examples/l3fwd-power/main.c

   1 /*-
   2  *   BSD LICENSE
   3  *
   4  *   Copyright(c) 2010-2013 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
  35 #include <stdio.h>
  36 #include <stdlib.h>
  37 #include <stdint.h>
  38 #include <inttypes.h>
  39 #include <sys/types.h>
  40 #include <string.h>
  41 #include <sys/queue.h>
  42 #include <stdarg.h>
  43 #include <errno.h>
  44 #include <getopt.h>
  45 #include <unistd.h>
  46 #include <signal.h>
  47
  48 #include <rte_common.h>
  49 #include <rte_byteorder.h>
  50 #include <rte_log.h>
  51 #include <rte_memory.h>
  52 #include <rte_memcpy.h>
  53 #include <rte_memzone.h>
  54 #include <rte_tailq.h>
  55 #include <rte_eal.h>
  56 #include <rte_per_lcore.h>
  57 #include <rte_launch.h>
  58 #include <rte_atomic.h>
  59 #include <rte_cycles.h>
  60 #include <rte_prefetch.h>
  61 #include <rte_lcore.h>
  62 #include <rte_per_lcore.h>
  63 #include <rte_branch_prediction.h>
  64 #include <rte_interrupts.h>
  65 #include <rte_pci.h>
  66 #include <rte_random.h>
  67 #include <rte_debug.h>
  68 #include <rte_ether.h>
  69 #include <rte_ethdev.h>
  70 #include <rte_ring.h>
  71 #include <rte_mempool.h>
  72 #include <rte_mbuf.h>
  73 #include <rte_ip.h>
  74 #include <rte_tcp.h>
  75 #include <rte_udp.h>
  76 #include <rte_string_fns.h>
  77 #include <rte_timer.h>
  78 #include <rte_power.h>
  79
  80 #include "main.h"
  81
  82 #define RTE_LOGTYPE_L3FWD_POWER RTE_LOGTYPE_USER1
  83
  84 #define MAX_PKT_BURST 32
  85
  86 #define MIN_ZERO_POLL_COUNT 5
  87
  88 /* around 100ms at 2 Ghz */
  89 #define TIMER_RESOLUTION_CYCLES           200000000ULL
  90 /* 100 ms interval */
  91 #define TIMER_NUMBER_PER_SECOND           10
  92 /* 100000 us */
  93 #define SCALING_PERIOD                    (1000000/TIMER_NUMBER_PER_SECOND)
  94 #define SCALING_DOWN_TIME_RATIO_THRESHOLD 0.25
  95
  96 #define APP_LOOKUP_EXACT_MATCH          0
  97 #define APP_LOOKUP_LPM                  1
  98 #define DO_RFC_1812_CHECKS
  99
 100 #ifndef APP_LOOKUP_METHOD
 101 #define APP_LOOKUP_METHOD             APP_LOOKUP_LPM
 102 #endif
 103
 104 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 105 #include <rte_hash.h>
 106 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 107 #include <rte_lpm.h>
 108 #else
 109 #error "APP_LOOKUP_METHOD set to incorrect value"
 110 #endif
 111
 112 #ifndef IPv6_BYTES
 113 #define IPv6_BYTES_FMT "%02x%02x:%02x%02x:%02x%02x:%02x%02x:"\
 114                        "%02x%02x:%02x%02x:%02x%02x:%02x%02x"
 115 #define IPv6_BYTES(addr) \
 116         addr[0],  addr[1], addr[2],  addr[3], \
 117         addr[4],  addr[5], addr[6],  addr[7], \
 118         addr[8],  addr[9], addr[10], addr[11],\
 119         addr[12], addr[13],addr[14], addr[15]
 120 #endif
 121
 122 #define MAX_JUMBO_PKT_LEN  9600
 123
 124 #define IPV6_ADDR_LEN 16
 125
 126 #define MEMPOOL_CACHE_SIZE 256
 127
 128 #define MBUF_SIZE (2048 + sizeof(struct rte_mbuf) + RTE_PKTMBUF_HEADROOM)
 129
 130 /*
 131  * This expression is used to calculate the number of mbufs needed depending on
 132  * user input, taking into account memory for rx and tx hardware rings, cache
 133  * per lcore and mtable per port per lcore. RTE_MAX is used to ensure that
 134  * NB_MBUF never goes below a minimum value of 8192.
 135  */
 136
 137 #define NB_MBUF RTE_MAX ( \
 138         (nb_ports*nb_rx_queue*RTE_TEST_RX_DESC_DEFAULT + \
 139         nb_ports*nb_lcores*MAX_PKT_BURST + \
 140         nb_ports*n_tx_queue*RTE_TEST_TX_DESC_DEFAULT + \
 141         nb_lcores*MEMPOOL_CACHE_SIZE), \
 142         (unsigned)8192)
 143
 144 /*
 145  * RX and TX Prefetch, Host, and Write-back threshold values should be
 146  * carefully set for optimal performance. Consult the network
 147  * controller's datasheet and supporting DPDK documentation for guidance
 148  * on how these parameters should be set.
 149  */
 150 #define RX_PTHRESH 8 /**< Default values of RX prefetch threshold reg. */
 151 #define RX_HTHRESH 8 /**< Default values of RX host threshold reg. */
 152 #define RX_WTHRESH 4 /**< Default values of RX write-back threshold reg. */
 153
 154 /*
 155  * These default values are optimized for use with the Intel(R) 82599 10 GbE
 156  * Controller and the DPDK ixgbe PMD. Consider using other values for other
 157  * network controllers and/or network drivers.
 158  */
 159 #define TX_PTHRESH 36 /**< Default values of TX prefetch threshold reg. */
 160 #define TX_HTHRESH 0  /**< Default values of TX host threshold reg. */
 161 #define TX_WTHRESH 0  /**< Default values of TX write-back threshold reg. */
 162
 163 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
 164
 165 #define NB_SOCKETS 8
 166
 167 /* Configure how many packets ahead to prefetch, when reading packets */
 168 #define PREFETCH_OFFSET 3
 169
 170 /*
 171  * Configurable number of RX/TX ring descriptors
 172  */
 173 #define RTE_TEST_RX_DESC_DEFAULT 128
 174 #define RTE_TEST_TX_DESC_DEFAULT 512
 175 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
 176 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
 177
 178 /* ethernet addresses of ports */
 179 static struct ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
 180
 181 /* mask of enabled ports */
 182 static uint32_t enabled_port_mask = 0;
 183 /* Ports set in promiscuous mode off by default. */
 184 static int promiscuous_on = 0;
 185 /* NUMA is enabled by default. */
 186 static int numa_on = 1;
 187
 188 enum freq_scale_hint_t
 189 {
 190         FREQ_LOWER    =      -1,
 191         FREQ_CURRENT  =       0,
 192         FREQ_HIGHER   =       1,
 193         FREQ_HIGHEST  =       2
 194 };
 195
 196 struct mbuf_table {
 197         uint16_t len;
 198         struct rte_mbuf *m_table[MAX_PKT_BURST];
 199 };
 200
 201 struct lcore_rx_queue {
 202         uint8_t port_id;
 203         uint8_t queue_id;
 204         enum freq_scale_hint_t freq_up_hint;
 205         uint32_t zero_rx_packet_count;
 206         uint32_t idle_hint;
 207 } __rte_cache_aligned;
 208
 209 #define MAX_RX_QUEUE_PER_LCORE 16
 210 #define MAX_TX_QUEUE_PER_PORT RTE_MAX_ETHPORTS
 211 #define MAX_RX_QUEUE_PER_PORT 128
 212
 213 #define MAX_LCORE_PARAMS 1024
 214 struct lcore_params {
 215         uint8_t port_id;
 216         uint8_t queue_id;
 217         uint8_t lcore_id;
 218 } __rte_cache_aligned;
 219
 220 static struct lcore_params lcore_params_array[MAX_LCORE_PARAMS];
 221 static struct lcore_params lcore_params_array_default[] = {
 222         {0, 0, 2},
 223         {0, 1, 2},
 224         {0, 2, 2},
 225         {1, 0, 2},
 226         {1, 1, 2},
 227         {1, 2, 2},
 228         {2, 0, 2},
 229         {3, 0, 3},
 230         {3, 1, 3},
 231 };
 232
 233 static struct lcore_params * lcore_params = lcore_params_array_default;
 234 static uint16_t nb_lcore_params = sizeof(lcore_params_array_default) /
 235                                 sizeof(lcore_params_array_default[0]);
 236
 237 static struct rte_eth_conf port_conf = {
 238         .rxmode = {
 239                 .max_rx_pkt_len = ETHER_MAX_LEN,
 240                 .split_hdr_size = 0,
 241                 .header_split   = 0, /**< Header Split disabled */
 242                 .hw_ip_checksum = 1, /**< IP checksum offload enabled */
 243                 .hw_vlan_filter = 0, /**< VLAN filtering disabled */
 244                 .jumbo_frame    = 0, /**< Jumbo Frame Support disabled */
