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  */
  34
  35 #include <stdio.h>
  36 #include <stdlib.h>
  37 #include <stdint.h>
  38 #include <inttypes.h>
  39 #include <sys/types.h>
  40 #include <string.h>
  41 #include <sys/queue.h>
  42 #include <stdarg.h>
  43 #include <errno.h>
  44 #include <getopt.h>
  45
  46 #include <rte_common.h>
  47 #include <rte_byteorder.h>
  48 #include <rte_log.h>
  49 #include <rte_memory.h>
  50 #include <rte_memcpy.h>
  51 #include <rte_memzone.h>
  52 #include <rte_tailq.h>
  53 #include <rte_eal.h>
  54 #include <rte_per_lcore.h>
  55 #include <rte_launch.h>
  56 #include <rte_atomic.h>
  57 #include <rte_cycles.h>
  58 #include <rte_prefetch.h>
  59 #include <rte_lcore.h>
  60 #include <rte_per_lcore.h>
  61 #include <rte_branch_prediction.h>
  62 #include <rte_interrupts.h>
  63 #include <rte_pci.h>
  64 #include <rte_random.h>
  65 #include <rte_debug.h>
  66 #include <rte_ether.h>
  67 #include <rte_ethdev.h>
  68 #include <rte_ring.h>
  69 #include <rte_mempool.h>
  70 #include <rte_mbuf.h>
  71 #include <rte_ip.h>
  72 #include <rte_tcp.h>
  73 #include <rte_udp.h>
  74 #include <rte_string_fns.h>
  75
  76 #include "main.h"
  77
  78 #define APP_LOOKUP_EXACT_MATCH          0
  79 #define APP_LOOKUP_LPM                  1
  80 #define DO_RFC_1812_CHECKS
  81
  82 //#define APP_LOOKUP_METHOD             APP_LOOKUP_EXACT_MATCH
  83 #ifndef APP_LOOKUP_METHOD
  84 #define APP_LOOKUP_METHOD             APP_LOOKUP_LPM
  85 #endif
  86
  87 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
  88 #include <rte_hash.h>
  89 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
  90 #include <rte_lpm.h>
  91 #else
  92 #error "APP_LOOKUP_METHOD set to incorrect value"
  93 #endif
  94
  95 #define RTE_LOGTYPE_L3FWD RTE_LOGTYPE_USER1
  96
  97 #define MAX_PORTS 32
  98
  99 #define MBUF_SIZE (2048 + sizeof(struct rte_mbuf) + RTE_PKTMBUF_HEADROOM)
 100 #define NB_MBUF   8192
 101
 102 /*
 103  * RX and TX Prefetch, Host, and Write-back threshold values should be
 104  * carefully set for optimal performance. Consult the network
 105  * controller's datasheet and supporting DPDK documentation for guidance
 106  * on how these parameters should be set.
 107  */
 108 #define RX_PTHRESH 8 /**< Default values of RX prefetch threshold reg. */
 109 #define RX_HTHRESH 8 /**< Default values of RX host threshold reg. */
 110 #define RX_WTHRESH 4 /**< Default values of RX write-back threshold reg. */
 111
 112 /*
 113  * These default values are optimized for use with the Intel(R) 82599 10 GbE
 114  * Controller and the DPDK ixgbe PMD. Consider using other values for other
 115  * network controllers and/or network drivers.
