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                 .mq_mode        = ETH_MQ_RX_RSS,
 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_IP,
 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, uint8_t port_id, uint16_t queue_id);
 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_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,
 773                              uint8_t port_id,
 774                              uint16_t queue_id)
 775 {
 776 /**
 777  * HW Rx queue size is 128 by default, Rx burst read at maximum 32 entries
 778  * per iteration
 779  */
 780 #define FREQ_GEAR1_RX_PACKET_THRESHOLD             MAX_PKT_BURST
 781 #define FREQ_GEAR2_RX_PACKET_THRESHOLD             (MAX_PKT_BURST*2)
 782 #define FREQ_GEAR3_RX_PACKET_THRESHOLD             (MAX_PKT_BURST*3)
 783 #define FREQ_UP_TREND1_ACC   1
 784 #define FREQ_UP_TREND2_ACC   100
 785 #define FREQ_UP_THRESHOLD    10000
 786
 787         if (likely(rte_eth_rx_descriptor_done(port_id, queue_id,
 788                         FREQ_GEAR3_RX_PACKET_THRESHOLD) > 0)) {
 789                 stats[lcore_id].trend = 0;
 790                 return FREQ_HIGHEST;
 791         } else if (likely(rte_eth_rx_descriptor_done(port_id, queue_id,
 792                         FREQ_GEAR2_RX_PACKET_THRESHOLD) > 0))
 793                 stats[lcore_id].trend += FREQ_UP_TREND2_ACC;
 794         else if (likely(rte_eth_rx_descriptor_done(port_id, queue_id,
 795                         FREQ_GEAR1_RX_PACKET_THRESHOLD) > 0))
 796                 stats[lcore_id].trend += FREQ_UP_TREND1_ACC;
 797
 798         if (likely(stats[lcore_id].trend > FREQ_UP_THRESHOLD)) {
 799                 stats[lcore_id].trend = 0;
 800                 return FREQ_HIGHER;
 801         }
 802
 803         return FREQ_CURRENT;
 804 }
 805
 806 /* main processing loop */
 807 static int
 808 main_loop(__attribute__((unused)) void *dummy)
 809 {
 810         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
 811         unsigned lcore_id;
 812         uint64_t prev_tsc, diff_tsc, cur_tsc;
 813         uint64_t prev_tsc_power = 0, cur_tsc_power, diff_tsc_power;
 814         int i, j, nb_rx;
 815         uint8_t portid, queueid;
 816         struct lcore_conf *qconf;
 817         struct lcore_rx_queue *rx_queue;
 818         enum freq_scale_hint_t lcore_scaleup_hint;
 819
 820         uint32_t lcore_rx_idle_count = 0;
 821         uint32_t lcore_idle_hint = 0;
 822
 823         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
 824
 825         prev_tsc = 0;
 826
 827         lcore_id = rte_lcore_id();
 828         qconf = &lcore_conf[lcore_id];
 829
 830         if (qconf->n_rx_queue == 0) {
 831                 RTE_LOG(INFO, L3FWD_POWER, "lcore %u has nothing to do\n", lcore_id);
 832                 return 0;
 833         }
 834
 835         RTE_LOG(INFO, L3FWD_POWER, "entering main loop on lcore %u\n", lcore_id);
 836
 837         for (i = 0; i < qconf->n_rx_queue; i++) {
 838
 839                 portid = qconf->rx_queue_list[i].port_id;
 840                 queueid = qconf->rx_queue_list[i].queue_id;
 841                 RTE_LOG(INFO, L3FWD_POWER, " -- lcoreid=%u portid=%hhu "
 842                         "rxqueueid=%hhu\n", lcore_id, portid, queueid);
 843         }
 844
 845         while (1) {
 846                 stats[lcore_id].nb_iteration_looped++;
 847
 848                 cur_tsc = rte_rdtsc();
 849                 cur_tsc_power = cur_tsc;
 850
 851                 /*
 852                  * TX burst queue drain
 853                  */
 854                 diff_tsc = cur_tsc - prev_tsc;
 855                 if (unlikely(diff_tsc > drain_tsc)) {
 856
 857                         /*
 858                          * This could be optimized (use queueid instead of
 859                          * portid), but it is not called so often
 860                          */
 861                         for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
 862                                 if (qconf->tx_mbufs[portid].len == 0)
 863                                         continue;
 864                                 send_burst(&lcore_conf[lcore_id],
 865                                         qconf->tx_mbufs[portid].len,
 866                                         portid);
 867                                 qconf->tx_mbufs[portid].len = 0;
 868                         }
 869
 870                         prev_tsc = cur_tsc;
 871                 }
 872
 873                 diff_tsc_power = cur_tsc_power - prev_tsc_power;
 874                 if (diff_tsc_power > TIMER_RESOLUTION_CYCLES) {
 875                         rte_timer_manage();
 876                         prev_tsc_power = cur_tsc_power;
 877                 }
 878
 879                 /*
 880                  * Read packet from RX queues
 881                  */
 882                 lcore_scaleup_hint = FREQ_CURRENT;
 883                 lcore_rx_idle_count = 0;
 884                 for (i = 0; i < qconf->n_rx_queue; ++i) {
 885                         rx_queue = &(qconf->rx_queue_list[i]);
 886                         rx_queue->idle_hint = 0;
 887                         portid = rx_queue->port_id;
 888                         queueid = rx_queue->queue_id;
 889
 890                         nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
 891                                                                 MAX_PKT_BURST);
 892                         stats[lcore_id].nb_rx_processed += nb_rx;
 893                         if (unlikely(nb_rx == 0)) {
 894                                 /**
 895                                  * no packet received from rx queue, try to
 896                                  * sleep for a while forcing CPU enter deeper
 897                                  * C states.
