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_tailq.h>
  54 #include <rte_eal.h>
  55 #include <rte_per_lcore.h>
  56 #include <rte_launch.h>
  57 #include <rte_atomic.h>
  58 #include <rte_cycles.h>
  59 #include <rte_prefetch.h>
  60 #include <rte_lcore.h>
  61 #include <rte_per_lcore.h>
  62 #include <rte_branch_prediction.h>
  63 #include <rte_interrupts.h>
  64 #include <rte_pci.h>
  65 #include <rte_random.h>
  66 #include <rte_debug.h>
  67 #include <rte_ether.h>
  68 #include <rte_ethdev.h>
  69 #include <rte_ring.h>
  70 #include <rte_mempool.h>
  71 #include <rte_mbuf.h>
  72 #include <rte_ip.h>
  73 #include <rte_tcp.h>
  74 #include <rte_udp.h>
  75 #include <rte_string_fns.h>
  76 #include <rte_timer.h>
  77 #include <rte_power.h>
  78
  79 #define RTE_LOGTYPE_L3FWD_POWER RTE_LOGTYPE_USER1
  80
  81 #define MAX_PKT_BURST 32
  82
  83 #define MIN_ZERO_POLL_COUNT 5
  84
  85 /* around 100ms at 2 Ghz */
  86 #define TIMER_RESOLUTION_CYCLES           200000000ULL
  87 /* 100 ms interval */
  88 #define TIMER_NUMBER_PER_SECOND           10
  89 /* 100000 us */
  90 #define SCALING_PERIOD                    (1000000/TIMER_NUMBER_PER_SECOND)
  91 #define SCALING_DOWN_TIME_RATIO_THRESHOLD 0.25
  92
  93 #define APP_LOOKUP_EXACT_MATCH          0
  94 #define APP_LOOKUP_LPM                  1
  95 #define DO_RFC_1812_CHECKS
  96
  97 #ifndef APP_LOOKUP_METHOD
  98 #define APP_LOOKUP_METHOD             APP_LOOKUP_LPM
  99 #endif
 100
 101 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 102 #include <rte_hash.h>
 103 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 104 #include <rte_lpm.h>
 105 #else
 106 #error "APP_LOOKUP_METHOD set to incorrect value"
 107 #endif
 108
 109 #ifndef IPv6_BYTES
 110 #define IPv6_BYTES_FMT "%02x%02x:%02x%02x:%02x%02x:%02x%02x:"\
 111                        "%02x%02x:%02x%02x:%02x%02x:%02x%02x"
 112 #define IPv6_BYTES(addr) \
 113         addr[0],  addr[1], addr[2],  addr[3], \
 114         addr[4],  addr[5], addr[6],  addr[7], \
 115         addr[8],  addr[9], addr[10], addr[11],\
 116         addr[12], addr[13],addr[14], addr[15]
 117 #endif
 118
 119 #define MAX_JUMBO_PKT_LEN  9600
 120
 121 #define IPV6_ADDR_LEN 16
 122
 123 #define MEMPOOL_CACHE_SIZE 256
 124
 125 #define MBUF_SIZE (2048 + sizeof(struct rte_mbuf) + RTE_PKTMBUF_HEADROOM)
 126
 127 /*
 128  * This expression is used to calculate the number of mbufs needed depending on
 129  * user input, taking into account memory for rx and tx hardware rings, cache
 130  * per lcore and mtable per port per lcore. RTE_MAX is used to ensure that
 131  * NB_MBUF never goes below a minimum value of 8192.
 132  */
 133
 134 #define NB_MBUF RTE_MAX ( \
 135         (nb_ports*nb_rx_queue*RTE_TEST_RX_DESC_DEFAULT + \
 136         nb_ports*nb_lcores*MAX_PKT_BURST + \
 137         nb_ports*n_tx_queue*RTE_TEST_TX_DESC_DEFAULT + \
 138         nb_lcores*MEMPOOL_CACHE_SIZE), \
 139         (unsigned)8192)
 140
 141 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
 142
 143 #define NB_SOCKETS 8
 144
 145 /* Configure how many packets ahead to prefetch, when reading packets */
 146 #define PREFETCH_OFFSET 3
 147
 148 /*
 149  * Configurable number of RX/TX ring descriptors
 150  */
 151 #define RTE_TEST_RX_DESC_DEFAULT 128
 152 #define RTE_TEST_TX_DESC_DEFAULT 512
 153 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
 154 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
 155
 156 /* ethernet addresses of ports */
 157 static struct ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
 158
 159 /* mask of enabled ports */
 160 static uint32_t enabled_port_mask = 0;
 161 /* Ports set in promiscuous mode off by default. */
 162 static int promiscuous_on = 0;
 163 /* NUMA is enabled by default. */
 164 static int numa_on = 1;
 165
 166 enum freq_scale_hint_t
 167 {
 168         FREQ_LOWER    =      -1,
 169         FREQ_CURRENT  =       0,
 170         FREQ_HIGHER   =       1,
 171         FREQ_HIGHEST  =       2
 172 };
 173
 174 struct mbuf_table {
 175         uint16_t len;
 176         struct rte_mbuf *m_table[MAX_PKT_BURST];
 177 };
 178
 179 struct lcore_rx_queue {
 180         uint8_t port_id;
 181         uint8_t queue_id;
 182         enum freq_scale_hint_t freq_up_hint;
 183         uint32_t zero_rx_packet_count;
 184         uint32_t idle_hint;
 185 } __rte_cache_aligned;
 186
 187 #define MAX_RX_QUEUE_PER_LCORE 16
 188 #define MAX_TX_QUEUE_PER_PORT RTE_MAX_ETHPORTS
 189 #define MAX_RX_QUEUE_PER_PORT 128
 190
 191 #define MAX_LCORE_PARAMS 1024
 192 struct lcore_params {
 193         uint8_t port_id;
 194         uint8_t queue_id;
 195         uint8_t lcore_id;
 196 } __rte_cache_aligned;
 197
 198 static struct lcore_params lcore_params_array[MAX_LCORE_PARAMS];
 199 static struct lcore_params lcore_params_array_default[] = {
 200         {0, 0, 2},
 201         {0, 1, 2},
 202         {0, 2, 2},
 203         {1, 0, 2},
 204         {1, 1, 2},
 205         {1, 2, 2},
 206         {2, 0, 2},
 207         {3, 0, 3},
 208         {3, 1, 3},
 209 };
 210
 211 static struct lcore_params * lcore_params = lcore_params_array_default;
 212 static uint16_t nb_lcore_params = sizeof(lcore_params_array_default) /
 213                                 sizeof(lcore_params_array_default[0]);
 214
 215 static struct rte_eth_conf port_conf = {
 216         .rxmode = {
 217                 .mq_mode        = ETH_MQ_RX_RSS,
 218                 .max_rx_pkt_len = ETHER_MAX_LEN,
 219                 .split_hdr_size = 0,
 220                 .header_split   = 0, /**< Header Split disabled */
 221                 .hw_ip_checksum = 1, /**< IP checksum offload enabled */
 222                 .hw_vlan_filter = 0, /**< VLAN filtering disabled */
 223                 .jumbo_frame    = 0, /**< Jumbo Frame Support disabled */
 224                 .hw_strip_crc   = 0, /**< CRC stripped by hardware */
 225         },
 226         .rx_adv_conf = {
 227                 .rss_conf = {
 228                         .rss_key = NULL,
 229                         .rss_hf = ETH_RSS_IP,
 230                 },
 231         },
 232         .txmode = {
 233                 .mq_mode = ETH_DCB_NONE,
 234         },
 235 };
 236
 237 static struct rte_mempool * pktmbuf_pool[NB_SOCKETS];
 238
 239
 240 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 241
 242 #ifdef RTE_MACHINE_CPUFLAG_SSE4_2
 243 #include <rte_hash_crc.h>
 244 #define DEFAULT_HASH_FUNC       rte_hash_crc
 245 #else
 246 #include <rte_jhash.h>
 247 #define DEFAULT_HASH_FUNC       rte_jhash
 248 #endif
 249
 250 struct ipv4_5tuple {
 251         uint32_t ip_dst;
