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_launch.h>
  54 #include <rte_atomic.h>
  55 #include <rte_spinlock.h>
  56 #include <rte_cycles.h>
  57 #include <rte_prefetch.h>
  58 #include <rte_lcore.h>
  59 #include <rte_per_lcore.h>
  60 #include <rte_branch_prediction.h>
  61 #include <rte_interrupts.h>
  62 #include <rte_pci.h>
  63 #include <rte_random.h>
  64 #include <rte_debug.h>
  65 #include <rte_ether.h>
  66 #include <rte_ethdev.h>
  67 #include <rte_mempool.h>
  68 #include <rte_mbuf.h>
  69 #include <rte_ip.h>
  70 #include <rte_tcp.h>
  71 #include <rte_udp.h>
  72 #include <rte_string_fns.h>
  73
  74 #define APP_LOOKUP_EXACT_MATCH          0
  75 #define APP_LOOKUP_LPM                  1
  76 #define DO_RFC_1812_CHECKS
  77
  78 //#define APP_LOOKUP_METHOD             APP_LOOKUP_EXACT_MATCH
  79 #ifndef APP_LOOKUP_METHOD
  80 #define APP_LOOKUP_METHOD             APP_LOOKUP_LPM
  81 #endif
  82
  83 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
  84 #include <rte_hash.h>
  85 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
  86 #include <rte_lpm.h>
  87 #else
  88 #error "APP_LOOKUP_METHOD set to incorrect value"
  89 #endif
  90
  91 #define RTE_LOGTYPE_L3FWD RTE_LOGTYPE_USER1
  92
  93 #define MEMPOOL_CACHE_SIZE 256
  94
  95 /*
  96  * This expression is used to calculate the number of mbufs needed depending on user input, taking
  97  *  into account memory for rx and tx hardware rings, cache per lcore and mtable per port per lcore.
  98  *  RTE_MAX is used to ensure that NB_MBUF never goes below a minimum value of 8192
  99  */
 100
 101 #define NB_MBUF RTE_MAX (                                               \
 102                                 (nb_ports*nb_rx_queue*nb_rxd +          \
 103                                 nb_ports*nb_lcores*MAX_PKT_BURST +      \
 104                                 nb_ports*n_tx_queue*nb_txd +            \
 105                                 nb_lcores*MEMPOOL_CACHE_SIZE),          \
 106                                 (unsigned)8192)
 107
 108 /*
 109  * RX and TX Prefetch, Host, and Write-back threshold values should be
 110  * carefully set for optimal performance. Consult the network
 111  * controller's datasheet and supporting DPDK documentation for guidance
 112  * on how these parameters should be set.
 113  */
 114 #define RX_PTHRESH 8 /**< Default values of RX prefetch threshold reg. */
 115 #define RX_HTHRESH 8 /**< Default values of RX host threshold reg. */
 116 #define RX_WTHRESH 4 /**< Default values of RX write-back threshold reg. */
 117
 118 /*
 119  * These default values are optimized for use with the Intel(R) 82599 10 GbE
 120  * Controller and the DPDK ixgbe PMD. Consider using other values for other
 121  * network controllers and/or network drivers.