 898                                  */
 899                                 rx_queue->zero_rx_packet_count++;
 900
 901                                 if (rx_queue->zero_rx_packet_count <=
 902                                                         MIN_ZERO_POLL_COUNT)
 903                                         continue;
 904
 905                                 rx_queue->idle_hint = power_idle_heuristic(\
 906                                         rx_queue->zero_rx_packet_count);
 907                                 lcore_rx_idle_count++;
 908                         } else {
 909                                 rx_queue->zero_rx_packet_count = 0;
 910
 911                                 /**
 912                                  * do not scale up frequency immediately as
 913                                  * user to kernel space communication is costly
 914                                  * which might impact packet I/O for received
 915                                  * packets.
 916                                  */
 917                                 rx_queue->freq_up_hint =
 918                                         power_freq_scaleup_heuristic(lcore_id,
 919                                                         portid, queueid);
 920                         }
 921
 922                         /* Prefetch first packets */
 923                         for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
 924                                 rte_prefetch0(rte_pktmbuf_mtod(
 925                                                 pkts_burst[j], void *));
 926                         }
 927
 928                         /* Prefetch and forward already prefetched packets */
 929                         for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
 930                                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
 931                                                 j + PREFETCH_OFFSET], void *));
 932                                 l3fwd_simple_forward(pkts_burst[j], portid,
 933                                                                 qconf);
 934                         }
 935
 936                         /* Forward remaining prefetched packets */
 937                         for (; j < nb_rx; j++) {
 938                                 l3fwd_simple_forward(pkts_burst[j], portid,
 939                                                                 qconf);
 940                         }
 941                 }
 942
 943                 if (likely(lcore_rx_idle_count != qconf->n_rx_queue)) {
 944                         for (i = 1, lcore_scaleup_hint =
 945                                 qconf->rx_queue_list[0].freq_up_hint;
 946                                         i < qconf->n_rx_queue; ++i) {
 947                                 rx_queue = &(qconf->rx_queue_list[i]);
 948                                 if (rx_queue->freq_up_hint >
 949                                                 lcore_scaleup_hint)
 950                                         lcore_scaleup_hint =
 951                                                 rx_queue->freq_up_hint;
 952                         }
 953
 954                         if (lcore_scaleup_hint == FREQ_HIGHEST)
 955                                 rte_power_freq_max(lcore_id);
 956                         else if (lcore_scaleup_hint == FREQ_HIGHER)
 957                                 rte_power_freq_up(lcore_id);
 958                 } else {
 959                         /**
 960                          * All Rx queues empty in recent consecutive polls,
 961                          * sleep in a conservative manner, meaning sleep as
 962                          * less as possible.
 963                          */
 964                         for (i = 1, lcore_idle_hint =
 965                                 qconf->rx_queue_list[0].idle_hint;
 966                                         i < qconf->n_rx_queue; ++i) {
 967                                 rx_queue = &(qconf->rx_queue_list[i]);
 968                                 if (rx_queue->idle_hint < lcore_idle_hint)
 969                                         lcore_idle_hint = rx_queue->idle_hint;
 970                         }
 971
 972                         if ( lcore_idle_hint < SLEEP_GEAR1_THRESHOLD)
 973                                 /**
 974                                  * execute "pause" instruction to avoid context
 975                                  * switch for short sleep.