 116  */
 117 #define TX_PTHRESH 36 /**< Default values of TX prefetch threshold reg. */
 118 #define TX_HTHRESH 0  /**< Default values of TX host threshold reg. */
 119 #define TX_WTHRESH 0  /**< Default values of TX write-back threshold reg. */
 120
 121 #define MAX_PKT_BURST 32
 122 #define BURST_TX_DRAIN 200000ULL /* around 100us at 2 Ghz */
 123
 124 #define NB_SOCKETS 8
 125
 126 #define SOCKET0 0
 127
 128 /* Configure how many packets ahead to prefetch, when reading packets */
 129 #define PREFETCH_OFFSET 3
 130
 131 /*
 132  * Configurable number of RX/TX ring descriptors
 133  */
 134 #define RTE_TEST_RX_DESC_DEFAULT 128
 135 #define RTE_TEST_TX_DESC_DEFAULT 512
 136 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
 137 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
 138
 139 /* ethernet addresses of ports */
 140 static struct ether_addr ports_eth_addr[MAX_PORTS];
 141
 142 /* mask of enabled ports */
 143 static uint32_t enabled_port_mask = 0;
 144 static int promiscuous_on = 0; /**< Ports set in promiscuous mode off by default. */
 145 static int numa_on = 1; /**< NUMA is enabled by default. */
 146
 147 struct mbuf_table {
 148         uint16_t len;
 149         struct rte_mbuf *m_table[MAX_PKT_BURST];
 150 };
 151
 152 struct lcore_rx_queue {
 153         uint8_t port_id;
 154         uint8_t queue_id;
 155 } __rte_cache_aligned;
 156
 157 #define MAX_RX_QUEUE_PER_LCORE 16
 158 #define MAX_TX_QUEUE_PER_PORT MAX_PORTS
 159 #define MAX_RX_QUEUE_PER_PORT 128
 160
 161 #define MAX_LCORE_PARAMS 1024
 162 struct lcore_params {
 163         uint8_t port_id;
 164         uint8_t queue_id;
 165         uint8_t lcore_id;
 166 } __rte_cache_aligned;
 167
 168 static struct lcore_params lcore_params_array[MAX_LCORE_PARAMS];
 169 static struct lcore_params lcore_params_array_default[] = {
 170         {0, 0, 2},
 171         {0, 1, 2},
 172         {0, 2, 2},
 173         {1, 0, 2},
 174         {1, 1, 2},
 175         {1, 2, 2},
 176         {2, 0, 2},
 177         {3, 0, 3},
 178         {3, 1, 3},
 179 };
 180
 181 static struct lcore_params * lcore_params = lcore_params_array_default;
 182 static uint16_t nb_lcore_params = sizeof(lcore_params_array_default) /
 183                                 sizeof(lcore_params_array_default[0]);
 184
 185 static struct rte_eth_conf port_conf = {
 186         .rxmode = {
 187                 .split_hdr_size = 0,
 188                 .header_split   = 0, /**< Header Split disabled */
 189                 .hw_ip_checksum = 1, /**< IP checksum offload enabled */
 190                 .hw_vlan_filter = 0, /**< VLAN filtering disabled */
 191                 .jumbo_frame    = 0, /**< Jumbo Frame Support disabled */
 192                 .hw_strip_crc   = 0, /**< CRC stripped by hardware */
 193         },
 194         .rx_adv_conf = {
 195                 .rss_conf = {
 196                         .rss_key = NULL,
 197                         .rss_hf = ETH_RSS_IPV4,
 198                 },
 199         },
 200         .txmode = {
 201         },
 202 };
 203
 204 static const struct rte_eth_rxconf rx_conf = {
 205         .rx_thresh = {
 206                 .pthresh = RX_PTHRESH,
 207                 .hthresh = RX_HTHRESH,
 208                 .wthresh = RX_WTHRESH,
 209         },
 210 };
 211
 212 static const struct rte_eth_txconf tx_conf = {
 213         .tx_thresh = {
 214                 .pthresh = TX_PTHRESH,
 215                 .hthresh = TX_HTHRESH,
 216                 .wthresh = TX_WTHRESH,
 217         },
 218         .tx_free_thresh = 0, /* Use PMD default values */
 219         .tx_rs_thresh = 0, /* Use PMD default values */
 220 };
 221