 122  */
 123 #define TX_PTHRESH 36 /**< Default values of TX prefetch threshold reg. */
 124 #define TX_HTHRESH 0  /**< Default values of TX host threshold reg. */
 125 #define TX_WTHRESH 0  /**< Default values of TX write-back threshold reg. */
 126
 127 #define MAX_PKT_BURST 32
 128 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
 129
 130 #define NB_SOCKETS 8
 131
 132 #define SOCKET0 0
 133
 134 /* Configure how many packets ahead to prefetch, when reading packets */
 135 #define PREFETCH_OFFSET 3
 136
 137 /*
 138  * Configurable number of RX/TX ring descriptors
 139  */
 140 #define RTE_TEST_RX_DESC_DEFAULT 128
 141 #define RTE_TEST_TX_DESC_DEFAULT 512
 142 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
 143 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
 144
 145 /* ethernet addresses of ports */
 146 static struct ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
 147
 148 /* mask of enabled ports */
 149 static uint32_t enabled_port_mask = 0;
 150 static int numa_on = 1; /**< NUMA is enabled by default. */
 151
 152 struct mbuf_table {
 153         uint16_t len;
 154         struct rte_mbuf *m_table[MAX_PKT_BURST];
 155 };
 156
 157 struct lcore_rx_queue {
 158         uint16_t port_id;
 159         uint8_t queue_id;
 160 } __rte_cache_aligned;
 161
 162 #define MAX_RX_QUEUE_PER_LCORE 16
 163 #define MAX_TX_QUEUE_PER_PORT 1
 164 #define MAX_RX_QUEUE_PER_PORT 1
 165
 166 #define MAX_LCORE_PARAMS 1024
 167 struct lcore_params {
 168         uint16_t port_id;
 169         uint8_t queue_id;
 170         uint8_t lcore_id;
 171 } __rte_cache_aligned;
 172
 173 static struct lcore_params lcore_params_array[MAX_LCORE_PARAMS];
 174 static struct lcore_params lcore_params_array_default[] = {
 175         {0, 0, 2},
 176         {0, 1, 2},
 177         {0, 2, 2},
 178         {1, 0, 2},
 179         {1, 1, 2},
 180         {1, 2, 2},
 181         {2, 0, 2},
 182         {3, 0, 3},
 183         {3, 1, 3},
 184 };
 185
 186 static struct lcore_params * lcore_params = lcore_params_array_default;
 187 static uint16_t nb_lcore_params = sizeof(lcore_params_array_default) /
 188                                 sizeof(lcore_params_array_default[0]);
 189
 190 static struct rte_eth_conf port_conf = {
 191         .rxmode = {
 192                 .mq_mode        = ETH_MQ_RX_RSS,
 193                 .max_rx_pkt_len = ETHER_MAX_LEN,
 194                 .split_hdr_size = 0,
 195                 .header_split   = 0, /**< Header Split disabled */
 196                 .hw_ip_checksum = 1, /**< IP checksum offload enabled */
 197                 .hw_vlan_filter = 0, /**< VLAN filtering disabled */
 198                 .jumbo_frame    = 0, /**< Jumbo Frame Support disabled */
 199                 .hw_strip_crc   = 1, /**< CRC stripped by hardware */
 200         },
 201         .rx_adv_conf = {
 202                 .rss_conf = {
 203                         .rss_key = NULL,
 204                         .rss_hf = ETH_RSS_IP,
 205                 },
 206         },
 207         .txmode = {
 208                 .mq_mode = ETH_MQ_TX_NONE,
 209         },
 210 };
 211
 212 static struct rte_mempool * pktmbuf_pool[NB_SOCKETS];
 213
 214
 215 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 216
 217 #ifdef RTE_ARCH_X86
 218 #include <rte_hash_crc.h>
 219 #define DEFAULT_HASH_FUNC       rte_hash_crc
 220 #else
 221 #include <rte_jhash.h>
 222 #define DEFAULT_HASH_FUNC       rte_jhash
 223 #endif
 224
 225 struct ipv4_5tuple {
 226         uint32_t ip_dst;
 227         uint32_t ip_src;
 228         uint16_t port_dst;
 229         uint16_t port_src;
 230         uint8_t proto;
 231 } __attribute__((__packed__));
 232
 233 struct l3fwd_route {
 234         struct ipv4_5tuple key;
 235         uint8_t if_out;
 236 };
 237
 238 static struct l3fwd_route l3fwd_route_array[] = {
 239         {{IPv4(100,10,0,1), IPv4(200,10,0,1), 101, 11, IPPROTO_TCP}, 0},
 240         {{IPv4(100,20,0,2), IPv4(200,20,0,2), 102, 12, IPPROTO_TCP}, 1},