 252         uint32_t ip_src;
 253         uint16_t port_dst;
 254         uint16_t port_src;
 255         uint8_t  proto;
 256 } __attribute__((__packed__));
 257
 258 struct ipv6_5tuple {
 259         uint8_t  ip_dst[IPV6_ADDR_LEN];
 260         uint8_t  ip_src[IPV6_ADDR_LEN];
 261         uint16_t port_dst;
 262         uint16_t port_src;
 263         uint8_t  proto;
 264 } __attribute__((__packed__));
 265
 266 struct ipv4_l3fwd_route {
 267         struct ipv4_5tuple key;
 268         uint8_t if_out;
 269 };
 270
 271 struct ipv6_l3fwd_route {
 272         struct ipv6_5tuple key;
 273         uint8_t if_out;
 274 };
 275
 276 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
 277         {{IPv4(100,10,0,1), IPv4(200,10,0,1), 101, 11, IPPROTO_TCP}, 0},
 278         {{IPv4(100,20,0,2), IPv4(200,20,0,2), 102, 12, IPPROTO_TCP}, 1},
 279         {{IPv4(100,30,0,3), IPv4(200,30,0,3), 103, 13, IPPROTO_TCP}, 2},
 280         {{IPv4(100,40,0,4), IPv4(200,40,0,4), 104, 14, IPPROTO_TCP}, 3},
 281 };
 282
 283 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
 284         {
 285                 {
 286                         {0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
 287                          0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
 288                         {0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
 289                          0x02, 0x1e, 0x67, 0xff, 0xfe, 0x0d, 0xb6, 0x0a},
 290                          1, 10, IPPROTO_UDP
 291                 }, 4
 292         },
 293 };
 294
 295 typedef struct rte_hash lookup_struct_t;
 296 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
 297 static lookup_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
 298
 299 #define L3FWD_HASH_ENTRIES      1024
 300
 301 #define IPV4_L3FWD_NUM_ROUTES \
 302         (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
 303
 304 #define IPV6_L3FWD_NUM_ROUTES \
 305         (sizeof(ipv6_l3fwd_route_array) / sizeof(ipv6_l3fwd_route_array[0]))
 306
 307 static uint8_t ipv4_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 308 static uint8_t ipv6_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 309 #endif
 310
 311 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 312 struct ipv4_l3fwd_route {
 313         uint32_t ip;
 314         uint8_t  depth;
 315         uint8_t  if_out;
 316 };
 317
 318 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
 319         {IPv4(1,1,1,0), 24, 0},
 320         {IPv4(2,1,1,0), 24, 1},
 321         {IPv4(3,1,1,0), 24, 2},
 322         {IPv4(4,1,1,0), 24, 3},
 323         {IPv4(5,1,1,0), 24, 4},
 324         {IPv4(6,1,1,0), 24, 5},
 325         {IPv4(7,1,1,0), 24, 6},
 326         {IPv4(8,1,1,0), 24, 7},
 327 };
 328
 329 #define IPV4_L3FWD_NUM_ROUTES \
 330         (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
 331
 332 #define IPV4_L3FWD_LPM_MAX_RULES     1024
 333
 334 typedef struct rte_lpm lookup_struct_t;
 335 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
 336 #endif
 337
 338 struct lcore_conf {
 339         uint16_t n_rx_queue;
 340         struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
 341         uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
 342         struct mbuf_table tx_mbufs[RTE_MAX_ETHPORTS];
 343         lookup_struct_t * ipv4_lookup_struct;
 344         lookup_struct_t * ipv6_lookup_struct;
 345 } __rte_cache_aligned;
 346
 347 struct lcore_stats {
 348         /* total sleep time in ms since last frequency scaling down */
 349         uint32_t sleep_time;
 350         /* number of long sleep recently */
 351         uint32_t nb_long_sleep;
 352         /* freq. scaling up trend */
 353         uint32_t trend;
 354         /* total packet processed recently */
 355         uint64_t nb_rx_processed;
 356         /* total iterations looped recently */
 357         uint64_t nb_iteration_looped;
 358         uint32_t padding[9];
 359 } __rte_cache_aligned;
 360
 361 static struct lcore_conf lcore_conf[RTE_MAX_LCORE] __rte_cache_aligned;
 362 static struct lcore_stats stats[RTE_MAX_LCORE] __rte_cache_aligned;
 363 static struct rte_timer power_timers[RTE_MAX_LCORE];
 364
 365 static inline uint32_t power_idle_heuristic(uint32_t zero_rx_packet_count);
 366 static inline enum freq_scale_hint_t power_freq_scaleup_heuristic( \
 367                         unsigned lcore_id, uint8_t port_id, uint16_t queue_id);
 368
 369 /* exit signal handler */
 370 static void
 371 signal_exit_now(int sigtype)
 372 {
 373         unsigned lcore_id;
 374         int ret;
 375
 376         if (sigtype == SIGINT) {
 377                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
 378                         if (rte_lcore_is_enabled(lcore_id) == 0)
 379                                 continue;
 380
 381                         /* init power management library */
 382                         ret = rte_power_exit(lcore_id);
 383                         if (ret)
 384                                 rte_exit(EXIT_FAILURE, "Power management "
 385                                         "library de-initialization failed on "
 386                                                         "core%u\n", lcore_id);
 387                 }
 388         }
 389
 390         rte_exit(EXIT_SUCCESS, "User forced exit\n");
 391 }
 392
 393 /*  Freqency scale down timer callback */
 394 static void
 395 power_timer_cb(__attribute__((unused)) struct rte_timer *tim,
 396                           __attribute__((unused)) void *arg)
 397 {
 398         uint64_t hz;
 399         float sleep_time_ratio;
 400         unsigned lcore_id = rte_lcore_id();
 401
 402         /* accumulate total execution time in us when callback is invoked */
 403         sleep_time_ratio = (float)(stats[lcore_id].sleep_time) /
 404                                         (float)SCALING_PERIOD;
 405
 406         /**
 407          * check whether need to scale down frequency a step if it sleep a lot.
 408          */
 409         if (sleep_time_ratio >= SCALING_DOWN_TIME_RATIO_THRESHOLD)
 410                 rte_power_freq_down(lcore_id);
 411         else if ( (unsigned)(stats[lcore_id].nb_rx_processed /
 412                 stats[lcore_id].nb_iteration_looped) < MAX_PKT_BURST)
 413                 /**
 414                  * scale down a step if average packet per iteration less
 415                  * than expectation.
 416                  */
 417                 rte_power_freq_down(lcore_id);
 418
 419         /**
 420          * initialize another timer according to current frequency to ensure
 421          * timer interval is relatively fixed.
