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 #include "main.h"
  80
  81 #define RTE_LOGTYPE_L3FWD_POWER RTE_LOGTYPE_USER1
  82
  83 #define MAX_PKT_BURST 32
  84
  85 #define MIN_ZERO_POLL_COUNT 5
  86
  87 /* around 100ms at 2 Ghz */
  88 #define TIMER_RESOLUTION_CYCLES           200000000ULL
  89 /* 100 ms interval */
  90 #define TIMER_NUMBER_PER_SECOND           10
  91 /* 100000 us */
  92 #define SCALING_PERIOD                    (1000000/TIMER_NUMBER_PER_SECOND)
  93 #define SCALING_DOWN_TIME_RATIO_THRESHOLD 0.25
  94
  95 #define APP_LOOKUP_EXACT_MATCH          0
  96 #define APP_LOOKUP_LPM                  1
  97 #define DO_RFC_1812_CHECKS
  98
  99 #ifndef APP_LOOKUP_METHOD
 100 #define APP_LOOKUP_METHOD             APP_LOOKUP_LPM
 101 #endif
 102
 103 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 104 #include <rte_hash.h>
 105 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 106 #include <rte_lpm.h>
 107 #else
 108 #error "APP_LOOKUP_METHOD set to incorrect value"
 109 #endif
 110
 111 #ifndef IPv6_BYTES
 112 #define IPv6_BYTES_FMT "%02x%02x:%02x%02x:%02x%02x:%02x%02x:"\
 113                        "%02x%02x:%02x%02x:%02x%02x:%02x%02x"
 114 #define IPv6_BYTES(addr) \
 115         addr[0],  addr[1], addr[2],  addr[3], \
 116         addr[4],  addr[5], addr[6],  addr[7], \
 117         addr[8],  addr[9], addr[10], addr[11],\
 118         addr[12], addr[13],addr[14], addr[15]
 119 #endif
 120
 121 #define MAX_JUMBO_PKT_LEN  9600
 122
 123 #define IPV6_ADDR_LEN 16
 124
 125 #define MEMPOOL_CACHE_SIZE 256
 126
 127 #define MBUF_SIZE (2048 + sizeof(struct rte_mbuf) + RTE_PKTMBUF_HEADROOM)
 128
 129 /*
 130  * This expression is used to calculate the number of mbufs needed depending on
 131  * user input, taking into account memory for rx and tx hardware rings, cache
 132  * per lcore and mtable per port per lcore. RTE_MAX is used to ensure that
 133  * NB_MBUF never goes below a minimum value of 8192.
 134  */
 135
 136 #define NB_MBUF RTE_MAX ( \
 137         (nb_ports*nb_rx_queue*RTE_TEST_RX_DESC_DEFAULT + \
 138         nb_ports*nb_lcores*MAX_PKT_BURST + \
 139         nb_ports*n_tx_queue*RTE_TEST_TX_DESC_DEFAULT + \
 140         nb_lcores*MEMPOOL_CACHE_SIZE), \
 141         (unsigned)8192)
 142
 143 /*
 144  * RX and TX Prefetch, Host, and Write-back threshold values should be
 145  * carefully set for optimal performance. Consult the network
 146  * controller's datasheet and supporting DPDK documentation for guidance
 147  * on how these parameters should be set.
 148  */
 149 #define RX_PTHRESH 8 /**< Default values of RX prefetch threshold reg. */
 150 #define RX_HTHRESH 8 /**< Default values of RX host threshold reg. */
 151 #define RX_WTHRESH 4 /**< Default values of RX write-back threshold reg. */
 152
 153 /*
 154  * These default values are optimized for use with the Intel(R) 82599 10 GbE
 155  * Controller and the DPDK ixgbe PMD. Consider using other values for other
 156  * network controllers and/or network drivers.
 157  */
 158 #define TX_PTHRESH 36 /**< Default values of TX prefetch threshold reg. */
 159 #define TX_HTHRESH 0  /**< Default values of TX host threshold reg. */
 160 #define TX_WTHRESH 0  /**< Default values of TX write-back threshold reg. */
 161
 162 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
 163
 164 #define NB_SOCKETS 8
 165
 166 /* Configure how many packets ahead to prefetch, when reading packets */
 167 #define PREFETCH_OFFSET 3
 168
 169 /*
 170  * Configurable number of RX/TX ring descriptors
 171  */
 172 #define RTE_TEST_RX_DESC_DEFAULT 128
 173 #define RTE_TEST_TX_DESC_DEFAULT 512
 174 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
 175 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
 176
 177 /* ethernet addresses of ports */
 178 static struct ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
 179
 180 /* mask of enabled ports */
 181 static uint32_t enabled_port_mask = 0;
 182 /* Ports set in promiscuous mode off by default. */
 183 static int promiscuous_on = 0;
 184 /* NUMA is enabled by default. */
 185 static int numa_on = 1;
 186
 187 enum freq_scale_hint_t
 188 {
 189         FREQ_LOWER    =      -1,
 190         FREQ_CURRENT  =       0,
 191         FREQ_HIGHER   =       1,
 192         FREQ_HIGHEST  =       2
 193 };
 194
 195 struct mbuf_table {
 196         uint16_t len;
 197         struct rte_mbuf *m_table[MAX_PKT_BURST];
 198 };
 199
 200 struct lcore_rx_queue {
 201         uint8_t port_id;
 202         uint8_t queue_id;
 203         enum freq_scale_hint_t freq_up_hint;
 204         uint32_t zero_rx_packet_count;
 205         uint32_t idle_hint;
 206 } __rte_cache_aligned;
 207
 208 #define MAX_RX_QUEUE_PER_LCORE 16
 209 #define MAX_TX_QUEUE_PER_PORT RTE_MAX_ETHPORTS
 210 #define MAX_RX_QUEUE_PER_PORT 128
 211
 212 #define MAX_LCORE_PARAMS 1024
 213 struct lcore_params {
 214         uint8_t port_id;
 215         uint8_t queue_id;
 216         uint8_t lcore_id;
 217 } __rte_cache_aligned;
 218
 219 static struct lcore_params lcore_params_array[MAX_LCORE_PARAMS];
 220 static struct lcore_params lcore_params_array_default[] = {
 221         {0, 0, 2},
 222         {0, 1, 2},
 223         {0, 2, 2},
 224         {1, 0, 2},
 225         {1, 1, 2},
 226         {1, 2, 2},
 227         {2, 0, 2},
 228         {3, 0, 3},
 229         {3, 1, 3},
 230 };
 231
 232 static struct lcore_params * lcore_params = lcore_params_array_default;
 233 static uint16_t nb_lcore_params = sizeof(lcore_params_array_default) /
 234                                 sizeof(lcore_params_array_default[0]);
 235
 236 static struct rte_eth_conf port_conf = {
 237         .rxmode = {
 238                 .max_rx_pkt_len = ETHER_MAX_LEN,
 239                 .split_hdr_size = 0,
 240                 .header_split   = 0, /**< Header Split disabled */
 241                 .hw_ip_checksum = 1, /**< IP checksum offload enabled */
 242                 .hw_vlan_filter = 0, /**< VLAN filtering disabled */
 243                 .jumbo_frame    = 0, /**< Jumbo Frame Support disabled */
 244                 .hw_strip_crc   = 0, /**< CRC stripped by hardware */
 245         },
 246         .rx_adv_conf = {
 247                 .rss_conf = {
 248                         .rss_key = NULL,
 249                         .rss_hf = ETH_RSS_IPV4 | ETH_RSS_IPV6,
 250                 },
 251         },
 252         .txmode = {
 253                 .mq_mode = ETH_DCB_NONE,
 254         },
 255 };
 256
 257 static const struct rte_eth_rxconf rx_conf = {
 258         .rx_thresh = {
 259                 .pthresh = RX_PTHRESH,
 260                 .hthresh = RX_HTHRESH,
 261                 .wthresh = RX_WTHRESH,
 262         },
 263         .rx_free_thresh = 32,
 264 };
 265
 266 static const struct rte_eth_txconf tx_conf = {
 267         .tx_thresh = {
 268                 .pthresh = TX_PTHRESH,
 269                 .hthresh = TX_HTHRESH,
 270                 .wthresh = TX_WTHRESH,
 271         },
 272         .tx_free_thresh = 0, /* Use PMD default values */
 273         .tx_rs_thresh = 0, /* Use PMD default values */