 241         {{IPv4(100,30,0,3), IPv4(200,30,0,3), 103, 13, IPPROTO_TCP}, 2},
 242         {{IPv4(100,40,0,4), IPv4(200,40,0,4), 104, 14, IPPROTO_TCP}, 3},
 243 };
 244
 245 typedef struct rte_hash lookup_struct_t;
 246 static lookup_struct_t *l3fwd_lookup_struct[NB_SOCKETS];
 247
 248 #define L3FWD_HASH_ENTRIES      1024
 249 struct rte_hash_parameters l3fwd_hash_params = {
 250         .name = "l3fwd_hash_0",
 251         .entries = L3FWD_HASH_ENTRIES,
 252         .key_len = sizeof(struct ipv4_5tuple),
 253         .hash_func = DEFAULT_HASH_FUNC,
 254         .hash_func_init_val = 0,
 255         .socket_id = SOCKET0,
 256 };
 257
 258 #define L3FWD_NUM_ROUTES \
 259         (sizeof(l3fwd_route_array) / sizeof(l3fwd_route_array[0]))
 260
 261 static uint8_t l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 262 #endif
 263
 264 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 265 struct l3fwd_route {
 266         uint32_t ip;
 267         uint8_t  depth;
 268         uint8_t  if_out;
 269 };
 270
 271 static struct l3fwd_route l3fwd_route_array[] = {
 272         {IPv4(1,1,1,0), 24, 0},
 273         {IPv4(2,1,1,0), 24, 1},
 274         {IPv4(3,1,1,0), 24, 2},
 275         {IPv4(4,1,1,0), 24, 3},
 276         {IPv4(5,1,1,0), 24, 4},
 277         {IPv4(6,1,1,0), 24, 5},
 278         {IPv4(7,1,1,0), 24, 6},
 279         {IPv4(8,1,1,0), 24, 7},
 280 };
 281
 282 #define L3FWD_NUM_ROUTES \
 283         (sizeof(l3fwd_route_array) / sizeof(l3fwd_route_array[0]))
 284
 285 #define L3FWD_LPM_MAX_RULES     1024
 286
 287 typedef struct rte_lpm lookup_struct_t;
 288 static lookup_struct_t *l3fwd_lookup_struct[NB_SOCKETS];
 289 #endif
 290
 291 struct lcore_conf {
 292         uint16_t n_rx_queue;
 293         struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
 294         uint16_t tx_queue_id;
 295         struct mbuf_table tx_mbufs[RTE_MAX_ETHPORTS];
 296         lookup_struct_t * lookup_struct;
 297 } __rte_cache_aligned;
 298
 299 static struct lcore_conf lcore_conf[RTE_MAX_LCORE];
 300 static rte_spinlock_t spinlock_conf[RTE_MAX_ETHPORTS] = {RTE_SPINLOCK_INITIALIZER};
 301 /* Send burst of packets on an output interface */
 302 static inline int
 303 send_burst(struct lcore_conf *qconf, uint16_t n, uint16_t port)
 304 {
 305         struct rte_mbuf **m_table;
 306         int ret;
 307         uint16_t queueid;
 308
 309         queueid = qconf->tx_queue_id;
 310         m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
 311
 312         rte_spinlock_lock(&spinlock_conf[port]);
 313         ret = rte_eth_tx_burst(port, queueid, m_table, n);
 314         rte_spinlock_unlock(&spinlock_conf[port]);
 315
 316         if (unlikely(ret < n)) {
 317                 do {
 318                         rte_pktmbuf_free(m_table[ret]);
 319                 } while (++ret < n);
 320         }
 321
 322         return 0;
 323 }
 324
 325 /* Enqueue a single packet, and send burst if queue is filled */
 326 static inline int
 327 send_single_packet(struct rte_mbuf *m, uint16_t port)
 328 {
 329         uint32_t lcore_id;
 330         uint16_t len;
 331         struct lcore_conf *qconf;
 332
 333         lcore_id = rte_lcore_id();
 334
 335         qconf = &lcore_conf[lcore_id];
 336         len = qconf->tx_mbufs[port].len;
 337         qconf->tx_mbufs[port].m_table[len] = m;
 338         len++;
 339
 340         /* enough pkts to be sent */
 341         if (unlikely(len == MAX_PKT_BURST)) {
 342                 send_burst(qconf, MAX_PKT_BURST, port);
 343                 len = 0;
 344         }
 345
 346         qconf->tx_mbufs[port].len = len;
 347         return 0;
 348 }
 349
 350 #ifdef DO_RFC_1812_CHECKS
 351 static inline int
 352 is_valid_ipv4_pkt(struct ipv4_hdr *pkt, uint32_t link_len)
 353 {
 354         /* From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2 */
 355         /*
 356          * 1. The packet length reported by the Link Layer must be large
 357          * enough to hold the minimum length legal IP datagram (20 bytes).