 222 static struct rte_mempool * pktmbuf_pool[NB_SOCKETS];
 223
 224
 225 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 226
 227 #ifdef RTE_MACHINE_CPUFLAG_SSE4_2
 228 #include <rte_hash_crc.h>
 229 #define DEFAULT_HASH_FUNC       rte_hash_crc
 230 #else
 231 #include <rte_jhash.h>
 232 #define DEFAULT_HASH_FUNC       rte_jhash
 233 #endif
 234
 235 struct ipv4_5tuple {
 236         uint32_t ip_dst;
 237         uint32_t ip_src;
 238         uint16_t port_dst;
 239         uint16_t port_src;
 240         uint8_t proto;
 241 } __attribute__((__packed__));
 242
 243 struct l3fwd_route {
 244         struct ipv4_5tuple key;
 245         uint8_t if_out;
 246 };
 247
 248 static struct l3fwd_route l3fwd_route_array[] = {
 249         {{IPv4(100,10,0,1), IPv4(200,10,0,1), 101, 11, IPPROTO_TCP}, 0},
 250         {{IPv4(100,20,0,2), IPv4(200,20,0,2), 102, 12, IPPROTO_TCP}, 1},
 251         {{IPv4(100,30,0,3), IPv4(200,30,0,3), 103, 13, IPPROTO_TCP}, 2},
 252         {{IPv4(100,40,0,4), IPv4(200,40,0,4), 104, 14, IPPROTO_TCP}, 3},
 253 };
 254
 255 typedef struct rte_hash lookup_struct_t;
 256 static lookup_struct_t *l3fwd_lookup_struct[NB_SOCKETS];
 257
 258 #define L3FWD_HASH_ENTRIES      1024
 259 struct rte_hash_parameters l3fwd_hash_params = {
 260         .name = "l3fwd_hash_0",
 261         .entries = L3FWD_HASH_ENTRIES,
 262         .bucket_entries = 4,
 263         .key_len = sizeof(struct ipv4_5tuple),
 264         .hash_func = rte_hash_crc,
 265         .hash_func_init_val = 0,
 266         .socket_id = SOCKET0,
 267 };
 268
 269 #define L3FWD_NUM_ROUTES \
 270         (sizeof(l3fwd_route_array) / sizeof(l3fwd_route_array[0]))
 271
 272 static uint8_t l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 273 #endif
 274
 275 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 276 struct l3fwd_route {
 277         uint32_t ip;
 278         uint8_t  depth;
 279         uint8_t  if_out;
 280 };
 281
 282 static struct l3fwd_route l3fwd_route_array[] = {
 283         {IPv4(1,1,1,0), 24, 0},
 284         {IPv4(2,1,1,0), 24, 1},
 285         {IPv4(3,1,1,0), 24, 2},
 286         {IPv4(4,1,1,0), 24, 3},
 287         {IPv4(5,1,1,0), 24, 4},
 288         {IPv4(6,1,1,0), 24, 5},
 289         {IPv4(7,1,1,0), 24, 6},
 290         {IPv4(8,1,1,0), 24, 7},
 291 };
 292
 293 #define L3FWD_NUM_ROUTES \
 294         (sizeof(l3fwd_route_array) / sizeof(l3fwd_route_array[0]))
 295
 296 #define L3FWD_LPM_MAX_RULES     1024
 297
 298 typedef struct rte_lpm lookup_struct_t;
 299 static lookup_struct_t *l3fwd_lookup_struct[NB_SOCKETS];
 300 #endif
 301
 302 struct lcore_conf {
 303         uint16_t n_rx_queue;
 304         struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
 305         uint16_t tx_queue_id[MAX_PORTS];
 306         struct mbuf_table tx_mbufs[MAX_PORTS];
 307         lookup_struct_t * lookup_struct;
 308 } __rte_cache_aligned;
 309
 310 static struct lcore_conf lcore_conf[RTE_MAX_LCORE];
 311
 312 /* Send burst of packets on an output interface */
 313 static inline int
 314 send_burst(struct lcore_conf *qconf, uint16_t n, uint8_t port)
 315 {
 316         struct rte_mbuf **m_table;
 317         int ret;
 318         uint16_t queueid;
 319
 320         queueid = qconf->tx_queue_id[port];
 321         m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
 322
 323         ret = rte_eth_tx_burst(port, queueid, m_table, n);
 324         if (unlikely(ret < n)) {
 325                 do {
 326                         rte_pktmbuf_free(m_table[ret]);
 327                 } while (++ret < n);
 328         }
 329
 330         return 0;
 331 }
 332
 333 /* Enqueue a single packet, and send burst if queue is filled */