 274         .txq_flags = 0x0,
 275 };
 276
 277 static struct rte_mempool * pktmbuf_pool[NB_SOCKETS];
 278
 279
 280 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 281
 282 #ifdef RTE_MACHINE_CPUFLAG_SSE4_2
 283 #include <rte_hash_crc.h>
 284 #define DEFAULT_HASH_FUNC       rte_hash_crc
 285 #else
 286 #include <rte_jhash.h>
 287 #define DEFAULT_HASH_FUNC       rte_jhash
 288 #endif
 289
 290 struct ipv4_5tuple {
 291         uint32_t ip_dst;
 292         uint32_t ip_src;
 293         uint16_t port_dst;
 294         uint16_t port_src;
 295         uint8_t  proto;
 296 } __attribute__((__packed__));
 297
 298 struct ipv6_5tuple {
 299         uint8_t  ip_dst[IPV6_ADDR_LEN];
 300         uint8_t  ip_src[IPV6_ADDR_LEN];
 301         uint16_t port_dst;
 302         uint16_t port_src;
 303         uint8_t  proto;
 304 } __attribute__((__packed__));
 305
 306 struct ipv4_l3fwd_route {
 307         struct ipv4_5tuple key;
 308         uint8_t if_out;
 309 };
 310
 311 struct ipv6_l3fwd_route {
 312         struct ipv6_5tuple key;
 313         uint8_t if_out;
 314 };
 315
 316 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
 317         {{IPv4(100,10,0,1), IPv4(200,10,0,1), 101, 11, IPPROTO_TCP}, 0},
 318         {{IPv4(100,20,0,2), IPv4(200,20,0,2), 102, 12, IPPROTO_TCP}, 1},
 319         {{IPv4(100,30,0,3), IPv4(200,30,0,3), 103, 13, IPPROTO_TCP}, 2},
 320         {{IPv4(100,40,0,4), IPv4(200,40,0,4), 104, 14, IPPROTO_TCP}, 3},
 321 };
 322
 323 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
 324         {
 325                 {
 326                         {0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
 327                          0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
 328                         {0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
 329                          0x02, 0x1e, 0x67, 0xff, 0xfe, 0x0d, 0xb6, 0x0a},
 330                          1, 10, IPPROTO_UDP
 331                 }, 4
 332         },
 333 };
 334
 335 typedef struct rte_hash lookup_struct_t;
 336 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
 337 static lookup_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
 338
 339 #define L3FWD_HASH_ENTRIES      1024
 340
 341 #define IPV4_L3FWD_NUM_ROUTES \
 342         (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
 343
 344 #define IPV6_L3FWD_NUM_ROUTES \
 345         (sizeof(ipv6_l3fwd_route_array) / sizeof(ipv6_l3fwd_route_array[0]))
 346
 347 static uint8_t ipv4_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 348 static uint8_t ipv6_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 349 #endif
 350
 351 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 352 struct ipv4_l3fwd_route {
 353         uint32_t ip;
 354         uint8_t  depth;
 355         uint8_t  if_out;
 356 };
 357
 358 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
 359         {IPv4(1,1,1,0), 24, 0},
 360         {IPv4(2,1,1,0), 24, 1},
 361         {IPv4(3,1,1,0), 24, 2},
 362         {IPv4(4,1,1,0), 24, 3},
 363         {IPv4(5,1,1,0), 24, 4},
 364         {IPv4(6,1,1,0), 24, 5},
 365         {IPv4(7,1,1,0), 24, 6},
 366         {IPv4(8,1,1,0), 24, 7},
 367 };
 368
 369 #define IPV4_L3FWD_NUM_ROUTES \
 370         (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
 371
 372 #define IPV4_L3FWD_LPM_MAX_RULES     1024
 373
 374 typedef struct rte_lpm lookup_struct_t;
 375 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
 376 #endif
 377
 378 struct lcore_conf {
 379         uint16_t n_rx_queue;
 380         struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
 381         uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
 382         struct mbuf_table tx_mbufs[RTE_MAX_ETHPORTS];
 383         lookup_struct_t * ipv4_lookup_struct;
 384         lookup_struct_t * ipv6_lookup_struct;
 385 } __rte_cache_aligned;
 386
 387 struct lcore_stats {
 388         /* total sleep time in ms since last frequency scaling down */
 389         uint32_t sleep_time;
 390         /* number of long sleep recently */
 391         uint32_t nb_long_sleep;
 392         /* freq. scaling up trend */
 393         uint32_t trend;
 394         /* total packet processed recently */
 395         uint64_t nb_rx_processed;
 396         /* total iterations looped recently */
 397         uint64_t nb_iteration_looped;
 398         uint32_t padding[9];
 399 } __rte_cache_aligned;
 400
 401 static struct lcore_conf lcore_conf[RTE_MAX_LCORE] __rte_cache_aligned;
 402 static struct lcore_stats stats[RTE_MAX_LCORE] __rte_cache_aligned;
 403 static struct rte_timer power_timers[RTE_MAX_LCORE];
 404
 405 static inline uint32_t power_idle_heuristic(uint32_t zero_rx_packet_count);
 406 static inline enum freq_scale_hint_t power_freq_scaleup_heuristic( \
 407                         unsigned lcore_id, uint8_t port_id, uint16_t queue_id);
 408
 409 /* exit signal handler */
 410 static void
 411 signal_exit_now(int sigtype)
 412 {
 413         unsigned lcore_id;
 414         int ret;
 415
 416         if (sigtype == SIGINT) {
 417                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
 418                         if (rte_lcore_is_enabled(lcore_id) == 0)
 419                                 continue;
 420
 421                         /* init power management library */
 422                         ret = rte_power_exit(lcore_id);
 423                         if (ret)
 424                                 rte_exit(EXIT_FAILURE, "Power management "
 425                                         "library de-initialization failed on "
 426                                                         "core%u\n", lcore_id);
 427                 }
 428         }
 429
 430         rte_exit(EXIT_SUCCESS, "User forced exit\n");
 431 }
 432
 433 /*  Freqency scale down timer callback */
 434 static void
 435 power_timer_cb(__attribute__((unused)) struct rte_timer *tim,
 436                           __attribute__((unused)) void *arg)
 437 {
 438         uint64_t hz;
 439         float sleep_time_ratio;
 440         unsigned lcore_id = rte_lcore_id();
 441
 442         /* accumulate total execution time in us when callback is invoked */
 443         sleep_time_ratio = (float)(stats[lcore_id].sleep_time) /
 444                                         (float)SCALING_PERIOD;
 445
 446         /**
 447          * check whether need to scale down frequency a step if it sleep a lot.
 448          */
 449         if (sleep_time_ratio >= SCALING_DOWN_TIME_RATIO_THRESHOLD)
 450                 rte_power_freq_down(lcore_id);
 451         else if ( (unsigned)(stats[lcore_id].nb_rx_processed /
 452                 stats[lcore_id].nb_iteration_looped) < MAX_PKT_BURST)
 453                 /**
 454                  * scale down a step if average packet per iteration less
 455                  * than expectation.
 456                  */
 457                 rte_power_freq_down(lcore_id);
 458
 459         /**
 460          * initialize another timer according to current frequency to ensure
 461          * timer interval is relatively fixed.
