n",
 989                                                                 portid);
 990                         return -1;
 991                 }
 992                 if (portid >= nb_ports) {
 993                         printf("port %u is not present on the board\n",
 994                                                                 portid);
 995                         return -1;
 996                 }
 997         }
 998         return 0;
 999 }
1000
1001 static uint8_t
1002 get_port_n_rx_queues(const uint8_t port)
1003 {
1004         int queue = -1;
1005         uint16_t i;
1006
1007         for (i = 0; i < nb_lcore_params; ++i) {
1008                 if (lcore_params[i].port_id == port &&
1009                                 lcore_params[i].queue_id > queue)
1010                         queue = lcore_params[i].queue_id;
1011         }
1012         return (uint8_t)(++queue);
1013 }
1014
1015 static int
1016 init_lcore_rx_queues(void)
1017 {
1018         uint16_t i, nb_rx_queue;
1019         uint8_t lcore;
1020
1021         for (i = 0; i < nb_lcore_params; ++i) {
1022                 lcore = lcore_params[i].lcore_id;
1023                 nb_rx_queue = lcore_conf[lcore].n_rx_queue;
1024                 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
1025                         printf("error: too many queues (%u) for lcore: %u\n",
1026                                 (unsigned)nb_rx_queue + 1, (unsigned)lcore);
1027                         return -1;
1028                 } else {
1029                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
1030                                 lcore_params[i].port_id;
1031                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
1032                                 lcore_params[i].queue_id;
1033                         lcore_conf[lcore].n_rx_queue++;
1034                 }
1035         }
1036         return 0;
1037 }
1038
1039 /* display usage */
1040 static void
1041 print_usage(const char *prgname)
1042 {
1043         printf ("%s [EAL options] -- -p PORTMASK -P"
1044                 "  [--config (port,queue,lcore)[,(port,queue,lcore]]"
1045                 "  [--enable-jumbo [--max-pkt-len PKTLEN]]\n"
1046                 "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
1047                 "  -P : enable promiscuous mode\n"
1048                 "  --config (port,queue,lcore): rx queues configuration\n"
1049                 "  --no-numa: optional, disable numa awareness\n"
1050                 "  --enable-jumbo: enable jumbo frame"
1051                 " which max packet len is PKTLEN in decimal (64-9600)\n",
1052                 prgname);
1053 }
1054
1055 static int parse_max_pkt_len(const char *pktlen)
1056 {
1057         char *end = NULL;
1058         unsigned long len;
1059
1060         /* parse decimal string */
1061         len = strtoul(pktlen, &end, 10);
1062         if ((pktlen[0] == '\0') || (end == NULL) || (*end != '\0'))
1063                 return -1;
1064
1065         if (len == 0)
1066                 return -1;
1067
1068         return len;
1069 }
1070
1071 static int
1072 parse_portmask(const char *portmask)
1073 {
1074         char *end = NULL;
1075         unsigned long pm;
1076
1077         /* parse hexadecimal string */
1078         pm = strtoul(portmask, &end, 16);
1079         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
1080                 return -1;
1081
1082         if (pm == 0)
1083                 return -1;
1084
1085         return pm;
1086 }
1087
1088 static int
1089 parse_config(const char *q_arg)
1090 {
1091         char s[256];
1092         const char *p, *p0 = q_arg;
1093         char *end;
1094         enum fieldnames {
1095                 FLD_PORT = 0,
1096                 FLD_QUEUE,
1097                 FLD_LCORE,
1098                 _NUM_FLD
1099         };
1100         unsigned long int_fld[_NUM_FLD];
1101         char *str_fld[_NUM_FLD];
1102         int i;
1103         unsigned size;
1104
1105         nb_lcore_params = 0;
1106
1107         while ((p = strchr(p0,'(')) != NULL) {
1108                 ++p;
1109                 if((p0 = strchr(p,')')) == NULL)
1110                         return -1;
1111
1112                 size = p0 - p;
1113                 if(size >= sizeof(s))
1114                         return -1;
1115
1116                 snprintf(s, sizeof(s), "%.*s", size, p);
1117                 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') !=
1118                                                                 _NUM_FLD)
1119                         return -1;
1120                 for (i = 0; i < _NUM_FLD; i++){
1121                         errno = 0;
1122                         int_fld[i] = strtoul(str_fld[i], &end, 0);
1123                         if (errno != 0 || end == str_fld[i] || int_fld[i] >