 976                                  */
 977                                 rte_delay_us(lcore_idle_hint);
 978                         else
 979                                 /* long sleep force runing thread to suspend */
 980                                 usleep(lcore_idle_hint);
 981
 982                         stats[lcore_id].sleep_time += lcore_idle_hint;
 983                 }
 984         }
 985 }
 986
 987 static int
 988 check_lcore_params(void)
 989 {
 990         uint8_t queue, lcore;
 991         uint16_t i;
 992         int socketid;
 993
 994         for (i = 0; i < nb_lcore_params; ++i) {
 995                 queue = lcore_params[i].queue_id;
 996                 if (queue >= MAX_RX_QUEUE_PER_PORT) {
 997                         printf("invalid queue number: %hhu\n", queue);
 998                         return -1;
 999                 }
1000                 lcore = lcore_params[i].lcore_id;
1001                 if (!rte_lcore_is_enabled(lcore)) {
1002                         printf("error: lcore %hhu is not enabled in lcore "
1003                                                         "mask\n", lcore);
1004                         return -1;
1005                 }
1006                 if ((socketid = rte_lcore_to_socket_id(lcore) != 0) &&
1007                                                         (numa_on == 0)) {
1008                         printf("warning: lcore %hhu is on socket %d with numa "
1009                                                 "off\n", lcore, socketid);
1010                 }
1011         }
1012         return 0;
1013 }
1014
1015 static int
1016 check_port_config(const unsigned nb_ports)
1017 {
1018         unsigned portid;
1019         uint16_t i;
1020
1021         for (i = 0; i < nb_lcore_params; ++i) {
1022                 portid = lcore_params[i].port_id;
1023                 if ((enabled_port_mask & (1 << portid)) == 0) {
1024                         printf("port %u is not enabled in port mask\n",
1025                                                                 portid);
1026                         return -1;
1027                 }
1028                 if (portid >= nb_ports) {
1029                         printf("port %u is not present on the board\n",
1030                                                                 portid);
1031                         return -1;
1032                 }
1033         }
1034         return 0;
1035 }
1036
1037 static uint8_t
1038 get_port_n_rx_queues(const uint8_t port)
1039 {
1040         int queue = -1;
1041         uint16_t i;
1042
1043         for (i = 0; i < nb_lcore_params; ++i) {
1044                 if (lcore_params[i].port_id == port &&
1045                                 lcore_params[i].queue_id > queue)
1046                         queue = lcore_params[i].queue_id;
1047         }
1048         return (uint8_t)(++queue);
1049 }
1050
1051 static int
1052 init_lcore_rx_queues(void)
1053 {
1054         uint16_t i, nb_rx_queue;
1055         uint8_t lcore;
1056
1057         for (i = 0; i < nb_lcore_params; ++i) {
1058                 lcore = lcore_params[i].lcore_id;
1059                 nb_rx_queue = lcore_conf[lcore].n_rx_queue;
1060                 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
1061                         printf("error: too many queues (%u) for lcore: %u\n",
1062                                 (unsigned)nb_rx_queue + 1, (unsigned)lcore);
1063                         return -1;
1064                 } else {
1065                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
1066                                 lcore_params[i].port_id;
1067                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
1068                                 lcore_params[i].queue_id;
1069                         lcore_conf[lcore].n_rx_queue++;
1070                 }
1071         }
1072         return 0;
1073 }
1074
1075 /* display usage */
1076 static void
1077 print_usage(const char *prgname)
1078 {
1079         printf ("%s [EAL options] -- -p PORTMASK -P"
1080                 "  [--config (port,queue,lcore)[,(port,queue,lcore]]"
1081                 "  [--enable-jumbo [--max-pkt-len PKTLEN]]\n"
1082                 "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
1083                 "  -P : enable promiscuous mode\n"
1084                 "  --config (port,queue,lcore): rx queues configuration\n"
1085                 "  --no-numa: optional, disable numa awareness\n"
1086                 "  --enable-jumbo: enable jumbo frame"
1087                 " which max packet len is PKTLEN in decimal (64-9600)\n",
1088                 prgname);
1089 }
1090
1091 static int parse_max_pkt_len(const char *pktlen)
1092 {
1093         char *end = NULL;
1094         unsigned long len;
1095
1096         /* parse decimal string */
1097         len = strtoul(pktlen, &end, 10);
1098         if ((pktlen[0] == '\0') || (end == NULL) || (*end != '\0'))
1099                 return -1;
1100
1101         if (len == 0)
1102                 return -1;
1103
1104         return len;
1105 }
1106
1107 static int
1108 parse_portmask(const char *portmask)
1109 {
1110         char *end = NULL;
1111         unsigned long pm;