 422          */
 423         hz = rte_get_timer_hz();
 424         rte_timer_reset(&power_timers[lcore_id], hz/TIMER_NUMBER_PER_SECOND,
 425                                 SINGLE, lcore_id, power_timer_cb, NULL);
 426
 427         stats[lcore_id].nb_rx_processed = 0;
 428         stats[lcore_id].nb_iteration_looped = 0;
 429
 430         stats[lcore_id].sleep_time = 0;
 431 }
 432
 433 /* Send burst of packets on an output interface */
 434 static inline int
 435 send_burst(struct lcore_conf *qconf, uint16_t n, uint8_t port)
 436 {
 437         struct rte_mbuf **m_table;
 438         int ret;
 439         uint16_t queueid;
 440
 441         queueid = qconf->tx_queue_id[port];
 442         m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
 443
 444         ret = rte_eth_tx_burst(port, queueid, m_table, n);
 445         if (unlikely(ret < n)) {
 446                 do {
 447                         rte_pktmbuf_free(m_table[ret]);
 448                 } while (++ret < n);
 449         }
 450
 451         return 0;
 452 }
 453
 454 /* Enqueue a single packet, and send burst if queue is filled */
 455 static inline int
 456 send_single_packet(struct rte_mbuf *m, uint8_t port)
 457 {
 458         uint32_t lcore_id;
 459         uint16_t len;
 460         struct lcore_conf *qconf;
 461
 462         lcore_id = rte_lcore_id();
 463
 464         qconf = &lcore_conf[lcore_id];
 465         len = qconf->tx_mbufs[port].len;
 466         qconf->tx_mbufs[port].m_table[len] = m;
 467         len++;
 468
 469         /* enough pkts to be sent */
 470         if (unlikely(len == MAX_PKT_BURST)) {
 471                 send_burst(qconf, MAX_PKT_BURST, port);
 472                 len = 0;
 473         }
 474
 475         qconf->tx_mbufs[port].len = len;
 476         return 0;
 477 }
 478
 479 #ifdef DO_RFC_1812_CHECKS
 480 static inline int
 481 is_valid_ipv4_pkt(struct ipv4_hdr *pkt, uint32_t link_len)
 482 {
 483         /* From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2 */
 484         /*
 485          * 1. The packet length reported by the Link Layer must be large
 486          * enough to hold the minimum length legal IP datagram (20 bytes).
 487          */
 488         if (link_len < sizeof(struct ipv4_hdr))
 489                 return -1;
 490
 491         /* 2. The IP checksum must be correct. */
 492         /* this is checked in H/W */
 493
 494         /*
 495          * 3. The IP version number must be 4. If the version number is not 4
 496          * then the packet may be another version of IP, such as IPng or
 497          * ST-II.
 498          */
 499         if (((pkt->version_ihl) >> 4) != 4)
 500                 return -3;
 501         /*
 502          * 4. The IP header length field must be large enough to hold the
 503          * minimum length legal IP datagram (20 bytes = 5 words).
 504          */
 505         if ((pkt->version_ihl & 0xf) < 5)
 506                 return -4;
 507
 508         /*
 509          * 5. The IP total length field must be large enough to hold the IP
 510          * datagram header, whose length is specified in the IP header length
 511          * field.
 512          */
 513         if (rte_cpu_to_be_16(pkt->total_length) < sizeof(struct ipv4_hdr))
 514                 return -5;
 515
 516         return 0;
 517 }
 518 #endif
 519
 520 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 521 static void
 522 print_ipv4_key(struct ipv4_5tuple key)
 523 {
 524         printf("IP dst = %08x, IP src = %08x, port dst = %d, port src = %d, "
 525                 "proto = %d\n", (unsigned)key.ip_dst, (unsigned)key.ip_src,
 526                                 key.port_dst, key.port_src, key.proto);
 527 }
 528 static void
 529 print_ipv6_key(struct ipv6_5tuple key)
 530 {
 531         printf( "IP dst = " IPv6_BYTES_FMT ", IP src = " IPv6_BYTES_FMT ", "
 532                 "port dst = %d, port src = %d, proto = %d\n",
 533                 IPv6_BYTES(key.ip_dst), IPv6_BYTES(key.ip_src),
 534                 key.port_dst, key.port_src, key.proto);
 535 }
 536
 537 static inline uint8_t
 538 get_ipv4_dst_port(struct ipv4_hdr *ipv4_hdr, uint8_t portid,
 539                 lookup_struct_t * ipv4_l3fwd_lookup_struct)
 540 {
 541         struct ipv4_5tuple key;
 542         struct tcp_hdr *tcp;
 543         struct udp_hdr *udp;
 544         int ret = 0;
 545
 546         key.ip_dst = rte_be_to_cpu_32(ipv4_hdr->dst_addr);
 547         key.ip_src = rte_be_to_cpu_32(ipv4_hdr->src_addr);
 548         key.proto = ipv4_hdr->next_proto_id;
 549
 550         switch (ipv4_hdr->next_proto_id) {
 551         case IPPROTO_TCP:
 552                 tcp = (struct tcp_hdr *)((unsigned char *)ipv4_hdr +
 553                                         sizeof(struct ipv4_hdr));
 554                 key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
 555                 key.port_src = rte_be_to_cpu_16(tcp->src_port);
 556                 break;
 557
 558         case IPPROTO_UDP:
 559                 udp = (struct udp_hdr *)((unsigned char *)ipv4_hdr +
 560                                         sizeof(struct ipv4_hdr));
 561                 key.port_dst = rte_be_to_cpu_16(udp->dst_port);
 562                 key.port_src = rte_be_to_cpu_16(udp->src_port);
 563                 break;
 564
 565         default:
 566                 key.port_dst = 0;
 567                 key.port_src = 0;
 568                 break;
 569         }
 570
 571         /* Find destination port */
 572         ret = rte_hash_lookup(ipv4_l3fwd_lookup_struct, (const void *)&key);
 573         return (uint8_t)((ret < 0)? portid : ipv4_l3fwd_out_if[ret]);
 574 }
 575
 576 static inline uint8_t
 577 get_ipv6_dst_port(struct ipv6_hdr *ipv6_hdr,  uint8_t portid,
 578                         lookup_struct_t *ipv6_l3fwd_lookup_struct)
 579 {
 580         struct ipv6_5tuple key;
 581         struct tcp_hdr *tcp;
 582         struct udp_hdr *udp;
 583         int ret = 0;
 584
 585         memcpy(key.ip_dst, ipv6_hdr->dst_addr, IPV6_ADDR_LEN);
 586         memcpy(key.ip_src, ipv6_hdr->src_addr, IPV6_ADDR_LEN);
 587
 588         key.proto = ipv6_hdr->proto;
 589
 590         switch (ipv6_hdr->proto) {
 591         case IPPROTO_TCP:
 592                 tcp = (struct tcp_hdr *)((unsigned char *) ipv6_hdr +
 593                                         sizeof(struct ipv6_hdr));
 594                 key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
 595                 key.port_src = rte_be_to_cpu_16(tcp->src_port);
 596                 break;
 597
 598         case IPPROTO_UDP:
 599                 udp = (struct udp_hdr *)((unsigned char *) ipv6_hdr +
 600                                         sizeof(struct ipv6_hdr));
 601                 key.port_dst = rte_be_to_cpu_16(udp->dst_port);
 602                 key.port_src = rte_be_to_cpu_16(udp->src_port);
 603                 break;
 604
 605         default:
 606                 key.port_dst = 0;
 607                 key.port_src = 0;
 608                 break;
 609         }
 610
 611         /* Find destination port */
 612         ret = rte_hash_lookup(ipv6_l3fwd_lookup_struct, (const void *)&key);
 613         return (uint8_t)((ret < 0)? portid : ipv6_l3fwd_out_if[ret]);
 614 }
 615 #endif
 616
 617 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 618 static inline uint8_t
 619 get_ipv4_dst_port(struct ipv4_hdr *ipv4_hdr, uint8_t portid,
 620                 lookup_struct_t *ipv4_l3fwd_lookup_struct)
 621 {
 622         uint8_t next_hop;
 623
 624         return (uint8_t) ((rte_lpm_lookup(ipv4_l3fwd_lookup_struct,
 625                         rte_be_to_cpu_32(ipv4_hdr->dst_addr), &next_hop) == 0)?