1124                                                                         255)
1125                                 return -1;
1126                 }
1127                 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
1128                         printf("exceeded max number of lcore params: %hu\n",
1129                                 nb_lcore_params);
1130                         return -1;
1131                 }
1132                 lcore_params_array[nb_lcore_params].port_id =
1133                                 (uint8_t)int_fld[FLD_PORT];
1134                 lcore_params_array[nb_lcore_params].queue_id =
1135                                 (uint8_t)int_fld[FLD_QUEUE];
1136                 lcore_params_array[nb_lcore_params].lcore_id =
1137                                 (uint8_t)int_fld[FLD_LCORE];
1138                 ++nb_lcore_params;
1139         }
1140         lcore_params = lcore_params_array;
1141
1142         return 0;
1143 }
1144
1145 /* Parse the argument given in the command line of the application */
1146 static int
1147 parse_args(int argc, char **argv)
1148 {
1149         int opt, ret;
1150         char **argvopt;
1151         int option_index;
1152         char *prgname = argv[0];
1153         static struct option lgopts[] = {
1154                 {"config", 1, 0, 0},
1155                 {"no-numa", 0, 0, 0},
1156                 {"enable-jumbo", 0, 0, 0},
1157                 {NULL, 0, 0, 0}
1158         };
1159
1160         argvopt = argv;
1161
1162         while ((opt = getopt_long(argc, argvopt, "p:P",
1163                                 lgopts, &option_index)) != EOF) {
1164
1165                 switch (opt) {
1166                 /* portmask */
1167                 case 'p':
1168                         enabled_port_mask = parse_portmask(optarg);
1169                         if (enabled_port_mask == 0) {
1170                                 printf("invalid portmask\n");
1171                                 print_usage(prgname);
1172                                 return -1;
1173                         }
1174                         break;
1175                 case 'P':
1176                         printf("Promiscuous mode selected\n");
1177                         promiscuous_on = 1;
1178                         break;
1179
1180                 /* long options */
1181                 case 0:
1182                         if (!strncmp(lgopts[option_index].name, "config", 6)) {
1183                                 ret = parse_config(optarg);
1184                                 if (ret) {
1185                                         printf("invalid config\n");
1186                                         print_usage(prgname);
1187                                         return -1;
1188                                 }
1189                         }
1190
1191                         if (!strncmp(lgopts[option_index].name,
1192                                                 "no-numa", 7)) {
1193                                 printf("numa is disabled \n");
1194                                 numa_on = 0;
1195                         }
1196
1197                         if (!strncmp(lgopts[option_index].name,
1198                                         "enable-jumbo", 12)) {
1199                                 struct option lenopts =
1200                                         {"max-pkt-len", required_argument, \
1201                                                                         0, 0};
1202
1203                                 printf("jumbo frame is enabled \n");
1204                                 port_conf.rxmode.jumbo_frame = 1;
1205
1206                                 /**
1207                                  * if no max-pkt-len set, use the default value
1208                                  * ETHER_MAX_LEN
1209                                  */
1210                                 if (0 == getopt_long(argc, argvopt, "",
1211                                                 &lenopts, &option_index)) {
1212                                         ret = parse_max_pkt_len(optarg);
1213                                         if ((ret < 64) ||
1214                                                 (ret > MAX_JUMBO_PKT_LEN)){
1215                                                 printf("invalid packet "
1216                                                                 "length\n");
1217                                                 print_usage(prgname);
1218                                                 return -1;
1219                                         }
1220                                         port_conf.rxmode.max_rx_pkt_len = ret;
1221                                 }
1222                                 printf("set jumbo frame "
1223                                         "max packet length to %u\n",