1112
1113         /* parse hexadecimal string */
1114         pm = strtoul(portmask, &end, 16);
1115         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
1116                 return -1;
1117
1118         if (pm == 0)
1119                 return -1;
1120
1121         return pm;
1122 }
1123
1124 static int
1125 parse_config(const char *q_arg)
1126 {
1127         char s[256];
1128         const char *p, *p0 = q_arg;
1129         char *end;
1130         enum fieldnames {
1131                 FLD_PORT = 0,
1132                 FLD_QUEUE,
1133                 FLD_LCORE,
1134                 _NUM_FLD
1135         };
1136         unsigned long int_fld[_NUM_FLD];
1137         char *str_fld[_NUM_FLD];
1138         int i;
1139         unsigned size;
1140
1141         nb_lcore_params = 0;
1142
1143         while ((p = strchr(p0,'(')) != NULL) {
1144                 ++p;
1145                 if((p0 = strchr(p,')')) == NULL)
1146                         return -1;
1147
1148                 size = p0 - p;
1149                 if(size >= sizeof(s))
1150                         return -1;
1151
1152                 snprintf(s, sizeof(s), "%.*s", size, p);
1153                 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') !=
1154                                                                 _NUM_FLD)
1155                         return -1;
1156                 for (i = 0; i < _NUM_FLD; i++){
1157                         errno = 0;
1158                         int_fld[i] = strtoul(str_fld[i], &end, 0);
1159                         if (errno != 0 || end == str_fld[i] || int_fld[i] >
1160                                                                         255)
1161                                 return -1;
1162                 }
1163                 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
1164                         printf("exceeded max number of lcore params: %hu\n",
1165                                 nb_lcore_params);
1166                         return -1;
1167                 }
1168                 lcore_params_array[nb_lcore_params].port_id =
1169                                 (uint8_t)int_fld[FLD_PORT];
1170                 lcore_params_array[nb_lcore_params].queue_id =
1171                                 (uint8_t)int_fld[FLD_QUEUE];
1172                 lcore_params_array[nb_lcore_params].lcore_id =
1173                                 (uint8_t)int_fld[FLD_LCORE];
1174                 ++nb_lcore_params;
1175         }
1176         lcore_params = lcore_params_array;
1177
1178         return 0;
1179 }
1180
1181 /* Parse the argument given in the command line of the application */
1182 static int
1183 parse_args(int argc, char **argv)
1184 {
1185         int opt, ret;
1186         char **argvopt;
1187         int option_index;
1188         char *prgname = argv[0];
1189         static struct option lgopts[] = {
1190                 {"config", 1, 0, 0},
1191                 {"no-numa", 0, 0, 0},
1192                 {"enable-jumbo", 0, 0, 0},
1193                 {NULL, 0, 0, 0}
1194         };
1195
1196         argvopt = argv;
1197
1198         while ((opt = getopt_long(argc, argvopt, "p:P",
1199                                 lgopts, &option_index)) != EOF) {
1200
1201                 switch (opt) {
1202                 /* portmask */
1203                 case 'p':
1204                         enabled_port_mask = parse_portmask(optarg);
1205                         if (enabled_port_mask == 0) {
1206                                 printf("invalid portmask\n");
1207                                 print_usage(prgname);
1208                                 return -1;
1209                         }
1210                         break;
1211                 case 'P':
1212                         printf("Promiscuous mode selected\n");
1213                         promiscuous_on = 1;
1214                         break;
1215
1216                 /* long options */
1217                 case 0:
1218                         if (!strncmp(lgopts[option_index].name, "config", 6)) {
1219                                 ret = parse_config(optarg);
1220                                 if (ret) {
1221                                         printf("invalid config\n");
1222                                         print_usage(prgname);
1223                                         return -1;
1224                                 }
1225                         }
1226
1227                         if (!strncmp(lgopts[option_index].name,
1228                                                 "no-numa", 7)) {
1229                                 printf("numa is disabled \n");
1230                                 numa_on = 0;
1231                         }
1232
1233                         if (!strncmp(lgopts[option_index].name,
1234                                         "enable-jumbo", 12)) {
1235                                 struct option lenopts =
1236                                         {"max-pkt-len", required_argument, \
1237                                                                         0, 0};
1238
1239                                 printf("jumbo frame is enabled \n");