 462          */
 463         hz = rte_get_timer_hz();
 464         rte_timer_reset(&power_timers[lcore_id], hz/TIMER_NUMBER_PER_SECOND,
 465                                 SINGLE, lcore_id, power_timer_cb, NULL);
 466
 467         stats[lcore_id].nb_rx_processed = 0;
 468         stats[lcore_id].nb_iteration_looped = 0;
 469
 470         stats[lcore_id].sleep_time = 0;
 471 }
 472
 473 /* Send burst of packets on an output interface */
 474 static inline int
 475 send_burst(struct lcore_conf *qconf, uint16_t n, uint8_t port)
 476 {
 477         struct rte_mbuf **m_table;
 478         int ret;
 479         uint16_t queueid;
 480
 481         queueid = qconf->tx_queue_id[port];
 482         m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
 483
 484         ret = rte_eth_tx_burst(port, queueid, m_table, n);
 485         if (unlikely(ret < n)) {
 486                 do {
 487                         rte_pktmbuf_free(m_table[ret]);
 488                 } while (++ret < n);
 489         }
 490
 491         return 0;
 492 }
 493
 494 /* Enqueue a single packet, and send burst if queue is filled */
 495 static inline int
 496 send_single_packet(struct rte_mbuf *m, uint8_t port)
 497 {
 498         uint32_t lcore_id;
 499         uint16_t len;
 500         struct lcore_conf *qconf;
 501
 502         lcore_id = rte_lcore_id();
 503
 504         qconf = &lcore_conf[lcore_id];
 505         len = qconf->tx_mbufs[port].len;
 506         qconf->tx_mbufs[port].m_table[len] = m;
 507         len++;
 508
 509         /* enough pkts to be sent */
 510         if (unlikely(len == MAX_PKT_BURST)) {
 511                 send_burst(qconf, MAX_PKT_BURST, port);
 512                 len = 0;
 513         }
 514
 515         qconf->tx_mbufs[port].len = len;
 516         return 0;
 517 }
 518
 519 #ifdef DO_RFC_1812_CHECKS
 520 static inline int
 521 is_valid_ipv4_pkt(struct ipv4_hdr *pkt, uint32_t link_len)
 522 {
 523         /* From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2 */
 524         /*
 525          * 1. The packet length reported by the Link Layer must be large
 526          * enough to hold the minimum length legal IP datagram (20 bytes).
 527          */
 528         if (link_len < sizeof(struct ipv4_hdr))
 529                 return -1;
 530
 531         /* 2. The IP checksum must be correct. */
 532         /* this is checked in H/W */
 533
 534         /*
 535          * 3. The IP version number must be 4. If the version number is not 4
 536          * then the packet may be another version of IP, such as IPng or
 537          * ST-II.
 538          */
 539         if (((pkt->version_ihl) >> 4) != 4)
 540                 return -3;
 541         /*
 542          * 4. The IP header length field must be large enough to hold the
 543          * minimum length legal IP datagram (20 bytes = 5 words).
 544          */
 545         if ((pkt->version_ihl & 0xf) < 5)
 546                 return -4;
 547
 548         /*
 549          * 5. The IP total length field must be large enough to hold the IP
 550          * datagram header, whose length is specified in the IP header length
 551          * field.
 552          */
 553         if (rte_cpu_to_be_16(pkt->total_length) < sizeof(struct ipv4_hdr))
 554                 return -5;
 555
 556         return 0;
 557 }
 558 #endif
 559
 560 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 561 static void
 562 print_ipv4_key(struct ipv4_5tuple key)
 563 {
 564         printf("IP dst = %08x, IP src = %08x, port dst = %d, port src = %d, "
 565                 "proto = %d\n", (unsigned)key.ip_dst, (unsigned)key.ip_src,
 566                                 key.port_dst, key.port_src, key.proto);
 567 }
 568 static void
 569 print_ipv6_key(struct ipv6_5tuple key)
 570 {
 571         printf( "IP dst = " IPv6_BYTES_FMT ", IP src = " IPv6_BYTES_FMT ", "
 572                 "port dst = %d, port src = %d, proto = %d\n",
 573                 IPv6_BYTES(key.ip_dst), IPv6_BYTES(key.ip_src),
 574                 key.port_dst, key.port_src, key.proto);
 575 }
 576
 577 static inline uint8_t
 578 get_ipv4_dst_port(struct ipv4_hdr *ipv4_hdr, uint8_t portid,
 579                 lookup_struct_t * ipv4_l3fwd_lookup_struct)
 580 {
 581         struct ipv4_5tuple key;
 582         struct tcp_hdr *tcp;
 583         struct udp_hdr *udp;
 584         int ret = 0;
 585
 586         key.ip_dst = rte_be_to_cpu_32(ipv4_hdr->dst_addr);
 587         key.ip_src = rte_be_to_cpu_32(ipv4_hdr->src_addr);
 588         key.proto = ipv4_hdr->next_proto_id;
 589
 590         switch (ipv4_hdr->next_proto_id) {
 591         case IPPROTO_TCP:
 592                 tcp = (struct tcp_hdr *)((unsigned char *)ipv4_hdr +
 593                                         sizeof(struct ipv4_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 *)ipv4_hdr +
 600                                         sizeof(struct ipv4_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(ipv4_l3fwd_lookup_struct, (const void *)&key);
 613         return (uint8_t)((ret < 0)? portid : ipv4_l3fwd_out_if[ret]);
 614 }
 615
 616 static inline uint8_t
 617 get_ipv6_dst_port(struct ipv6_hdr *ipv6_hdr,  uint8_t portid,
 618                         lookup_struct_t *ipv6_l3fwd_lookup_struct)
 619 {
 620         struct ipv6_5tuple key;
 621         struct tcp_hdr *tcp;
 622         struct udp_hdr *udp;
 623         int ret = 0;
 624
 625         memcpy(key.ip_dst, ipv6_hdr->dst_addr, IPV6_ADDR_LEN);
 626         memcpy(key.ip_src, ipv6_hdr->src_addr, IPV6_ADDR_LEN);
 627
 628         key.proto = ipv6_hdr->proto;
 629
 630         switch (ipv6_hdr->proto) {
 631         case IPPROTO_TCP:
 632                 tcp = (struct tcp_hdr *)((unsigned char *) ipv6_hdr +
 633                                         sizeof(struct ipv6_hdr));
 634                 key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
 635                 key.port_src = rte_be_to_cpu_16(tcp->src_port);
 636                 break;
 637
 638         case IPPROTO_UDP:
 639                 udp = (struct udp_hdr *)((unsigned char *) ipv6_hdr +
 640                                         sizeof(struct ipv6_hdr));
 641                 key.port_dst = rte_be_to_cpu_16(udp->dst_port);
 642                 key.port_src = rte_be_to_cpu_16(udp->src_port);
 643                 break;
 644
 645         default:
 646                 key.port_dst = 0;
 647                 key.port_src = 0;
 648                 break;
 649         }
 650
 651         /* Find destination port */
 652         ret = rte_hash_lookup(ipv6_l3fwd_lookup_struct, (const void *)&key);
 653         return (uint8_t)((ret < 0)? portid : ipv6_l3fwd_out_if[ret]);
 654 }
 655 #endif
 656
 657 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 658 static inline uint8_t
 659 get_ipv4_dst_port(struct ipv4_hdr *ipv4_hdr, uint8_t portid,
 660                 lookup_struct_t *ipv4_l3fwd_lookup_struct)
 661 {
 662         uint8_t next_hop;
 663
 664         return (uint8_t) ((rte_lpm_lookup(ipv4_l3fwd_lookup_struct,
 665                         rte_be_to_cpu_32(ipv4_hdr->dst_addr), &next_hop) == 0)?