1224                                 (unsigned int)port_conf.rxmode.max_rx_pkt_len);
1225                         }
1226
1227                         break;
1228
1229                 default:
1230                         print_usage(prgname);
1231                         return -1;
1232                 }
1233         }
1234
1235         if (optind >= 0)
1236                 argv[optind-1] = prgname;
1237
1238         ret = optind-1;
1239         optind = 0; /* reset getopt lib */
1240         return ret;
1241 }
1242
1243 static void
1244 print_ethaddr(const char *name, const struct ether_addr *eth_addr)
1245 {
1246         char buf[ETHER_ADDR_FMT_SIZE];
1247         ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr);
1248         printf("%s%s", name, buf);
1249 }
1250
1251 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1252 static void
1253 setup_hash(int socketid)
1254 {
1255         struct rte_hash_parameters ipv4_l3fwd_hash_params = {
1256                 .name = NULL,
1257                 .entries = L3FWD_HASH_ENTRIES,
1258                 .key_len = sizeof(struct ipv4_5tuple),
1259                 .hash_func = DEFAULT_HASH_FUNC,
1260                 .hash_func_init_val = 0,
1261         };
1262
1263         struct rte_hash_parameters ipv6_l3fwd_hash_params = {
1264                 .name = NULL,
1265                 .entries = L3FWD_HASH_ENTRIES,
1266                 .key_len = sizeof(struct ipv6_5tuple),
1267                 .hash_func = DEFAULT_HASH_FUNC,
1268                 .hash_func_init_val = 0,
1269         };
1270
1271         unsigned i;
1272         int ret;
1273         char s[64];
1274
1275         /* create ipv4 hash */
1276         snprintf(s, sizeof(s), "ipv4_l3fwd_hash_%d", socketid);
1277         ipv4_l3fwd_hash_params.name = s;
1278         ipv4_l3fwd_hash_params.socket_id = socketid;
1279         ipv4_l3fwd_lookup_struct[socketid] =
1280                 rte_hash_create(&ipv4_l3fwd_hash_params);
1281         if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
1282                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
1283                                 "socket %d\n", socketid);
1284
1285         /* create ipv6 hash */
1286         snprintf(s, sizeof(s), "ipv6_l3fwd_hash_%d", socketid);
1287         ipv6_l3fwd_hash_params.name = s;
1288         ipv6_l3fwd_hash_params.socket_id = socketid;
1289         ipv6_l3fwd_lookup_struct[socketid] =
1290                 rte_hash_create(&ipv6_l3fwd_hash_params);
1291         if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
1292                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
1293                                 "socket %d\n", socketid);
1294
1295
1296         /* populate the ipv4 hash */
1297         for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
1298                 ret = rte_hash_add_key (ipv4_l3fwd_lookup_struct[socketid],
1299                                 (void *) &ipv4_l3fwd_route_array[i].key);
1300                 if (ret < 0) {
1301                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
1302                                 "l3fwd hash on socket %d\n", i, socketid);
1303                 }
1304                 ipv4_l3fwd_out_if[ret] = ipv4_l3fwd_route_array[i].if_out;
1305                 printf("Hash: Adding key\n");
1306                 print_ipv4_key(ipv4_l3fwd_route_array[i].key);
1307         }
1308
1309         /* populate the ipv6 hash */
1310         for (i = 0; i < IPV6_L3FWD_NUM_ROUTES; i++) {
1311                 ret = rte_hash_add_key (ipv6_l3fwd_lookup_struct[socketid],
1312                                 (void *) &ipv6_l3fwd_route_array[i].key);
1313                 if (ret < 0) {
1314                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
1315                                 "l3fwd hash on socket %d\n", i, socketid);
1316                 }
1317                 ipv6_l3fwd_out_if[ret] = ipv6_l3fwd_route_array[i].if_out;
1318                 printf("Hash: Adding key\n");
1319                 print_ipv6_key(ipv6_l3fwd_route_array[i].key);
1320         }
1321 }
1322 #endif
1323
1324 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1325 static void
1326 setup_lpm(int socketid)
1327 {
1328         unsigned i;
1329         int ret;
1330         char s[64];
1331
1332         /* create the LPM table */
1333         snprintf(s, sizeof(s), "IPV4_L3FWD_LPM_%d", socketid);
1334         ipv4_l3fwd_lookup_struct[socketid] = rte_lpm_create(s, socketid,
1335                                 IPV4_L3FWD_LPM_MAX_RULES, 0);
1336         if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
1337                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
1338                                 " on socket %d\n", socketid);
1339
1340         /* populate the LPM table */