1240                                 port_conf.rxmode.jumbo_frame = 1;
1241
1242                                 /**
1243                                  * if no max-pkt-len set, use the default value
1244                                  * ETHER_MAX_LEN
1245                                  */
1246                                 if (0 == getopt_long(argc, argvopt, "",
1247                                                 &lenopts, &option_index)) {
1248                                         ret = parse_max_pkt_len(optarg);
1249                                         if ((ret < 64) ||
1250                                                 (ret > MAX_JUMBO_PKT_LEN)){
1251                                                 printf("invalid packet "
1252                                                                 "length\n");
1253                                                 print_usage(prgname);
1254                                                 return -1;
1255                                         }
1256                                         port_conf.rxmode.max_rx_pkt_len = ret;
1257                                 }
1258                                 printf("set jumbo frame "
1259                                         "max packet length to %u\n",
1260                                 (unsigned int)port_conf.rxmode.max_rx_pkt_len);
1261                         }
1262
1263                         break;
1264
1265                 default:
1266                         print_usage(prgname);
1267                         return -1;
1268                 }
1269         }
1270
1271         if (optind >= 0)
1272                 argv[optind-1] = prgname;
1273
1274         ret = optind-1;
1275         optind = 0; /* reset getopt lib */
1276         return ret;
1277 }
1278
1279 static void
1280 print_ethaddr(const char *name, const struct ether_addr *eth_addr)
1281 {
1282         printf ("%s%02X:%02X:%02X:%02X:%02X:%02X", name,
1283                 eth_addr->addr_bytes[0],
1284                 eth_addr->addr_bytes[1],
1285                 eth_addr->addr_bytes[2],
1286                 eth_addr->addr_bytes[3],
1287                 eth_addr->addr_bytes[4],
1288                 eth_addr->addr_bytes[5]);
1289 }
1290
1291 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1292 static void
1293 setup_hash(int socketid)
1294 {
1295         struct rte_hash_parameters ipv4_l3fwd_hash_params = {
1296                 .name = NULL,
1297                 .entries = L3FWD_HASH_ENTRIES,
1298                 .bucket_entries = 4,
1299                 .key_len = sizeof(struct ipv4_5tuple),
1300                 .hash_func = DEFAULT_HASH_FUNC,
1301                 .hash_func_init_val = 0,
1302         };
1303
1304         struct rte_hash_parameters ipv6_l3fwd_hash_params = {
1305                 .name = NULL,
1306                 .entries = L3FWD_HASH_ENTRIES,
1307                 .bucket_entries = 4,
1308                 .key_len = sizeof(struct ipv6_5tuple),
1309                 .hash_func = DEFAULT_HASH_FUNC,
1310                 .hash_func_init_val = 0,
1311         };
1312
1313         unsigned i;
1314         int ret;
1315         char s[64];
1316
1317         /* create ipv4 hash */
1318         snprintf(s, sizeof(s), "ipv4_l3fwd_hash_%d", socketid);
1319         ipv4_l3fwd_hash_params.name = s;
1320         ipv4_l3fwd_hash_params.socket_id = socketid;
1321         ipv4_l3fwd_lookup_struct[socketid] =
1322                 rte_hash_create(&ipv4_l3fwd_hash_params);
1323         if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
1324                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
1325                                 "socket %d\n", socketid);
1326
1327         /* create ipv6 hash */
1328         snprintf(s, sizeof(s), "ipv6_l3fwd_hash_%d", socketid);
1329         ipv6_l3fwd_hash_params.name = s;
1330         ipv6_l3fwd_hash_params.socket_id = socketid;
1331         ipv6_l3fwd_lookup_struct[socketid] =
1332                 rte_hash_create(&ipv6_l3fwd_hash_params);
1333         if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
1334                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
1335                                 "socket %d\n", socketid);
1336
1337
1338         /* populate the ipv4 hash */
1339         for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
1340                 ret = rte_hash_add_key (ipv4_l3fwd_lookup_struct[socketid],
1341                                 (void *) &ipv4_l3fwd_route_array[i].key);
1342                 if (ret < 0) {
1343                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
1344                                 "l3fwd hash on socket %d\n", i, socketid);
1345                 }
1346                 ipv4_l3fwd_out_if[ret] = ipv4_l3fwd_route_array[i].if_out;
1347                 printf("Hash: Adding key\n");
1348                 print_ipv4_key(ipv4_l3fwd_route_array[i].key);
1349         }
1350
1351         /* populate the ipv6 hash */
1352         for (i = 0; i < IPV6_L3FWD_NUM_ROUTES; i++) {
1353                 ret = rte_hash_add_key (ipv6_l3fwd_lookup_struct[socketid],
1354                                 (void *) &ipv6_l3fwd_route_array[i].key);