 666                         next_hop : portid);
 667 }
 668 #endif
 669
 670 static inline void
 671 l3fwd_simple_forward(struct rte_mbuf *m, uint8_t portid,
 672                                 struct lcore_conf *qconf)
 673 {
 674         struct ether_hdr *eth_hdr;
 675         struct ipv4_hdr *ipv4_hdr;
 676         void *d_addr_bytes;
 677         uint8_t dst_port;
 678
 679         eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
 680
 681         if (m->ol_flags & PKT_RX_IPV4_HDR) {
 682                 /* Handle IPv4 headers.*/
 683                 ipv4_hdr =
 684                         (struct ipv4_hdr *)(rte_pktmbuf_mtod(m, unsigned char*)
 685                                                 + sizeof(struct ether_hdr));
 686
 687 #ifdef DO_RFC_1812_CHECKS
 688                 /* Check to make sure the packet is valid (RFC1812) */
 689                 if (is_valid_ipv4_pkt(ipv4_hdr, m->pkt.pkt_len) < 0) {
 690                         rte_pktmbuf_free(m);
 691                         return;
 692                 }
 693 #endif
 694
 695                 dst_port = get_ipv4_dst_port(ipv4_hdr, portid,
 696                                         qconf->ipv4_lookup_struct);
 697                 if (dst_port >= RTE_MAX_ETHPORTS ||
 698                                 (enabled_port_mask & 1 << dst_port) == 0)
 699                         dst_port = portid;
 700
 701                 /* 02:00:00:00:00:xx */
 702                 d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
 703                 *((uint64_t *)d_addr_bytes) =
 704                         0x000000000002 + ((uint64_t)dst_port << 40);
 705
 706 #ifdef DO_RFC_1812_CHECKS
 707                 /* Update time to live and header checksum */
 708                 --(ipv4_hdr->time_to_live);
 709                 ++(ipv4_hdr->hdr_checksum);
 710 #endif
 711
 712                 /* src addr */
 713                 ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
 714
 715                 send_single_packet(m, dst_port);
 716         }
 717         else {
 718                 /* Handle IPv6 headers.*/
 719 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 720                 struct ipv6_hdr *ipv6_hdr;
 721
 722                 ipv6_hdr =
 723                         (struct ipv6_hdr *)(rte_pktmbuf_mtod(m, unsigned char*)
 724                                                 + sizeof(struct ether_hdr));
 725
 726                 dst_port = get_ipv6_dst_port(ipv6_hdr, portid,
 727                                         qconf->ipv6_lookup_struct);
 728
 729                 if (dst_port >= RTE_MAX_ETHPORTS ||
 730                                 (enabled_port_mask & 1 << dst_port) == 0)
 731                         dst_port = portid;
 732
 733                 /* 02:00:00:00:00:xx */
 734                 d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
 735                 *((uint64_t *)d_addr_bytes) =
 736                         0x000000000002 + ((uint64_t)dst_port << 40);
 737
 738                 /* src addr */
 739                 ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
 740
 741                 send_single_packet(m, dst_port);
 742 #else
 743                 /* We don't currently handle IPv6 packets in LPM mode. */
 744                 rte_pktmbuf_free(m);
 745 #endif
 746         }
 747
 748 }
 749
 750 #define SLEEP_GEAR1_THRESHOLD            100
 751 #define SLEEP_GEAR2_THRESHOLD            1000
 752
 753 static inline uint32_t
 754 power_idle_heuristic(uint32_t zero_rx_packet_count)
 755 {
 756         /* If zero count is less than 100, use it as the sleep time in us */
 757         if (zero_rx_packet_count < SLEEP_GEAR1_THRESHOLD)
 758                 return zero_rx_packet_count;
 759         /* If zero count is less than 1000, sleep time should be 100 us */
 760         else if ((zero_rx_packet_count >= SLEEP_GEAR1_THRESHOLD) &&
 761                         (zero_rx_packet_count < SLEEP_GEAR2_THRESHOLD))
 762                 return SLEEP_GEAR1_THRESHOLD;
 763         /* If zero count is greater than 1000, sleep time should be 1000 us */
 764         else if (zero_rx_packet_count >= SLEEP_GEAR2_THRESHOLD)
 765                 return SLEEP_GEAR2_THRESHOLD;
 766
 767         return 0;
 768 }
 769
 770 static inline enum freq_scale_hint_t
 771 power_freq_scaleup_heuristic(unsigned lcore_id,
 772                              uint8_t port_id,
 773                              uint16_t queue_id)
 774 {
 775 /**
 776  * HW Rx queue size is 128 by default, Rx burst read at maximum 32 entries
 777  * per iteration
 778  */
 779 #define FREQ_GEAR1_RX_PACKET_THRESHOLD             MAX_PKT_BURST
 780 #define FREQ_GEAR2_RX_PACKET_THRESHOLD             (MAX_PKT_BURST*2)
 781 #define FREQ_GEAR3_RX_PACKET_THRESHOLD             (MAX_PKT_BURST*3)
 782 #define FREQ_UP_TREND1_ACC   1
 783 #define FREQ_UP_TREND2_ACC   100
 784 #define FREQ_UP_THRESHOLD    10000
 785
 786         if (likely(rte_eth_rx_descriptor_done(port_id, queue_id,
 787                         FREQ_GEAR3_RX_PACKET_THRESHOLD) > 0)) {
 788                 stats[lcore_id].trend = 0;
 789                 return FREQ_HIGHEST;
 790         } else if (likely(rte_eth_rx_descriptor_done(port_id, queue_id,
 791                         FREQ_GEAR2_RX_PACKET_THRESHOLD) > 0))
 792                 stats[lcore_id].trend += FREQ_UP_TREND2_ACC;
 793         else if (likely(rte_eth_rx_descriptor_done(port_id, queue_id,
 794                         FREQ_GEAR1_RX_PACKET_THRESHOLD) > 0))
 795                 stats[lcore_id].trend += FREQ_UP_TREND1_ACC;
 796
 797         if (likely(stats[lcore_id].trend > FREQ_UP_THRESHOLD)) {
 798                 stats[lcore_id].trend = 0;
 799                 return FREQ_HIGHER;
 800         }
 801
 802         return FREQ_CURRENT;
 803 }
 804
 805 /* main processing loop */
 806 static int
 807 main_loop(__attribute__((unused)) void *dummy)
 808 {
 809         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
 810         unsigned lcore_id;
 811         uint64_t prev_tsc, diff_tsc, cur_tsc;
 812         uint64_t prev_tsc_power = 0, cur_tsc_power, diff_tsc_power;
 813         int i, j, nb_rx;
 814         uint8_t portid, queueid;
 815         struct lcore_conf *qconf;
 816         struct lcore_rx_queue *rx_queue;
 817         enum freq_scale_hint_t lcore_scaleup_hint;
 818
 819         uint32_t lcore_rx_idle_count = 0;
 820         uint32_t lcore_idle_hint = 0;
 821
 822         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
 823
 824         prev_tsc = 0;
 825
 826         lcore_id = rte_lcore_id();
 827         qconf = &lcore_conf[lcore_id];
 828
 829         if (qconf->n_rx_queue == 0) {
 830                 RTE_LOG(INFO, L3FWD_POWER, "lcore %u has nothing to do\n", lcore_id);
 831                 return 0;
 832         }
 833
 834         RTE_LOG(INFO, L3FWD_POWER, "entering main loop on lcore %u\n", lcore_id);
 835
 836         for (i = 0; i < qconf->n_rx_queue; i++) {
 837
 838                 portid = qconf->rx_queue_list[i].port_id;
 839                 queueid = qconf->rx_queue_list[i].queue_id;
 840                 RTE_LOG(INFO, L3FWD_POWER, " -- lcoreid=%u portid=%hhu "
 841                         "rxqueueid=%hhu\n", lcore_id, portid, queueid);
 842         }
 843
 844         while (1) {
 845                 stats[lcore_id].nb_iteration_looped++;
 846
 847                 cur_tsc = rte_rdtsc();
 848                 cur_tsc_power = cur_tsc;
 849
 850                 /*
 851                  * TX burst queue drain
 852                  */
 853                 diff_tsc = cur_tsc - prev_tsc;
 854                 if (unlikely(diff_tsc > drain_tsc)) {
 855
 856                         /*
 857                          * This could be optimized (use queueid instead of
 858                          * portid), but it is not called so often
 859                          */
 860                         for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
 861                                 if (qconf->tx_mbufs[portid].len == 0)
 862                                         continue;
 863                                 send_burst(&lcore_conf[lcore_id],
 864                                         qconf->tx_mbufs[portid].len,
 865                                         portid);
 866                                 qconf->tx_mbufs[portid].len = 0;
 867                         }
 868
 869                         prev_tsc = cur_tsc;
 870                 }
 871
 872                 diff_tsc_power = cur_tsc_power - prev_tsc_power;
 873                 if (diff_tsc_power > TIMER_RESOLUTION_CYCLES) {
 874                         rte_timer_manage();
 875                         prev_tsc_power = cur_tsc_power;
 876                 }
 877
 878                 /*
 879                  * Read packet from RX queues
 880                  */
 881                 lcore_scaleup_hint = FREQ_CURRENT;
 882                 lcore_rx_idle_count = 0;
 883                 for (i = 0; i < qconf->n_rx_queue; ++i) {
 884                         rx_queue = &(qconf->rx_queue_list[i]);
 885                         rx_queue->idle_hint = 0;
 886                         portid = rx_queue->port_id;
 887                         queueid = rx_queue->queue_id;
 888
 889                         nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
 890                                                                 MAX_PKT_BURST);
 891                         stats[lcore_id].nb_rx_processed += nb_rx;
 892                         if (unlikely(nb_rx == 0)) {
 893                                 /**
 894                                  * no packet received from rx queue, try to
 895                                  * sleep for a while forcing CPU enter deeper
 896                                  * C states.
