examples/l3fwd/main.c

   1 /*-
   2  *   BSD LICENSE
   3  *
   4  *   Copyright(c) 2010-2012 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  *  version: DPDK.L.1.2.3-3
  34  */
  35
  36 #include <stdio.h>
  37 #include <stdlib.h>
  38 #include <stdint.h>
  39 #include <inttypes.h>
  40 #include <sys/types.h>
  41 #include <string.h>
  42 #include <sys/queue.h>
  43 #include <stdarg.h>
  44 #include <errno.h>
  45 #include <getopt.h>
  46
  47 #include <rte_common.h>
  48 #include <rte_byteorder.h>
  49 #include <rte_log.h>
  50 #include <rte_memory.h>
  51 #include <rte_memcpy.h>
  52 #include <rte_memzone.h>
  53 #include <rte_tailq.h>
  54 #include <rte_eal.h>
  55 #include <rte_per_lcore.h>
  56 #include <rte_launch.h>
  57 #include <rte_atomic.h>
  58 #include <rte_cycles.h>
  59 #include <rte_prefetch.h>
  60 #include <rte_lcore.h>
  61 #include <rte_per_lcore.h>
  62 #include <rte_branch_prediction.h>
  63 #include <rte_interrupts.h>
  64 #include <rte_pci.h>
  65 #include <rte_random.h>
  66 #include <rte_debug.h>
  67 #include <rte_ether.h>
  68 #include <rte_ethdev.h>
  69 #include <rte_ring.h>
  70 #include <rte_mempool.h>
  71 #include <rte_mbuf.h>
  72 #include <rte_ip.h>
  73 #include <rte_tcp.h>
  74 #include <rte_udp.h>
  75 #include <rte_string_fns.h>
  76
  77 #include "main.h"
  78
  79 #define APP_LOOKUP_EXACT_MATCH          0
  80 #define APP_LOOKUP_LPM                  1
  81 #define DO_RFC_1812_CHECKS
  82
  83 //#define APP_LOOKUP_METHOD             APP_LOOKUP_EXACT_MATCH
  84 #ifndef APP_LOOKUP_METHOD
  85 #define APP_LOOKUP_METHOD             APP_LOOKUP_LPM
  86 #endif
  87
  88 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
  89 #include <rte_hash.h>
  90 #include <rte_hash_crc.h>
  91 #include <rte_jhash.h>
  92 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
  93 #include <rte_lpm.h>
  94 #else
  95 #error "APP_LOOKUP_METHOD set to incorrect value"
  96 #endif
  97
  98 #define RTE_LOGTYPE_L3FWD RTE_LOGTYPE_USER1
  99
 100 #define MAX_PORTS 32
 101
 102 #define MBUF_SIZE (2048 + sizeof(struct rte_mbuf) + RTE_PKTMBUF_HEADROOM)
 103 #define NB_MBUF   8192
 104
 105 /*
 106  * RX and TX Prefetch, Host, and Write-back threshold values should be
 107  * carefully set for optimal performance. Consult the network
 108  * controller's datasheet and supporting DPDK documentation for guidance
 109  * on how these parameters should be set.
 110  */
 111 #define RX_PTHRESH 8 /**< Default values of RX prefetch threshold reg. */
 112 #define RX_HTHRESH 8 /**< Default values of RX host threshold reg. */
 113 #define RX_WTHRESH 4 /**< Default values of RX write-back threshold reg. */
 114
 115 /*
 116  * These default values are optimized for use with the Intel(R) 82599 10 GbE
 117  * Controller and the DPDK ixgbe PMD. Consider using other values for other
 118  * network controllers and/or network drivers.