1355                 if (ret < 0) {
1356                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
1357                                 "l3fwd hash on socket %d\n", i, socketid);
1358                 }
1359                 ipv6_l3fwd_out_if[ret] = ipv6_l3fwd_route_array[i].if_out;
1360                 printf("Hash: Adding key\n");
1361                 print_ipv6_key(ipv6_l3fwd_route_array[i].key);
1362         }
1363 }
1364 #endif
1365
1366 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1367 static void
1368 setup_lpm(int socketid)
1369 {
1370         unsigned i;
1371         int ret;
1372         char s[64];
1373
1374         /* create the LPM table */
1375         snprintf(s, sizeof(s), "IPV4_L3FWD_LPM_%d", socketid);
1376         ipv4_l3fwd_lookup_struct[socketid] = rte_lpm_create(s, socketid,
1377                                 IPV4_L3FWD_LPM_MAX_RULES, 0);
1378         if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
1379                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
1380                                 " on socket %d\n", socketid);
1381
1382         /* populate the LPM table */
1383         for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
1384                 ret = rte_lpm_add(ipv4_l3fwd_lookup_struct[socketid],
1385                         ipv4_l3fwd_route_array[i].ip,
1386                         ipv4_l3fwd_route_array[i].depth,
1387                         ipv4_l3fwd_route_array[i].if_out);
1388
1389                 if (ret < 0) {
1390                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
1391                                 "l3fwd LPM table on socket %d\n",
1392                                 i, socketid);
1393                 }
1394
1395                 printf("LPM: Adding route 0x%08x / %d (%d)\n",
1396                         (unsigned)ipv4_l3fwd_route_array[i].ip,
1397                         ipv4_l3fwd_route_array[i].depth,
1398                         ipv4_l3fwd_route_array[i].if_out);
1399         }
1400 }
1401 #endif
1402
1403 static int
1404 init_mem(unsigned nb_mbuf)
1405 {
1406         struct lcore_conf *qconf;
1407         int socketid;
1408         unsigned lcore_id;
1409         char s[64];
1410
1411         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1412                 if (rte_lcore_is_enabled(lcore_id) == 0)
1413                         continue;
1414
1415                 if (numa_on)
1416                         socketid = rte_lcore_to_socket_id(lcore_id);
1417                 else
1418                         socketid = 0;
1419
1420                 if (socketid >= NB_SOCKETS) {
1421                         rte_exit(EXIT_FAILURE, "Socket %d of lcore %u is "
1422                                         "out of range %d\n", socketid,
1423                                                 lcore_id, NB_SOCKETS);
1424                 }
1425                 if (pktmbuf_pool[socketid] == NULL) {
1426                         snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
1427                         pktmbuf_pool[socketid] =
1428                                 rte_mempool_create(s, nb_mbuf,
1429                                         MBUF_SIZE, MEMPOOL_CACHE_SIZE,
1430                                         sizeof(struct rte_pktmbuf_pool_private),
1431                                         rte_pktmbuf_pool_init, NULL,
1432                                         rte_pktmbuf_init, NULL,
1433                                         socketid, 0);
1434                         if (pktmbuf_pool[socketid] == NULL)
1435                                 rte_exit(EXIT_FAILURE,
1436                                         "Cannot init mbuf pool on socket %d\n",
1437                                                                 socketid);
1438                         else
1439                                 printf("Allocated mbuf pool on socket %d\n",
1440                                                                 socketid);
1441
1442 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1443                         setup_lpm(socketid);
1444 #else
1445                         setup_hash(socketid);
1446 #endif
1447                 }
1448                 qconf = &lcore_conf[lcore_id];
1449                 qconf->ipv4_lookup_struct = ipv4_l3fwd_lookup_struct[socketid];
1450 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1451                 qconf->ipv6_lookup_struct = ipv6_l3fwd_lookup_struct[socketid];
1452 #endif
1453         }
1454         return 0;
1455 }
1456
1457 /* Check the link status of all ports in up to 9s, and print them finally */
1458 static void
1459 check_all_ports_link_status(uint8_t port_num, uint32_t port_mask)
1460 {
1461 #define CHECK_INTERVAL 100 /* 100ms */
1462 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
1463         uint8_t portid, count, all_ports_up, print_flag = 0;
1464         struct rte_eth_link link;
1465
1466         printf("\nChecking link status");
1467         fflush(stdout);
1468         for (count = 0; count <= MAX_CHECK_TIME; count++) {
1469                 all_ports_up = 1;
1470                 for (portid = 0; portid < port_num; portid++) {