 245                 .hw_strip_crc   = 0, /**< CRC stripped by hardware */
 246         },
 247         .rx_adv_conf = {
 248                 .rss_conf = {
 249                         .rss_key = NULL,
 250                         .rss_hf = ETH_RSS_IPV4 | ETH_RSS_IPV6,
 251                 },
 252         },
 253         .txmode = {
 254                 .mq_mode = ETH_DCB_NONE,
 255         },
 256 };
 257
 258 static const struct rte_eth_rxconf rx_conf = {
 259         .rx_thresh = {
 260                 .pthresh = RX_PTHRESH,
 261                 .hthresh = RX_HTHRESH,
 262                 .wthresh = RX_WTHRESH,
 263         },
 264         .rx_free_thresh = 32,
 265 };
 266
 267 static const struct rte_eth_txconf tx_conf = {
 268         .tx_thresh = {
 269                 .pthresh = TX_PTHRESH,
 270                 .hthresh = TX_HTHRESH,
 271                 .wthresh = TX_WTHRESH,
 272         },
 273         .tx_free_thresh = 0, /* Use PMD default values */
 274         .tx_rs_thresh = 0, /* Use PMD default values */
 275         .txq_flags = 0x0,
 276 };
 277
 278 static struct rte_mempool * pktmbuf_pool[NB_SOCKETS];
 279
 280
 281 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 282
 283 #ifdef RTE_MACHINE_CPUFLAG_SSE4_2
 284 #include <rte_hash_crc.h>
 285 #define DEFAULT_HASH_FUNC       rte_hash_crc
 286 #else
 287 #include <rte_jhash.h>
 288 #define DEFAULT_HASH_FUNC       rte_jhash
 289 #endif
 290
 291 struct ipv4_5tuple {
 292         uint32_t ip_dst;
 293         uint32_t ip_src;
 294         uint16_t port_dst;
 295         uint16_t port_src;
 296         uint8_t  proto;
 297 } __attribute__((__packed__));
 298
 299 struct ipv6_5tuple {
 300         uint8_t  ip_dst[IPV6_ADDR_LEN];
 301         uint8_t  ip_src[IPV6_ADDR_LEN];
 302         uint16_t port_dst;
 303         uint16_t port_src;
 304         uint8_t  proto;
 305 } __attribute__((__packed__));
 306
 307 struct ipv4_l3fwd_route {
 308         struct ipv4_5tuple key;
 309         uint8_t if_out;
 310 };
 311
 312 struct ipv6_l3fwd_route {
 313         struct ipv6_5tuple key;
 314         uint8_t if_out;
 315 };
 316
 317 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
 318         {{IPv4(100,10,0,1), IPv4(200,10,0,1), 101, 11, IPPROTO_TCP}, 0},
 319         {{IPv4(100,20,0,2), IPv4(200,20,0,2), 102, 12, IPPROTO_TCP}, 1},
 320         {{IPv4(100,30,0,3), IPv4(200,30,0,3), 103, 13, IPPROTO_TCP}, 2},
 321         {{IPv4(100,40,0,4), IPv4(200,40,0,4), 104, 14, IPPROTO_TCP}, 3},
 322 };
 323
 324 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
 325         {
 326                 {
 327                         {0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
 328                          0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
 329                         {0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
 330                          0x02, 0x1e, 0x67, 0xff, 0xfe, 0x0d, 0xb6, 0x0a},
 331                          1, 10, IPPROTO_UDP
 332                 }, 4
 333         },
 334 };
 335
 336 typedef struct rte_hash lookup_struct_t;
 337 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
 338 static lookup_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
 339
 340 #define L3FWD_HASH_ENTRIES      1024
 341
 342 #define IPV4_L3FWD_NUM_ROUTES \
 343         (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
 344
 345 #define IPV6_L3FWD_NUM_ROUTES \
 346         (sizeof(ipv6_l3fwd_route_array) / sizeof(ipv6_l3fwd_route_array[0]))
 347
 348 static uint8_t ipv4_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 349 static uint8_t ipv6_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 350 #endif
 351
 352 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 353 struct ipv4_l3fwd_route {
 354         uint32_t ip;
 355         uint8_t  depth;
 356         uint8_t  if_out;
 357 };
 358
 359 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
 360         {IPv4(1,1,1,0), 24, 0},
 361         {IPv4(2,1,1,0), 24, 1},
 362         {IPv4(3,1,1,0), 24, 2},
 363         {IPv4(4,1,1,0), 24, 3},
 364         {IPv4(5,1,1,0), 24, 4},
 365         {IPv4(6,1,1,0), 24, 5},
 366         {IPv4(7,1,1,0), 24, 6},
 367         {IPv4(8,1,1,0), 24, 7},
 368 };
 369
 370 #define IPV4_L3FWD_NUM_ROUTES \
 371         (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
 372
 373 #define IPV4_L3FWD_LPM_MAX_RULES     1024
 374
 375 typedef struct rte_lpm lookup_struct_t;
 376 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
 377 #endif
 378
 379 struct lcore_conf {
 380         uint16_t n_rx_queue;
 381         struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
 382         uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
 383         struct mbuf_table tx_mbufs[RTE_MAX_ETHPORTS];
 384         lookup_struct_t * ipv4_lookup_struct;
 385         lookup_struct_t * ipv6_lookup_struct;
 386 } __rte_cache_aligned;
 387
 388 struct lcore_stats {
 389         /* total sleep time in ms since last frequency scaling down */
 390         uint32_t sleep_time;
 391         /* number of long sleep recently */
 392         uint32_t nb_long_sleep;
 393         /* freq. scaling up trend */
 394         uint32_t trend;
 395         /* total packet processed recently */
 396         uint64_t nb_rx_processed;
 397         /* total iterations looped recently */
 398         uint64_t nb_iteration_looped;
 399         uint32_t padding[9];
 400 } __rte_cache_aligned;
 401
 402 static struct lcore_conf lcore_conf[RTE_MAX_LCORE] __rte_cache_aligned;
 403 static struct lcore_stats stats[RTE_MAX_LCORE] __rte_cache_aligned;
 404 static struct rte_timer power_timers[RTE_MAX_LCORE];
 405
 406 static inline uint32_t power_idle_heuristic(uint32_t zero_rx_packet_count);
 407 static inline enum freq_scale_hint_t power_freq_scaleup_heuristic( \
 408                                 unsigned lcore_id, uint32_t rx_ring_length);
 409
 410 /* exit signal handler */
 411 static void
 412 signal_exit_now(int sigtype)
 413 {
 414         unsigned lcore_id;
 415         int ret;
 416
 417         if (sigtype == SIGINT) {
 418                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
 419                         if (rte_lcore_is_enabled(lcore_id) == 0)
 420                                 continue;
 421
 422                         /* init power management library */
 423                         ret = rte_power_exit(lcore_id);
 424                         if (ret)
 425                                 rte_exit(EXIT_FAILURE, "Power management "
 426                                         "library de-initialization failed on "
 427                                                         "core%u\n", lcore_id);
 428                 }
 429         }
 430
 431         rte_exit(EXIT_SUCCESS, "User forced exit\n");
 432 }
 433
 434 /*  Freqency scale down timer callback */
 435 static void
 436 power_timer_cb(__attribute__((unused)) struct rte_timer *tim,
 437                           __attribute__((unused)) void *arg)
 438 {
 439         uint64_t hz;
 440         float sleep_time_ratio;
 441         unsigned lcore_id = rte_lcore_id();
 442
 443         /* accumulate total execution time in us when callback is invoked */
 444         sleep_time_ratio = (float)(stats[lcore_id].sleep_time) /
 445                                         (float)SCALING_PERIOD;
 446
 447         /**
 448          * check whether need to scale down frequency a step if it sleep a lot.
