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