1471                         if ((port_mask & (1 << portid)) == 0)
1472                                 continue;
1473                         memset(&link, 0, sizeof(link));
1474                         rte_eth_link_get_nowait(portid, &link);
1475                         /* print link status if flag set */
1476                         if (print_flag == 1) {
1477                                 if (link.link_status)
1478                                         printf("Port %d Link Up - speed %u "
1479                                                 "Mbps - %s\n", (uint8_t)portid,
1480                                                 (unsigned)link.link_speed,
1481                                 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
1482                                         ("full-duplex") : ("half-duplex\n"));
1483                                 else
1484                                         printf("Port %d Link Down\n",
1485                                                 (uint8_t)portid);
1486                                 continue;
1487                         }
1488                         /* clear all_ports_up flag if any link down */
1489                         if (link.link_status == 0) {
1490                                 all_ports_up = 0;
1491                                 break;
1492                         }
1493                 }
1494                 /* after finally printing all link status, get out */
1495                 if (print_flag == 1)
1496                         break;
1497
1498                 if (all_ports_up == 0) {
1499                         printf(".");
1500                         fflush(stdout);
1501                         rte_delay_ms(CHECK_INTERVAL);
1502                 }
1503
1504                 /* set the print_flag if all ports up or timeout */
1505                 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
1506                         print_flag = 1;
1507                         printf("done\n");
1508                 }
1509         }
1510 }
1511
1512 int
1513 MAIN(int argc, char **argv)
1514 {
1515         struct lcore_conf *qconf;
1516         int ret;
1517         unsigned nb_ports;
1518         uint16_t queueid;
1519         unsigned lcore_id;
1520         uint64_t hz;
1521         uint32_t n_tx_queue, nb_lcores;
1522         uint8_t portid, nb_rx_queue, queue, socketid;
1523
1524         /* catch SIGINT and restore cpufreq governor to ondemand */
1525         signal(SIGINT, signal_exit_now);
1526
1527         /* init EAL */
1528         ret = rte_eal_init(argc, argv);
1529         if (ret < 0)
1530                 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
1531         argc -= ret;
1532         argv += ret;
1533
1534         /* init RTE timer library to be used late */
1535         rte_timer_subsystem_init();
1536
1537         /* parse application arguments (after the EAL ones) */
1538         ret = parse_args(argc, argv);
1539         if (ret < 0)
1540                 rte_exit(EXIT_FAILURE, "Invalid L3FWD parameters\n");
1541
1542         if (check_lcore_params() < 0)
1543                 rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
1544
1545         ret = init_lcore_rx_queues();
1546         if (ret < 0)
1547                 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
1548
1549
1550         if (rte_eal_pci_probe() < 0)
1551                 rte_exit(EXIT_FAILURE, "Cannot probe PCI\n");
1552
1553         nb_ports = rte_eth_dev_count();
1554         if (nb_ports > RTE_MAX_ETHPORTS)
1555                 nb_ports = RTE_MAX_ETHPORTS;
1556
1557         if (check_port_config(nb_ports) < 0)
1558                 rte_exit(EXIT_FAILURE, "check_port_config failed\n");
1559
1560         nb_lcores = rte_lcore_count();
1561
1562         /* initialize all ports */
1563         for (portid = 0; portid < nb_ports; portid++) {
1564                 /* skip ports that are not enabled */
1565                 if ((enabled_port_mask & (1 << portid)) == 0) {
1566                         printf("\nSkipping disabled port %d\n", portid);
1567                         continue;
1568                 }
1569
1570                 /* init port */
1571                 printf("Initializing port %d ... ", portid );
1572                 fflush(stdout);
1573
1574                 nb_rx_queue = get_port_n_rx_queues(portid);
1575                 n_tx_queue = nb_lcores;
1576                 if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
1577                         n_tx_queue = MAX_TX_QUEUE_PER_PORT;
1578                 printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
1579                         nb_rx_queue, (unsigned)n_tx_queue );
1580                 ret = rte_eth_dev_configure(portid, nb_rx_queue,
1581                                         (uint16_t)n_tx_queue, &port_conf);
1582                 if (ret < 0)
1583                         rte_exit(EXIT_FAILURE, "Cannot configure device: "
1584                                         "err=%d, port=%d\n", ret, portid);
1585
1586                 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
1587                 print_ethaddr(" Address:", &ports_eth_addr[portid]);