 449          */
 450         if (sleep_time_ratio >= SCALING_DOWN_TIME_RATIO_THRESHOLD)
 451                 rte_power_freq_down(lcore_id);
 452         else if ( (unsigned)(stats[lcore_id].nb_rx_processed /
 453                 stats[lcore_id].nb_iteration_looped) < MAX_PKT_BURST)
 454                 /**
 455                  * scale down a step if average packet per iteration less
 456                  * than expectation.
 457                  */
 458                 rte_power_freq_down(lcore_id);
 459
 460         /**
 461          * initialize another timer according to current frequency to ensure
 462          * timer interval is relatively fixed.
 463          */
 464         hz = rte_get_timer_hz();
 465         rte_timer_reset(&power_timers[lcore_id], hz/TIMER_NUMBER_PER_SECOND,
 466                                 SINGLE, lcore_id, power_timer_cb, NULL);
 467
 468         stats[lcore_id].nb_rx_processed = 0;
 469         stats[lcore_id].nb_iteration_looped = 0;
 470
 471         stats[lcore_id].sleep_time = 0;
 472 }
 473
 474 /* Send burst of packets on an output interface */
 475 static inline int
 476 send_burst(struct lcore_conf *qconf, uint16_t n, uint8_t port)
 477 {
 478         struct rte_mbuf **m_table;
 479         int ret;
 480         uint16_t queueid;
 481
 482         queueid = qconf->tx_queue_id[port];
 483         m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
 484
 485         ret = rte_eth_tx_burst(port, queueid, m_table, n);
 486         if (unlikely(ret < n)) {
 487                 do {
 488                         rte_pktmbuf_free(m_table[ret]);
 489                 } while (++ret < n);
 490         }
 491
 492         return 0;
 493 }
 494
 495 /* Enqueue a single packet, and send burst if queue is filled */
 496 static inline int
 497 send_single_packet(struct rte_mbuf *m, uint8_t port)
 498 {
 499         uint32_t lcore_id;
 500         uint16_t len;
 501         struct lcore_conf *qconf;
 502
 503         lcore_id = rte_lcore_id();
 504
 505         qconf = &lcore_conf[lcore_id];
 506         len = qconf->tx_mbufs[port].len;
 507         qconf->tx_mbufs[port].m_table[len] = m;
 508         len++;
 509
 510         /* enough pkts to be sent */
 511         if (unlikely(len == MAX_PKT_BURST)) {
 512                 send_burst(qconf, MAX_PKT_BURST, port);
 513                 len = 0;
 514         }
 515
 516         qconf->tx_mbufs[port].len = len;
 517         return 0;
 518 }
 519
 520 #ifdef DO_RFC_1812_CHECKS
 521 static inline int
 522 is_valid_ipv4_pkt(struct ipv4_hdr *pkt, uint32_t link_len)
 523 {
 524         /* From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2 */
 525         /*
 526          * 1. The packet length reported by the Link Layer must be large
 527          * enough to hold the minimum length legal IP datagram (20 bytes).
 528          */
 529         if (link_len < sizeof(struct ipv4_hdr))
 530                 return -1;
 531
 532         /* 2. The IP checksum must be correct. */
 533         /* this is checked in H/W */
 534
 535         /*
 536          * 3. The IP version number must be 4. If the version number is not 4
 537          * then the packet may be another version of IP, such as IPng or
 538          * ST-II.
 539          */
 540         if (((pkt->version_ihl) >> 4) != 4)
 541                 return -3;
 542         /*
 543          * 4. The IP header length field must be large enough to hold the
 544          * minimum length legal IP datagram (20 bytes = 5 words).
 545          */
 546         if ((pkt->version_ihl & 0xf) < 5)
 547                 return -4;
 548
 549         /*
 550          * 5. The IP total length field must be large enough to hold the IP
 551          * datagram header, whose length is specified in the IP header length
 552          * field.
 553          */
 554         if (rte_cpu_to_be_16(pkt->total_length) < sizeof(struct ipv4_hdr))
 555                 return -5;
 556
 557         return 0;
 558 }
 559 #endif
 560
 561 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 562 static void
 563 print_ipv4_key(struct ipv4_5tuple key)
 564 {
 565         printf("IP dst = %08x, IP src = %08x, port dst = %d, port src = %d, "
 566                 "proto = %d\n", (unsigned)key.ip_dst, (unsigned)key.ip_src,
 567                                 key.port_dst, key.port_src, key.proto);
 568 }
 569 static void
 570 print_ipv6_key(struct ipv6_5tuple key)
 571 {
 572         printf( "IP dst = " IPv6_BYTES_FMT ", IP src = " IPv6_BYTES_FMT ", "
 573                 "port dst = %d, port src = %d, proto = %d\n",
 574                 IPv6_BYTES(key.ip_dst), IPv6_BYTES(key.ip_src),
 575                 key.port_dst, key.port_src, key.proto);
 576 }
 577
 578 static inline uint8_t
 579 get_ipv4_dst_port(struct ipv4_hdr *ipv4_hdr, uint8_t portid,
 580                 lookup_struct_t * ipv4_l3fwd_lookup_struct)
 581 {
 582         struct ipv4_5tuple key;
 583         struct tcp_hdr *tcp;
 584         struct udp_hdr *udp;
 585         int ret = 0;
 586
 587         key.ip_dst = rte_be_to_cpu_32(ipv4_hdr->dst_addr);
 588         key.ip_src = rte_be_to_cpu_32(ipv4_hdr->src_addr);
 589         key.proto = ipv4_hdr->next_proto_id;
 590
 591         switch (ipv4_hdr->next_proto_id) {
 592         case IPPROTO_TCP:
 593                 tcp = (struct tcp_hdr *)((unsigned char *)ipv4_hdr +
 594                                         sizeof(struct ipv4_hdr));
 595                 key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
 596                 key.port_src = rte_be_to_cpu_16(tcp->src_port);
 597                 break;
 598
 599         case IPPROTO_UDP:
 600                 udp = (struct udp_hdr *)((unsigned char *)ipv4_hdr +
 601                                         sizeof(struct ipv4_hdr));
 602                 key.port_dst = rte_be_to_cpu_16(udp->dst_port);
 603                 key.port_src = rte_be_to_cpu_16(udp->src_port);
 604                 break;
 605
 606         default:
 607                 key.port_dst = 0;
 608                 key.port_src = 0;
 609                 break;
 610         }
 611
 612         /* Find destination port */
 613         ret = rte_hash_lookup(ipv4_l3fwd_lookup_struct, (const void *)&key);
 614         return (uint8_t)((ret < 0)? portid : ipv4_l3fwd_out_if[ret]);
 615 }
 616
 617 static inline uint8_t
 618 get_ipv6_dst_port(struct ipv6_hdr *ipv6_hdr,  uint8_t portid,
 619                         lookup_struct_t *ipv6_l3fwd_lookup_struct)
 620 {
 621         struct ipv6_5tuple key;
 622         struct tcp_hdr *tcp;
 623         struct udp_hdr *udp;
 624         int ret = 0;
 625
 626         memcpy(key.ip_dst, ipv6_hdr->dst_addr, IPV6_ADDR_LEN);
 627         memcpy(key.ip_src, ipv6_hdr->src_addr, IPV6_ADDR_LEN);
 628
 629         key.proto = ipv6_hdr->proto;
 630
 631         switch (ipv6_hdr->proto) {
 632         case IPPROTO_TCP:
 633                 tcp = (struct tcp_hdr *)((unsigned char *) ipv6_hdr +
 634                                         sizeof(struct ipv6_hdr));
 635                 key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
 636                 key.port_src = rte_be_to_cpu_16(tcp->src_port);
 637                 break;
 638
 639         case IPPROTO_UDP:
 640                 udp = (struct udp_hdr *)((unsigned char *) ipv6_hdr +
 641                                         sizeof(struct ipv6_hdr));
 642                 key.port_dst = rte_be_to_cpu_16(udp->dst_port);
 643                 key.port_src = rte_be_to_cpu_16(udp->src_port);
 644                 break;
 645
 646         default:
 647                 key.port_dst = 0;
 648                 key.port_src = 0;
 649                 break;
 650         }
 651
 652         /* Find destination port */
 653         ret = rte_hash_lookup(ipv6_l3fwd_lookup_struct, (const void *)&key);
 654         return (uint8_t)((ret < 0)? portid : ipv6_l3fwd_out_if[ret]);
 655 }
 656 #endif
 657
 658 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 659 static inline uint8_t
 660 get_ipv4_dst_port(struct ipv4_hdr *ipv4_hdr, uint8_t portid,
 661                 lookup_struct_t *ipv4_l3fwd_lookup_struct)
 662 {
 663         uint8_t next_hop;
 664
 665         return (uint8_t) ((rte_lpm_lookup(ipv4_l3fwd_lookup_struct,
 666                         rte_be_to_cpu_32(ipv4_hdr->dst_addr), &next_hop) == 0)?