 626                         next_hop : portid);
 627 }
 628 #endif
 629
 630 static inline void
 631 l3fwd_simple_forward(struct rte_mbuf *m, uint8_t portid,
 632                                 struct lcore_conf *qconf)
 633 {
 634         struct ether_hdr *eth_hdr;
 635         struct ipv4_hdr *ipv4_hdr;
 636         void *d_addr_bytes;
 637         uint8_t dst_port;
 638
 639         eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
 640
 641         if (m->ol_flags & PKT_RX_IPV4_HDR) {
 642                 /* Handle IPv4 headers.*/
 643                 ipv4_hdr =
 644                         (struct ipv4_hdr *)(rte_pktmbuf_mtod(m, unsigned char*)
 645                                                 + sizeof(struct ether_hdr));
 646
 647 #ifdef DO_RFC_1812_CHECKS
 648                 /* Check to make sure the packet is valid (RFC1812) */
 649                 if (is_valid_ipv4_pkt(ipv4_hdr, m->pkt_len) < 0) {
 650                         rte_pktmbuf_free(m);
 651                         return;
 652                 }
 653 #endif
 654
 655                 dst_port = get_ipv4_dst_port(ipv4_hdr, portid,
 656                                         qconf->ipv4_lookup_struct);
 657                 if (dst_port >= RTE_MAX_ETHPORTS ||
 658                                 (enabled_port_mask & 1 << dst_port) == 0)
 659                         dst_port = portid;
 660
 661                 /* 02:00:00:00:00:xx */
 662                 d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
 663                 *((uint64_t *)d_addr_bytes) =
 664                         0x000000000002 + ((uint64_t)dst_port << 40);
 665
 666 #ifdef DO_RFC_1812_CHECKS
 667                 /* Update time to live and header checksum */
 668                 --(ipv4_hdr->time_to_live);
 669                 ++(ipv4_hdr->hdr_checksum);
 670 #endif
 671
 672                 /* src addr */
 673                 ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
 674
 675                 send_single_packet(m, dst_port);
 676         }
 677         else {
 678                 /* Handle IPv6 headers.*/
 679 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 680                 struct ipv6_hdr *ipv6_hdr;
 681
 682                 ipv6_hdr =
 683                         (struct ipv6_hdr *)(rte_pktmbuf_mtod(m, unsigned char*)
 684                                                 + sizeof(struct ether_hdr));
 685
 686                 dst_port = get_ipv6_dst_port(ipv6_hdr, portid,
 687                                         qconf->ipv6_lookup_struct);
 688
 689                 if (dst_port >= RTE_MAX_ETHPORTS ||
 690                                 (enabled_port_mask & 1 << dst_port) == 0)
 691                         dst_port = portid;
 692
 693                 /* 02:00:00:00:00:xx */
 694                 d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
 695                 *((uint64_t *)d_addr_bytes) =
 696                         0x000000000002 + ((uint64_t)dst_port << 40);
 697
 698                 /* src addr */
 699                 ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
 700
 701                 send_single_packet(m, dst_port);
 702 #else
 703                 /* We don't currently handle IPv6 packets in LPM mode. */
 704                 rte_pktmbuf_free(m);
 705 #endif
 706         }
 707
 708 }
 709
 710 #define SLEEP_GEAR1_THRESHOLD            100
 711 #define SLEEP_GEAR2_THRESHOLD            1000
 712
 713 static inline uint32_t
 714 power_idle_heuristic(uint32_t zero_rx_packet_count)
 715 {
 716         /* If zero count is less than 100, use it as the sleep time in us */
 717         if (zero_rx_packet_count < SLEEP_GEAR1_THRESHOLD)
 718                 return zero_rx_packet_count;
 719         /* If zero count is less than 1000, sleep time should be 100 us */
 720         else if ((zero_rx_packet_count >= SLEEP_GEAR1_THRESHOLD) &&
 721                         (zero_rx_packet_count < SLEEP_GEAR2_THRESHOLD))
 722                 return SLEEP_GEAR1_THRESHOLD;
 723         /* If zero count is greater than 1000, sleep time should be 1000 us */
 724         else if (zero_rx_packet_count >= SLEEP_GEAR2_THRESHOLD)
 725                 return SLEEP_GEAR2_THRESHOLD;
 726
 727         return 0;
 728 }
 729
 730 static inline enum freq_scale_hint_t
 731 power_freq_scaleup_heuristic(unsigned lcore_id,
 732                              uint8_t port_id,
 733                              uint16_t queue_id)
 734 {
 735 /**
 736  * HW Rx queue size is 128 by default, Rx burst read at maximum 32 entries
 737  * per iteration
 738  */
 739 #define FREQ_GEAR1_RX_PACKET_THRESHOLD             MAX_PKT_BURST
 740 #define FREQ_GEAR2_RX_PACKET_THRESHOLD             (MAX_PKT_BURST*2)
 741 #define FREQ_GEAR3_RX_PACKET_THRESHOLD             (MAX_PKT_BURST*3)
 742 #define FREQ_UP_TREND1_ACC   1
 743 #define FREQ_UP_TREND2_ACC   100
 744 #define FREQ_UP_THRESHOLD    10000
 745
 746         if (likely(rte_eth_rx_descriptor_done(port_id, queue_id,
 747                         FREQ_GEAR3_RX_PACKET_THRESHOLD) > 0)) {
 748                 stats[lcore_id].trend = 0;
 749                 return FREQ_HIGHEST;
 750         } else if (likely(rte_eth_rx_descriptor_done(port_id, queue_id,
 751                         FREQ_GEAR2_RX_PACKET_THRESHOLD) > 0))
 752                 stats[lcore_id].trend += FREQ_UP_TREND2_ACC;
 753         else if (likely(rte_eth_rx_descriptor_done(port_id, queue_id,
 754                         FREQ_GEAR1_RX_PACKET_THRESHOLD) > 0))
 755                 stats[lcore_id].trend += FREQ_UP_TREND1_ACC;
 756
 757         if (likely(stats[lcore_id].trend > FREQ_UP_THRESHOLD)) {
 758                 stats[lcore_id].trend = 0;
 759                 return FREQ_HIGHER;
 760         }
 761
 762         return FREQ_CURRENT;
 763 }
 764
 765 /* main processing loop */
 766 static int
 767 main_loop(__attribute__((unused)) void *dummy)
 768 {
 769         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
 770         unsigned lcore_id;
 771         uint64_t prev_tsc, diff_tsc, cur_tsc;
 772         uint64_t prev_tsc_power = 0, cur_tsc_power, diff_tsc_power;
 773         int i, j, nb_rx;
 774         uint8_t portid, queueid;
 775         struct lcore_conf *qconf;
 776         struct lcore_rx_queue *rx_queue;
 777         enum freq_scale_hint_t lcore_scaleup_hint;
 778
 779         uint32_t lcore_rx_idle_count = 0;
 780         uint32_t lcore_idle_hint = 0;
 781
 782         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
 783
 784         prev_tsc = 0;
 785
 786         lcore_id = rte_lcore_id();
 787         qconf = &lcore_conf[lcore_id];
 788
 789         if (qconf->n_rx_queue == 0) {
 790                 RTE_LOG(INFO, L3FWD_POWER, "lcore %u has nothing to do\n", lcore_id);
 791                 return 0;
 792         }
 793
 794         RTE_LOG(INFO, L3FWD_POWER, "entering main loop on lcore %u\n", lcore_id);
 795
 796         for (i = 0; i < qconf->n_rx_queue; i++) {
 797
 798                 portid = qconf->rx_queue_list[i].port_id;
 799                 queueid = qconf->rx_queue_list[i].queue_id;
 800                 RTE_LOG(INFO, L3FWD_POWER, " -- lcoreid=%u portid=%hhu "
 801                         "rxqueueid=%hhu\n", lcore_id, portid, queueid);
 802         }
 803
 804         while (1) {
 805                 stats[lcore_id].nb_iteration_looped++;
 806
 807                 cur_tsc = rte_rdtsc();
 808                 cur_tsc_power = cur_tsc;
 809
 810                 /*
 811                  * TX burst queue drain
 812                  */
 813                 diff_tsc = cur_tsc - prev_tsc;
 814                 if (unlikely(diff_tsc > drain_tsc)) {
 815
 816                         /*
 817                          * This could be optimized (use queueid instead of
 818                          * portid), but it is not called so often
 819                          */
 820                         for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
 821                                 if (qconf->tx_mbufs[portid].len == 0)
 822                                         continue;
 823                                 send_burst(&lcore_conf[lcore_id],
 824                                         qconf->tx_mbufs[portid].len,
 825                                         portid);
 826                                 qconf->tx_mbufs[portid].len = 0;
 827                         }
 828
 829                         prev_tsc = cur_tsc;
 830                 }
 831
 832                 diff_tsc_power = cur_tsc_power - prev_tsc_power;
 833                 if (diff_tsc_power > TIMER_RESOLUTION_CYCLES) {
 834                         rte_timer_manage();
 835                         prev_tsc_power = cur_tsc_power;
 836                 }
 837
 838                 /*
 839                  * Read packet from RX queues
 840                  */
 841                 lcore_scaleup_hint = FREQ_CURRENT;
 842                 lcore_rx_idle_count = 0;
 843                 for (i = 0; i < qconf->n_rx_queue; ++i) {
 844                         rx_queue = &(qconf->rx_queue_list[i]);
 845                         rx_queue->idle_hint = 0;
 846                         portid = rx_queue->port_id;
 847                         queueid = rx_queue->queue_id;
 848
 849                         nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
 850                                                                 MAX_PKT_BURST);
 851                         stats[lcore_id].nb_rx_processed += nb_rx;
 852                         if (unlikely(nb_rx == 0)) {
 853                                 /**
 854                                  * no packet received from rx queue, try to
 855                                  * sleep for a while forcing CPU enter deeper
 856                                  * C states.