 358          */
 359         if (link_len < sizeof(struct ipv4_hdr))
 360                 return -1;
 361
 362         /* 2. The IP checksum must be correct. */
 363         /* this is checked in H/W */
 364
 365         /*
 366          * 3. The IP version number must be 4. If the version number is not 4
 367          * then the packet may be another version of IP, such as IPng or
 368          * ST-II.
 369          */
 370         if (((pkt->version_ihl) >> 4) != 4)
 371                 return -3;
 372         /*
 373          * 4. The IP header length field must be large enough to hold the
 374          * minimum length legal IP datagram (20 bytes = 5 words).
 375          */
 376         if ((pkt->version_ihl & 0xf) < 5)
 377                 return -4;
 378
 379         /*
 380          * 5. The IP total length field must be large enough to hold the IP
 381          * datagram header, whose length is specified in the IP header length
 382          * field.
 383          */
 384         if (rte_cpu_to_be_16(pkt->total_length) < sizeof(struct ipv4_hdr))
 385                 return -5;
 386
 387         return 0;
 388 }
 389 #endif
 390
 391 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 392 static void
 393 print_key(struct ipv4_5tuple key)
 394 {
 395         printf("IP dst = %08x, IP src = %08x, port dst = %d, port src = %d, proto = %d\n",
 396                (unsigned)key.ip_dst, (unsigned)key.ip_src, key.port_dst, key.port_src, key.proto);
 397 }
 398
 399 static inline uint16_t
 400 get_dst_port(struct ipv4_hdr *ipv4_hdr, uint16_t portid,
 401               lookup_struct_t *l3fwd_lookup_struct)
 402 {
 403         struct ipv4_5tuple key;
 404         struct tcp_hdr *tcp;
 405         struct udp_hdr *udp;
 406         int ret = 0;
 407
 408         key.ip_dst = rte_be_to_cpu_32(ipv4_hdr->dst_addr);
 409         key.ip_src = rte_be_to_cpu_32(ipv4_hdr->src_addr);
 410         key.proto = ipv4_hdr->next_proto_id;
 411
 412         switch (ipv4_hdr->next_proto_id) {
 413         case IPPROTO_TCP:
 414                 tcp = (struct tcp_hdr *)((unsigned char *) ipv4_hdr +
 415                                         sizeof(struct ipv4_hdr));
 416                 key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
 417                 key.port_src = rte_be_to_cpu_16(tcp->src_port);
 418                 break;
 419
 420         case IPPROTO_UDP:
 421                 udp = (struct udp_hdr *)((unsigned char *) ipv4_hdr +
 422                                         sizeof(struct ipv4_hdr));
 423                 key.port_dst = rte_be_to_cpu_16(udp->dst_port);
 424                 key.port_src = rte_be_to_cpu_16(udp->src_port);
 425                 break;
 426
 427         default:
 428                 key.port_dst = 0;
 429                 key.port_src = 0;
 430         }
 431
 432         /* Find destination port */
 433         ret = rte_hash_lookup(l3fwd_lookup_struct, (const void *)&key);
 434         return ((ret < 0) ? portid : l3fwd_out_if[ret]);
 435 }
 436 #endif
 437
 438 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 439 static inline uint32_t
 440 get_dst_port(struct ipv4_hdr *ipv4_hdr, uint16_t portid,
 441               lookup_struct_t *l3fwd_lookup_struct)
 442 {
 443         uint32_t next_hop;
 444
 445         return ((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, uint16_t portid,
 453                       lookup_struct_t *l3fwd_lookup_struct)
 454 {
 455         struct ether_hdr *eth_hdr;
 456         struct ipv4_hdr *ipv4_hdr;
 457         void *tmp;
 458         uint16_t dst_port;
 459
 460         eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
 461
 462         ipv4_hdr = rte_pktmbuf_mtod_offset(m, struct ipv4_hdr *,
 463                                            sizeof(struct ether_hdr));
 464
 465 #ifdef DO_RFC_1812_CHECKS
 466         /* Check to make sure the packet is valid (RFC1812) */
 467         if (is_valid_ipv4_pkt(ipv4_hdr, m->pkt_len) < 0) {
 468                 rte_pktmbuf_free(m);
 469                 return;
 470         }
 471 #endif
 472
 473         dst_port = get_dst_port(ipv4_hdr, portid, l3fwd_lookup_struct);
 474         if (dst_port >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port) == 0)
 475                 dst_port = portid;
 476