 667                         next_hop : portid);
 668 }
 669 #endif
 670
 671 static inline void
 672 l3fwd_simple_forward(struct rte_mbuf *m, uint8_t portid,
 673                                 struct lcore_conf *qconf)
 674 {
 675         struct ether_hdr *eth_hdr;
 676         struct ipv4_hdr *ipv4_hdr;
 677         void *d_addr_bytes;
 678         uint8_t dst_port;
 679
 680         eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
 681
 682         if (m->ol_flags & PKT_RX_IPV4_HDR) {
 683                 /* Handle IPv4 headers.*/
 684                 ipv4_hdr =
 685                         (struct ipv4_hdr *)(rte_pktmbuf_mtod(m, unsigned char*)
 686                                                 + sizeof(struct ether_hdr));
 687
 688 #ifdef DO_RFC_1812_CHECKS
 689                 /* Check to make sure the packet is valid (RFC1812) */
 690                 if (is_valid_ipv4_pkt(ipv4_hdr, m->pkt.pkt_len) < 0) {
 691                         rte_pktmbuf_free(m);
 692                         return;
 693                 }
 694 #endif
 695
 696                 dst_port = get_ipv4_dst_port(ipv4_hdr, portid,
 697                                         qconf->ipv4_lookup_struct);
 698                 if (dst_port >= RTE_MAX_ETHPORTS ||
 699                                 (enabled_port_mask & 1 << dst_port) == 0)
 700                         dst_port = portid;
 701
 702                 /* 02:00:00:00:00:xx */
 703                 d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
 704                 *((uint64_t *)d_addr_bytes) =
 705                         0x000000000002 + ((uint64_t)dst_port << 40);
 706
 707 #ifdef DO_RFC_1812_CHECKS
 708                 /* Update time to live and header checksum */
 709                 --(ipv4_hdr->time_to_live);
 710                 ++(ipv4_hdr->hdr_checksum);
 711 #endif
 712
 713                 /* src addr */
 714                 ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
 715
 716                 send_single_packet(m, dst_port);
 717         }
 718         else {
 719                 /* Handle IPv6 headers.*/
 720 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 721                 struct ipv6_hdr *ipv6_hdr;
 722
 723                 ipv6_hdr =
 724                         (struct ipv6_hdr *)(rte_pktmbuf_mtod(m, unsigned char*)
 725                                                 + sizeof(struct ether_hdr));
 726
 727                 dst_port = get_ipv6_dst_port(ipv6_hdr, portid,
 728                                         qconf->ipv6_lookup_struct);
 729
 730                 if (dst_port >= RTE_MAX_ETHPORTS ||
 731                                 (enabled_port_mask & 1 << dst_port) == 0)
 732                         dst_port = portid;
 733
 734                 /* 02:00:00:00:00:xx */
 735                 d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
 736                 *((uint64_t *)d_addr_bytes) =
 737                         0x000000000002 + ((uint64_t)dst_port << 40);
 738
 739                 /* src addr */
 740                 ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
 741
 742                 send_single_packet(m, dst_port);
 743 #else
 744                 /* We don't currently handle IPv6 packets in LPM mode. */
 745                 rte_pktmbuf_free(m);
 746 #endif
 747         }
 748
 749 }
 750
 751 #define SLEEP_GEAR1_THRESHOLD            100
 752 #define SLEEP_GEAR2_THRESHOLD            1000
 753
 754 static inline uint32_t
 755 power_idle_heuristic(uint32_t zero_rx_packet_count)
 756 {
 757         /* If zero count is less than 100, use it as the sleep time in us */
 758         if (zero_rx_packet_count < SLEEP_GEAR1_THRESHOLD)
 759                 return zero_rx_packet_count;
 760         /* If zero count is less than 1000, sleep time should be 100 us */
 761         else if ((zero_rx_packet_count >= SLEEP_GEAR1_THRESHOLD) &&
 762                         (zero_rx_packet_count < SLEEP_GEAR2_THRESHOLD))
 763                 return SLEEP_GEAR1_THRESHOLD;
 764         /* If zero count is greater than 1000, sleep time should be 1000 us */
 765         else if (zero_rx_packet_count >= SLEEP_GEAR2_THRESHOLD)
 766                 return SLEEP_GEAR2_THRESHOLD;
 767
 768         return 0;
 769 }
 770
 771 static inline enum freq_scale_hint_t
 772 power_freq_scaleup_heuristic(unsigned lcore_id, uint32_t rx_ring_length)
 773 {
 774 /**
 775  * HW Rx queue size is 128 by default, Rx burst read at maximum 32 entries
 776  * per iteration
 777  */
 778 #define FREQ_GEAR1_RX_PACKET_THRESHOLD             MAX_PKT_BURST
 779 #define FREQ_GEAR2_RX_PACKET_THRESHOLD             64
 780 #define FREQ_GEAR3_RX_PACKET_THRESHOLD             96
 781 #define FREQ_UP_TREND1_ACC   1
 782 #define FREQ_UP_TREND2_ACC   100
 783 #define FREQ_UP_THRESHOLD    10000
 784
 785         /**
 786          * there are received packets to process, staying at C0 state while
 787          * trying to scale up frequency depending on how many entries on h/w
 788          * queue. Determine frequency scaleup trend based on availiable entries
 789          * on Rx queues.