 857                                  */
 858                                 rx_queue->zero_rx_packet_count++;
 859
 860                                 if (rx_queue->zero_rx_packet_count <=
 861                                                         MIN_ZERO_POLL_COUNT)
 862                                         continue;
 863
 864                                 rx_queue->idle_hint = power_idle_heuristic(\
 865                                         rx_queue->zero_rx_packet_count);
 866                                 lcore_rx_idle_count++;
 867                         } else {
 868                                 rx_queue->zero_rx_packet_count = 0;
 869
 870                                 /**
 871                                  * do not scale up frequency immediately as
 872                                  * user to kernel space communication is costly
 873                                  * which might impact packet I/O for received
 874                                  * packets.
 875                                  */
 876                                 rx_queue->freq_up_hint =
 877                                         power_freq_scaleup_heuristic(lcore_id,
 878                                                         portid, queueid);
 879                         }
 880
 881                         /* Prefetch first packets */
 882                         for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
 883                                 rte_prefetch0(rte_pktmbuf_mtod(
 884                                                 pkts_burst[j], void *));
 885                         }
 886
 887                         /* Prefetch and forward already prefetched packets */
 888                         for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
 889                                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
 890                                                 j + PREFETCH_OFFSET], void *));
 891                                 l3fwd_simple_forward(pkts_burst[j], portid,
 892                                                                 qconf);
 893                         }
 894
 895                         /* Forward remaining prefetched packets */
 896                         for (; j < nb_rx; j++) {
 897                                 l3fwd_simple_forward(pkts_burst[j], portid,
 898                                                                 qconf);
 899                         }
 900                 }
 901
 902                 if (likely(lcore_rx_idle_count != qconf->n_rx_queue)) {
 903                         for (i = 1, lcore_scaleup_hint =
 904                                 qconf->rx_queue_list[0].freq_up_hint;
 905                                         i < qconf->n_rx_queue; ++i) {
 906                                 rx_queue = &(qconf->rx_queue_list[i]);
 907                                 if (rx_queue->freq_up_hint >
 908                                                 lcore_scaleup_hint)
 909                                         lcore_scaleup_hint =
 910                                                 rx_queue->freq_up_hint;
 911                         }
 912
 913                         if (lcore_scaleup_hint == FREQ_HIGHEST)
 914                                 rte_power_freq_max(lcore_id);
 915                         else if (lcore_scaleup_hint == FREQ_HIGHER)
 916                                 rte_power_freq_up(lcore_id);
 917                 } else {
 918                         /**
 919                          * All Rx queues empty in recent consecutive polls,
 920                          * sleep in a conservative manner, meaning sleep as
 921                          * less as possible.
 922                          */
 923                         for (i = 1, lcore_idle_hint =
 924                                 qconf->rx_queue_list[0].idle_hint;
 925                                         i < qconf->n_rx_queue; ++i) {
 926                                 rx_queue = &(qconf->rx_queue_list[i]);
 927                                 if (rx_queue->idle_hint < lcore_idle_hint)
 928                                         lcore_idle_hint = rx_queue->idle_hint;
 929                         }
 930
 931                         if ( lcore_idle_hint < SLEEP_GEAR1_THRESHOLD)
 932                                 /**
 933                                  * execute "pause" instruction to avoid context
 934                                  * switch for short sleep.
 935                                  */
 936                                 rte_delay_us(lcore_idle_hint);
 937                         else
 938                                 /* long sleep force runing thread to suspend */
 939                                 usleep(lcore_idle_hint);
 940
 941                         stats[lcore_id].sleep_time += lcore_idle_hint;
 942                 }
 943         }
 944 }
 945
 946 static int
 947 check_lcore_params(void)
 948 {
 949         uint8_t queue, lcore;
 950         uint16_t i;
 951         int socketid;
 952
 953         for (i = 0; i < nb_lcore_params; ++i) {
 954                 queue = lcore_params[i].queue_id;
 955                 if (queue >= MAX_RX_QUEUE_PER_PORT) {
 956                         printf("invalid queue number: %hhu\n", queue);
 957                         return -1;
 958                 }
 959                 lcore = lcore_params[i].lcore_id;
 960                 if (!rte_lcore_is_enabled(lcore)) {
 961                         printf("error: lcore %hhu is not enabled in lcore "
 962                                                         "mask\n", lcore);
 963                         return -1;
 964                 }
 965                 if ((socketid = rte_lcore_to_socket_id(lcore) != 0) &&
 966                                                         (numa_on == 0)) {
 967                         printf("warning: lcore %hhu is on socket %d with numa "
 968                                                 "off\n", lcore, socketid);
 969                 }
 970         }
 971         return 0;
 972 }
 973
 974 static int
 975 check_port_config(const unsigned nb_ports)
 976 {
 977         unsigned portid;
 978         uint16_t i;
 979
 980         for (i = 0; i < nb_lcore_params; ++i) {
 981                 portid = lcore_params[i].port_id;
 982                 if ((enabled_port_mask & (1 << portid)) == 0) {
 983                         printf("port %u is not enabled in port mask\n",
 984                                                                 portid);
 985                         return -1;
 986                 }
 987                 if (portid >= nb_ports) {
 988                         printf("port %u is not present on the board\n",
 989                                                                 portid);
 990                         return -1;
 991                 }
 992         }
 993         return 0;
 994 }
 995
 996 static uint8_t
 997 get_port_n_rx_queues(const uint8_t port)
 998 {
 999         int queue = -1;
1000         uint16_t i;
1001
1002         for (i = 0; i < nb_lcore_params; ++i) {
1003                 if (lcore_params[i].port_id == port &&
1004                                 lcore_params[i].queue_id > queue)
1005                         queue = lcore_params[i].queue_id;
1006         }
1007         return (uint8_t)(++queue);
1008 }
1009
1010 static int
1011 init_lcore_rx_queues(void)
1012 {
1013         uint16_t i, nb_rx_queue;
1014         uint8_t lcore;
1015
1016         for (i = 0; i < nb_lcore_params; ++i) {
1017                 lcore = lcore_params[i].lcore_id;
1018                 nb_rx_queue = lcore_conf[lcore].n_rx_queue;
1019                 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
1020                         printf("error: too many queues (%u) for lcore: %u\n",
1021                                 (unsigned)nb_rx_queue + 1, (unsigned)lcore);
1022                         return -1;
1023                 } else {
1024                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
1025                                 lcore_params[i].port_id;
1026                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
1027                                 lcore_params[i].queue_id;
1028                         lcore_conf[lcore].n_rx_queue++;
1029                 }
1030         }
1031         return 0;
1032 }
1033
1034 /* display usage */
1035 static void
1036 print_usage(const char *prgname)
1037 {
1038         printf ("%s [EAL options] -- -p PORTMASK -P"
1039                 "  [--config (port,queue,lcore)[,(port,queue,lcore]]"
1040                 "  [--enable-jumbo [--max-pkt-len PKTLEN]]\n"
1041                 "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
1042                 "  -P : enable promiscuous mode\n"
1043                 "  --config (port,queue,lcore): rx queues configuration\n"