 477         /* 02:00:00:00:00:xx */
 478         tmp = &eth_hdr->d_addr.addr_bytes[0];
 479         *((uint64_t *)tmp) = 0x000000000002 + ((uint64_t)dst_port << 40);
 480
 481 #ifdef DO_RFC_1812_CHECKS
 482         /* Update time to live and header checksum */
 483         --(ipv4_hdr->time_to_live);
 484         ++(ipv4_hdr->hdr_checksum);
 485 #endif
 486
 487         /* src addr */
 488         ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
 489
 490         send_single_packet(m, dst_port);
 491
 492 }
 493
 494 /* main processing loop */
 495 static int
 496 main_loop(__attribute__((unused)) void *dummy)
 497 {
 498         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
 499         unsigned lcore_id;
 500         uint64_t prev_tsc, diff_tsc, cur_tsc;
 501         int i, j, nb_rx;
 502         uint8_t queueid;
 503         uint16_t portid;
 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=%u rxqueueid=%hhu\n",
 524                 lcore_id, 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 uint16_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         uint16_t portid;
 684         uint16_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 = 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 = 1; /* 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         char buf[ETHER_ADDR_FMT_SIZE];
 830         ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr);
 831         printf("%s%s", name, buf);
 832 }
 833
 834 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 835 static void
 836 setup_hash(int socketid)
 837 {
 838         unsigned i;
 839         int ret;
 840         char s[64];
 841
 842         /* create  hashes */
 843         snprintf(s, sizeof(s), "l3fwd_hash_%d", socketid);
 844         l3fwd_hash_params.name = s;
 845         l3fwd_hash_params.socket_id = socketid;
 846         l3fwd_lookup_struct[socketid] = rte_hash_create(&l3fwd_hash_params);
 847         if (l3fwd_lookup_struct[socketid] == NULL)
 848                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
 849                                 "socket %d\n", socketid);
 850
 851         /* populate the hash */
 852         for (i = 0; i < L3FWD_NUM_ROUTES; i++) {
 853                 ret = rte_hash_add_key (l3fwd_lookup_struct[socketid],
 854                                 (void *) &l3fwd_route_array[i].key);
 855                 if (ret < 0) {
 856                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
 857                                 "l3fwd hash on socket %d\n", i, socketid);
 858                 }
 859                 l3fwd_out_if[ret] = l3fwd_route_array[i].if_out;
 860                 printf("Hash: Adding key\n");
 861                 print_key(l3fwd_route_array[i].key);
 862         }
 863 }
 864 #endif
 865
 866 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 867 static void
 868 setup_lpm(int socketid)
 869 {
 870         unsigned i;
 871         int ret;
 872         char s[64];
 873
 874         struct rte_lpm_config lpm_ipv4_config;
 875
 876         lpm_ipv4_config.max_rules = L3FWD_LPM_MAX_RULES;
 877         lpm_ipv4_config.number_tbl8s = 256;
 878         lpm_ipv4_config.flags = 0;
 879
 880         /* create the LPM table */
 881         snprintf(s, sizeof(s), "L3FWD_LPM_%d", socketid);
 882         l3fwd_lookup_struct[socketid] =
 883                         rte_lpm_create(s, socketid, &lpm_ipv4_config);
 884         if (l3fwd_lookup_struct[socketid] == NULL)
 885                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
 886                                 " on socket %d\n", socketid);
 887
 888         /* populate the LPM table */
 889         for (i = 0; i < L3FWD_NUM_ROUTES; i++) {
 890                 ret = rte_lpm_add(l3fwd_lookup_struct[socketid],
 891                         l3fwd_route_array[i].ip,
 892                         l3fwd_route_array[i].depth,
 893                         l3fwd_route_array[i].if_out);
 894
 895                 if (ret < 0) {
 896                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
 897                                 "l3fwd LPM table on socket %d\n",
 898                                 i, socketid);
 899                 }
 900
 901                 printf("LPM: Adding route 0x%08x / %d (%d)\n",