 119  */
 120 #define TX_PTHRESH 36 /**< Default values of TX prefetch threshold reg. */
 121 #define TX_HTHRESH 0  /**< Default values of TX host threshold reg. */
 122 #define TX_WTHRESH 0  /**< Default values of TX write-back threshold reg. */
 123
 124 #define MAX_PKT_BURST 32
 125 #define BURST_TX_DRAIN 200000ULL /* around 100us at 2 Ghz */
 126
 127 #define NB_SOCKETS 8
 128
 129 #define SOCKET0 0
 130
 131 /* Configure how many packets ahead to prefetch, when reading packets */
 132 #define PREFETCH_OFFSET 3
 133
 134 /*
 135  * Configurable number of RX/TX ring descriptors
 136  */
 137 #define RTE_TEST_RX_DESC_DEFAULT 128
 138 #define RTE_TEST_TX_DESC_DEFAULT 512
 139 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
 140 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
 141
 142 /* ethernet addresses of ports */
 143 static struct ether_addr ports_eth_addr[MAX_PORTS];
 144
 145 /* mask of enabled ports */
 146 static uint32_t enabled_port_mask = 0;
 147 static int promiscuous_on = 0; /**< Ports set in promiscuous mode off by default. */
 148 static int numa_on = 1; /**< NUMA is enabled by default. */
 149
 150 struct mbuf_table {
 151         uint16_t len;
 152         struct rte_mbuf *m_table[MAX_PKT_BURST];
 153 };
 154
 155 struct lcore_rx_queue {
 156         uint8_t port_id;
 157         uint8_t queue_id;
 158 } __rte_cache_aligned;
 159
 160 #define MAX_RX_QUEUE_PER_LCORE 16
 161 #define MAX_TX_QUEUE_PER_PORT MAX_PORTS
 162 #define MAX_RX_QUEUE_PER_PORT 128
 163
 164 #define MAX_LCORE_PARAMS 1024
 165 struct lcore_params {
 166         uint8_t port_id;
 167         uint8_t queue_id;
 168         uint8_t lcore_id;
 169 } __rte_cache_aligned;
 170
 171 static struct lcore_params lcore_params_array[MAX_LCORE_PARAMS];
 172 static struct lcore_params lcore_params_array_default[] = {
 173         {0, 0, 2},
 174         {0, 1, 2},
 175         {0, 2, 2},
 176         {1, 0, 2},
 177         {1, 1, 2},
 178         {1, 2, 2},
 179         {2, 0, 2},
 180         {3, 0, 3},
 181         {3, 1, 3},
 182 };
 183
 184 static struct lcore_params * lcore_params = lcore_params_array_default;
 185 static uint16_t nb_lcore_params = sizeof(lcore_params_array_default) /
 186                                 sizeof(lcore_params_array_default[0]);
 187
 188 static struct rte_eth_conf port_conf = {
 189         .rxmode = {
 190                 .split_hdr_size = 0,
 191                 .header_split   = 0, /**< Header Split disabled */
 192                 .hw_ip_checksum = 1, /**< IP checksum offload enabled */
 193                 .hw_vlan_filter = 0, /**< VLAN filtering disabled */
 194                 .jumbo_frame    = 0, /**< Jumbo Frame Support disabled */
 195                 .hw_strip_crc   = 0, /**< CRC stripped by hardware */
 196         },
 197         .rx_adv_conf = {
 198                 .rss_conf = {
 199                         .rss_key = NULL,
 200                         .rss_hf = ETH_RSS_IPV4,
 201                 },
 202         },
 203         .txmode = {
 204         },
 205 };
 206
 207 static const struct rte_eth_rxconf rx_conf = {
 208         .rx_thresh = {
 209                 .pthresh = RX_PTHRESH,
 210                 .hthresh = RX_HTHRESH,
 211                 .wthresh = RX_WTHRESH,
 212         },
 213 };
 214
 215 static const struct rte_eth_txconf tx_conf = {
 216         .tx_thresh = {
 217                 .pthresh = TX_PTHRESH,
 218                 .hthresh = TX_HTHRESH,
 219                 .wthresh = TX_WTHRESH,
 220         },
 221         .tx_free_thresh = 0, /* Use PMD default values */
 222         .tx_rs_thresh = 0, /* Use PMD default values */
 223 };
 224
 225 static struct rte_mempool * pktmbuf_pool[NB_SOCKETS];
 226
 227
 228 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 229 struct ipv4_5tuple {
 230         uint32_t ip_dst;
 231         uint32_t ip_src;
 232         uint16_t port_dst;
 233         uint16_t port_src;
 234         uint8_t proto;
 235 } __attribute__((__packed__));
 236
 237 struct l3fwd_route {
 238         struct ipv4_5tuple key;
 239         uint8_t if_out;
 240 };
 241
 242 static struct l3fwd_route l3fwd_route_array[] = {
 243         {{IPv4(100,10,0,1), IPv4(200,10,0,1), 101, 11, IPPROTO_TCP}, 0},
 244         {{IPv4(100,20,0,2), IPv4(200,20,0,2), 102, 12, IPPROTO_TCP}, 1},
 245         {{IPv4(100,30,0,3), IPv4(200,30,0,3), 103, 13, IPPROTO_TCP}, 2},
 246         {{IPv4(100,40,0,4), IPv4(200,40,0,4), 104, 14, IPPROTO_TCP}, 3},
 247 };
 248
 249 typedef struct rte_hash lookup_struct_t;
 250 static lookup_struct_t *l3fwd_lookup_struct[NB_SOCKETS];
 251