 334 static inline int
 335 send_single_packet(struct rte_mbuf *m, uint8_t port)
 336 {
 337         uint32_t lcore_id;
 338         uint16_t len;
 339         struct lcore_conf *qconf;
 340
 341         lcore_id = rte_lcore_id();
 342
 343         qconf = &lcore_conf[lcore_id];
 344         len = qconf->tx_mbufs[port].len;
 345         qconf->tx_mbufs[port].m_table[len] = m;
 346         len++;
 347
 348         /* enough pkts to be sent */
 349         if (unlikely(len == MAX_PKT_BURST)) {
 350                 send_burst(qconf, MAX_PKT_BURST, port);
 351                 len = 0;
 352         }
 353
 354         qconf->tx_mbufs[port].len = len;
 355         return 0;
 356 }
 357
 358 #ifdef DO_RFC_1812_CHECKS
 359 static inline int
 360 is_valid_ipv4_pkt(struct ipv4_hdr *pkt, uint32_t link_len)
 361 {
 362         /* From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2 */
 363         /*
 364          * 1. The packet length reported by the Link Layer must be large
 365          * enough to hold the minimum length legal IP datagram (20 bytes).
 366          */
 367         if (link_len < sizeof(struct ipv4_hdr))
 368                 return -1;
 369
 370         /* 2. The IP checksum must be correct. */
 371         /* this is checked in H/W */
 372
 373         /*
 374          * 3. The IP version number must be 4. If the version number is not 4
 375          * then the packet may be another version of IP, such as IPng or
 376          * ST-II.
 377          */
 378         if (((pkt->version_ihl) >> 4) != 4)
 379                 return -3;
 380         /*
 381          * 4. The IP header length field must be large enough to hold the
 382          * minimum length legal IP datagram (20 bytes = 5 words).
 383          */
 384         if ((pkt->version_ihl & 0xf) < 5)
 385                 return -4;
 386
 387         /*
 388          * 5. The IP total length field must be large enough to hold the IP
 389          * datagram header, whose length is specified in the IP header length
 390          * field.
 391          */
 392         if (rte_cpu_to_be_16(pkt->total_length) < sizeof(struct ipv4_hdr))
 393                 return -5;
 394
 395         return 0;
 396 }
 397 #endif
 398
 399 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 400 static void
 401 print_key(struct ipv4_5tuple key)
 402 {
 403         printf("IP dst = %08x, IP src = %08x, port dst = %d, port src = %d, proto = %d\n",
 404                (unsigned)key.ip_dst, (unsigned)key.ip_src, key.port_dst, key.port_src, key.proto);
 405 }
 406
 407 static inline uint8_t
 408 get_dst_port(struct ipv4_hdr *ipv4_hdr,  uint8_t portid, lookup_struct_t * l3fwd_lookup_struct)
 409 {
 410         struct ipv4_5tuple key;
 411         struct tcp_hdr *tcp;
 412         struct udp_hdr *udp;
 413         int ret = 0;
 414
 415         key.ip_dst = rte_be_to_cpu_32(ipv4_hdr->dst_addr);
 416         key.ip_src = rte_be_to_cpu_32(ipv4_hdr->src_addr);
 417         key.proto = ipv4_hdr->next_proto_id;
 418
 419         switch (ipv4_hdr->next_proto_id) {
 420         case IPPROTO_TCP:
 421                 tcp = (struct tcp_hdr *)((unsigned char *) ipv4_hdr +
 422                                         sizeof(struct ipv4_hdr));
 423                 key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
 424                 key.port_src = rte_be_to_cpu_16(tcp->src_port);
 425                 break;
 426
 427         case IPPROTO_UDP:
 428                 udp = (struct udp_hdr *)((unsigned char *) ipv4_hdr +
 429                                         sizeof(struct ipv4_hdr));
 430                 key.port_dst = rte_be_to_cpu_16(udp->dst_port);