1044                 "  --no-numa: optional, disable numa awareness\n"
1045                 "  --enable-jumbo: enable jumbo frame"
1046                 " which max packet len is PKTLEN in decimal (64-9600)\n",
1047                 prgname);
1048 }
1049
1050 static int parse_max_pkt_len(const char *pktlen)
1051 {
1052         char *end = NULL;
1053         unsigned long len;
1054
1055         /* parse decimal string */
1056         len = strtoul(pktlen, &end, 10);
1057         if ((pktlen[0] == '\0') || (end == NULL) || (*end != '\0'))
1058                 return -1;
1059
1060         if (len == 0)
1061                 return -1;
1062
1063         return len;
1064 }
1065
1066 static int
1067 parse_portmask(const char *portmask)
1068 {
1069         char *end = NULL;
1070         unsigned long pm;
1071
1072         /* parse hexadecimal string */
1073         pm = strtoul(portmask, &end, 16);
1074         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
1075                 return -1;
1076
1077         if (pm == 0)
1078                 return -1;
1079
1080         return pm;
1081 }
1082
1083 static int
1084 parse_config(const char *q_arg)
1085 {
1086         char s[256];
1087         const char *p, *p0 = q_arg;
1088         char *end;
1089         enum fieldnames {
1090                 FLD_PORT = 0,
1091                 FLD_QUEUE,
1092                 FLD_LCORE,
1093                 _NUM_FLD
1094         };
1095         unsigned long int_fld[_NUM_FLD];
1096         char *str_fld[_NUM_FLD];
1097         int i;
1098         unsigned size;
1099
1100         nb_lcore_params = 0;
1101
1102         while ((p = strchr(p0,'(')) != NULL) {
1103                 ++p;
1104                 if((p0 = strchr(p,')')) == NULL)
1105                         return -1;
1106
1107                 size = p0 - p;
1108                 if(size >= sizeof(s))
1109                         return -1;
1110
1111                 snprintf(s, sizeof(s), "%.*s", size, p);
1112                 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') !=
1113                                                                 _NUM_FLD)
1114                         return -1;
1115                 for (i = 0; i < _NUM_FLD; i++){
1116                         errno = 0;
1117                         int_fld[i] = strtoul(str_fld[i], &end, 0);
1118                         if (errno != 0 || end == str_fld[i] || int_fld[i] >
1119                                                                         255)
1120                                 return -1;
1121                 }
1122                 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
1123                         printf("exceeded max number of lcore params: %hu\n",
1124                                 nb_lcore_params);
1125                         return -1;
1126                 }
1127                 lcore_params_array[nb_lcore_params].port_id =
1128                                 (uint8_t)int_fld[FLD_PORT];
1129                 lcore_params_array[nb_lcore_params].queue_id =
1130                                 (uint8_t)int_fld[FLD_QUEUE];
1131                 lcore_params_array[nb_lcore_params].lcore_id =
1132                                 (uint8_t)int_fld[FLD_LCORE];
1133                 ++nb_lcore_params;
1134         }
1135         lcore_params = lcore_params_array;
1136
1137         return 0;
1138 }
1139
1140 /* Parse the argument given in the command line of the application */
1141 static int
1142 parse_args(int argc, char **argv)
1143 {
1144         int opt, ret;
1145         char **argvopt;
1146         int option_index;
1147         char *prgname = argv[0];
1148         static struct option lgopts[] = {
1149                 {"config", 1, 0, 0},
1150                 {"no-numa", 0, 0, 0},
1151                 {"enable-jumbo", 0, 0, 0},
1152                 {NULL, 0, 0, 0}
1153         };
1154
1155         argvopt = argv;
1156
1157         while ((opt = getopt_long(argc, argvopt, "p:P",
1158                                 lgopts, &option_index)) != EOF) {
1159
1160                 switch (opt) {
1161                 /* portmask */
1162                 case 'p':
1163                         enabled_port_mask = parse_portmask(optarg);
1164                         if (enabled_port_mask == 0) {
1165                                 printf("invalid portmask\n");
1166                                 print_usage(prgname);
1167                                 return -1;
1168                         }
1169                         break;
1170                 case 'P':
1171                         printf("Promiscuous mode selected\n");
1172                         promiscuous_on = 1;
1173                         break;
1174
1175                 /* long options */
1176                 case 0:
1177                         if (!strncmp(lgopts[option_index].name, "config", 6)) {
1178                                 ret = parse_config(optarg);
1179                                 if (ret) {
1180                                         printf("invalid config\n");
1181                                         print_usage(prgname);
1182                                         return -1;
1183                                 }
1184                         }
1185
1186                         if (!strncmp(lgopts[option_index].name,
1187                                                 "no-numa", 7)) {
1188                                 printf("numa is disabled \n");
1189                                 numa_on = 0;
1190                         }
1191
1192                         if (!strncmp(lgopts[option_index].name,
1193                                         "enable-jumbo", 12)) {
1194                                 struct option lenopts =
1195                                         {"max-pkt-len", required_argument, \
1196                                                                         0, 0};
1197
1198                                 printf("jumbo frame is enabled \n");
1199                                 port_conf.rxmode.jumbo_frame = 1;
1200
1201                                 /**
1202                                  * if no max-pkt-len set, use the default value
1203                                  * ETHER_MAX_LEN
1204                                  */
1205                                 if (0 == getopt_long(argc, argvopt, "",
1206                                                 &lenopts, &option_index)) {
1207                                         ret = parse_max_pkt_len(optarg);
1208                                         if ((ret < 64) ||
1209                                                 (ret > MAX_JUMBO_PKT_LEN)){
1210                                                 printf("invalid packet "
1211                                                                 "length\n");
1212                                                 print_usage(prgname);
1213                                                 return -1;
1214                                         }
1215                                         port_conf.rxmode.max_rx_pkt_len = ret;
1216                                 }
1217                                 printf("set jumbo frame "
1218                                         "max packet length to %u\n",
1219                                 (unsigned int)port_conf.rxmode.max_rx_pkt_len);
1220                         }
1221
1222                         break;
1223
1224                 default:
1225                         print_usage(prgname);
1226                         return -1;
1227                 }
1228         }
1229
1230         if (optind >= 0)
1231                 argv[optind-1] = prgname;
1232
1233         ret = optind-1;
1234         optind = 0; /* reset getopt lib */
1235         return ret;
1236 }
1237
1238 static void
1239 print_ethaddr(const char *name, const struct ether_addr *eth_addr)
1240 {
1241         char buf[ETHER_ADDR_FMT_SIZE];
1242         ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr);
1243         printf("%s%s", name, buf);
1244 }
1245
1246 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1247 static void
1248 setup_hash(int socketid)
1249 {
1250         struct rte_hash_parameters ipv4_l3fwd_hash_params = {
1251                 .name = NULL,
1252                 .entries = L3FWD_HASH_ENTRIES,
1253                 .bucket_entries = 4,
1254                 .key_len = sizeof(struct ipv4_5tuple),
1255                 .hash_func = DEFAULT_HASH_FUNC,
1256                 .hash_func_init_val = 0,
1257         };
1258
1259         struct rte_hash_parameters ipv6_l3fwd_hash_params = {
1260                 .name = NULL,
1261                 .entries = L3FWD_HASH_ENTRIES,
1262                 .bucket_entries = 4,
1263                 .key_len = sizeof(struct ipv6_5tuple),
1264                 .hash_func = DEFAULT_HASH_FUNC,
1265                 .hash_func_init_val = 0,
1266         };
1267
1268         unsigned i;
1269         int ret;