 790          */
 791         if (rx_ring_length > FREQ_GEAR3_RX_PACKET_THRESHOLD) {
 792                 stats[lcore_id].trend = 0;
 793                 return FREQ_HIGHEST;
 794         } else if (rx_ring_length > FREQ_GEAR2_RX_PACKET_THRESHOLD)
 795                 stats[lcore_id].trend += FREQ_UP_TREND2_ACC;
 796         else if (rx_ring_length > FREQ_GEAR1_RX_PACKET_THRESHOLD)
 797                 stats[lcore_id].trend += FREQ_UP_TREND1_ACC;
 798
 799         if (stats[lcore_id].trend > FREQ_UP_THRESHOLD) {
 800                 stats[lcore_id].trend = 0;
 801                 return FREQ_HIGHER;
 802         }
 803
 804         return FREQ_CURRENT;
 805 }
 806
 807 /* main processing loop */
 808 static int
 809 main_loop(__attribute__((unused)) void *dummy)
 810 {
 811         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
 812         unsigned lcore_id;
 813         uint64_t prev_tsc, diff_tsc, cur_tsc;
 814         uint64_t prev_tsc_power = 0, cur_tsc_power, diff_tsc_power;
 815         int i, j, nb_rx;
 816         uint8_t portid, queueid;
 817         struct lcore_conf *qconf;
 818         struct lcore_rx_queue *rx_queue;
 819         uint32_t rx_ring_length;
 820         enum freq_scale_hint_t lcore_scaleup_hint;
 821
 822         uint32_t lcore_rx_idle_count = 0;
 823         uint32_t lcore_idle_hint = 0;
 824
 825         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
 826
 827         prev_tsc = 0;
 828
 829         lcore_id = rte_lcore_id();
 830         qconf = &lcore_conf[lcore_id];
 831
 832         if (qconf->n_rx_queue == 0) {
 833                 RTE_LOG(INFO, L3FWD_POWER, "lcore %u has nothing to do\n", lcore_id);
 834                 return 0;
 835         }
 836
 837         RTE_LOG(INFO, L3FWD_POWER, "entering main loop on lcore %u\n", lcore_id);
 838
 839         for (i = 0; i < qconf->n_rx_queue; i++) {
 840
 841                 portid = qconf->rx_queue_list[i].port_id;
 842                 queueid = qconf->rx_queue_list[i].queue_id;
 843                 RTE_LOG(INFO, L3FWD_POWER, " -- lcoreid=%u portid=%hhu "
 844                         "rxqueueid=%hhu\n", lcore_id, portid, queueid);
 845         }
 846
 847         while (1) {
 848                 stats[lcore_id].nb_iteration_looped++;
 849
 850                 cur_tsc = rte_rdtsc();
 851                 cur_tsc_power = cur_tsc;
 852
 853                 /*
 854                  * TX burst queue drain
 855                  */
 856                 diff_tsc = cur_tsc - prev_tsc;
 857                 if (unlikely(diff_tsc > drain_tsc)) {
 858
 859                         /*
 860                          * This could be optimized (use queueid instead of
 861                          * portid), but it is not called so often
 862                          */
 863                         for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
 864                                 if (qconf->tx_mbufs[portid].len == 0)
 865                                         continue;
 866                                 send_burst(&lcore_conf[lcore_id],
 867                                         qconf->tx_mbufs[portid].len,
 868                                         portid);
 869                                 qconf->tx_mbufs[portid].len = 0;
 870                         }
 871
 872                         prev_tsc = cur_tsc;
 873                 }
 874
 875                 diff_tsc_power = cur_tsc_power - prev_tsc_power;
 876                 if (diff_tsc_power > TIMER_RESOLUTION_CYCLES) {
 877                         rte_timer_manage();
 878                         prev_tsc_power = cur_tsc_power;
 879                 }
 880
 881                 /*
 882                  * Read packet from RX queues
 883                  */
 884                 lcore_scaleup_hint = FREQ_CURRENT;
 885                 lcore_rx_idle_count = 0;
 886                 for (i = 0; i < qconf->n_rx_queue; ++i) {
 887                         rx_queue = &(qconf->rx_queue_list[i]);
 888                         rx_queue->idle_hint = 0;
 889                         portid = rx_queue->port_id;
 890                         queueid = rx_queue->queue_id;
 891
 892                         nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
 893                                                                 MAX_PKT_BURST);
 894                         stats[lcore_id].nb_rx_processed += nb_rx;
 895                         if (unlikely(nb_rx == 0)) {
 896                                 /**
 897                                  * no packet received from rx queue, try to
 898                                  * sleep for a while forcing CPU enter deeper
 899                                  * C states.
 900                                  */
 901                                 rx_queue->zero_rx_packet_count++;
 902
 903                                 if (rx_queue->zero_rx_packet_count <=
 904                                                         MIN_ZERO_POLL_COUNT)
 905                                         continue;
 906
 907                                 rx_queue->idle_hint = power_idle_heuristic(\
 908                                         rx_queue->zero_rx_packet_count);
 909                                 lcore_rx_idle_count++;
 910                         } else {
 911                                 /**
 912                                  * get availiable descriptor number via MMIO read is costly,
 913                                  * so only do it when recent poll returns maximum number.
 914                                  */
 915                                 if (nb_rx >= MAX_PKT_BURST)
 916                                      rx_ring_length = rte_eth_rx_queue_count(portid, queueid);
 917                                 else
 918                                      rx_ring_length = 0;
 919
 920                                 rx_queue->zero_rx_packet_count = 0;
 921
 922                                 /**
 923                                  * do not scale up frequency immediately as
 924                                  * user to kernel space communication is costly
 925                                  * which might impact packet I/O for received
 926                                  * packets.
 927                                  */
 928                                 rx_queue->freq_up_hint =
 929                                         power_freq_scaleup_heuristic(lcore_id,
 930                                                         rx_ring_length);
 931                         }
 932
 933                         /* Prefetch first packets */
 934                         for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
 935                                 rte_prefetch0(rte_pktmbuf_mtod(
 936                                                 pkts_burst[j], void *));
 937                         }
 938
 939                         /* Prefetch and forward already prefetched packets */
 940                         for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
 941                                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
 942                                                 j + PREFETCH_OFFSET], void *));
 943                                 l3fwd_simple_forward(pkts_burst[j], portid,
 944                                                                 qconf);
 945                         }
 946
 947                         /* Forward remaining prefetched packets */
 948                         for (; j < nb_rx; j++) {
 949                                 l3fwd_simple_forward(pkts_burst[j], portid,
 950                                                                 qconf);
 951                         }
 952                 }
 953
 954                 if (likely(lcore_rx_idle_count != qconf->n_rx_queue)) {
 955                         for (i = 1, lcore_scaleup_hint =
 956                                 qconf->rx_queue_list[0].freq_up_hint;
 957                                         i < qconf->n_rx_queue; ++i) {
 958                                 rx_queue = &(qconf->rx_queue_list[i]);
 959                                 if (rx_queue->freq_up_hint >
 960                                                 lcore_scaleup_hint)
 961                                         lcore_scaleup_hint =
 962                                                 rx_queue->freq_up_hint;
 963                         }
 964
 965                         if (lcore_scaleup_hint == FREQ_HIGHEST)
 966                                 rte_power_freq_max(lcore_id);
 967                         else if (lcore_scaleup_hint == FREQ_HIGHER)
 968                                 rte_power_freq_up(lcore_id);
 969                 } else {
 970                         /**
 971                          * All Rx queues empty in recent consecutive polls,
 972                          * sleep in a conservative manner, meaning sleep as
 973                          * less as possible.
 974                          */
 975                         for (i = 1, lcore_idle_hint =
 976                                 qconf->rx_queue_list[0].idle_hint;
 977                                         i < qconf->n_rx_queue; ++i) {
 978                                 rx_queue = &(qconf->rx_queue_list[i]);
 979                                 if (rx_queue->idle_hint < lcore_idle_hint)
 980                                         lcore_idle_hint = rx_queue->idle_hint;
 981                         }
 982
 983                         if ( lcore_idle_hint < SLEEP_GEAR1_THRESHOLD)
 984                                 /**
 985                                  * execute "pause" instruction to avoid context
 986                                  * switch for short sleep.