 431                 key.port_src = rte_be_to_cpu_16(udp->src_port);
 432                 break;
 433
 434         default:
 435                 key.port_dst = 0;
 436                 key.port_src = 0;
 437         }
 438
 439         /* Find destination port */
 440         ret = rte_hash_lookup(l3fwd_lookup_struct, (const void *)&key);
 441         return (uint8_t)((ret < 0)? portid : l3fwd_out_if[ret]);
 442 }
 443 #endif
 444
 445 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 446 static inline uint8_t
 447 get_dst_port(struct ipv4_hdr *ipv4_hdr,  uint8_t portid, lookup_struct_t * l3fwd_lookup_struct)
 448 {
 449         uint8_t next_hop;
 450
 451         return (uint8_t) ((rte_lpm_lookup(l3fwd_lookup_struct,
 452                         rte_be_to_cpu_32(ipv4_hdr->dst_addr), &next_hop) == 0)?
 453                         next_hop : portid);
 454 }
 455 #endif
 456
 457 static inline void
 458 l3fwd_simple_forward(struct rte_mbuf *m, uint8_t portid, lookup_struct_t * l3fwd_lookup_struct)
 459 {
 460         struct ether_hdr *eth_hdr;
 461         struct ipv4_hdr *ipv4_hdr;
 462         void *tmp;
 463         uint8_t dst_port;
 464
 465         eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
 466
 467         ipv4_hdr = (struct ipv4_hdr *)(rte_pktmbuf_mtod(m, unsigned char *) +
 468                                 sizeof(struct ether_hdr));
 469
 470 #ifdef DO_RFC_1812_CHECKS
 471         /* Check to make sure the packet is valid (RFC1812) */
 472         if (is_valid_ipv4_pkt(ipv4_hdr, m->pkt.pkt_len) < 0) {
 473                 rte_pktmbuf_free(m);
 474                 return;
 475         }
 476 #endif
 477
 478         dst_port = get_dst_port(ipv4_hdr, portid, l3fwd_lookup_struct);
 479         if (dst_port >= MAX_PORTS || (enabled_port_mask & 1 << dst_port) == 0)
 480                 dst_port = portid;
 481
 482         /* 00:09:c0:00:00:xx */
 483         tmp = &eth_hdr->d_addr.addr_bytes[0];
 484         *((uint64_t *)tmp) = 0x000000c00900 + (dst_port << 24);
 485
 486 #ifdef DO_RFC_1812_CHECKS
 487         /* Update time to live and header checksum */
 488         --(ipv4_hdr->time_to_live);
 489         ++(ipv4_hdr->hdr_checksum);
 490 #endif
 491
 492         /* src addr */
 493         ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
 494
 495         send_single_packet(m, dst_port);
 496
 497 }
 498
 499 /* main processing loop */
 500 static __attribute__((noreturn)) int
 501 main_loop(__attribute__((unused)) void *dummy)
 502 {
 503         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
 504         unsigned lcore_id;
 505         uint64_t prev_tsc = 0;
 506         uint64_t diff_tsc, cur_tsc;
 507         int i, j, nb_rx;
 508         uint8_t portid, queueid;
 509         struct lcore_conf *qconf;
 510
 511         lcore_id = rte_lcore_id();
 512         qconf = &lcore_conf[lcore_id];
 513
 514         if (qconf->n_rx_queue == 0) {
 515                 RTE_LOG(INFO, L3FWD, "lcore %u has nothing to do\n", lcore_id);
 516                 while(1);
 517         }
 518
 519         RTE_LOG(INFO, L3FWD, "entering main loop on lcore %u\n", lcore_id);
 520
 521         for (i = 0; i < qconf->n_rx_queue; i++) {
 522
 523                 portid = qconf->rx_queue_list[i].port_id;
 524                 queueid = qconf->rx_queue_list[i].queue_id;
 525                 RTE_LOG(INFO, L3FWD, " -- lcoreid=%u portid=%hhu rxqueueid=%hhu\n", lcore_id,
 526                         portid, queueid);
 527         }
 528
 529         while (1) {
 530
 531                 cur_tsc = rte_rdtsc();
 532
 533                 /*
 534                  * TX burst queue drain
 535                  */
