examples/l3fwd-vf/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 <signal.h>
  45
  46 #include <rte_common.h>
  47 #include <rte_byteorder.h>
  48 #include <rte_log.h>
  49 #include <rte_memory.h>
  50 #include <rte_memcpy.h>
  51 #include <rte_memzone.h>
  52 #include <rte_tailq.h>
  53 #include <rte_eal.h>
  54 #include <rte_per_lcore.h>
  55 #include <rte_launch.h>
  56 #include <rte_atomic.h>
  57 #include <rte_cycles.h>
  58 #include <rte_prefetch.h>
  59 #include <rte_lcore.h>
  60 #include <rte_per_lcore.h>
  61 #include <rte_branch_prediction.h>
  62 #include <rte_interrupts.h>
  63 #include <rte_pci.h>
  64 #include <rte_random.h>
  65 #include <rte_debug.h>
  66 #include <rte_ether.h>
  67 #include <rte_ethdev.h>
  68 #include <rte_ring.h>
  69 #include <rte_mempool.h>
  70 #include <rte_mbuf.h>
  71 #include <rte_ip.h>
  72 #include <rte_tcp.h>
  73 #include <rte_udp.h>
  74 #include <rte_string_fns.h>
  75
  76 #include "main.h"
  77
  78 #define APP_LOOKUP_EXACT_MATCH          0
  79 #define APP_LOOKUP_LPM                  1
  80 #define DO_RFC_1812_CHECKS
  81
  82 //#define APP_LOOKUP_METHOD             APP_LOOKUP_EXACT_MATCH
  83 #ifndef APP_LOOKUP_METHOD
  84 #define APP_LOOKUP_METHOD             APP_LOOKUP_LPM
  85 #endif
  86
  87 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
  88 #include <rte_hash.h>
  89 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
  90 #include <rte_lpm.h>
  91 #else
  92 #error "APP_LOOKUP_METHOD set to incorrect value"
  93 #endif
  94
  95 #define RTE_LOGTYPE_L3FWD RTE_LOGTYPE_USER1
  96
  97 #define MEMPOOL_CACHE_SIZE 256
  98
  99 #define MBUF_SIZE (2048 + sizeof(struct rte_mbuf) + RTE_PKTMBUF_HEADROOM)
 100
 101 /*
 102  * This expression is used to calculate the number of mbufs needed depending on user input, taking
 103  *  into account memory for rx and tx hardware rings, cache per lcore and mtable per port per lcore.
 104  *  RTE_MAX is used to ensure that NB_MBUF never goes below a minimum value of 8192
 105  */
 106
 107 #define NB_MBUF RTE_MAX (                                                                                                                                       \
 108                                 (nb_ports*nb_rx_queue*RTE_TEST_RX_DESC_DEFAULT +                                                        \
 109                                 nb_ports*nb_lcores*MAX_PKT_BURST +                                                                                      \
 110                                 nb_ports*n_tx_queue*RTE_TEST_TX_DESC_DEFAULT +                                                          \
 111                                 nb_lcores*MEMPOOL_CACHE_SIZE),                                                                                          \
 112                                 (unsigned)8192)
 113
 114 /*
 115  * RX and TX Prefetch, Host, and Write-back threshold values should be
 116  * carefully set for optimal performance. Consult the network
 117  * controller's datasheet and supporting DPDK documentation for guidance
 118  * on how these parameters should be set.
 119  */
 120 #define RX_PTHRESH 8 /**< Default values of RX prefetch threshold reg. */
 121 #define RX_HTHRESH 8 /**< Default values of RX host threshold reg. */
 122 #define RX_WTHRESH 4 /**< Default values of RX write-back threshold reg. */
 123
 124 /*
 125  * These default values are optimized for use with the Intel(R) 82599 10 GbE
 126  * Controller and the DPDK ixgbe PMD. Consider using other values for other
 127  * network controllers and/or network drivers.