 902                         (unsigned)l3fwd_route_array[i].ip,
 903                         l3fwd_route_array[i].depth,
 904                         l3fwd_route_array[i].if_out);
 905         }
 906 }
 907 #endif
 908
 909 static int
 910 init_mem(unsigned nb_mbuf)
 911 {
 912         struct lcore_conf *qconf;
 913         int socketid;
 914         unsigned lcore_id;
 915         char s[64];
 916
 917         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
 918                 if (rte_lcore_is_enabled(lcore_id) == 0)
 919                         continue;
 920
 921                 if (numa_on)
 922                         socketid = rte_lcore_to_socket_id(lcore_id);
 923                 else
 924                         socketid = 0;
 925
 926                 if (socketid >= NB_SOCKETS) {
 927                         rte_exit(EXIT_FAILURE, "Socket %d of lcore %u is out of range %d\n",
 928                                 socketid, lcore_id, NB_SOCKETS);
 929                 }
 930                 if (pktmbuf_pool[socketid] == NULL) {
 931                         snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
 932                         pktmbuf_pool[socketid] = rte_pktmbuf_pool_create(s,
 933                                 nb_mbuf, MEMPOOL_CACHE_SIZE, 0,
 934                                 RTE_MBUF_DEFAULT_BUF_SIZE, socketid);
 935                         if (pktmbuf_pool[socketid] == NULL)
 936                                 rte_exit(EXIT_FAILURE, "Cannot init mbuf pool on socket %d\n", socketid);
 937                         else
 938                                 printf("Allocated mbuf pool on socket %d\n", socketid);
 939
 940 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 941                         setup_lpm(socketid);
 942 #else
 943                         setup_hash(socketid);
 944 #endif
 945                 }
 946                 qconf = &lcore_conf[lcore_id];
 947                 qconf->lookup_struct = l3fwd_lookup_struct[socketid];
 948         }
 949         return 0;
 950 }
 951
 952 int
 953 main(int argc, char **argv)
 954 {
 955         struct lcore_conf *qconf;
 956         struct rte_eth_dev_info dev_info;
 957         struct rte_eth_txconf *txconf;
 958         int ret;
 959         unsigned nb_ports;
 960         uint16_t queueid, portid;
 961         unsigned lcore_id;
 962         uint32_t nb_lcores;
 963         uint16_t n_tx_queue;
 964         uint8_t nb_rx_queue, queue, socketid;
 965
 966         signal(SIGINT, signal_handler);
 967         /* init EAL */
 968         ret = rte_eal_init(argc, argv);
 969         if (ret < 0)
 970                 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
 971         argc -= ret;
 972         argv += ret;
 973
 974         /* parse application arguments (after the EAL ones) */
 975         ret = parse_args(argc, argv);
 976         if (ret < 0)
 977                 rte_exit(EXIT_FAILURE, "Invalid L3FWD-VF parameters\n");
 978
 979         if (check_lcore_params() < 0)
 980                 rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
 981
 982         ret = init_lcore_rx_queues();
 983         if (ret < 0)
 984                 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
 985
 986         nb_ports = rte_eth_dev_count();
 987
 988         if (check_port_config(nb_ports) < 0)
 989                 rte_exit(EXIT_FAILURE, "check_port_config failed\n");
 990
 991         nb_lcores = rte_lcore_count();
 992
 993         /* initialize all ports */
 994         for (portid = 0; portid < nb_ports; portid++) {
 995                 /* skip ports that are not enabled */
 996                 if ((enabled_port_mask & (1 << portid)) == 0) {
 997                         printf("\nSkipping disabled port %d\n", portid);
 998                         continue;
 999                 }
1000
1001                 /* init port */
1002                 printf("Initializing port %d ... ", portid );
1003                 fflush(stdout);
1004
1005                 /* must always equal(=1) */
1006                 nb_rx_queue = get_port_n_rx_queues(portid);
1007                 n_tx_queue = MAX_TX_QUEUE_PER_PORT;
1008
1009                 printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
1010                         nb_rx_queue, (unsigned)1 );