1270         char s[64];
1271
1272         /* create ipv4 hash */
1273         snprintf(s, sizeof(s), "ipv4_l3fwd_hash_%d", socketid);
1274         ipv4_l3fwd_hash_params.name = s;
1275         ipv4_l3fwd_hash_params.socket_id = socketid;
1276         ipv4_l3fwd_lookup_struct[socketid] =
1277                 rte_hash_create(&ipv4_l3fwd_hash_params);
1278         if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
1279                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
1280                                 "socket %d\n", socketid);
1281
1282         /* create ipv6 hash */
1283         snprintf(s, sizeof(s), "ipv6_l3fwd_hash_%d", socketid);
1284         ipv6_l3fwd_hash_params.name = s;
1285         ipv6_l3fwd_hash_params.socket_id = socketid;
1286         ipv6_l3fwd_lookup_struct[socketid] =
1287                 rte_hash_create(&ipv6_l3fwd_hash_params);
1288         if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
1289                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
1290                                 "socket %d\n", socketid);
1291
1292
1293         /* populate the ipv4 hash */
1294         for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
1295                 ret = rte_hash_add_key (ipv4_l3fwd_lookup_struct[socketid],
1296                                 (void *) &ipv4_l3fwd_route_array[i].key);
1297                 if (ret < 0) {
1298                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
1299                                 "l3fwd hash on socket %d\n", i, socketid);
1300                 }
1301                 ipv4_l3fwd_out_if[ret] = ipv4_l3fwd_route_array[i].if_out;
1302                 printf("Hash: Adding key\n");
1303                 print_ipv4_key(ipv4_l3fwd_route_array[i].key);
1304         }
1305
1306         /* populate the ipv6 hash */
1307         for (i = 0; i < IPV6_L3FWD_NUM_ROUTES; i++) {
1308                 ret = rte_hash_add_key (ipv6_l3fwd_lookup_struct[socketid],
1309                                 (void *) &ipv6_l3fwd_route_array[i].key);
1310                 if (ret < 0) {
1311                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
1312                                 "l3fwd hash on socket %d\n", i, socketid);
1313                 }
1314                 ipv6_l3fwd_out_if[ret] = ipv6_l3fwd_route_array[i].if_out;
1315                 printf("Hash: Adding key\n");
1316                 print_ipv6_key(ipv6_l3fwd_route_array[i].key);
1317         }
1318 }
1319 #endif
1320
1321 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1322 static void
1323 setup_lpm(int socketid)
1324 {
1325         unsigned i;
1326         int ret;
1327         char s[64];
1328
1329         /* create the LPM table */
1330         snprintf(s, sizeof(s), "IPV4_L3FWD_LPM_%d", socketid);
1331         ipv4_l3fwd_lookup_struct[socketid] = rte_lpm_create(s, socketid,
1332                                 IPV4_L3FWD_LPM_MAX_RULES, 0);
1333         if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
1334                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
1335                                 " on socket %d\n", socketid);
1336
1337         /* populate the LPM table */
1338         for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
1339                 ret = rte_lpm_add(ipv4_l3fwd_lookup_struct[socketid],
1340                         ipv4_l3fwd_route_array[i].ip,
1341                         ipv4_l3fwd_route_array[i].depth,
1342                         ipv4_l3fwd_route_array[i].if_out);
1343
1344                 if (ret < 0) {
1345                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
1346                                 "l3fwd LPM table on socket %d\n",
1347                                 i, socketid);
1348                 }
1349
1350                 printf("LPM: Adding route 0x%08x / %d (%d)\n",
1351                         (unsigned)ipv4_l3fwd_route_array[i].ip,
1352                         ipv4_l3fwd_route_array[i].depth,
1353                         ipv4_l3fwd_route_array[i].if_out);
1354         }
1355 }
1356 #endif
1357
1358 static int
1359 init_mem(unsigned nb_mbuf)
1360 {
1361         struct lcore_conf *qconf;
1362         int socketid;
1363         unsigned lcore_id;
1364         char s[64];
1365
1366         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1367                 if (rte_lcore_is_enabled(lcore_id) == 0)
1368                         continue;
1369
1370                 if (numa_on)
1371                         socketid = rte_lcore_to_socket_id(lcore_id);
1372                 else
1373                         socketid = 0;
1374
1375                 if (socketid >= NB_SOCKETS) {
1376                         rte_exit(EXIT_FAILURE, "Socket %d of lcore %u is "
1377                                         "out of range %d\n", socketid,
1378                                                 lcore_id, NB_SOCKETS);
1379                 }
1380                 if (pktmbuf_pool[socketid] == NULL) {
1381                         snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
1382                         pktmbuf_pool[socketid] =
1383                                 rte_mempool_create(s, nb_mbuf,
1384                                         MBUF_SIZE, MEMPOOL_CACHE_SIZE,
1385                                         sizeof(struct rte_pktmbuf_pool_private),
1386                                         rte_pktmbuf_pool_init, NULL,
1387                                         rte_pktmbuf_init, NULL,
1388                                         socketid, 0);
1389                         if (pktmbuf_pool[socketid] == NULL)
1390                                 rte_exit(EXIT_FAILURE,
1391                                         "Cannot init mbuf pool on socket %d\n",
1392                                                                 socketid);
1393                         else
1394                                 printf("Allocated mbuf pool on socket %d\n",
1395                                                                 socketid);
1396
1397 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1398                         setup_lpm(socketid);
1399 #else
1400                         setup_hash(socketid);
1401 #endif
1402                 }
1403                 qconf = &lcore_conf[lcore_id];
1404                 qconf->ipv4_lookup_struct = ipv4_l3fwd_lookup_struct[socketid];
1405 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1406                 qconf->ipv6_lookup_struct = ipv6_l3fwd_lookup_struct[socketid];
1407 #endif
1408         }
1409         return 0;
1410 }
1411
1412 /* Check the link status of all ports in up to 9s, and print them finally */
1413 static void
1414 check_all_ports_link_status(uint8_t port_num, uint32_t port_mask)
1415 {
1416 #define CHECK_INTERVAL 100 /* 100ms */
1417 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
1418         uint8_t portid, count, all_ports_up, print_flag = 0;
1419         struct rte_eth_link link;
1420
1421         printf("\nChecking link status");
1422         fflush(stdout);
1423         for (count = 0; count <= MAX_CHECK_TIME; count++) {
1424                 all_ports_up = 1;
1425                 for (portid = 0; portid < port_num; portid++) {
1426                         if ((port_mask & (1 << portid)) == 0)
1427                                 continue;
1428                         memset(&link, 0, sizeof(link));
1429                         rte_eth_link_get_nowait(portid, &link);
1430                         /* print link status if flag set */
1431                         if (print_flag == 1) {
1432                                 if (link.link_status)
1433                                         printf("Port %d Link Up - speed %u "
1434                                                 "Mbps - %s\n", (uint8_t)portid,
1435                                                 (unsigned)link.link_speed,
1436                                 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
1437                                         ("full-duplex") : ("half-duplex\n"));
1438                                 else
1439                                         printf("Port %d Link Down\n",
1440                                                 (uint8_t)portid);
1441                                 continue;
1442                         }
1443                         /* clear all_ports_up flag if any link down */
1444                         if (link.link_status == 0) {
1445                                 all_ports_up = 0;
1446                                 break;
1447                         }
1448                 }
1449                 /* after finally printing all link status, get out */
1450                 if (print_flag == 1)
1451                         break;
1452
1453                 if (all_ports_up == 0) {
1454                         printf(".");
1455                         fflush(stdout);