 128  */
 129 #define TX_PTHRESH 36 /**< Default values of TX prefetch threshold reg. */
 130 #define TX_HTHRESH 0  /**< Default values of TX host threshold reg. */
 131 #define TX_WTHRESH 0  /**< Default values of TX write-back threshold reg. */
 132
 133 #define MAX_PKT_BURST 32
 134 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
 135
 136 #define NB_SOCKETS 8
 137
 138 #define SOCKET0 0
 139
 140 /* Configure how many packets ahead to prefetch, when reading packets */
 141 #define PREFETCH_OFFSET 3
 142
 143 /*
 144  * Configurable number of RX/TX ring descriptors
 145  */
 146 #define RTE_TEST_RX_DESC_DEFAULT 128
 147 #define RTE_TEST_TX_DESC_DEFAULT 512
 148 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
 149 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
 150
 151 /* ethernet addresses of ports */
 152 static struct ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
 153
 154 /* mask of enabled ports */
 155 static uint32_t enabled_port_mask = 0;
 156 static int numa_on = 1; /**< NUMA is enabled by default. */
 157
 158 struct mbuf_table {
 159         uint16_t len;
 160         struct rte_mbuf *m_table[MAX_PKT_BURST];
 161 };
 162
 163 struct lcore_rx_queue {
 164         uint8_t port_id;
 165         uint8_t queue_id;
 166 } __rte_cache_aligned;
 167
 168 #define MAX_RX_QUEUE_PER_LCORE 16
 169 #define MAX_TX_QUEUE_PER_PORT 1
 170 #define MAX_RX_QUEUE_PER_PORT 1
 171
 172 #define MAX_LCORE_PARAMS 1024
 173 struct lcore_params {
 174         uint8_t port_id;
 175         uint8_t queue_id;
 176         uint8_t lcore_id;
 177 } __rte_cache_aligned;
 178
 179 static struct lcore_params lcore_params_array[MAX_LCORE_PARAMS];
 180 static struct lcore_params lcore_params_array_default[] = {
 181         {0, 0, 2},
 182         {0, 1, 2},
 183         {0, 2, 2},
 184         {1, 0, 2},
 185         {1, 1, 2},
 186         {1, 2, 2},
 187         {2, 0, 2},
 188         {3, 0, 3},
 189         {3, 1, 3},
 190 };
 191
 192 static struct lcore_params * lcore_params = lcore_params_array_default;
 193 static uint16_t nb_lcore_params = sizeof(lcore_params_array_default) /
 194                                 sizeof(lcore_params_array_default[0]);
 195
 196 static struct rte_eth_conf port_conf = {
 197         .rxmode = {
 198                 .mq_mode        = ETH_MQ_RX_RSS,
 199                 .max_rx_pkt_len = ETHER_MAX_LEN,
 200                 .split_hdr_size = 0,
 201                 .header_split   = 0, /**< Header Split disabled */
 202                 .hw_ip_checksum = 1, /**< IP checksum offload enabled */
 203                 .hw_vlan_filter = 0, /**< VLAN filtering disabled */
 204                 .jumbo_frame    = 0, /**< Jumbo Frame Support disabled */
 205                 .hw_strip_crc   = 0, /**< CRC stripped by hardware */
 206         },
 207         .rx_adv_conf = {
 208                 .rss_conf = {
 209                         .rss_key = NULL,
 210                         .rss_hf = ETH_RSS_IP,
 211                 },
 212         },
 213         .txmode = {
 214                 .mq_mode = ETH_MQ_TX_NONE,
 215         },
 216 };
 217
 218 static struct rte_mempool * pktmbuf_pool[NB_SOCKETS];
 219
 220
 221 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 222
 223 #ifdef RTE_MACHINE_CPUFLAG_SSE4_2
 224 #include <rte_hash_crc.h>
 225 #define DEFAULT_HASH_FUNC       rte_hash_crc
 226 #else
 227 #include <rte_jhash.h>
 228 #define DEFAULT_HASH_FUNC       rte_jhash
 229 #endif
 230
 231 struct ipv4_5tuple {
 232         uint32_t ip_dst;
 233         uint32_t ip_src;
 234         uint16_t port_dst;
 235         uint16_t port_src;
 236         uint8_t proto;
 237 } __attribute__((__packed__));
 238
 239 struct l3fwd_route {
 240         struct ipv4_5tuple key;
 241         uint8_t if_out;
 242 };
 243
 244 static struct l3fwd_route l3fwd_route_array[] = {
 245         {{IPv4(100,10,0,1), IPv4(200,10,0,1), 101, 11, IPPROTO_TCP}, 0},
 246         {{IPv4(100,20,0,2), IPv4(200,20,0,2), 102, 12, IPPROTO_TCP}, 1},
 247         {{IPv4(100,30,0,3), IPv4(200,30,0,3), 103, 13, IPPROTO_TCP}, 2},
 248         {{IPv4(100,40,0,4), IPv4(200,40,0,4), 104, 14, IPPROTO_TCP}, 3},
 249 };
 250
 251 typedef struct rte_hash lookup_struct_t;
 252 static lookup_struct_t *l3fwd_lookup_struct[NB_SOCKETS];
 253
 254 #define L3FWD_HASH_ENTRIES      1024
 255 struct rte_hash_parameters l3fwd_hash_params = {
 256         .name = "l3fwd_hash_0",