 897                                  */
 898                                 rx_queue->zero_rx_packet_count++;
 899
 900                                 if (rx_queue->zero_rx_packet_count <=
 901                                                         MIN_ZERO_POLL_COUNT)
 902                                         continue;
 903
 904                                 rx_queue->idle_hint = power_idle_heuristic(\
 905                                         rx_queue->zero_rx_packet_count);
 906                                 lcore_rx_idle_count++;
 907                         } else {
 908                                 rx_queue->zero_rx_packet_count = 0;
 909
 910                                 /**
 911                                  * do not scale up frequency immediately as
 912                                  * user to kernel space communication is costly
 913                                  * which might impact packet I/O for received
 914                                  * packets.
 915                                  */
 916                                 rx_queue->freq_up_hint =
 917                                         power_freq_scaleup_heuristic(lcore_id,
 918                                                         portid, queueid);
 919                         }
 920
 921                         /* Prefetch first packets */
 922                         for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
 923                                 rte_prefetch0(rte_pktmbuf_mtod(
 924                                                 pkts_burst[j], void *));
 925                         }
 926
 927                         /* Prefetch and forward already prefetched packets */
 928                         for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
 929                                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
 930                                                 j + PREFETCH_OFFSET], void *));
 931                                 l3fwd_simple_forward(pkts_burst[j], portid,
 932                                                                 qconf);
 933                         }
 934
 935                         /* Forward remaining prefetched packets */
 936                         for (; j < nb_rx; j++) {
 937                                 l3fwd_simple_forward(pkts_burst[j], portid,
 938                                                                 qconf);
 939                         }
 940                 }
 941
 942                 if (likely(lcore_rx_idle_count != qconf->n_rx_queue)) {
 943                         for (i = 1, lcore_scaleup_hint =
 944                                 qconf->rx_queue_list[0].freq_up_hint;
 945                                         i < qconf->n_rx_queue; ++i) {
 946                                 rx_queue = &(qconf->rx_queue_list[i]);
 947                                 if (rx_queue->freq_up_hint >
 948                                                 lcore_scaleup_hint)
 949                                         lcore_scaleup_hint =
 950                                                 rx_queue->freq_up_hint;
 951                         }
 952
 953                         if (lcore_scaleup_hint == FREQ_HIGHEST)
 954                                 rte_power_freq_max(lcore_id);
 955                         else if (lcore_scaleup_hint == FREQ_HIGHER)
 956                                 rte_power_freq_up(lcore_id);
 957                 } else {
 958                         /**
 959                          * All Rx queues empty in recent consecutive polls,
 960                          * sleep in a conservative manner, meaning sleep as
 961                          * less as possible.
 962                          */
 963                         for (i = 1, lcore_idle_hint =
 964                                 qconf->rx_queue_list[0].idle_hint;
 965                                         i < qconf->n_rx_queue; ++i) {
 966                                 rx_queue = &(qconf->rx_queue_list[i]);
 967                                 if (rx_queue->idle_hint < lcore_idle_hint)
 968                                         lcore_idle_hint = rx_queue->idle_hint;
 969                         }
 970
 971                         if ( lcore_idle_hint < SLEEP_GEAR1_THRESHOLD)
 972                                 /**
 973                                  * execute "pause" instruction to avoid context
 974                                  * switch for short sleep.
 975                                  */
 976                                 rte_delay_us(lcore_idle_hint);
 977                         else
 978                                 /* long sleep force runing thread to suspend */
 979                                 usleep(lcore_idle_hint);
 980
 981                         stats[lcore_id].sleep_time += lcore_idle_hint;
 982                 }
 983         }
 984 }
 985
 986 static int
 987 check_lcore_params(void)
 988 {
 989         uint8_t queue, lcore;
 990         uint16_t i;
 991         int socketid;
 992
 993         for (i = 0; i < nb_lcore_params; ++i) {
 994                 queue = lcore_params[i].queue_id;
 995                 if (queue >= MAX_RX_QUEUE_PER_PORT) {
 996                         printf("invalid queue number: %hhu\n", queue);
 997                         return -1;
 998                 }
 999                 lcore = lcore_params[i].lcore_id;
1000                 if (!rte_lcore_is_enabled(lcore)) {
1001                         printf("error: lcore %hhu is not enabled in lcore "
1002                                                         "mask\n", lcore);
1003                         return -1;
1004                 }
1005                 if ((socketid = rte_lcore_to_socket_id(lcore) != 0) &&
1006                                                         (numa_on == 0)) {
1007                         printf("warning: lcore %hhu is on socket %d with numa "
1008                                                 "off\n", lcore, socketid);
1009                 }
1010         }
1011         return 0;
1012 }
1013
1014 static int
1015 check_port_config(const unsigned nb_ports)
1016 {
1017         unsigned portid;
1018         uint16_t i;
1019
1020         for (i = 0; i < nb_lcore_params; ++i) {
1021                 portid = lcore_params[i].port_id;
1022                 if ((enabled_port_mask & (1 << portid)) == 0) {
1023                         printf("port %u is not enabled in port mask\n",
1024                                                                 portid);
1025                         return -1;
1026                 }
1027                 if (portid >= nb_ports) {
1028                         printf("port %u is not present on the board\n",
1029                                                                 portid);
1030                         return -1;
1031                 }
1032         }
1033         return 0;
1034 }
1035
1036 static uint8_t
1037 get_port_n_rx_queues(const uint8_t port)
1038 {
1039         int queue = -1;
1040         uint16_t i;
1041
1042         for (i = 0; i < nb_lcore_params; ++i) {
1043                 if (lcore_params[i].port_id == port &&
1044                                 lcore_params[i].queue_id > queue)
1045                         queue = lcore_params[i].queue_id;
1046         }
1047         return (uint8_t)(++queue);
1048 }
1049
1050 static int
1051 init_lcore_rx_queues(void)
1052 {
1053         uint16_t i, nb_rx_queue;
1054         uint8_t lcore;
1055
1056         for (i = 0; i < nb_lcore_params; ++i) {
1057                 lcore = lcore_params[i].lcore_id;
1058                 nb_rx_queue = lcore_conf[lcore].n_rx_queue;
1059                 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
1060                         printf("error: too many queues (%u) for lcore: %u\n",
1061                                 (unsigned)nb_rx_queue + 1, (unsigned)lcore);
1062                         return -1;
1063                 } else {
1064                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
1065                                 lcore_params[i].port_id;
1066                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
1067                                 lcore_params[i].queue_id;
1068                         lcore_conf[lcore].n_rx_queue++;
1069                 }
1070         }
1071         return 0;
1072 }
1073
1074 /* display usage */
1075 static void
1076 print_usage(const char *prgname)
1077 {
1078         printf ("%s [EAL options] -- -p PORTMASK -P"
1079                 "  [--config (port,queue,lcore)[,(port,queue,lcore]]"
1080                 "  [--enable-jumbo [--max-pkt-len PKTLEN]]\n"
1081                 "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
1082                 "  -P : enable promiscuous mode\n"
1083                 "  --config (port,queue,lcore): rx queues configuration\n"
1084                 "  --no-numa: optional, disable numa awareness\n"