 252 #define L3FWD_HASH_ENTRIES      1024
 253 struct rte_hash_parameters l3fwd_hash_params = {
 254         .name = "l3fwd_hash_0",
 255         .entries = L3FWD_HASH_ENTRIES,
 256         .bucket_entries = 4,
 257         .key_len = sizeof(struct ipv4_5tuple),
 258         .hash_func = rte_hash_crc,
 259         .hash_func_init_val = 0,
 260         .socket_id = SOCKET0,
 261 };
 262
 263 #define L3FWD_NUM_ROUTES \
 264         (sizeof(l3fwd_route_array) / sizeof(l3fwd_route_array[0]))
 265
 266 static uint8_t l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 267 #endif
 268
 269 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 270 struct l3fwd_route {
 271         uint32_t ip;
 272         uint8_t  depth;
 273         uint8_t  if_out;
 274 };
 275
 276 static struct l3fwd_route l3fwd_route_array[] = {
 277         {IPv4(1,1,1,0), 24, 0},
 278         {IPv4(2,1,1,0), 24, 1},
 279         {IPv4(3,1,1,0), 24, 2},
 280         {IPv4(4,1,1,0), 24, 3},
 281         {IPv4(5,1,1,0), 24, 4},
 282         {IPv4(6,1,1,0), 24, 5},
 283         {IPv4(7,1,1,0), 24, 6},
 284         {IPv4(8,1,1,0), 24, 7},
 285 };
 286
 287 #define L3FWD_NUM_ROUTES \
 288         (sizeof(l3fwd_route_array) / sizeof(l3fwd_route_array[0]))
 289
 290 #define L3FWD_LPM_MAX_RULES     1024
 291
 292 typedef struct rte_lpm lookup_struct_t;
 293 static lookup_struct_t *l3fwd_lookup_struct[NB_SOCKETS];
 294 #endif
 295
 296 struct lcore_conf {
 297         uint16_t n_rx_queue;
 298         struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
 299         uint16_t tx_queue_id[MAX_PORTS];
 300         struct mbuf_table tx_mbufs[MAX_PORTS];
 301         lookup_struct_t * lookup_struct;
 302 } __rte_cache_aligned;
 303
 304 static struct lcore_conf lcore_conf[RTE_MAX_LCORE];
 305
 306 /* Send burst of packets on an output interface */
 307 static inline int
 308 send_burst(struct lcore_conf *qconf, uint16_t n, uint8_t port)
 309 {
 310         struct rte_mbuf **m_table;
 311         int ret;
 312         uint16_t queueid;
 313
 314         queueid = qconf->tx_queue_id[port];
 315         m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
 316
 317         ret = rte_eth_tx_burst(port, queueid, m_table, n);
 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 = (struct ipv4_hdr *)(rte_pktmbuf_mtod(m, unsigned char *) +
 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.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 >= MAX_PORTS || (enabled_port_mask & 1 << dst_port) == 0)
 474                 dst_port = portid;
 475
 476         /* 00:09:c0:00:00:xx */
 477         tmp = &eth_hdr->d_addr.addr_bytes[0];
 478         *((uint64_t *)tmp) = 0x000000c00900 + (dst_port << 24);
 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 __attribute__((noreturn)) 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 = 0;
 500         uint64_t diff_tsc, cur_tsc;
 501         int i, j, nb_rx;
 502         uint8_t portid, queueid;
 503         struct lcore_conf *qconf;
 504
 505         lcore_id = rte_lcore_id();
 506         qconf = &lcore_conf[lcore_id];
 507
 508         if (qconf->n_rx_queue == 0) {
 509                 RTE_LOG(INFO, L3FWD, "lcore %u has nothing to do\n", lcore_id);
 510                 while(1);
 511         }
 512
 513         RTE_LOG(INFO, L3FWD, "entering main loop on lcore %u\n", lcore_id);
 514
 515         for (i = 0; i < qconf->n_rx_queue; i++) {
 516
 517                 portid = qconf->rx_queue_list[i].port_id;
 518                 queueid = qconf->rx_queue_list[i].queue_id;
 519                 RTE_LOG(INFO, L3FWD, " -- lcoreid=%u portid=%hhu rxqueueid=%hhu\n", lcore_id,
 520                         portid, queueid);
 521         }
 522
 523         while (1) {
 524
 525                 cur_tsc = rte_rdtsc();
 526
 527                 /*
 528                  * TX burst queue drain
 529                  */
 530                 diff_tsc = cur_tsc - prev_tsc;
 531                 if (unlikely(diff_tsc > BURST_TX_DRAIN)) {
 532
 533                         /*
 534                          * This could be optimized (use queueid instead of
 535                          * portid), but it is not called so often
 536                          */
 537                         for (portid = 0; portid < MAX_PORTS; portid++) {
 538                                 if (qconf->tx_mbufs[portid].len == 0)
 539                                         continue;
 540                                 send_burst(&lcore_conf[lcore_id],
 541                                         qconf->tx_mbufs[portid].len,
 542                                         portid);