 257         .entries = L3FWD_HASH_ENTRIES,
 258         .bucket_entries = 4,
 259         .key_len = sizeof(struct ipv4_5tuple),
 260         .hash_func = DEFAULT_HASH_FUNC,
 261         .hash_func_init_val = 0,
 262         .socket_id = SOCKET0,
 263 };
 264
 265 #define L3FWD_NUM_ROUTES \
 266         (sizeof(l3fwd_route_array) / sizeof(l3fwd_route_array[0]))
 267
 268 static uint8_t l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 269 #endif
 270
 271 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 272 struct l3fwd_route {
 273         uint32_t ip;
 274         uint8_t  depth;
 275         uint8_t  if_out;
 276 };
 277
 278 static struct l3fwd_route l3fwd_route_array[] = {
 279         {IPv4(1,1,1,0), 24, 0},
 280         {IPv4(2,1,1,0), 24, 1},
 281         {IPv4(3,1,1,0), 24, 2},
 282         {IPv4(4,1,1,0), 24, 3},
 283         {IPv4(5,1,1,0), 24, 4},
 284         {IPv4(6,1,1,0), 24, 5},
 285         {IPv4(7,1,1,0), 24, 6},
 286         {IPv4(8,1,1,0), 24, 7},
 287 };
 288
 289 #define L3FWD_NUM_ROUTES \
 290         (sizeof(l3fwd_route_array) / sizeof(l3fwd_route_array[0]))
 291
 292 #define L3FWD_LPM_MAX_RULES     1024
 293
 294 typedef struct rte_lpm lookup_struct_t;
 295 static lookup_struct_t *l3fwd_lookup_struct[NB_SOCKETS];
 296 #endif
 297
 298 struct lcore_conf {
 299         uint16_t n_rx_queue;
 300         struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
 301         uint16_t tx_queue_id;
 302         struct mbuf_table tx_mbufs[RTE_MAX_ETHPORTS];
 303         lookup_struct_t * lookup_struct;
 304 } __rte_cache_aligned;
 305
 306 static struct lcore_conf lcore_conf[RTE_MAX_LCORE];
 307
 308 /* Send burst of packets on an output interface */
 309 static inline int
 310 send_burst(struct lcore_conf *qconf, uint16_t n, uint8_t port)
 311 {
 312         struct rte_mbuf **m_table;
 313         int ret;
 314         uint16_t queueid;
 315
 316         queueid = qconf->tx_queue_id;
 317         m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
 318
 319         ret = rte_eth_tx_burst(port, queueid, m_table, n);
 320         if (unlikely(ret < n)) {
 321                 do {
 322                         rte_pktmbuf_free(m_table[ret]);
 323                 } while (++ret < n);
 324         }
 325
 326         return 0;
 327 }
 328
 329 /* Enqueue a single packet, and send burst if queue is filled */
 330 static inline int
 331 send_single_packet(struct rte_mbuf *m, uint8_t port)
 332 {
 333         uint32_t lcore_id;
 334         uint16_t len;
 335         struct lcore_conf *qconf;
 336
 337         lcore_id = rte_lcore_id();
 338
 339         qconf = &lcore_conf[lcore_id];
 340         len = qconf->tx_mbufs[port].len;
 341         qconf->tx_mbufs[port].m_table[len] = m;
 342         len++;
 343
 344         /* enough pkts to be sent */
 345         if (unlikely(len == MAX_PKT_BURST)) {
 346                 send_burst(qconf, MAX_PKT_BURST, port);
 347                 len = 0;
 348         }
 349
 350         qconf->tx_mbufs[port].len = len;
 351         return 0;
 352 }
 353
 354 #ifdef DO_RFC_1812_CHECKS
 355 static inline int
 356 is_valid_ipv4_pkt(struct ipv4_hdr *pkt, uint32_t link_len)
 357 {
 358         /* From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2 */
 359         /*
 360          * 1. The packet length reported by the Link Layer must be large
 361          * enough to hold the minimum length legal IP datagram (20 bytes).
 362          */
 363         if (link_len < sizeof(struct ipv4_hdr))
 364                 return -1;
 365
 366         /* 2. The IP checksum must be correct. */
 367         /* this is checked in H/W */
 368
 369         /*
 370          * 3. The IP version number must be 4. If the version number is not 4
 371          * then the packet may be another version of IP, such as IPng or
 372          * ST-II.
 373          */
 374         if (((pkt->version_ihl) >> 4) != 4)
 375                 return -3;
 376         /*
 377          * 4. The IP header length field must be large enough to hold the
 378          * minimum length legal IP datagram (20 bytes = 5 words).
 379          */
 380         if ((pkt->version_ihl & 0xf) < 5)
 381                 return -4;
 382
 383         /*
 384          * 5. The IP total length field must be large enough to hold the IP
 385          * datagram header, whose length is specified in the IP header length
 386          * field.