1085                 "  --enable-jumbo: enable jumbo frame"
1086                 " which max packet len is PKTLEN in decimal (64-9600)\n",
1087                 prgname);
1088 }
1089
1090 static int parse_max_pkt_len(const char *pktlen)
1091 {
1092         char *end = NULL;
1093         unsigned long len;
1094
1095         /* parse decimal string */
1096         len = strtoul(pktlen, &end, 10);
1097         if ((pktlen[0] == '\0') || (end == NULL) || (*end != '\0'))
1098                 return -1;
1099
1100         if (len == 0)
1101                 return -1;
1102
1103         return len;
1104 }
1105
1106 static int
1107 parse_portmask(const char *portmask)
1108 {
1109         char *end = NULL;
1110         unsigned long pm;
1111
1112         /* parse hexadecimal string */
1113         pm = strtoul(portmask, &end, 16);
1114         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
1115                 return -1;
1116
1117         if (pm == 0)
1118                 return -1;
1119
1120         return pm;
1121 }
1122
1123 static int
1124 parse_config(const char *q_arg)
1125 {
1126         char s[256];
1127         const char *p, *p0 = q_arg;
1128         char *end;
1129         enum fieldnames {
1130                 FLD_PORT = 0,
1131                 FLD_QUEUE,
1132                 FLD_LCORE,
1133                 _NUM_FLD
1134         };
1135         unsigned long int_fld[_NUM_FLD];
1136         char *str_fld[_NUM_FLD];
1137         int i;
1138         unsigned size;
1139
1140         nb_lcore_params = 0;
1141
1142         while ((p = strchr(p0,'(')) != NULL) {
1143                 ++p;
1144                 if((p0 = strchr(p,')')) == NULL)
1145                         return -1;
1146
1147                 size = p0 - p;
1148                 if(size >= sizeof(s))
1149                         return -1;
1150
1151                 rte_snprintf(s, sizeof(s), "%.*s", size, p);
1152                 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') !=
1153                                                                 _NUM_FLD)
1154                         return -1;
1155                 for (i = 0; i < _NUM_FLD; i++){
1156                         errno = 0;
1157                         int_fld[i] = strtoul(str_fld[i], &end, 0);
1158                         if (errno != 0 || end == str_fld[i] || int_fld[i] >
1159                                                                         255)
1160                                 return -1;
1161                 }
1162                 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
1163                         printf("exceeded max number of lcore params: %hu\n",
1164                                 nb_lcore_params);
1165                         return -1;
1166                 }
1167                 lcore_params_array[nb_lcore_params].port_id =
1168                                 (uint8_t)int_fld[FLD_PORT];
1169                 lcore_params_array[nb_lcore_params].queue_id =
1170                                 (uint8_t)int_fld[FLD_QUEUE];
1171                 lcore_params_array[nb_lcore_params].lcore_id =
1172                                 (uint8_t)int_fld[FLD_LCORE];
1173                 ++nb_lcore_params;
1174         }
1175         lcore_params = lcore_params_array;
1176
1177         return 0;
1178 }
1179
1180 /* Parse the argument given in the command line of the application */
1181 static int
1182 parse_args(int argc, char **argv)
1183 {
1184         int opt, ret;
1185         char **argvopt;
1186         int option_index;
1187         char *prgname = argv[0];
1188         static struct option lgopts[] = {
1189                 {"config", 1, 0, 0},
1190                 {"no-numa", 0, 0, 0},
1191                 {"enable-jumbo", 0, 0, 0},
1192                 {NULL, 0, 0, 0}
1193         };
1194
1195         argvopt = argv;
1196
1197         while ((opt = getopt_long(argc, argvopt, "p:P",
1198                                 lgopts, &option_index)) != EOF) {
1199
1200                 switch (opt) {
1201                 /* portmask */
1202                 case 'p':
1203                         enabled_port_mask = parse_portmask(optarg);
1204                         if (enabled_port_mask == 0) {
1205                                 printf("invalid portmask\n");
1206                                 print_usage(prgname);
1207                                 return -1;
1208                         }
1209                         break;
1210                 case 'P':
1211                         printf("Promiscuous mode selected\n");
1212                         promiscuous_on = 1;
1213                         break;
1214
1215                 /* long options */
1216                 case 0:
1217                         if (!strncmp(lgopts[option_index].name, "config", 6)) {
1218                                 ret = parse_config(optarg);
1219                                 if (ret) {
1220                                         printf("invalid config\n");
1221                                         print_usage(prgname);
1222                                         return -1;
1223                                 }
1224                         }
1225
1226                         if (!strncmp(lgopts[option_index].name,
1227                                                 "no-numa", 7)) {
1228                                 printf("numa is disabled \n");
1229                                 numa_on = 0;
1230                         }
1231
1232                         if (!strncmp(lgopts[option_index].name,
1233                                         "enable-jumbo", 12)) {
1234                                 struct option lenopts =
1235                                         {"max-pkt-len", required_argument, \
1236                                                                         0, 0};
1237
1238                                 printf("jumbo frame is enabled \n");
1239                                 port_conf.rxmode.jumbo_frame = 1;
1240
1241                                 /**
1242                                  * if no max-pkt-len set, use the default value
1243                                  * ETHER_MAX_LEN
1244                                  */
1245                                 if (0 == getopt_long(argc, argvopt, "",
1246                                                 &lenopts, &option_index)) {
1247                                         ret = parse_max_pkt_len(optarg);
1248                                         if ((ret < 64) ||
1249                                                 (ret > MAX_JUMBO_PKT_LEN)){
1250                                                 printf("invalid packet "
1251                                                                 "length\n");
1252                                                 print_usage(prgname);
1253                                                 return -1;
1254                                         }
1255                                         port_conf.rxmode.max_rx_pkt_len = ret;
1256                                 }
1257                                 printf("set jumbo frame "
1258                                         "max packet length to %u\n",
1259                                 (unsigned int)port_conf.rxmode.max_rx_pkt_len);
1260                         }
1261
1262                         break;
1263
1264                 default:
1265                         print_usage(prgname);
1266                         return -1;
1267                 }
1268         }
1269
1270         if (optind >= 0)
1271                 argv[optind-1] = prgname;
1272
1273         ret = optind-1;
1274         optind = 0; /* reset getopt lib */
1275         return ret;
1276 }
1277
1278 static void
1279 print_ethaddr(const char *name, const struct ether_addr *eth_addr)
1280 {
1281         printf ("%s%02X:%02X:%02X:%02X:%02X:%02X", name,
1282                 eth_addr->addr_bytes[0],
1283                 eth_addr->addr_bytes[1],
1284                 eth_addr->addr_bytes[2],
1285                 eth_addr->addr_bytes[3],
1286                 eth_addr->addr_bytes[4],
1287                 eth_addr->addr_bytes[5]);
1288 }
1289
1290 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1291 static void
1292 setup_hash(int socketid)
1293 {
1294         struct rte_hash_parameters ipv4_l3fwd_hash_params = {
1295                 .name = NULL,
1296                 .entries = L3FWD_HASH_ENTRIES,
1297                 .bucket_entries = 4,
1298                 .key_len = sizeof(struct ipv4_5tuple),
1299                 .hash_func = DEFAULT_HASH_FUNC,
1300                 .hash_func_init_val = 0,
1301         };
1302
1303         struct rte_hash_parameters ipv6_l3fwd_hash_params = {
1304                 .name = NULL,
1305                 .entries = L3FWD_HASH_ENTRIES,
1306                 .bucket_entries = 4,
1307                 .key_len = sizeof(struct ipv6_5tuple),
1308                 .hash_func = DEFAULT_HASH_FUNC,