 543                                 qconf->tx_mbufs[portid].len = 0;
 544                         }
 545
 546                         prev_tsc = cur_tsc;
 547                 }
 548
 549                 /*
 550                  * Read packet from RX queues
 551                  */
 552                 for (i = 0; i < qconf->n_rx_queue; ++i) {
 553
 554                         portid = qconf->rx_queue_list[i].port_id;
 555                         queueid = qconf->rx_queue_list[i].queue_id;
 556                         nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst, MAX_PKT_BURST);
 557
 558                         /* Prefetch first packets */
 559                         for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
 560                                 rte_prefetch0(rte_pktmbuf_mtod(
 561                                                 pkts_burst[j], void *));
 562                         }
 563
 564                         /* Prefetch and forward already prefetched packets */
 565                         for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
 566                                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
 567                                                 j + PREFETCH_OFFSET], void *));
 568                                 l3fwd_simple_forward(pkts_burst[j], portid, qconf->lookup_struct);
 569                         }
 570
 571                         /* Forward remaining prefetched packets */
 572                         for (; j < nb_rx; j++) {
 573                                 l3fwd_simple_forward(pkts_burst[j], portid, qconf->lookup_struct);
 574                         }
 575                 }
 576         }
 577 }
 578
 579 static int
 580 check_lcore_params(void)
 581 {
 582         uint8_t queue, lcore;
 583         uint16_t i;
 584         int socketid;
 585
 586         for (i = 0; i < nb_lcore_params; ++i) {
 587                 queue = lcore_params[i].queue_id;
 588                 if (queue >= MAX_RX_QUEUE_PER_PORT) {
 589                         printf("invalid queue number: %hhu\n", queue);
 590                         return -1;
 591                 }
 592                 lcore = lcore_params[i].lcore_id;
 593                 if (!rte_lcore_is_enabled(lcore)) {
 594                         printf("error: lcore %hhu is not enabled in lcore mask\n", lcore);
 595                         return -1;
 596                 }
 597                 if ((socketid = rte_lcore_to_socket_id(lcore) != 0) &&
 598                         (numa_on == 0)) {
 599                         printf("warning: lcore %hhu is on socket %d with numa off \n",
 600                                 lcore, socketid);
 601                 }
 602         }
 603         return 0;
 604 }
 605
 606 static int
 607 check_port_config(const unsigned nb_ports)
 608 {
 609         unsigned portid;
 610         uint16_t i;
 611
 612         for (i = 0; i < nb_lcore_params; ++i) {
 613                 portid = lcore_params[i].port_id;
 614                 if ((enabled_port_mask & (1 << portid)) == 0) {
 615                         printf("port %u is not enabled in port mask\n", portid);
 616                         return -1;
 617                 }
 618                 if (portid >= nb_ports) {
 619                         printf("port %u is not present on the board\n", portid);
 620                         return -1;
 621                 }
 622         }
 623         return 0;
 624 }
 625
 626 static uint8_t
 627 get_port_n_rx_queues(const uint8_t port)
 628 {
 629         int queue = -1;
 630         uint16_t i;
 631
 632         for (i = 0; i < nb_lcore_params; ++i) {
 633                 if (lcore_params[i].port_id == port && lcore_params[i].queue_id > queue)
 634                         queue = lcore_params[i].queue_id;
 635         }
 636         return (uint8_t)(++queue);
 637 }
 638
 639 static int
 640 init_lcore_rx_queues(void)
 641 {
 642         uint16_t i, nb_rx_queue;
 643         uint8_t lcore;
 644
 645         for (i = 0; i < nb_lcore_params; ++i) {
 646                 lcore = lcore_params[i].lcore_id;
 647                 nb_rx_queue = lcore_conf[lcore].n_rx_queue;
 648                 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
 649                         printf("error: too many queues (%u) for lcore: %u\n",
 650                                 (unsigned)nb_rx_queue + 1, (unsigned)lcore);
 651                         return -1;
 652                 } else {
 653                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