 987                                  */
 988                                 rte_delay_us(lcore_idle_hint);
 989                         else
 990                                 /* long sleep force runing thread to suspend */
 991                                 usleep(lcore_idle_hint);
 992
 993                         stats[lcore_id].sleep_time += lcore_idle_hint;
 994                 }
 995         }
 996 }
 997
 998 static int
 999 check_lcore_params(void)
1000 {
1001         uint8_t queue, lcore;
1002         uint16_t i;
1003         int socketid;
1004
1005         for (i = 0; i < nb_lcore_params; ++i) {
1006                 queue = lcore_params[i].queue_id;
1007                 if (queue >= MAX_RX_QUEUE_PER_PORT) {
1008                         printf("invalid queue number: %hhu\n", queue);
1009                         return -1;
1010                 }
1011                 lcore = lcore_params[i].lcore_id;
1012                 if (!rte_lcore_is_enabled(lcore)) {
1013                         printf("error: lcore %hhu is not enabled in lcore "
1014                                                         "mask\n", lcore);
1015                         return -1;
1016                 }
1017                 if ((socketid = rte_lcore_to_socket_id(lcore) != 0) &&
1018                                                         (numa_on == 0)) {
1019                         printf("warning: lcore %hhu is on socket %d with numa "
1020                                                 "off\n", lcore, socketid);
1021                 }
1022         }
1023         return 0;
1024 }
1025
1026 static int
1027 check_port_config(const unsigned nb_ports)
1028 {
1029         unsigned portid;
1030         uint16_t i;
1031
1032         for (i = 0; i < nb_lcore_params; ++i) {
1033                 portid = lcore_params[i].port_id;
1034                 if ((enabled_port_mask & (1 << portid)) == 0) {
1035                         printf("port %u is not enabled in port mask\n",
1036                                                                 portid);
1037                         return -1;
1038                 }
1039                 if (portid >= nb_ports) {
1040                         printf("port %u is not present on the board\n",
1041                                                                 portid);
1042                         return -1;
1043                 }
1044         }
1045         return 0;
1046 }
1047
1048 static uint8_t
1049 get_port_n_rx_queues(const uint8_t port)
1050 {
1051         int queue = -1;
1052         uint16_t i;
1053
1054         for (i = 0; i < nb_lcore_params; ++i) {
1055                 if (lcore_params[i].port_id == port &&
1056                                 lcore_params[i].queue_id > queue)
1057                         queue = lcore_params[i].queue_id;
1058         }
1059         return (uint8_t)(++queue);
1060 }
1061
1062 static int
1063 init_lcore_rx_queues(void)
1064 {
1065         uint16_t i, nb_rx_queue;
1066         uint8_t lcore;
1067
1068         for (i = 0; i < nb_lcore_params; ++i) {
1069                 lcore = lcore_params[i].lcore_id;
1070                 nb_rx_queue = lcore_conf[lcore].n_rx_queue;
1071                 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
1072                         printf("error: too many queues (%u) for lcore: %u\n",
1073                                 (unsigned)nb_rx_queue + 1, (unsigned)lcore);
1074                         return -1;
1075                 } else {
1076                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
1077                                 lcore_params[i].port_id;
1078                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
1079                                 lcore_params[i].queue_id;
1080                         lcore_conf[lcore].n_rx_queue++;
1081                 }
1082         }
1083         return 0;
1084 }
1085
1086 /* display usage */
1087 static void
1088 print_usage(const char *prgname)
1089 {
1090         printf ("%s [EAL options] -- -p PORTMASK -P"
1091                 "  [--config (port,queue,lcore)[,(port,queue,lcore]]"
1092                 "  [--enable-jumbo [--max-pkt-len PKTLEN]]\n"
1093                 "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
1094                 "  -P : enable promiscuous mode\n"
1095                 "  --config (port,queue,lcore): rx queues configuration\n"
1096                 "  --no-numa: optional, disable numa awareness\n"
1097                 "  --enable-jumbo: enable jumbo frame"
1098                 " which max packet len is PKTLEN in decimal (64-9600)\n",
1099                 prgname);
1100 }
1101
1102 static int parse_max_pkt_len(const char *pktlen)
1103 {
1104         char *end = NULL;
1105         unsigned long len;
1106
1107         /* parse decimal string */
1108         len = strtoul(pktlen, &end, 10);
1109         if ((pktlen[0] == '\0') || (end == NULL) || (*end != '\0'))
1110                 return -1;
1111
1112         if (len == 0)
1113                 return -1;
1114
1115         return len;
1116 }
1117
1118 static int
1119 parse_portmask(const char *portmask)
1120 {
1121         char *end = NULL;
1122         unsigned long pm;
1123
1124         /* parse hexadecimal string */
1125         pm = strtoul(portmask, &end, 16);
1126         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
1127                 return -1;
1128
1129         if (pm == 0)
1130                 return -1;
1131
1132         return pm;
1133 }
1134
1135 static int
1136 parse_config(const char *q_arg)
1137 {
1138         char s[256];
1139         const char *p, *p0 = q_arg;
1140         char *end;
1141         enum fieldnames {
1142                 FLD_PORT = 0,
1143                 FLD_QUEUE,
1144                 FLD_LCORE,
1145                 _NUM_FLD
1146         };
1147         unsigned long int_fld[_NUM_FLD];
1148         char *str_fld[_NUM_FLD];
1149         int i;
1150         unsigned size;
1151
1152         nb_lcore_params = 0;
1153
1154         while ((p = strchr(p0,'(')) != NULL) {
1155                 ++p;
1156                 if((p0 = strchr(p,')')) == NULL)
1157                         return -1;
1158
1159                 size = p0 - p;
1160                 if(size >= sizeof(s))
1161                         return -1;
1162
1163                 rte_snprintf(s, sizeof(s), "%.*s", size, p);
1164                 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') !=
1165                                                                 _NUM_FLD)
1166                         return -1;
1167                 for (i = 0; i < _NUM_FLD; i++){
1168                         errno = 0;
1169                         int_fld[i] = strtoul(str_fld[i], &end, 0);
1170                         if (errno != 0 || end == str_fld[i] || int_fld[i] >
1171                                                                         255)
1172                                 return -1;
1173                 }
1174                 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
1175                         printf("exceeded max number of lcore params: %hu\n",
1176                                 nb_lcore_params);
1177                         return -1;
1178                 }
1179                 lcore_params_array[nb_lcore_params].port_id =
1180                                 (uint8_t)int_fld[FLD_PORT];
1181                 lcore_params_array[nb_lcore_params].queue_id =
1182                                 (uint8_t)int_fld[FLD_QUEUE];
1183                 lcore_params_array[nb_lcore_params].lcore_id =
1184                                 (uint8_t)int_fld[FLD_LCORE];
1185                 ++nb_lcore_params;
1186         }
1187         lcore_params = lcore_params_array;
1188
1189         return 0;
1190 }
1191
1192 /* Parse the argument given in the command line of the application */
1193 static int
1194 parse_args(int argc, char **argv)
1195 {
1196         int opt, ret;
1197         char **argvopt;
1198         int option_index;
1199         char *prgname = argv[0];
1200         static struct option lgopts[] = {
1201                 {"config", 1, 0, 0},
1202                 {"no-numa", 0, 0, 0},
1203                 {"enable-jumbo", 0, 0, 0},
1204                 {NULL, 0, 0, 0}
1205         };
1206
1207         argvopt = argv;
1208
1209         while ((opt = getopt_long(argc, argvopt, "p:P",
1210                                 lgopts, &option_index)) != EOF) {
1211
1212                 switch (opt) {
1213                 /* portmask */
1214                 case 'p':
1215                         enabled_port_mask = parse_portmask(optarg);
1216                         if (enabled_port_mask == 0) {
1217                                 printf("invalid portmask\n");
1218                                 print_usage(prgname);
1219                                 return -1;
1220                         }
1221                         break;
1222                 case 'P':
1223                         printf("Promiscuous mode selected\n");
1224                         promiscuous_on = 1;
1225                         break;
1226
1227                 /* long options */
1228                 case 0:
1229                         if (!strncmp(lgopts[option_index].name, "config", 6)) {
1230                                 ret = parse_config(optarg);
1231                                 if (ret) {
1232                                         printf("invalid config\n");
1233                                         print_usage(prgname);
1234                                         return -1;
1235                                 }
1236                         }
1237
1238                         if (!strncmp(lgopts[option_index].name,
1239                                                 "no-numa", 7)) {
1240                                 printf("numa is disabled \n");
1241                                 numa_on = 0;
1242                         }
1243
1244                         if (!strncmp(lgopts[option_index].name,
1245                                         "enable-jumbo", 12)) {
1246                                 struct option lenopts =
1247                                         {"max-pkt-len", required_argument, \
1248                                                                         0, 0};
1249
1250                                 printf("jumbo frame is enabled \n");
1251                                 port_conf.rxmode.jumbo_frame = 1;