 387          */
 388         if (rte_cpu_to_be_16(pkt->total_length) < sizeof(struct ipv4_hdr))
 389                 return -5;
 390
 391         return 0;
 392 }
 393 #endif
 394
 395 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 396 static void
 397 print_key(struct ipv4_5tuple key)
 398 {
 399         printf("IP dst = %08x, IP src = %08x, port dst = %d, port src = %d, proto = %d\n",
 400                (unsigned)key.ip_dst, (unsigned)key.ip_src, key.port_dst, key.port_src, key.proto);
 401 }
 402
 403 static inline uint8_t
 404 get_dst_port(struct ipv4_hdr *ipv4_hdr,  uint8_t portid, lookup_struct_t * l3fwd_lookup_struct)
 405 {
 406         struct ipv4_5tuple key;
 407         struct tcp_hdr *tcp;
 408         struct udp_hdr *udp;
 409         int ret = 0;
 410
 411         key.ip_dst = rte_be_to_cpu_32(ipv4_hdr->dst_addr);
 412         key.ip_src = rte_be_to_cpu_32(ipv4_hdr->src_addr);
 413         key.proto = ipv4_hdr->next_proto_id;
 414
 415         switch (ipv4_hdr->next_proto_id) {
 416         case IPPROTO_TCP:
 417                 tcp = (struct tcp_hdr *)((unsigned char *) ipv4_hdr +
 418                                         sizeof(struct ipv4_hdr));
 419                 key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
 420                 key.port_src = rte_be_to_cpu_16(tcp->src_port);
 421                 break;
 422
 423         case IPPROTO_UDP:
 424                 udp = (struct udp_hdr *)((unsigned char *) ipv4_hdr +
 425                                         sizeof(struct ipv4_hdr));
 426                 key.port_dst = rte_be_to_cpu_16(udp->dst_port);
 427                 key.port_src = rte_be_to_cpu_16(udp->src_port);
 428                 break;
 429
 430         default:
 431                 key.port_dst = 0;
 432                 key.port_src = 0;
 433         }
 434
 435         /* Find destination port */
 436         ret = rte_hash_lookup(l3fwd_lookup_struct, (const void *)&key);
 437         return (uint8_t)((ret < 0)? portid : l3fwd_out_if[ret]);
 438 }
 439 #endif
 440
 441 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 442 static inline uint8_t
 443 get_dst_port(struct ipv4_hdr *ipv4_hdr,  uint8_t portid, lookup_struct_t * l3fwd_lookup_struct)
 444 {
 445         uint8_t next_hop;
 446
 447         return (uint8_t) ((rte_lpm_lookup(l3fwd_lookup_struct,
 448                         rte_be_to_cpu_32(ipv4_hdr->dst_addr), &next_hop) == 0)?
 449                         next_hop : portid);
 450 }
 451 #endif
 452
 453 static inline void
 454 l3fwd_simple_forward(struct rte_mbuf *m, uint8_t portid, lookup_struct_t * l3fwd_lookup_struct)
 455 {
 456         struct ether_hdr *eth_hdr;
 457         struct ipv4_hdr *ipv4_hdr;
 458         void *tmp;
 459         uint8_t dst_port;
 460
 461         eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
 462
 463         ipv4_hdr = (struct ipv4_hdr *)(rte_pktmbuf_mtod(m, unsigned char *) +
 464                                 sizeof(struct ether_hdr));
 465
 466 #ifdef DO_RFC_1812_CHECKS
 467         /* Check to make sure the packet is valid (RFC1812) */
 468         if (is_valid_ipv4_pkt(ipv4_hdr, m->pkt_len) < 0) {
 469                 rte_pktmbuf_free(m);
 470                 return;
 471         }
 472 #endif
 473
 474         dst_port = get_dst_port(ipv4_hdr, portid, l3fwd_lookup_struct);
 475         if (dst_port >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port) == 0)
 476                 dst_port = portid;
 477
 478         /* 02:00:00:00:00:xx */
 479         tmp = &eth_hdr->d_addr.addr_bytes[0];
 480         *((uint64_t *)tmp) = 0x000000000002 + ((uint64_t)dst_port << 40);
 481
 482 #ifdef DO_RFC_1812_CHECKS
 483         /* Update time to live and header checksum */
 484         --(ipv4_hdr->time_to_live);
 485         ++(ipv4_hdr->hdr_checksum);
 486 #endif
 487
 488         /* src addr */
 489         ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
 490
 491         send_single_packet(m, dst_port);
 492
 493 }
 494
 495 /* main processing loop */
 496 static int
 497 main_loop(__attribute__((unused)) void *dummy)
 498 {
 499         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
 500         unsigned lcore_id;
 501         uint64_t prev_tsc, diff_tsc, cur_tsc;
 502         int i, j, nb_rx;
 503         uint8_t portid, queueid;
 504         struct lcore_conf *qconf;
 505         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
 506
 507         prev_tsc = 0;
 508
 509         lcore_id = rte_lcore_id();
 510         qconf = &lcore_conf[lcore_id];
 511
 512         if (qconf->n_rx_queue == 0) {
 513                 RTE_LOG(INFO, L3FWD, "lcore %u has nothing to do\n", lcore_id);
 514                 return 0;
 515         }
 516
 517         RTE_LOG(INFO, L3FWD, "entering main loop on lcore %u\n", lcore_id);
 518
 519         for (i = 0; i < qconf->n_rx_queue; i++) {
 520
 521                 portid = qconf->rx_queue_list[i].port_id;
 522                 queueid = qconf->rx_queue_list[i].queue_id;
 523                 RTE_LOG(INFO, L3FWD, " -- lcoreid=%u portid=%hhu rxqueueid=%hhu\n", lcore_id,
 524                         portid, queueid);
 525         }
 526
 527         while (1) {
 528
 529                 cur_tsc = rte_rdtsc();
 530
 531                 /*
 532                  * TX burst queue drain
 533                  */
 534                 diff_tsc = cur_tsc - prev_tsc;
 535                 if (unlikely(diff_tsc > drain_tsc)) {
 536
 537                         /*
 538                          * This could be optimized (use queueid instead of
 539                          * portid), but it is not called so often
 540                          */