 654                                 lcore_params[i].port_id;
 655                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
 656                                 lcore_params[i].queue_id;
 657                         lcore_conf[lcore].n_rx_queue++;
 658                 }
 659         }
 660         return 0;
 661 }
 662
 663 /* display usage */
 664 static void
 665 print_usage(const char *prgname)
 666 {
 667         printf ("%s [EAL options] -- -p PORTMASK -P"
 668                 "  [--config (port,queue,lcore)[,(port,queue,lcore]]\n"
 669                 "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
 670                 "  --config (port,queue,lcore): rx queues configuration\n"
 671                 "  --no-numa: optional, disable numa awareness\n",
 672                 prgname);
 673 }
 674
 675 static int
 676 parse_portmask(const char *portmask)
 677 {
 678         char *end = NULL;
 679         unsigned long pm;
 680
 681         /* parse hexadecimal string */
 682         pm = strtoul(portmask, &end, 16);
 683         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
 684                 return -1;
 685
 686         if (pm == 0)
 687                 return -1;
 688
 689         return pm;
 690 }
 691
 692 static int
 693 parse_config(const char *q_arg)
 694 {
 695         char s[256];
 696         const char *p, *p0 = q_arg;
 697         char *end;
 698         enum fieldnames {
 699                 FLD_PORT = 0,
 700                 FLD_QUEUE,
 701                 FLD_LCORE,
 702                 _NUM_FLD
 703         };
 704         unsigned long int_fld[_NUM_FLD];
 705         char *str_fld[_NUM_FLD];
 706         int i;
 707         unsigned size;
 708
 709         nb_lcore_params = 0;
 710
 711         while ((p = strchr(p0,'(')) != NULL) {
 712                 ++p;
 713                 if((p0 = strchr(p,')')) == NULL)
 714                         return -1;
 715
 716                 size = p0 - p;
 717                 if(size >= sizeof(s))
 718                         return -1;
 719
 720                 rte_snprintf(s, sizeof(s), "%.*s", size, p);
 721                 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') != _NUM_FLD)
 722                         return -1;
 723                 for (i = 0; i < _NUM_FLD; i++){
 724                         errno = 0;
 725                         int_fld[i] = strtoul(str_fld[i], &end, 0);
 726                         if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
 727                                 return -1;
 728                 }
 729                 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
 730                         printf("exceeded max number of lcore params: %hu\n",
 731                                 nb_lcore_params);
 732                         return -1;
 733                 }
 734                 lcore_params_array[nb_lcore_params].port_id = (uint8_t)int_fld[FLD_PORT];
 735                 lcore_params_array[nb_lcore_params].queue_id = (uint8_t)int_fld[FLD_QUEUE];
 736                 lcore_params_array[nb_lcore_params].lcore_id = (uint8_t)int_fld[FLD_LCORE];
 737                 ++nb_lcore_params;
 738         }
 739         lcore_params = lcore_params_array;
 740         return 0;
 741 }
 742
 743 /* Parse the argument given in the command line of the application */
 744 static int
 745 parse_args(int argc, char **argv)
 746 {
 747         int opt, ret;
 748         char **argvopt;
 749         int option_index;
 750         char *prgname = argv[0];
 751         static struct option lgopts[] = {
 752                 {"config", 1, 0, 0},
 753                 {"no-numa", 0, 0, 0},
 754                 {NULL, 0, 0, 0}
 755         };
 756
 757         argvopt = argv;
 758
 759         while ((opt = getopt_long(argc, argvopt, "p:P",
 760                                 lgopts, &option_index)) != EOF) {
 761
 762                 switch (opt) {
 763                 /* portmask */
 764                 case 'p':
 765                         enabled_port_mask = parse_portmask(optarg);
 766                         if (enabled_port_mask == 0) {
 767                                 printf("invalid portmask\n");
 768                                 print_usage(prgname);
 769                                 return -1;
 770                         }
 771                         break;
 772                 case 'P':
 773                         printf("Promiscuous mode selected\n");
 774                         promiscuous_on = 1;
 775                         break;
 776
 777                 /* long options */
 778                 case 0:
 779                         if (!strcmp(lgopts[option_index].name, "config")) {
 780                                 ret = parse_config(optarg);
 781                                 if (ret) {
 782                                         printf("invalid config\n");
 783                                         print_usage(prgname);
 784                                         return -1;
 785                                 }
 786                         }
 787
 788                         if (!strcmp(lgopts[option_index].name, "no-numa")) {
 789                                 printf("numa is disabled \n");
 790                                 numa_on = 0;
 791                         }
 792                         break;
 793
 794                 default:
 795                         print_usage(prgname);
 796                         return -1;
 797                 }
 798         }
 799
 800         if (optind >= 0)
 801                 argv[optind-1] = prgname;
 802
 803         ret = optind-1;
 804         optind = 0; /* reset getopt lib */
 805         return ret;
 806 }
 807
 808 static void
 809 print_ethaddr(const char *name, const struct ether_addr *eth_addr)
 810 {
 811         printf ("%s%02X:%02X:%02X:%02X:%02X:%02X", name,
 812                 eth_addr->addr_bytes[0],
 813                 eth_addr->addr_bytes[1],
 814                 eth_addr->addr_bytes[2],
 815                 eth_addr->addr_bytes[3],
 816                 eth_addr->addr_bytes[4],
 817                 eth_addr->addr_bytes[5]);
 818 }
 819
 820 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 821 static void
 822 setup_hash(int socketid)
 823 {
 824         unsigned i;
 825         int ret;
 826         char s[64];
 827
 828         /* create  hashes */
 829         rte_snprintf(s, sizeof(s), "l3fwd_hash_%d", socketid);
 830         l3fwd_hash_params.name = s;
 831         l3fwd_hash_params.socket_id = socketid;
 832         l3fwd_lookup_struct[socketid] = rte_hash_create(&l3fwd_hash_params);
 833         if (l3fwd_lookup_struct[socketid] == NULL)
 834                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
 835                                 "socket %d\n", socketid);
 836
 837         /* populate the hash */
 838         for (i = 0; i < L3FWD_NUM_ROUTES; i++) {
 839                 ret = rte_hash_add_key (l3fwd_lookup_struct[socketid],
 840                                 (void *) &l3fwd_route_array[i].key);
 841                 if (ret < 0) {
 842                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
 843                                 "l3fwd hash on socket %d\n", i, socketid);
 844                 }
 845                 l3fwd_out_if[ret] = l3fwd_route_array[i].if_out;
 846                 printf("Hash: Adding key\n");
 847                 print_key(l3fwd_route_array[i].key);
 848         }
 849 }
 850 #endif
 851
 852 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 853 static void
 854 setup_lpm(int socketid)
 855 {
 856         unsigned i;
 857         int ret;
 858         char s[64];
 859
 860         /* create the LPM table */
 861         rte_snprintf(s, sizeof(s), "L3FWD_LPM_%d", socketid);
 862         l3fwd_lookup_struct[socketid] = rte_lpm_create(s, socketid,
 863                                 L3FWD_LPM_MAX_RULES, RTE_LPM_MEMZONE);
 864         if (l3fwd_lookup_struct[socketid] == NULL)
 865                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
 866                                 " on socket %d\n", socketid);
 867
 868         /* populate the LPM table */
 869         for (i = 0; i < L3FWD_NUM_ROUTES; i++) {
 870                 ret = rte_lpm_add(l3fwd_lookup_struct[socketid],
 871                         l3fwd_route_array[i].ip,
 872                         l3fwd_route_array[i].depth,
 873                         l3fwd_route_array[i].if_out);
 874
 875                 if (ret < 0) {
 876                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
 877                                 "l3fwd LPM table on socket %d\n",
 878                                 i, socketid);
 879                 }
 880
 881                 printf("LPM: Adding route 0x%08x / %d (%d)\n",
 882                         (unsigned)l3fwd_route_array[i].ip,
 883                         l3fwd_route_array[i].depth,
 884                         l3fwd_route_array[i].if_out);
 885         }
 886 }
 887 #endif
 888
 889 static int
 890 init_mem(void)
 891 {
 892         struct lcore_conf *qconf;
 893         int socketid;
 894         unsigned lcore_id;
 895         char s[64];
 896
 897         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