1309                 .hash_func_init_val = 0,
1310         };
1311
1312         unsigned i;
1313         int ret;
1314         char s[64];
1315
1316         /* create ipv4 hash */
1317         rte_snprintf(s, sizeof(s), "ipv4_l3fwd_hash_%d", socketid);
1318         ipv4_l3fwd_hash_params.name = s;
1319         ipv4_l3fwd_hash_params.socket_id = socketid;
1320         ipv4_l3fwd_lookup_struct[socketid] =
1321                 rte_hash_create(&ipv4_l3fwd_hash_params);
1322         if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
1323                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
1324                                 "socket %d\n", socketid);
1325
1326         /* create ipv6 hash */
1327         rte_snprintf(s, sizeof(s), "ipv6_l3fwd_hash_%d", socketid);
1328         ipv6_l3fwd_hash_params.name = s;
1329         ipv6_l3fwd_hash_params.socket_id = socketid;
1330         ipv6_l3fwd_lookup_struct[socketid] =
1331                 rte_hash_create(&ipv6_l3fwd_hash_params);
1332         if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
1333                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
1334                                 "socket %d\n", socketid);
1335
1336
1337         /* populate the ipv4 hash */
1338         for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
1339                 ret = rte_hash_add_key (ipv4_l3fwd_lookup_struct[socketid],
1340                                 (void *) &ipv4_l3fwd_route_array[i].key);
1341                 if (ret < 0) {
1342                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
1343                                 "l3fwd hash on socket %d\n", i, socketid);
1344                 }
1345                 ipv4_l3fwd_out_if[ret] = ipv4_l3fwd_route_array[i].if_out;
1346                 printf("Hash: Adding key\n");
1347                 print_ipv4_key(ipv4_l3fwd_route_array[i].key);
1348         }
1349
1350         /* populate the ipv6 hash */
1351         for (i = 0; i < IPV6_L3FWD_NUM_ROUTES; i++) {
1352                 ret = rte_hash_add_key (ipv6_l3fwd_lookup_struct[socketid],
1353                                 (void *) &ipv6_l3fwd_route_array[i].key);
1354                 if (ret < 0) {
1355                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
1356                                 "l3fwd hash on socket %d\n", i, socketid);
1357                 }
1358                 ipv6_l3fwd_out_if[ret] = ipv6_l3fwd_route_array[i].if_out;
1359                 printf("Hash: Adding key\n");
1360                 print_ipv6_key(ipv6_l3fwd_route_array[i].key);
1361         }
1362 }
1363 #endif
1364
1365 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1366 static void
1367 setup_lpm(int socketid)
1368 {
1369         unsigned i;
1370         int ret;
1371         char s[64];
1372
1373         /* create the LPM table */
1374         rte_snprintf(s, sizeof(s), "IPV4_L3FWD_LPM_%d", socketid);
1375         ipv4_l3fwd_lookup_struct[socketid] = rte_lpm_create(s, socketid,
1376                                 IPV4_L3FWD_LPM_MAX_RULES, 0);
1377         if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
1378                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
1379                                 " on socket %d\n", socketid);
1380
1381         /* populate the LPM table */
1382         for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
1383                 ret = rte_lpm_add(ipv4_l3fwd_lookup_struct[socketid],
1384                         ipv4_l3fwd_route_array[i].ip,
1385                         ipv4_l3fwd_route_array[i].depth,
1386                         ipv4_l3fwd_route_array[i].if_out);
1387
1388                 if (ret < 0) {
1389                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
1390                                 "l3fwd LPM table on socket %d\n",
1391                                 i, socketid);
1392                 }
1393
1394                 printf("LPM: Adding route 0x%08x / %d (%d)\n",
1395                         (unsigned)ipv4_l3fwd_route_array[i].ip,
1396                         ipv4_l3fwd_route_array[i].depth,
1397                         ipv4_l3fwd_route_array[i].if_out);
1398         }
1399 }
1400 #endif
1401
1402 static int
1403 init_mem(unsigned nb_mbuf)
1404 {
1405         struct lcore_conf *qconf;
1406         int socketid;
1407         unsigned lcore_id;
1408         char s[64];
1409
1410         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1411                 if (rte_lcore_is_enabled(lcore_id) == 0)
1412                         continue;
1413
1414                 if (numa_on)
1415                         socketid = rte_lcore_to_socket_id(lcore_id);
1416                 else
1417                         socketid = 0;
1418
1419                 if (socketid >= NB_SOCKETS) {
1420                         rte_exit(EXIT_FAILURE, "Socket %d of lcore %u is "
1421                                         "out of range %d\n", socketid,
1422                                                 lcore_id, NB_SOCKETS);
1423                 }
1424                 if (pktmbuf_pool[socketid] == NULL) {
1425                         rte_snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
1426                         pktmbuf_pool[socketid] =
1427                                 rte_mempool_create(s, nb_mbuf,
1428                                         MBUF_SIZE, MEMPOOL_CACHE_SIZE,
1429                                         sizeof(struct rte_pktmbuf_pool_private),
1430                                         rte_pktmbuf_pool_init, NULL,
1431                                         rte_pktmbuf_init, NULL,
1432                                         socketid, 0);
1433                         if (pktmbuf_pool[socketid] == NULL)
1434                                 rte_exit(EXIT_FAILURE,
1435                                         "Cannot init mbuf pool on socket %d\n",
1436                                                                 socketid);
1437                         else
1438                                 printf("Allocated mbuf pool on socket %d\n",
1439                                                                 socketid);
1440
1441 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1442                         setup_lpm(socketid);
1443 #else
1444                         setup_hash(socketid);
1445 #endif
1446                 }
1447                 qconf = &lcore_conf[lcore_id];
1448                 qconf->ipv4_lookup_struct = ipv4_l3fwd_lookup_struct[socketid];
1449 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1450                 qconf->ipv6_lookup_struct = ipv6_l3fwd_lookup_struct[socketid];
1451 #endif
1452         }
1453         return 0;
1454 }
1455
1456 /* Check the link status of all ports in up to 9s, and print them finally */
1457 static void
1458 check_all_ports_link_status(uint8_t port_num, uint32_t port_mask)
1459 {
1460 #define CHECK_INTERVAL 100 /* 100ms */
1461 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
1462         uint8_t portid, count, all_ports_up, print_flag = 0;
1463         struct rte_eth_link link;
1464
1465         printf("\nChecking link status");
1466         fflush(stdout);
1467         for (count = 0; count <= MAX_CHECK_TIME; count++) {
1468                 all_ports_up = 1;
1469                 for (portid = 0; portid < port_num; portid++) {
1470                         if ((port_mask & (1 << portid)) == 0)
1471                                 continue;
1472                         memset(&link, 0, sizeof(link));
1473                         rte_eth_link_get_nowait(portid, &link);
1474                         /* print link status if flag set */
1475                         if (print_flag == 1) {
1476                                 if (link.link_status)
1477                                         printf("Port %d Link Up - speed %u "
1478                                                 "Mbps - %s\n", (uint8_t)portid,
1479                                                 (unsigned)link.link_speed,
1480                                 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
1481                                         ("full-duplex") : ("half-duplex\n"));
1482                                 else
1483                                         printf("Port %d Link Down\n",
1484                                                 (uint8_t)portid);
1485                                 continue;
1486                         }
1487                         /* clear all_ports_up flag if any link down */
1488                         if (link.link_status == 0) {
1489                                 all_ports_up = 0;
1490                                 break;
1491                         }
1492                 }
1493                 /* after finally printing all link status, get out */
1494                 if (print_flag == 1)
1495                         break;
1496