 541                         for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
 542                                 if (qconf->tx_mbufs[portid].len == 0)
 543                                         continue;
 544                                 send_burst(&lcore_conf[lcore_id],
 545                                         qconf->tx_mbufs[portid].len,
 546                                         portid);
 547                                 qconf->tx_mbufs[portid].len = 0;
 548                         }
 549
 550                         prev_tsc = cur_tsc;
 551                 }
 552
 553                 /*
 554                  * Read packet from RX queues
 555                  */
 556                 for (i = 0; i < qconf->n_rx_queue; ++i) {
 557
 558                         portid = qconf->rx_queue_list[i].port_id;
 559                         queueid = qconf->rx_queue_list[i].queue_id;
 560                         nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst, MAX_PKT_BURST);
 561
 562                         /* Prefetch first packets */
 563                         for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
 564                                 rte_prefetch0(rte_pktmbuf_mtod(
 565                                                 pkts_burst[j], void *));
 566                         }
 567
 568                         /* Prefetch and forward already prefetched packets */
 569                         for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
 570                                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
 571                                                 j + PREFETCH_OFFSET], void *));
 572                                 l3fwd_simple_forward(pkts_burst[j], portid, qconf->lookup_struct);
 573                         }
 574
 575                         /* Forward remaining prefetched packets */
 576                         for (; j < nb_rx; j++) {
 577                                 l3fwd_simple_forward(pkts_burst[j], portid, qconf->lookup_struct);
 578                         }
 579                 }
 580         }
 581 }
 582
 583 static int
 584 check_lcore_params(void)
 585 {
 586         uint8_t queue, lcore;
 587         uint16_t i;
 588         int socketid;
 589
 590         for (i = 0; i < nb_lcore_params; ++i) {
 591                 queue = lcore_params[i].queue_id;
 592                 if (queue >= MAX_RX_QUEUE_PER_PORT) {
 593                         printf("invalid queue number: %hhu\n", queue);
 594                         return -1;
 595                 }
 596                 lcore = lcore_params[i].lcore_id;
 597                 if (!rte_lcore_is_enabled(lcore)) {
 598                         printf("error: lcore %hhu is not enabled in lcore mask\n", lcore);
 599                         return -1;
 600                 }
 601                 if ((socketid = rte_lcore_to_socket_id(lcore) != 0) &&
 602                         (numa_on == 0)) {
 603                         printf("warning: lcore %hhu is on socket %d with numa off \n",
 604                                 lcore, socketid);
 605                 }
 606         }
 607         return 0;
 608 }
 609
 610 static int
 611 check_port_config(const unsigned nb_ports)
 612 {
 613         unsigned portid;
 614         uint16_t i;
 615
 616         for (i = 0; i < nb_lcore_params; ++i) {
 617                 portid = lcore_params[i].port_id;
 618                 if ((enabled_port_mask & (1 << portid)) == 0) {
 619                         printf("port %u is not enabled in port mask\n", portid);
 620                         return -1;
 621                 }
 622                 if (portid >= nb_ports) {
 623                         printf("port %u is not present on the board\n", portid);
 624                         return -1;
 625                 }
 626         }
 627         return 0;
 628 }
 629
 630 static uint8_t
 631 get_port_n_rx_queues(const uint8_t port)
 632 {
 633         int queue = -1;
 634         uint16_t i;
 635
 636         for (i = 0; i < nb_lcore_params; ++i) {
 637                 if (lcore_params[i].port_id == port && lcore_params[i].queue_id > queue)
 638                         queue = lcore_params[i].queue_id;
 639         }
 640         return (uint8_t)(++queue);
 641 }
 642
 643 static int
 644 init_lcore_rx_queues(void)
 645 {
 646         uint16_t i, nb_rx_queue;
 647         uint8_t lcore;
 648
 649         for (i = 0; i < nb_lcore_params; ++i) {
 650                 lcore = lcore_params[i].lcore_id;
 651                 nb_rx_queue = lcore_conf[lcore].n_rx_queue;
 652                 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
 653                         printf("error: too many queues (%u) for lcore: %u\n",
 654                                 (unsigned)nb_rx_queue + 1, (unsigned)lcore);
 655                         return -1;
 656                 } else {
 657                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
 658                                 lcore_params[i].port_id;
 659                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
 660                                 lcore_params[i].queue_id;
 661                         lcore_conf[lcore].n_rx_queue++;
 662                 }
 663         }
 664         return 0;
 665 }
 666
 667 /* display usage */
 668 static void
 669 print_usage(const char *prgname)
 670 {
 671         printf ("%s [EAL options] -- -p PORTMASK"
 672                 "  [--config (port,queue,lcore)[,(port,queue,lcore]]\n"
 673                 "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
 674                 "  --config (port,queue,lcore): rx queues configuration\n"
 675                 "  --no-numa: optional, disable numa awareness\n",