 898                 if (rte_lcore_is_enabled(lcore_id) == 0)
 899                         continue;
 900
 901                 if (numa_on)
 902                         socketid = rte_lcore_to_socket_id(lcore_id);
 903                 else
 904                         socketid = 0;
 905
 906                 if (socketid >= NB_SOCKETS) {
 907                         rte_exit(EXIT_FAILURE, "Socket %d of lcore %u is out of range %d\n",
 908                                 socketid, lcore_id, NB_SOCKETS);
 909                 }
 910                 if (pktmbuf_pool[socketid] == NULL) {
 911                         rte_snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
 912                         pktmbuf_pool[socketid] =
 913                                 rte_mempool_create(s, NB_MBUF, MBUF_SIZE, 32,
 914                                         sizeof(struct rte_pktmbuf_pool_private),
 915                                         rte_pktmbuf_pool_init, NULL,
 916                                         rte_pktmbuf_init, NULL,
 917                                         socketid, 0);
 918                         if (pktmbuf_pool[socketid] == NULL)
 919                                 rte_exit(EXIT_FAILURE,
 920                                                 "Cannot init mbuf pool on socket %d\n", socketid);
 921                         else
 922                                 printf("Allocated mbuf pool on socket %d\n", socketid);
 923
 924 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 925                         setup_lpm(socketid);
 926 #else
 927                         setup_hash(socketid);
 928 #endif
 929                 }
 930                 qconf = &lcore_conf[lcore_id];
 931                 qconf->lookup_struct = l3fwd_lookup_struct[socketid];
 932         }
 933         return 0;
 934 }
 935
 936 int
 937 MAIN(int argc, char **argv)
 938 {
 939         struct lcore_conf *qconf;
 940         struct rte_eth_link link;
 941         int ret;
 942         unsigned nb_ports;
 943         uint16_t queueid;
 944         unsigned lcore_id;
 945         uint32_t n_tx_queue, nb_lcores;
 946         uint8_t portid, nb_rx_queue, queue, socketid;
 947
 948         /* init EAL */
 949         ret = rte_eal_init(argc, argv);
 950         if (ret < 0)
 951                 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
 952         argc -= ret;
 953         argv += ret;
 954
 955         /* parse application arguments (after the EAL ones) */
 956         ret = parse_args(argc, argv);
 957         if (ret < 0)
 958                 rte_exit(EXIT_FAILURE, "Invalid L3FWD parameters\n");
 959
 960         if (check_lcore_params() < 0)
 961                 rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
 962
 963         ret = init_lcore_rx_queues();
 964         if (ret < 0)
 965                 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
 966
 967         ret = init_mem();
 968         if (ret < 0)
 969                 rte_exit(EXIT_FAILURE, "init_mem failed\n");
 970
 971         /* init driver */
 972 #ifdef RTE_LIBRTE_IGB_PMD
 973         if (rte_igb_pmd_init() < 0)
 974                 rte_exit(EXIT_FAILURE, "Cannot init igb pmd\n");
 975 #endif
 976 #ifdef RTE_LIBRTE_IXGBE_PMD
 977         if (rte_ixgbe_pmd_init() < 0)
 978                 rte_exit(EXIT_FAILURE, "Cannot init ixgbe pmd\n");
 979 #endif
 980
 981         if (rte_eal_pci_probe() < 0)
 982                 rte_exit(EXIT_FAILURE, "Cannot probe PCI\n");
 983
 984         nb_ports = rte_eth_dev_count();
 985         if (nb_ports > MAX_PORTS)
 986                 nb_ports = MAX_PORTS;
 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                 nb_rx_queue = get_port_n_rx_queues(portid);
1006                 n_tx_queue = nb_lcores;
1007                 if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
1008                         n_tx_queue = MAX_TX_QUEUE_PER_PORT;
1009                 printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
1010                         nb_rx_queue, (unsigned)n_tx_queue );
1011                 ret = rte_eth_dev_configure(portid, nb_rx_queue,
1012                                         (uint16_t)n_tx_queue, &port_conf);
1013                 if (ret < 0)
1014                         rte_exit(EXIT_FAILURE, "Cannot configure device: err=%d, port=%d\n",
1015                                 ret, portid);
1016
1017                 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
1018                 print_ethaddr(" Address:", &ports_eth_addr[portid]);