1456                         rte_delay_ms(CHECK_INTERVAL);
1457                 }
1458
1459                 /* set the print_flag if all ports up or timeout */
1460                 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
1461                         print_flag = 1;
1462                         printf("done\n");
1463                 }
1464         }
1465 }
1466
1467 int
1468 main(int argc, char **argv)
1469 {
1470         struct lcore_conf *qconf;
1471         struct rte_eth_dev_info dev_info;
1472         struct rte_eth_txconf *txconf;
1473         int ret;
1474         unsigned nb_ports;
1475         uint16_t queueid;
1476         unsigned lcore_id;
1477         uint64_t hz;
1478         uint32_t n_tx_queue, nb_lcores;
1479         uint8_t portid, nb_rx_queue, queue, socketid;
1480
1481         /* catch SIGINT and restore cpufreq governor to ondemand */
1482         signal(SIGINT, signal_exit_now);
1483
1484         /* init EAL */
1485         ret = rte_eal_init(argc, argv);
1486         if (ret < 0)
1487                 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
1488         argc -= ret;
1489         argv += ret;
1490
1491         /* init RTE timer library to be used late */
1492         rte_timer_subsystem_init();
1493
1494         /* parse application arguments (after the EAL ones) */
1495         ret = parse_args(argc, argv);
1496         if (ret < 0)
1497                 rte_exit(EXIT_FAILURE, "Invalid L3FWD parameters\n");
1498
1499         if (check_lcore_params() < 0)
1500                 rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
1501
1502         ret = init_lcore_rx_queues();
1503         if (ret < 0)
1504                 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
1505
1506
1507         nb_ports = rte_eth_dev_count();
1508         if (nb_ports > RTE_MAX_ETHPORTS)
1509                 nb_ports = RTE_MAX_ETHPORTS;
1510
1511         if (check_port_config(nb_ports) < 0)
1512                 rte_exit(EXIT_FAILURE, "check_port_config failed\n");
1513
1514         nb_lcores = rte_lcore_count();
1515
1516         /* initialize all ports */
1517         for (portid = 0; portid < nb_ports; portid++) {
1518                 /* skip ports that are not enabled */
1519                 if ((enabled_port_mask & (1 << portid)) == 0) {
1520                         printf("\nSkipping disabled port %d\n", portid);
1521                         continue;
1522                 }
1523
1524                 /* init port */
1525                 printf("Initializing port %d ... ", portid );
1526                 fflush(stdout);
1527
1528                 nb_rx_queue = get_port_n_rx_queues(portid);
1529                 n_tx_queue = nb_lcores;
1530                 if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
1531                         n_tx_queue = MAX_TX_QUEUE_PER_PORT;
1532                 printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
1533                         nb_rx_queue, (unsigned)n_tx_queue );
1534                 ret = rte_eth_dev_configure(portid, nb_rx_queue,
1535                                         (uint16_t)n_tx_queue, &port_conf);
1536                 if (ret < 0)
1537                         rte_exit(EXIT_FAILURE, "Cannot configure device: "
1538                                         "err=%d, port=%d\n", ret, portid);
1539
1540                 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
1541                 print_ethaddr(" Address:", &ports_eth_addr[portid]);
1542                 printf(", ");
1543
1544                 /* init memory */
1545                 ret = init_mem(NB_MBUF);
1546                 if (ret < 0)
1547                         rte_exit(EXIT_FAILURE, "init_mem failed\n");
1548
1549                 /* init one TX queue per couple (lcore,port) */
1550                 queueid = 0;
1551                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1552                         if (rte_lcore_is_enabled(lcore_id) == 0)
1553                                 continue;
1554
1555                         if (numa_on)
1556                                 socketid = \
1557                                 (uint8_t)rte_lcore_to_socket_id(lcore_id);
1558                         else
1559                                 socketid = 0;
1560
1561                         printf("txq=%u,%d,%d ", lcore_id, queueid, socketid);
1562                         fflush(stdout);
1563
1564                         rte_eth_dev_info_get(portid, &dev_info);
1565                         txconf = &dev_info.default_txconf;
1566                         if (port_conf.rxmode.jumbo_frame)
1567                                 txconf->txq_flags = 0;
1568                         ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
1569                                                      socketid, txconf);
1570                         if (ret < 0)
1571                                 rte_exit(EXIT_FAILURE,
1572                                         "rte_eth_tx_queue_setup: err=%d, "
1573                                                 "port=%d\n", ret, portid);
1574
1575                         qconf = &lcore_conf[lcore_id];
1576                         qconf->tx_queue_id[portid] = queueid;
1577                         queueid++;
1578                 }
1579                 printf("\n");
1580         }
1581
1582         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1583                 if (rte_lcore_is_enabled(lcore_id) == 0)
1584                         continue;
1585
1586                 /* init power management library */
1587                 ret = rte_power_init(lcore_id);
1588                 if (ret)
1589                         rte_exit(EXIT_FAILURE, "Power management library "
1590                                 "initialization failed on core%u\n", lcore_id);
1591
1592                 /* init timer structures for each enabled lcore */
1593                 rte_timer_init(&power_timers[lcore_id]);
1594                 hz = rte_get_timer_hz();
1595                 rte_timer_reset(&power_timers[lcore_id],
1596                         hz/TIMER_NUMBER_PER_SECOND, SINGLE, lcore_id,
1597                                                 power_timer_cb, NULL);
1598
1599                 qconf = &lcore_conf[lcore_id];
1600                 printf("\nInitializing rx queues on lcore %u ... ", lcore_id );
1601                 fflush(stdout);
1602                 /* init RX queues */
1603                 for(queue = 0; queue < qconf->n_rx_queue; ++queue) {
1604                         portid = qconf->rx_queue_list[queue].port_id;
1605                         queueid = qconf->rx_queue_list[queue].queue_id;
1606
1607                         if (numa_on)
1608                                 socketid = \
1609                                 (uint8_t)rte_lcore_to_socket_id(lcore_id);
1610                         else
1611                                 socketid = 0;
1612
1613                         printf("rxq=%d,%d,%d ", portid, queueid, socketid);
1614                         fflush(stdout);
1615
1616                         ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
1617                                 socketid, NULL,
1618                                 pktmbuf_pool[socketid]);
1619                         if (ret < 0)
1620                                 rte_exit(EXIT_FAILURE,
1621                                         "rte_eth_rx_queue_setup: err=%d, "
1622                                                 "port=%d\n", ret, portid);
1623                 }
1624         }
1625
1626         printf("\n");
1627
1628         /* start ports */
1629         for (portid = 0; portid < nb_ports; portid++) {
1630                 if ((enabled_port_mask & (1 << portid)) == 0) {
1631                         continue;
1632                 }
1633                 /* Start device */
1634                 ret = rte_eth_dev_start(portid);
1635                 if (ret < 0)
1636                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, "
1637                                                 "port=%d\n", ret, portid);
1638
1639                 /*
1640                  * If enabled, put device in promiscuous mode.
1641                  * This allows IO forwarding mode to forward packets
1642                  * to itself through 2 cross-connected  ports of the
1643                  * target machine.
1644                  */
1645                 if (promiscuous_on)
1646                         rte_eth_promiscuous_enable(portid);
1647         }
1648
1649         check_all_ports_link_status((uint8_t)nb_ports, enabled_port_mask);
1650
1651         /* launch per-lcore init on every lcore */
1652         rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
1653         RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1654                 if (rte_eal_wait_lcore(lcore_id) < 0)
1655                         return -1;
1656         }
1657
1658         return 0;
1659 }