 676                 prgname);
 677 }
 678
 679 /* Custom handling of signals to handle process terminal */
 680 static void
 681 signal_handler(int signum)
 682 {
 683         uint8_t portid;
 684         uint8_t nb_ports = rte_eth_dev_count();
 685
 686         /* When we receive a SIGINT signal */
 687         if (signum == SIGINT) {
 688                 for (portid = 0; portid < nb_ports; portid++) {
 689                         /* skip ports that are not enabled */
 690                         if ((enabled_port_mask & (1 << portid)) == 0)
 691                                 continue;
 692                         rte_eth_dev_close(portid);
 693                 }
 694         }
 695         rte_exit(EXIT_SUCCESS, "\n User forced exit\n");
 696 }
 697 static int
 698 parse_portmask(const char *portmask)
 699 {
 700         char *end = NULL;
 701         unsigned long pm;
 702
 703         /* parse hexadecimal string */
 704         pm = strtoul(portmask, &end, 16);
 705         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
 706                 return -1;
 707
 708         if (pm == 0)
 709                 return -1;
 710
 711         return pm;
 712 }
 713
 714 static int
 715 parse_config(const char *q_arg)
 716 {
 717         char s[256];
 718         const char *p, *p0 = q_arg;
 719         char *end;
 720         enum fieldnames {
 721                 FLD_PORT = 0,
 722                 FLD_QUEUE,
 723                 FLD_LCORE,
 724                 _NUM_FLD
 725         };
 726         unsigned long int_fld[_NUM_FLD];
 727         char *str_fld[_NUM_FLD];
 728         int i;
 729         unsigned size;
 730
 731         nb_lcore_params = 0;
 732
 733         while ((p = strchr(p0,'(')) != NULL) {
 734                 ++p;
 735                 if((p0 = strchr(p,')')) == NULL)
 736                         return -1;
 737
 738                 size = p0 - p;
 739                 if(size >= sizeof(s))
 740                         return -1;
 741
 742                 snprintf(s, sizeof(s), "%.*s", size, p);
 743                 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') != _NUM_FLD)
 744                         return -1;
 745                 for (i = 0; i < _NUM_FLD; i++){
 746                         errno = 0;
 747                         int_fld[i] = strtoul(str_fld[i], &end, 0);
 748                         if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
 749                                 return -1;
 750                 }
 751                 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
 752                         printf("exceeded max number of lcore params: %hu\n",
 753                                 nb_lcore_params);
 754                         return -1;
 755                 }
 756                 lcore_params_array[nb_lcore_params].port_id = (uint8_t)int_fld[FLD_PORT];
 757                 lcore_params_array[nb_lcore_params].queue_id = (uint8_t)int_fld[FLD_QUEUE];
 758                 lcore_params_array[nb_lcore_params].lcore_id = (uint8_t)int_fld[FLD_LCORE];
 759                 ++nb_lcore_params;
 760         }
 761         lcore_params = lcore_params_array;
 762         return 0;
 763 }
 764
 765 /* Parse the argument given in the command line of the application */
 766 static int
 767 parse_args(int argc, char **argv)
 768 {
 769         int opt, ret;
 770         char **argvopt;
 771         int option_index;
 772         char *prgname = argv[0];
 773         static struct option lgopts[] = {
 774                 {"config", 1, 0, 0},
 775                 {"no-numa", 0, 0, 0},
 776                 {NULL, 0, 0, 0}
 777         };
 778
 779         argvopt = argv;
 780
 781         while ((opt = getopt_long(argc, argvopt, "p:",
 782                                 lgopts, &option_index)) != EOF) {
 783
 784                 switch (opt) {
 785                 /* portmask */
 786                 case 'p':
 787                         enabled_port_mask = parse_portmask(optarg);
 788                         if (enabled_port_mask == 0) {
 789                                 printf("invalid portmask\n");
 790                                 print_usage(prgname);
 791                                 return -1;
 792                         }
 793                         break;
 794
 795                 /* long options */
 796                 case 0:
 797                         if (!strcmp(lgopts[option_index].name, "config")) {
 798                                 ret = parse_config(optarg);
 799                                 if (ret) {
 800                                         printf("invalid config\n");
 801                                         print_usage(prgname);
 802                                         return -1;
 803                                 }
 804                         }
 805
 806                         if (!strcmp(lgopts[option_index].name, "no-numa")) {
 807                                 printf("numa is disabled \n");
 808                                 numa_on = 0;
 809                         }
 810                         break;
 811
 812                 default:
 813                         print_usage(prgname);
 814                         return -1;
 815                 }
 816         }
 817
 818         if (optind >= 0)
 819                 argv[optind-1] = prgname;
 820
 821         ret = optind-1;
 822         optind = 0; /* reset getopt lib */
 823         return ret;
 824 }
 825
 826 static void
 827 print_ethaddr(const char *name, const struct ether_addr *eth_addr)
 828 {
 829         printf ("%s%02X:%02X:%02X:%02X:%02X:%02X", name,
 830                 eth_addr->addr_bytes[0],
 831                 eth_addr->addr_bytes[1],