1019                 printf(", ");
1020
1021
1022                 /* init one TX queue per couple (lcore,port) */
1023                 queueid = 0;
1024                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1025                         if (rte_lcore_is_enabled(lcore_id) == 0)
1026                                 continue;
1027
1028                         if (numa_on)
1029                                 socketid = (uint8_t)rte_lcore_to_socket_id(lcore_id);
1030                         else
1031                                 socketid = 0;
1032
1033                         printf("txq=%u,%d,%d ", lcore_id, queueid, socketid);
1034                         fflush(stdout);
1035                         ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
1036                                                      socketid, &tx_conf);
1037                         if (ret < 0)
1038                                 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: err=%d, "
1039                                         "port=%d\n", ret, portid);
1040
1041                         qconf = &lcore_conf[lcore_id];
1042                         qconf->tx_queue_id[portid] = queueid;
1043                         queueid++;
1044                 }
1045                 printf("\n");
1046         }
1047
1048         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1049                 if (rte_lcore_is_enabled(lcore_id) == 0)
1050                         continue;
1051                 qconf = &lcore_conf[lcore_id];
1052                 printf("\nInitializing rx queues on lcore %u ... ", lcore_id );
1053                 fflush(stdout);
1054                 /* init RX queues */
1055                 for(queue = 0; queue < qconf->n_rx_queue; ++queue) {
1056                         portid = qconf->rx_queue_list[queue].port_id;
1057                         queueid = qconf->rx_queue_list[queue].queue_id;
1058
1059                         if (numa_on)
1060                                 socketid = (uint8_t)rte_lcore_to_socket_id(lcore_id);
1061                         else
1062                                 socketid = 0;
1063
1064                         printf("rxq=%d,%d,%d ", portid, queueid, socketid);
1065                         fflush(stdout);
1066
1067                         ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
1068                                         socketid, &rx_conf, pktmbuf_pool[socketid]);
1069                         if (ret < 0)
1070                                 rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: err=%d,"
1071                                                 "port=%d\n", ret, portid);
1072                 }
1073         }
1074
1075         printf("\n");
1076
1077         /* start ports */
1078         for (portid = 0; portid < nb_ports; portid++) {
1079                 if ((enabled_port_mask & (1 << portid)) == 0) {
1080                         continue;
1081                 }
1082                 /* Start device */
1083                 ret = rte_eth_dev_start(portid);
1084                 if (ret < 0)
1085                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%d\n",
1086                                 ret, portid);
1087
1088                 printf("done: Port %d ", portid);
1089
1090                 /* get link status */
1091                 rte_eth_link_get(portid, &link);
1092                 if (link.link_status) {
1093                         printf(" Link Up - speed %u Mbps - %s\n",
1094                                (unsigned) link.link_speed,
1095                                (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
1096                                ("full-duplex") : ("half-duplex\n"));
1097                 } else {
1098                         printf(" Link Down\n");
1099                 }
1100                 /*
1101                  * If enabled, put device in promiscuous mode.
1102                  * This allows IO forwarding mode to forward packets
1103                  * to itself through 2 cross-connected  ports of the
1104                  * target machine.
1105                  */
1106                 if (promiscuous_on)
1107                         rte_eth_promiscuous_enable(portid);
1108         }
1109
1110         /* launch per-lcore init on every lcore */
1111         rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
1112         RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1113                 if (rte_eal_wait_lcore(lcore_id) < 0)
1114                         return -1;
1115         }
1116
1117         return 0;
1118 }