1252
1253                                 /**
1254                                  * if no max-pkt-len set, use the default value
1255                                  * ETHER_MAX_LEN
1256                                  */
1257                                 if (0 == getopt_long(argc, argvopt, "",
1258                                                 &lenopts, &option_index)) {
1259                                         ret = parse_max_pkt_len(optarg);
1260                                         if ((ret < 64) ||
1261                                                 (ret > MAX_JUMBO_PKT_LEN)){
1262                                                 printf("invalid packet "
1263                                                                 "length\n");
1264                                                 print_usage(prgname);
1265                                                 return -1;
1266                                         }
1267                                         port_conf.rxmode.max_rx_pkt_len = ret;
1268                                 }
1269                                 printf("set jumbo frame "
1270                                         "max packet length to %u\n",
1271                                 (unsigned int)port_conf.rxmode.max_rx_pkt_len);
1272                         }
1273
1274                         break;
1275
1276                 default:
1277                         print_usage(prgname);
1278                         return -1;
1279                 }
1280         }
1281
1282         if (optind >= 0)
1283                 argv[optind-1] = prgname;
1284
1285         ret = optind-1;
1286         optind = 0; /* reset getopt lib */
1287         return ret;
1288 }
1289
1290 static void
1291 print_ethaddr(const char *name, const struct ether_addr *eth_addr)
1292 {
1293         printf ("%s%02X:%02X:%02X:%02X:%02X:%02X", name,
1294                 eth_addr->addr_bytes[0],
1295                 eth_addr->addr_bytes[1],
1296                 eth_addr->addr_bytes[2],
1297                 eth_addr->addr_bytes[3],
1298                 eth_addr->addr_bytes[4],
1299                 eth_addr->addr_bytes[5]);
1300 }
1301
1302 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1303 static void
1304 setup_hash(int socketid)
1305 {
1306         struct rte_hash_parameters ipv4_l3fwd_hash_params = {
1307                 .name = NULL,
1308                 .entries = L3FWD_HASH_ENTRIES,
1309                 .bucket_entries = 4,
1310                 .key_len = sizeof(struct ipv4_5tuple),
1311                 .hash_func = DEFAULT_HASH_FUNC,
1312                 .hash_func_init_val = 0,
1313         };
1314
1315         struct rte_hash_parameters ipv6_l3fwd_hash_params = {
1316                 .name = NULL,
1317                 .entries = L3FWD_HASH_ENTRIES,
1318                 .bucket_entries = 4,
1319                 .key_len = sizeof(struct ipv6_5tuple),
1320                 .hash_func = DEFAULT_HASH_FUNC,
1321                 .hash_func_init_val = 0,
1322         };
1323
1324         unsigned i;
1325         int ret;
1326         char s[64];
1327
1328         /* create ipv4 hash */
1329         rte_snprintf(s, sizeof(s), "ipv4_l3fwd_hash_%d", socketid);
1330         ipv4_l3fwd_hash_params.name = s;
1331         ipv4_l3fwd_hash_params.socket_id = socketid;
1332         ipv4_l3fwd_lookup_struct[socketid] =
1333                 rte_hash_create(&ipv4_l3fwd_hash_params);
1334         if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
1335                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
1336                                 "socket %d\n", socketid);
1337
1338         /* create ipv6 hash */
1339         rte_snprintf(s, sizeof(s), "ipv6_l3fwd_hash_%d", socketid);
1340         ipv6_l3fwd_hash_params.name = s;
1341         ipv6_l3fwd_hash_params.socket_id = socketid;
1342         ipv6_l3fwd_lookup_struct[socketid] =
1343                 rte_hash_create(&ipv6_l3fwd_hash_params);
1344         if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
1345                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
1346                                 "socket %d\n", socketid);
1347
1348
1349         /* populate the ipv4 hash */
1350         for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
1351                 ret = rte_hash_add_key (ipv4_l3fwd_lookup_struct[socketid],
1352                                 (void *) &ipv4_l3fwd_route_array[i].key);
1353                 if (ret < 0) {
1354                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
1355                                 "l3fwd hash on socket %d\n", i, socketid);
1356                 }
1357                 ipv4_l3fwd_out_if[ret] = ipv4_l3fwd_route_array[i].if_out;
1358                 printf("Hash: Adding key\n");
1359                 print_ipv4_key(ipv4_l3fwd_route_array[i].key);
1360         }
1361
1362         /* populate the ipv6 hash */
1363         for (i = 0; i < IPV6_L3FWD_NUM_ROUTES; i++) {
1364                 ret = rte_hash_add_key (ipv6_l3fwd_lookup_struct[socketid],
1365                                 (void *) &ipv6_l3fwd_route_array[i].key);
1366                 if (ret < 0) {
1367                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
1368                                 "l3fwd hash on socket %d\n", i, socketid);
1369                 }
1370                 ipv6_l3fwd_out_if[ret] = ipv6_l3fwd_route_array[i].if_out;
1371                 printf("Hash: Adding key\n");
1372                 print_ipv6_key(ipv6_l3fwd_route_array[i].key);
1373         }
1374 }
1375 #endif
1376
1377 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1378 static void
1379 setup_lpm(int socketid)
1380 {
1381         unsigned i;
1382         int ret;
1383         char s[64];
1384
1385         /* create the LPM table */
1386         rte_snprintf(s, sizeof(s), "IPV4_L3FWD_LPM_%d", socketid);
1387         ipv4_l3fwd_lookup_struct[socketid] = rte_lpm_create(s, socketid,
1388                                 IPV4_L3FWD_LPM_MAX_RULES, 0);
1389         if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
1390                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
1391                                 " on socket %d\n", socketid);
1392
1393         /* populate the LPM table */
1394         for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
1395                 ret = rte_lpm_add(ipv4_l3fwd_lookup_struct[socketid],
1396                         ipv4_l3fwd_route_array[i].ip,
1397                         ipv4_l3fwd_route_array[i].depth,
1398                         ipv4_l3fwd_route_array[i].if_out);
1399
1400                 if (ret < 0) {
1401                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
1402                                 "l3fwd LPM table on socket %d\n",
1403                                 i, socketid);
1404                 }
1405
1406                 printf("LPM: Adding route 0x%08x / %d (%d)\n",
1407                         (unsigned)ipv4_l3fwd_route_array[i].ip,
1408                         ipv4_l3fwd_route_array[i].depth,
1409                         ipv4_l3fwd_route_array[i].if_out);
1410         }
1411 }
1412 #endif
1413
1414 static int
1415 init_mem(unsigned nb_mbuf)
1416 {
1417         struct lcore_conf *qconf;
1418         int socketid;
1419         unsigned lcore_id;
1420         char s[64];
1421
1422         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1423                 if (rte_lcore_is_enabled(lcore_id) == 0)
1424                         continue;
1425
1426                 if (numa_on)
1427                         socketid = rte_lcore_to_socket_id(lcore_id);
1428                 else
1429                         socketid = 0;
1430
1431                 if (socketid >= NB_SOCKETS) {
1432                         rte_exit(EXIT_FAILURE, "Socket %d of lcore %u is "
1433                                         "out of range %d\n", socketid,
1434                                                 lcore_id, NB_SOCKETS);
1435                 }
1436                 if (pktmbuf_pool[socketid] == NULL) {
1437                         rte_snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
1438                         pktmbuf_pool[socketid] =
1439                                 rte_mempool_create(s, nb_mbuf,
1440                                         MBUF_SIZE, MEMPOOL_CACHE_SIZE,
1441                                         sizeof(struct rte_pktmbuf_pool_private),
1442                                         rte_pktmbuf_pool_init, NULL,
1443                                         rte_pktmbuf_init, NULL,
1444                                         socketid, 0);
1445                         if (pktmbuf_pool[socketid] == NULL)
1446                                 rte_exit(EXIT_FAILURE,
1447                                         "Cannot init mbuf pool on socket %d\n",
1448                                                                 socketid);
1449                         else
1450                                 printf("Allocated mbuf pool on socket %d\n",
1451                                                                 socketid);
1452
1453 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1454                         setup_lpm(socketid);
1455 #else
1456                         setup_hash(socketid);
1457 #endif
1458                 }
1459                 qconf = &lcore_conf[lcore_id];
1460                 qconf->ipv4_lookup_struct = ipv4_l3fwd_lookup_struct[socketid];
1461 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1462                 qconf->ipv6_lookup_struct = ipv6_l3fwd_lookup_struct[socketid];
1463 #endif
1464         }
1465         return 0;
1466 }
1467
1468 /* Check the link status of all ports in up to 9s, and print them finally */
1469 static void
1470 check_all_ports_link_status(uint8_t port_num, uint32_t port_mask)
1471 {
1472 #define CHECK_INTERVAL 100 /* 100ms */
1473 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
1474         uint8_t portid, count, all_ports_up, print_flag = 0;
1475         struct rte_eth_link link;
1476
1477         printf("\nChecking link status");
1478         fflush(stdout);
1479         for (count = 0; count <= MAX_CHECK_TIME; count++) {
1480                 all_ports_up = 1;
1481                 for (portid = 0; portid < port_num; portid++) {
1482                         if ((port_mask & (1 << portid)) == 0)