1497                 if (all_ports_up == 0) {
1498                         printf(".");
1499                         fflush(stdout);
1500                         rte_delay_ms(CHECK_INTERVAL);
1501                 }
1502
1503                 /* set the print_flag if all ports up or timeout */
1504                 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
1505                         print_flag = 1;
1506                         printf("done\n");
1507                 }
1508         }
1509 }
1510
1511 int
1512 MAIN(int argc, char **argv)
1513 {
1514         struct lcore_conf *qconf;
1515         int ret;
1516         unsigned nb_ports;
1517         uint16_t queueid;
1518         unsigned lcore_id;
1519         uint64_t hz;
1520         uint32_t n_tx_queue, nb_lcores;
1521         uint8_t portid, nb_rx_queue, queue, socketid;
1522
1523         /* catch SIGINT and restore cpufreq governor to ondemand */
1524         signal(SIGINT, signal_exit_now);
1525
1526         /* init EAL */
1527         ret = rte_eal_init(argc, argv);
1528         if (ret < 0)
1529                 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
1530         argc -= ret;
1531         argv += ret;
1532
1533         /* init RTE timer library to be used late */
1534         rte_timer_subsystem_init();
1535
1536         /* parse application arguments (after the EAL ones) */
1537         ret = parse_args(argc, argv);
1538         if (ret < 0)
1539                 rte_exit(EXIT_FAILURE, "Invalid L3FWD parameters\n");
1540
1541         if (check_lcore_params() < 0)
1542                 rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
1543
1544         ret = init_lcore_rx_queues();
1545         if (ret < 0)
1546                 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
1547
1548
1549         /* init driver(s) */
1550         if (rte_pmd_init_all() < 0)
1551                 rte_exit(EXIT_FAILURE, "Cannot init pmd\n");
1552
1553         if (rte_eal_pci_probe() < 0)
1554                 rte_exit(EXIT_FAILURE, "Cannot probe PCI\n");
1555
1556         nb_ports = rte_eth_dev_count();
1557         if (nb_ports > RTE_MAX_ETHPORTS)
1558                 nb_ports = RTE_MAX_ETHPORTS;
1559
1560         if (check_port_config(nb_ports) < 0)
1561                 rte_exit(EXIT_FAILURE, "check_port_config failed\n");
1562
1563         nb_lcores = rte_lcore_count();
1564
1565         /* initialize all ports */
1566         for (portid = 0; portid < nb_ports; portid++) {
1567                 /* skip ports that are not enabled */
1568                 if ((enabled_port_mask & (1 << portid)) == 0) {
1569                         printf("\nSkipping disabled port %d\n", portid);
1570                         continue;
1571                 }
1572
1573                 /* init port */
1574                 printf("Initializing port %d ... ", portid );
1575                 fflush(stdout);
1576
1577                 nb_rx_queue = get_port_n_rx_queues(portid);
1578                 n_tx_queue = nb_lcores;
1579                 if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
1580                         n_tx_queue = MAX_TX_QUEUE_PER_PORT;
1581                 printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
1582                         nb_rx_queue, (unsigned)n_tx_queue );
1583                 ret = rte_eth_dev_configure(portid, nb_rx_queue,
1584                                         (uint16_t)n_tx_queue, &port_conf);
1585                 if (ret < 0)
1586                         rte_exit(EXIT_FAILURE, "Cannot configure device: "
1587                                         "err=%d, port=%d\n", ret, portid);
1588
1589                 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
1590                 print_ethaddr(" Address:", &ports_eth_addr[portid]);
1591                 printf(", ");
1592
1593                 /* init memory */
1594                 ret = init_mem(NB_MBUF);
1595                 if (ret < 0)
1596                         rte_exit(EXIT_FAILURE, "init_mem failed\n");
1597
1598                 /* init one TX queue per couple (lcore,port) */
1599                 queueid = 0;
1600                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1601                         if (rte_lcore_is_enabled(lcore_id) == 0)
1602                                 continue;
1603
1604                         if (numa_on)
1605                                 socketid = \
1606                                 (uint8_t)rte_lcore_to_socket_id(lcore_id);
1607                         else
1608                                 socketid = 0;
1609
1610                         printf("txq=%u,%d,%d ", lcore_id, queueid, socketid);
1611                         fflush(stdout);
1612                         ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
1613                                                      socketid, &tx_conf);
1614                         if (ret < 0)
1615                                 rte_exit(EXIT_FAILURE,
1616                                         "rte_eth_tx_queue_setup: err=%d, "
1617                                                 "port=%d\n", ret, portid);
1618
1619                         qconf = &lcore_conf[lcore_id];
1620                         qconf->tx_queue_id[portid] = queueid;
1621                         queueid++;
1622                 }
1623                 printf("\n");
1624         }
1625
1626         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1627                 if (rte_lcore_is_enabled(lcore_id) == 0)
1628                         continue;
1629
1630                 /* init power management library */
1631                 ret = rte_power_init(lcore_id);
1632                 if (ret)
1633                         rte_exit(EXIT_FAILURE, "Power management library "
1634                                 "initialization failed on core%u\n", lcore_id);
1635
1636                 /* init timer structures for each enabled lcore */
1637                 rte_timer_init(&power_timers[lcore_id]);
1638                 hz = rte_get_timer_hz();
1639                 rte_timer_reset(&power_timers[lcore_id],
1640                         hz/TIMER_NUMBER_PER_SECOND, SINGLE, lcore_id,
1641                                                 power_timer_cb, NULL);
1642
1643                 qconf = &lcore_conf[lcore_id];
1644                 printf("\nInitializing rx queues on lcore %u ... ", lcore_id );
1645                 fflush(stdout);
1646                 /* init RX queues */
1647                 for(queue = 0; queue < qconf->n_rx_queue; ++queue) {
1648                         portid = qconf->rx_queue_list[queue].port_id;
1649                         queueid = qconf->rx_queue_list[queue].queue_id;
1650
1651                         if (numa_on)
1652                                 socketid = \
1653                                 (uint8_t)rte_lcore_to_socket_id(lcore_id);
1654                         else
1655                                 socketid = 0;
1656
1657                         printf("rxq=%d,%d,%d ", portid, queueid, socketid);
1658                         fflush(stdout);
1659
1660                         ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
1661                                 socketid, &rx_conf, pktmbuf_pool[socketid]);
1662                         if (ret < 0)
1663                                 rte_exit(EXIT_FAILURE,
1664                                         "rte_eth_rx_queue_setup: err=%d, "
1665                                                 "port=%d\n", ret, portid);
1666                 }
1667         }
1668
1669         printf("\n");
1670
1671         /* start ports */
1672         for (portid = 0; portid < nb_ports; portid++) {
1673                 if ((enabled_port_mask & (1 << portid)) == 0) {
1674                         continue;
1675                 }
1676                 /* Start device */
1677                 ret = rte_eth_dev_start(portid);
1678                 if (ret < 0)
1679                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, "
1680                                                 "port=%d\n", ret, portid);
1681
1682                 /*
1683                  * If enabled, put device in promiscuous mode.
1684                  * This allows IO forwarding mode to forward packets
1685                  * to itself through 2 cross-connected  ports of the
1686                  * target machine.
1687                  */
1688                 if (promiscuous_on)
1689                         rte_eth_promiscuous_enable(portid);
1690         }
1691
1692         check_all_ports_link_status((uint8_t)nb_ports, enabled_port_mask);
1693
1694         /* launch per-lcore init on every lcore */
1695         rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
1696         RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1697                 if (rte_eal_wait_lcore(lcore_id) < 0)
1698                         return -1;
1699         }
1700
1701         return 0;
1702 }