1011                 ret = rte_eth_dev_configure(portid, nb_rx_queue, n_tx_queue, &port_conf);
1012                 if (ret < 0)
1013                         rte_exit(EXIT_FAILURE, "Cannot configure device: err=%d, port=%d\n",
1014                                 ret, portid);
1015
1016                 ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
1017                                                        &nb_txd);
1018                 if (ret < 0)
1019                         rte_exit(EXIT_FAILURE,
1020                                  "Cannot adjust number of descriptors: err=%d, port=%d\n",
1021                                  ret, portid);
1022
1023                 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
1024                 print_ethaddr(" Address:", &ports_eth_addr[portid]);
1025                 printf(", ");
1026
1027                 ret = init_mem(NB_MBUF);
1028                 if (ret < 0)
1029                         rte_exit(EXIT_FAILURE, "init_mem failed\n");
1030
1031                 /* init one TX queue */
1032                 socketid = (uint8_t)rte_lcore_to_socket_id(rte_get_master_lcore());
1033
1034                 printf("txq=%d,%d,%d ", portid, 0, socketid);
1035                 fflush(stdout);
1036
1037                 rte_eth_dev_info_get(portid, &dev_info);
1038                 txconf = &dev_info.default_txconf;
1039                 if (port_conf.rxmode.jumbo_frame)
1040                         txconf->txq_flags = 0;
1041                 ret = rte_eth_tx_queue_setup(portid, 0, nb_txd,
1042                                                  socketid, txconf);
1043                 if (ret < 0)
1044                         rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: err=%d, "
1045                                 "port=%d\n", ret, portid);
1046
1047                 printf("\n");
1048         }
1049
1050         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1051                 if (rte_lcore_is_enabled(lcore_id) == 0)
1052                         continue;
1053                 qconf = &lcore_conf[lcore_id];
1054                 qconf->tx_queue_id = 0;
1055
1056                 printf("\nInitializing rx queues on lcore %u ... ", lcore_id );
1057                 fflush(stdout);
1058                 /* init RX queues */
1059                 for(queue = 0; queue < qconf->n_rx_queue; ++queue) {
1060                         portid = qconf->rx_queue_list[queue].port_id;
1061                         queueid = qconf->rx_queue_list[queue].queue_id;
1062
1063                         if (numa_on)
1064                                 socketid = (uint8_t)rte_lcore_to_socket_id(lcore_id);
1065                         else
1066                                 socketid = 0;
1067
1068                         printf("rxq=%d,%d,%d ", portid, queueid, socketid);
1069                         fflush(stdout);
1070
1071                         ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
1072                                                 socketid, NULL,
1073                                                 pktmbuf_pool[socketid]);
1074                         if (ret < 0)
1075                                 rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: err=%d,"
1076                                                 "port=%d\n", ret, portid);
1077                 }
1078         }
1079         printf("\n");
1080
1081         /* start ports */
1082         for (portid = 0; portid < nb_ports; portid++) {
1083                 if ((enabled_port_mask & (1 << portid)) == 0) {
1084                         continue;
1085                 }
1086                 /* Start device */
1087                 ret = rte_eth_dev_start(portid);
1088                 if (ret < 0)
1089                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%d\n",
1090                                 ret, portid);
1091
1092                 printf("done: Port %d\n", portid);
1093
1094         }
1095
1096         /* launch per-lcore init on every lcore */
1097         rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
1098         RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1099                 if (rte_eal_wait_lcore(lcore_id) < 0)
1100                         return -1;
1101         }
1102
1103         return 0;
1104 }