1483                                 continue;
1484                         memset(&link, 0, sizeof(link));
1485                         rte_eth_link_get_nowait(portid, &link);
1486                         /* print link status if flag set */
1487                         if (print_flag == 1) {
1488                                 if (link.link_status)
1489                                         printf("Port %d Link Up - speed %u "
1490                                                 "Mbps - %s\n", (uint8_t)portid,
1491                                                 (unsigned)link.link_speed,
1492                                 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
1493                                         ("full-duplex") : ("half-duplex\n"));
1494                                 else
1495                                         printf("Port %d Link Down\n",
1496                                                 (uint8_t)portid);
1497                                 continue;
1498                         }
1499                         /* clear all_ports_up flag if any link down */
1500                         if (link.link_status == 0) {
1501                                 all_ports_up = 0;
1502                                 break;
1503                         }
1504                 }
1505                 /* after finally printing all link status, get out */
1506                 if (print_flag == 1)
1507                         break;
1508
1509                 if (all_ports_up == 0) {
1510                         printf(".");
1511                         fflush(stdout);
1512                         rte_delay_ms(CHECK_INTERVAL);
1513                 }
1514
1515                 /* set the print_flag if all ports up or timeout */
1516                 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
1517                         print_flag = 1;
1518                         printf("done\n");
1519                 }
1520         }
1521 }
1522
1523 int
1524 MAIN(int argc, char **argv)
1525 {
1526         struct lcore_conf *qconf;
1527         int ret;
1528         unsigned nb_ports;
1529         uint16_t queueid;
1530         unsigned lcore_id;
1531         uint64_t hz;
1532         uint32_t n_tx_queue, nb_lcores;
1533         uint8_t portid, nb_rx_queue, queue, socketid;
1534
1535         /* catch SIGINT and restore cpufreq governor to ondemand */
1536         signal(SIGINT, signal_exit_now);
1537
1538         /* init EAL */
1539         ret = rte_eal_init(argc, argv);
1540         if (ret < 0)
1541                 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
1542         argc -= ret;
1543         argv += ret;
1544
1545         /* init RTE timer library to be used late */
1546         rte_timer_subsystem_init();
1547
1548         /* parse application arguments (after the EAL ones) */
1549         ret = parse_args(argc, argv);
1550         if (ret < 0)
1551                 rte_exit(EXIT_FAILURE, "Invalid L3FWD parameters\n");
1552
1553         if (check_lcore_params() < 0)
1554                 rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
1555
1556         ret = init_lcore_rx_queues();
1557         if (ret < 0)
1558                 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
1559
1560
1561         /* init driver(s) */
1562         if (rte_pmd_init_all() < 0)
1563                 rte_exit(EXIT_FAILURE, "Cannot init pmd\n");
1564
1565         if (rte_eal_pci_probe() < 0)
1566                 rte_exit(EXIT_FAILURE, "Cannot probe PCI\n");
1567
1568         nb_ports = rte_eth_dev_count();
1569         if (nb_ports > RTE_MAX_ETHPORTS)
1570                 nb_ports = RTE_MAX_ETHPORTS;
1571
1572         if (check_port_config(nb_ports) < 0)
1573                 rte_exit(EXIT_FAILURE, "check_port_config failed\n");
1574
1575         nb_lcores = rte_lcore_count();
1576
1577         /* initialize all ports */
1578         for (portid = 0; portid < nb_ports; portid++) {
1579                 /* skip ports that are not enabled */
1580                 if ((enabled_port_mask & (1 << portid)) == 0) {
1581                         printf("\nSkipping disabled port %d\n", portid);
1582                         continue;
1583                 }
1584
1585                 /* init port */
1586                 printf("Initializing port %d ... ", portid );
1587                 fflush(stdout);
1588
1589                 nb_rx_queue = get_port_n_rx_queues(portid);
1590                 n_tx_queue = nb_lcores;
1591                 if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
1592                         n_tx_queue = MAX_TX_QUEUE_PER_PORT;
1593                 printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
1594                         nb_rx_queue, (unsigned)n_tx_queue );
1595                 ret = rte_eth_dev_configure(portid, nb_rx_queue,
1596                                         (uint16_t)n_tx_queue, &port_conf);
1597                 if (ret < 0)
1598                         rte_exit(EXIT_FAILURE, "Cannot configure device: "
1599                                         "err=%d, port=%d\n", ret, portid);
1600
1601                 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
1602                 print_ethaddr(" Address:", &ports_eth_addr[portid]);
1603                 printf(", ");
1604
1605                 /* init memory */
1606                 ret = init_mem(NB_MBUF);
1607                 if (ret < 0)
1608                         rte_exit(EXIT_FAILURE, "init_mem failed\n");
1609
1610                 /* init one TX queue per couple (lcore,port) */
1611                 queueid = 0;
1612                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1613                         if (rte_lcore_is_enabled(lcore_id) == 0)
1614                                 continue;
1615
1616                         if (numa_on)
1617                                 socketid = \
1618                                 (uint8_t)rte_lcore_to_socket_id(lcore_id);
1619                         else
1620                                 socketid = 0;
1621
1622                         printf("txq=%u,%d,%d ", lcore_id, queueid, socketid);
1623                         fflush(stdout);
1624                         ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
1625                                                      socketid, &tx_conf);
1626                         if (ret < 0)
1627                                 rte_exit(EXIT_FAILURE,
1628                                         "rte_eth_tx_queue_setup: err=%d, "
1629                                                 "port=%d\n", ret, portid);
1630
1631                         qconf = &lcore_conf[lcore_id];
1632                         qconf->tx_queue_id[portid] = queueid;
1633                         queueid++;
1634                 }
1635                 printf("\n");
1636         }
1637
1638         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1639                 if (rte_lcore_is_enabled(lcore_id) == 0)
1640                         continue;
1641
1642                 /* init power management library */
1643                 ret = rte_power_init(lcore_id);
1644                 if (ret)
1645                         rte_exit(EXIT_FAILURE, "Power management library "
1646                                 "initialization failed on core%u\n", lcore_id);
1647
1648                 /* init timer structures for each enabled lcore */
1649                 rte_timer_init(&power_timers[lcore_id]);
1650                 hz = rte_get_timer_hz();
1651                 rte_timer_reset(&power_timers[lcore_id],
1652                         hz/TIMER_NUMBER_PER_SECOND, SINGLE, lcore_id,
1653                                                 power_timer_cb, NULL);
1654
1655                 qconf = &lcore_conf[lcore_id];
1656                 printf("\nInitializing rx queues on lcore %u ... ", lcore_id );
1657                 fflush(stdout);
1658                 /* init RX queues */
1659                 for(queue = 0; queue < qconf->n_rx_queue; ++queue) {
1660                         portid = qconf->rx_queue_list[queue].port_id;
1661                         queueid = qconf->rx_queue_list[queue].queue_id;
1662
1663                         if (numa_on)
1664                                 socketid = \
1665                                 (uint8_t)rte_lcore_to_socket_id(lcore_id);
1666                         else
1667                                 socketid = 0;
1668
1669                         printf("rxq=%d,%d,%d ", portid, queueid, socketid);
1670                         fflush(stdout);
1671
1672                         ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
1673                                 socketid, &rx_conf, pktmbuf_pool[socketid]);
1674                         if (ret < 0)
1675                                 rte_exit(EXIT_FAILURE,
1676                                         "rte_eth_rx_queue_setup: err=%d, "
1677                                                 "port=%d\n", ret, portid);
1678                 }
1679         }
1680
1681         printf("\n");
1682
1683         /* start ports */
1684         for (portid = 0; portid < nb_ports; portid++) {
1685                 if ((enabled_port_mask & (1 << portid)) == 0) {
1686                         continue;
1687                 }
1688                 /* Start device */
1689                 ret = rte_eth_dev_start(portid);
1690                 if (ret < 0)
1691                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, "
1692                                                 "port=%d\n", ret, portid);
1693
1694                 /*
1695                  * If enabled, put device in promiscuous mode.
1696                  * This allows IO forwarding mode to forward packets
1697                  * to itself through 2 cross-connected  ports of the
1698                  * target machine.
1699                  */
1700                 if (promiscuous_on)
1701                         rte_eth_promiscuous_enable(portid);
1702         }
1703
1704         check_all_ports_link_status((uint8_t)nb_ports, enabled_port_mask);
1705
1706         /* launch per-lcore init on every lcore */
1707         rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
1708         RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1709                 if (rte_eal_wait_lcore(lcore_id) < 0)
1710                         return -1;
1711         }
1712
1713         return 0;
1714 }