1341         for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
1342                 ret = rte_lpm_add(ipv4_l3fwd_lookup_struct[socketid],
1343                         ipv4_l3fwd_route_array[i].ip,
1344                         ipv4_l3fwd_route_array[i].depth,
1345                         ipv4_l3fwd_route_array[i].if_out);
1346
1347                 if (ret < 0) {
1348                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
1349                                 "l3fwd LPM table on socket %d\n",
1350                                 i, socketid);
1351                 }
1352
1353                 printf("LPM: Adding route 0x%08x / %d (%d)\n",
1354                         (unsigned)ipv4_l3fwd_route_array[i].ip,
1355                         ipv4_l3fwd_route_array[i].depth,
1356                         ipv4_l3fwd_route_array[i].if_out);
1357         }
1358 }
1359 #endif
1360
1361 static int
1362 init_mem(unsigned nb_mbuf)
1363 {
1364         struct lcore_conf *qconf;
1365         int socketid;
1366         unsigned lcore_id;
1367         char s[64];
1368
1369         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1370                 if (rte_lcore_is_enabled(lcore_id) == 0)
1371                         continue;
1372
1373                 if (numa_on)
1374                         socketid = rte_lcore_to_socket_id(lcore_id);
1375                 else
1376                         socketid = 0;
1377
1378                 if (socketid >= NB_SOCKETS) {
1379                         rte_exit(EXIT_FAILURE, "Socket %d of lcore %u is "
1380                                         "out of range %d\n", socketid,
1381                                                 lcore_id, NB_SOCKETS);
1382                 }
1383                 if (pktmbuf_pool[socketid] == NULL) {
1384                         snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
1385                         pktmbuf_pool[socketid] =
1386                                 rte_pktmbuf_pool_create(s, nb_mbuf,
1387                                         MEMPOOL_CACHE_SIZE, 0,
1388                                         RTE_MBUF_DEFAULT_BUF_SIZE,
1389                                         socketid);
1390                         if (pktmbuf_pool[socketid] == NULL)
1391                                 rte_exit(EXIT_FAILURE,
1392                                         "Cannot init mbuf pool on socket %d\n",
1393                                                                 socketid);
1394                         else
1395                                 printf("Allocated mbuf pool on socket %d\n",
1396                                                                 socketid);
1397
1398 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1399                         setup_lpm(socketid);
1400 #else
1401                         setup_hash(socketid);
1402 #endif
1403                 }
1404                 qconf = &lcore_conf[lcore_id];
1405                 qconf->ipv4_lookup_struct = ipv4_l3fwd_lookup_struct[socketid];
1406 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1407                 qconf->ipv6_lookup_struct = ipv6_l3fwd_lookup_struct[socketid];
1408 #endif
1409         }
1410         return 0;
1411 }
1412
1413 /* Check the link status of all ports in up to 9s, and print them finally */
1414 static void
1415 check_all_ports_link_status(uint8_t port_num, uint32_t port_mask)
1416 {
1417 #define CHECK_INTERVAL 100 /* 100ms */
1418 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
1419         uint8_t portid, count, all_ports_up, print_flag = 0;
1420         struct rte_eth_link link;
1421
1422         printf("\nChecking link status");
1423         fflush(stdout);
1424         for (count = 0; count <= MAX_CHECK_TIME; count++) {
1425                 all_ports_up = 1;
1426                 for (portid = 0; portid < port_num; portid++) {
1427                         if ((port_mask & (1 << portid)) == 0)
1428                                 continue;
1429                         memset(&link, 0, sizeof(link));
1430                         rte_eth_link_get_nowait(portid, &link);
1431                         /* print link status if flag set */
1432                         if (print_flag == 1) {
1433                                 if (link.link_status)
1434                                         printf("Port %d Link Up - speed %u "
1435                                                 "Mbps - %s\n", (uint8_t)portid,
1436                                                 (unsigned)link.link_speed,
1437                                 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
1438                                         ("full-duplex") : ("half-duplex\n"));
1439                                 else
1440                                         printf("Port %d Link Down\n",