 832                 eth_addr->addr_bytes[2],
 833                 eth_addr->addr_bytes[3],
 834                 eth_addr->addr_bytes[4],
 835                 eth_addr->addr_bytes[5]);
 836 }
 837
 838 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 839 static void
 840 setup_hash(int socketid)
 841 {
 842         unsigned i;
 843         int ret;
 844         char s[64];
 845
 846         /* create  hashes */
 847         snprintf(s, sizeof(s), "l3fwd_hash_%d", socketid);
 848         l3fwd_hash_params.name = s;
 849         l3fwd_hash_params.socket_id = socketid;
 850         l3fwd_lookup_struct[socketid] = rte_hash_create(&l3fwd_hash_params);
 851         if (l3fwd_lookup_struct[socketid] == NULL)
 852                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
 853                                 "socket %d\n", socketid);
 854
 855         /* populate the hash */
 856         for (i = 0; i < L3FWD_NUM_ROUTES; i++) {
 857                 ret = rte_hash_add_key (l3fwd_lookup_struct[socketid],
 858                                 (void *) &l3fwd_route_array[i].key);
 859                 if (ret < 0) {
 860                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
 861                                 "l3fwd hash on socket %d\n", i, socketid);
 862                 }
 863                 l3fwd_out_if[ret] = l3fwd_route_array[i].if_out;
 864                 printf("Hash: Adding key\n");
 865                 print_key(l3fwd_route_array[i].key);
 866         }
 867 }
 868 #endif
 869
 870 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 871 static void
 872 setup_lpm(int socketid)
 873 {
 874         unsigned i;
 875         int ret;
 876         char s[64];
 877
 878         /* create the LPM table */
 879         snprintf(s, sizeof(s), "L3FWD_LPM_%d", socketid);
 880         l3fwd_lookup_struct[socketid] = rte_lpm_create(s, socketid,
 881                                 L3FWD_LPM_MAX_RULES, 0);
 882         if (l3fwd_lookup_struct[socketid] == NULL)
 883                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
 884                                 " on socket %d\n", socketid);
 885
 886         /* populate the LPM table */
 887         for (i = 0; i < L3FWD_NUM_ROUTES; i++) {
 888                 ret = rte_lpm_add(l3fwd_lookup_struct[socketid],
 889                         l3fwd_route_array[i].ip,
 890                         l3fwd_route_array[i].depth,
 891                         l3fwd_route_array[i].if_out);
 892
 893                 if (ret < 0) {
 894                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
 895                                 "l3fwd LPM table on socket %d\n",
 896                                 i, socketid);
 897                 }
 898
 899                 printf("LPM: Adding route 0x%08x / %d (%d)\n",
 900                         (unsigned)l3fwd_route_array[i].ip,
 901                         l3fwd_route_array[i].depth,
 902                         l3fwd_route_array[i].if_out);
 903         }
 904 }
 905 #endif
 906
 907 static int
 908 init_mem(unsigned nb_mbuf)
 909 {
 910         struct lcore_conf *qconf;
 911         int socketid;
 912         unsigned lcore_id;
 913         char s[64];
 914
 915         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
 916                 if (rte_lcore_is_enabled(lcore_id) == 0)
 917                         continue;
 918
 919                 if (numa_on)
 920                         socketid = rte_lcore_to_socket_id(lcore_id);
 921                 else
 922                         socketid = 0;
 923
 924                 if (socketid >= NB_SOCKETS) {
 925                         rte_exit(EXIT_FAILURE, "Socket %d of lcore %u is out of range %d\n",
 926                                 socketid, lcore_id, NB_SOCKETS);
 927                 }
 928                 if (pktmbuf_pool[socketid] == NULL) {
 929                         snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
 930                         pktmbuf_pool[socketid] =
 931                                 rte_mempool_create(s, nb_mbuf, MBUF_SIZE,
 932                                                    MEMPOOL_CACHE_SIZE,
 933                                         sizeof(struct rte_pktmbuf_pool_private),
 934                                         rte_pktmbuf_pool_init, NULL,
 935                                         rte_pktmbuf_init, NULL,
 936                                         socketid, 0);
 937                         if (pktmbuf_pool[socketid] == NULL)
 938                                 rte_exit(EXIT_FAILURE, "Cannot init mbuf pool on socket %d\n", socketid);
 939                         else
 940                                 printf("Allocated mbuf pool on socket %d\n", socketid);
 941
 942 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 943                         setup_lpm(socketid);
 944 #else
 945                         setup_hash(socketid);
 946 #endif
 947                 }
 948                 qconf = &lcore_conf[lcore_id];
 949                 qconf->lookup_struct = l3fwd_lookup_struct[socketid];
 950         }
 951         return 0;
 952 }
 953
 954 int
 955 MAIN(int argc, char **argv)
 956 {
 957         struct lcore_conf *qconf;
 958         struct rte_eth_dev_info dev_info;
 959         struct rte_eth_txconf *txconf;
 960         int ret;
 961         unsigned nb_ports;
 962         uint16_t queueid;
 963         unsigned lcore_id;
 964         uint32_t nb_lcores;
 965         uint16_t n_tx_queue;
 966         uint8_t portid, nb_rx_queue, queue, socketid;
 967
 968         signal(SIGINT, signal_handler);
 969         /* init EAL */
 970         ret = rte_eal_init(argc, argv);
 971         if (ret < 0)