1588                 printf(", ");
1589
1590                 /* init memory */
1591                 ret = init_mem(NB_MBUF);
1592                 if (ret < 0)
1593                         rte_exit(EXIT_FAILURE, "init_mem failed\n");
1594
1595                 /* init one TX queue per couple (lcore,port) */
1596                 queueid = 0;
1597                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1598                         if (rte_lcore_is_enabled(lcore_id) == 0)
1599                                 continue;
1600
1601                         if (numa_on)
1602                                 socketid = \
1603                                 (uint8_t)rte_lcore_to_socket_id(lcore_id);
1604                         else
1605                                 socketid = 0;
1606
1607                         printf("txq=%u,%d,%d ", lcore_id, queueid, socketid);
1608                         fflush(stdout);
1609                         ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
1610                                                      socketid, &tx_conf);
1611                         if (ret < 0)
1612                                 rte_exit(EXIT_FAILURE,
1613                                         "rte_eth_tx_queue_setup: err=%d, "
1614                                                 "port=%d\n", ret, portid);
1615
1616                         qconf = &lcore_conf[lcore_id];
1617                         qconf->tx_queue_id[portid] = queueid;
1618                         queueid++;
1619                 }
1620                 printf("\n");
1621         }
1622
1623         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1624                 if (rte_lcore_is_enabled(lcore_id) == 0)
1625                         continue;
1626
1627                 /* init power management library */
1628                 ret = rte_power_init(lcore_id);
1629                 if (ret)
1630                         rte_exit(EXIT_FAILURE, "Power management library "
1631                                 "initialization failed on core%u\n", lcore_id);
1632
1633                 /* init timer structures for each enabled lcore */
1634                 rte_timer_init(&power_timers[lcore_id]);
1635                 hz = rte_get_timer_hz();
1636                 rte_timer_reset(&power_timers[lcore_id],
1637                         hz/TIMER_NUMBER_PER_SECOND, SINGLE, lcore_id,
1638                                                 power_timer_cb, NULL);
1639
1640                 qconf = &lcore_conf[lcore_id];
1641                 printf("\nInitializing rx queues on lcore %u ... ", lcore_id );
1642                 fflush(stdout);
1643                 /* init RX queues */
1644                 for(queue = 0; queue < qconf->n_rx_queue; ++queue) {
1645                         portid = qconf->rx_queue_list[queue].port_id;
1646                         queueid = qconf->rx_queue_list[queue].queue_id;
1647
1648                         if (numa_on)
1649                                 socketid = \
1650                                 (uint8_t)rte_lcore_to_socket_id(lcore_id);
1651                         else
1652                                 socketid = 0;
1653
1654                         printf("rxq=%d,%d,%d ", portid, queueid, socketid);
1655                         fflush(stdout);
1656
1657                         ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
1658                                 socketid, &rx_conf, pktmbuf_pool[socketid]);
1659                         if (ret < 0)
1660                                 rte_exit(EXIT_FAILURE,
1661                                         "rte_eth_rx_queue_setup: err=%d, "
1662                                                 "port=%d\n", ret, portid);
1663                 }
1664         }
1665
1666         printf("\n");
1667
1668         /* start ports */
1669         for (portid = 0; portid < nb_ports; portid++) {
1670                 if ((enabled_port_mask & (1 << portid)) == 0) {
1671                         continue;
1672                 }
1673                 /* Start device */
1674                 ret = rte_eth_dev_start(portid);
1675                 if (ret < 0)
1676                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, "
1677                                                 "port=%d\n", ret, portid);
1678
1679                 /*
1680                  * If enabled, put device in promiscuous mode.
1681                  * This allows IO forwarding mode to forward packets
1682                  * to itself through 2 cross-connected  ports of the
1683                  * target machine.
1684                  */
1685                 if (promiscuous_on)
1686                         rte_eth_promiscuous_enable(portid);
1687         }
1688
1689         check_all_ports_link_status((uint8_t)nb_ports, enabled_port_mask);
1690
1691         /* launch per-lcore init on every lcore */
1692         rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
1693         RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1694                 if (rte_eal_wait_lcore(lcore_id) < 0)
1695                         return -1;
1696         }
1697
1698         return 0;
1699 }