1441                                                 (uint8_t)portid);
1442                                 continue;
1443                         }
1444                         /* clear all_ports_up flag if any link down */
1445                         if (link.link_status == 0) {
1446                                 all_ports_up = 0;
1447                                 break;
1448                         }
1449                 }
1450                 /* after finally printing all link status, get out */
1451                 if (print_flag == 1)
1452                         break;
1453
1454                 if (all_ports_up == 0) {
1455                         printf(".");
1456                         fflush(stdout);
1457                         rte_delay_ms(CHECK_INTERVAL);
1458                 }
1459
1460                 /* set the print_flag if all ports up or timeout */
1461                 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
1462                         print_flag = 1;
1463                         printf("done\n");
1464                 }
1465         }
1466 }
1467
1468 int
1469 main(int argc, char **argv)
1470 {
1471         struct lcore_conf *qconf;
1472         struct rte_eth_dev_info dev_info;
1473         struct rte_eth_txconf *txconf;
1474         int ret;
1475         unsigned nb_ports;
1476         uint16_t queueid;
1477         unsigned lcore_id;
1478         uint64_t hz;
1479         uint32_t n_tx_queue, nb_lcores;
1480         uint8_t portid, nb_rx_queue, queue, socketid;
1481
1482         /* catch SIGINT and restore cpufreq governor to ondemand */
1483         signal(SIGINT, signal_exit_now);
1484
1485         /* init EAL */
1486         ret = rte_eal_init(argc, argv);
1487         if (ret < 0)
1488                 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
1489         argc -= ret;
1490         argv += ret;
1491
1492         /* init RTE timer library to be used late */
1493         rte_timer_subsystem_init();
1494
1495         /* parse application arguments (after the EAL ones) */
1496         ret = parse_args(argc, argv);
1497         if (ret < 0)
1498                 rte_exit(EXIT_FAILURE, "Invalid L3FWD parameters\n");
1499
1500         if (check_lcore_params() < 0)
1501                 rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
1502
1503         ret = init_lcore_rx_queues();
1504         if (ret < 0)
1505                 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
1506
1507
1508         nb_ports = rte_eth_dev_count();
1509         if (nb_ports > RTE_MAX_ETHPORTS)
1510                 nb_ports = RTE_MAX_ETHPORTS;
1511
1512         if (check_port_config(nb_ports) < 0)
1513                 rte_exit(EXIT_FAILURE, "check_port_config failed\n");
1514
1515         nb_lcores = rte_lcore_count();
1516
1517         /* initialize all ports */
1518         for (portid = 0; portid < nb_ports; portid++) {
1519                 /* skip ports that are not enabled */
1520                 if ((enabled_port_mask & (1 << portid)) == 0) {
1521                         printf("\nSkipping disabled port %d\n", portid);
1522                         continue;
1523                 }
1524
1525                 /* init port */
1526                 printf("Initializing port %d ... ", portid );
1527                 fflush(stdout);
1528
1529                 nb_rx_queue = get_port_n_rx_queues(portid);
1530                 n_tx_queue = nb_lcores;
1531                 if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
1532                         n_tx_queue = MAX_TX_QUEUE_PER_PORT;
1533                 printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
1534                         nb_rx_queue, (unsigned)n_tx_queue );
1535                 ret = rte_eth_dev_configure(portid, nb_rx_queue,
1536                                         (uint16_t)n_tx_queue, &port_conf);
1537                 if (ret < 0)
1538                         rte_exit(EXIT_FAILURE, "Cannot configure device: "
1539                                         "err=%d, port=%d\n", ret, portid);
1540
1541                 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
1542                 print_ethaddr(" Address:", &ports_eth_addr[portid]);
1543                 printf(", ");
1544
1545                 /* init memory */
1546                 ret = init_mem(NB_MBUF);
1547                 if (ret < 0)
1548                         rte_exit(EXIT_FAILURE, "init_mem failed\n");
1549
1550                 /* init one TX queue per couple (lcore,port) */
1551                 queueid = 0;
1552                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1553                         if (rte_lcore_is_enabled(lcore_id) == 0)
1554                                 continue;
1555
1556                         if (numa_on)
1557                                 socketid = \