 972                 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
 973         argc -= ret;
 974         argv += ret;
 975
 976         /* parse application arguments (after the EAL ones) */
 977         ret = parse_args(argc, argv);
 978         if (ret < 0)
 979                 rte_exit(EXIT_FAILURE, "Invalid L3FWD-VF parameters\n");
 980
 981         if (check_lcore_params() < 0)
 982                 rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
 983
 984         ret = init_lcore_rx_queues();
 985         if (ret < 0)
 986                 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
 987
 988         nb_ports = rte_eth_dev_count();
 989         if (nb_ports > RTE_MAX_ETHPORTS)
 990                 nb_ports = RTE_MAX_ETHPORTS;
 991
 992         if (check_port_config(nb_ports) < 0)
 993                 rte_exit(EXIT_FAILURE, "check_port_config failed\n");
 994
 995         nb_lcores = rte_lcore_count();
 996
 997         /* initialize all ports */
 998         for (portid = 0; portid < nb_ports; portid++) {
 999                 /* skip ports that are not enabled */
1000                 if ((enabled_port_mask & (1 << portid)) == 0) {
1001                         printf("\nSkipping disabled port %d\n", portid);
1002                         continue;
1003                 }
1004
1005                 /* init port */
1006                 printf("Initializing port %d ... ", portid );
1007                 fflush(stdout);
1008
1009                 /* must always equal(=1) */
1010                 nb_rx_queue = get_port_n_rx_queues(portid);
1011                 n_tx_queue = MAX_TX_QUEUE_PER_PORT;
1012
1013                 printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
1014                         nb_rx_queue, (unsigned)1 );
1015                 ret = rte_eth_dev_configure(portid, nb_rx_queue, n_tx_queue, &port_conf);
1016                 if (ret < 0)
1017                         rte_exit(EXIT_FAILURE, "Cannot configure device: err=%d, port=%d\n",
1018                                 ret, portid);
1019
1020                 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
1021                 print_ethaddr(" Address:", &ports_eth_addr[portid]);
1022                 printf(", ");
1023
1024                 ret = init_mem(NB_MBUF);
1025                 if (ret < 0)
1026                         rte_exit(EXIT_FAILURE, "init_mem failed\n");
1027
1028                 /* init one TX queue */
1029                 socketid = (uint8_t)rte_lcore_to_socket_id(rte_get_master_lcore());
1030
1031                 printf("txq=%d,%d,%d ", portid, 0, socketid);
1032                 fflush(stdout);
1033
1034                 rte_eth_dev_info_get(portid, &dev_info);
1035                 txconf = &dev_info.default_txconf;
1036                 if (port_conf.rxmode.jumbo_frame)
1037                         txconf->txq_flags = 0;
1038                 ret = rte_eth_tx_queue_setup(portid, 0, nb_txd,
1039                                                  socketid, txconf);
1040                 if (ret < 0)
1041                         rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: err=%d, "
1042                                 "port=%d\n", ret, portid);
1043
1044                 printf("\n");
1045         }
1046
1047         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1048                 if (rte_lcore_is_enabled(lcore_id) == 0)
1049                         continue;
1050                 qconf = &lcore_conf[lcore_id];
1051                 qconf->tx_queue_id = 0;
1052
1053                 printf("\nInitializing rx queues on lcore %u ... ", lcore_id );
1054                 fflush(stdout);
1055                 /* init RX queues */
1056                 for(queue = 0; queue < qconf->n_rx_queue; ++queue) {
1057                         portid = qconf->rx_queue_list[queue].port_id;
1058                         queueid = qconf->rx_queue_list[queue].queue_id;
1059
1060                         if (numa_on)
1061                                 socketid = (uint8_t)rte_lcore_to_socket_id(lcore_id);
1062                         else
1063                                 socketid = 0;
1064
1065                         printf("rxq=%d,%d,%d ", portid, queueid, socketid);
1066                         fflush(stdout);
1067
1068                         ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
1069                                                 socketid, NULL,
1070                                                 pktmbuf_pool[socketid]);
1071                         if (ret < 0)
1072                                 rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: err=%d,"
1073                                                 "port=%d\n", ret, portid);
1074                 }
1075         }
1076         printf("\n");
1077
1078         /* start ports */
1079         for (portid = 0; portid < nb_ports; portid++) {
1080                 if ((enabled_port_mask & (1 << portid)) == 0) {
1081                         continue;
1082                 }
1083                 /* Start device */
1084                 ret = rte_eth_dev_start(portid);
1085                 if (ret < 0)
1086                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%d\n",
1087                                 ret, portid);
1088
1089                 printf("done: Port %d\n", portid);
1090
1091         }
1092
1093         /* launch per-lcore init on every lcore */
1094         rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
1095         RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1096                 if (rte_eal_wait_lcore(lcore_id) < 0)
1097                         return -1;
1098         }
1099
1100         return 0;
1101 }