1558                                 (uint8_t)rte_lcore_to_socket_id(lcore_id);
1559                         else
1560                                 socketid = 0;
1561
1562                         printf("txq=%u,%d,%d ", lcore_id, queueid, socketid);
1563                         fflush(stdout);
1564
1565                         rte_eth_dev_info_get(portid, &dev_info);
1566                         txconf = &dev_info.default_txconf;
1567                         if (port_conf.rxmode.jumbo_frame)
1568                                 txconf->txq_flags = 0;
1569                         ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
1570                                                      socketid, txconf);
1571                         if (ret < 0)
1572                                 rte_exit(EXIT_FAILURE,
1573                                         "rte_eth_tx_queue_setup: err=%d, "
1574                                                 "port=%d\n", ret, portid);
1575
1576                         qconf = &lcore_conf[lcore_id];
1577                         qconf->tx_queue_id[portid] = queueid;
1578                         queueid++;
1579                 }
1580                 printf("\n");
1581         }
1582
1583         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1584                 if (rte_lcore_is_enabled(lcore_id) == 0)
1585                         continue;
1586
1587                 /* init power management library */
1588                 ret = rte_power_init(lcore_id);
1589                 if (ret)
1590                         rte_exit(EXIT_FAILURE, "Power management library "
1591                                 "initialization failed on core%u\n", lcore_id);
1592
1593                 /* init timer structures for each enabled lcore */
1594                 rte_timer_init(&power_timers[lcore_id]);
1595                 hz = rte_get_timer_hz();
1596                 rte_timer_reset(&power_timers[lcore_id],
1597                         hz/TIMER_NUMBER_PER_SECOND, SINGLE, lcore_id,
1598                                                 power_timer_cb, NULL);
1599
1600                 qconf = &lcore_conf[lcore_id];
1601                 printf("\nInitializing rx queues on lcore %u ... ", lcore_id );
1602                 fflush(stdout);
1603                 /* init RX queues */
1604                 for(queue = 0; queue < qconf->n_rx_queue; ++queue) {
1605                         portid = qconf->rx_queue_list[queue].port_id;
1606                         queueid = qconf->rx_queue_list[queue].queue_id;
1607
1608                         if (numa_on)
1609                                 socketid = \
1610                                 (uint8_t)rte_lcore_to_socket_id(lcore_id);
1611                         else
1612                                 socketid = 0;
1613
1614                         printf("rxq=%d,%d,%d ", portid, queueid, socketid);
1615                         fflush(stdout);
1616
1617                         ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
1618                                 socketid, NULL,
1619                                 pktmbuf_pool[socketid]);
1620                         if (ret < 0)
1621                                 rte_exit(EXIT_FAILURE,
1622                                         "rte_eth_rx_queue_setup: err=%d, "
1623                                                 "port=%d\n", ret, portid);
1624                 }
1625         }
1626
1627         printf("\n");
1628
1629         /* start ports */
1630         for (portid = 0; portid < nb_ports; portid++) {
1631                 if ((enabled_port_mask & (1 << portid)) == 0) {
1632                         continue;
1633                 }
1634                 /* Start device */
1635                 ret = rte_eth_dev_start(portid);
1636                 if (ret < 0)
1637                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, "
1638                                                 "port=%d\n", ret, portid);
1639
1640                 /*
1641                  * If enabled, put device in promiscuous mode.
1642                  * This allows IO forwarding mode to forward packets
1643                  * to itself through 2 cross-connected  ports of the
1644                  * target machine.
1645                  */
1646                 if (promiscuous_on)
1647                         rte_eth_promiscuous_enable(portid);
1648         }
1649
1650         check_all_ports_link_status((uint8_t)nb_ports, enabled_port_mask);
1651
1652         /* launch per-lcore init on every lcore */
1653         rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
1654         RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1655                 if (rte_eal_wait_lcore(lcore_id) < 0)
1656                         return -1;
1657         }
1658
1659         return 0;
1660 }