 536                 diff_tsc = cur_tsc - prev_tsc;
 537                 if (unlikely(diff_tsc > BURST_TX_DRAIN)) {
 538
 539                         /*
 540                          * This could be optimized (use queueid instead of
 541                          * portid), but it is not called so often
 542                          */
 543                         for (portid = 0; portid < MAX_PORTS; portid++) {
 544                                 if (qconf->tx_mbufs[portid].len == 0)
 545                                         continue;
 546                                 send_burst(&lcore_conf[lcore_id],
 547                                         qconf->tx_mbufs[portid].len,
 548                                         portid);
 549                                 qconf->tx_mbufs[portid].len = 0;
 550                         }
 551
 552                         prev_tsc = cur_tsc;
 553                 }
 554
 555                 /*
 556                  * Read packet from RX queues
 557                  */
 558                 for (i = 0; i < qconf->n_rx_queue; ++i) {
 559
 560                         portid = qconf->rx_queue_list[i].port_id;
 561                         queueid = qconf->rx_queue_list[i].queue_id;
 562                         nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst, MAX_PKT_BURST);
 563
 564                         /* Prefetch first packets */
 565                         for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
 566                                 rte_prefetch0(rte_pktmbuf_mtod(
 567                                                 pkts_burst[j], void *));
 568                         }
 569
 570                         /* Prefetch and forward already prefetched packets */
 571                         for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
 572                                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
 573                                                 j + PREFETCH_OFFSET], void *));
 574                                 l3fwd_simple_forward(pkts_burst[j], portid, qconf->lookup_struct);
 575                         }
 576
 577                         /* Forward remaining prefetched packets */
 578                         for (; j < nb_rx; j++) {
 579                                 l3fwd_simple_forward(pkts_burst[j], portid, qconf->lookup_struct);
 580                         }
 581                 }
 582         }
 583 }
 584
 585 static int
 586 check_lcore_params(void)
 587 {
 588         uint8_t queue, lcore;
 589         uint16_t i;
 590         int socketid;
 591
 592         for (i = 0; i < nb_lcore_params; ++i) {
 593                 queue = lcore_params[i].queue_id;
 594                 if (queue >= MAX_RX_QUEUE_PER_PORT) {
 595                         printf("invalid queue number: %hhu\n", queue);
 596                         return -1;
 597                 }
 598                 lcore = lcore_params[i].lcore_id;
 599                 if (!rte_lcore_is_enabled(lcore)) {
 600                         printf("error: lcore %hhu is not enabled in lcore mask\n", lcore);
 601                         return -1;
 602                 }
 603                 if ((socketid = rte_lcore_to_socket_id(lcore) != 0) &&
 604                         (numa_on == 0)) {
 605                         printf("warning: lcore %hhu is on socket %d with numa off \n",
 606                                 lcore, socketid);
 607                 }
 608         }
 609         return 0;
 610 }
 611
 612 static int
 613 check_port_config(const unsigned nb_ports)
 614 {
 615         unsigned portid;
 616         uint16_t i;
 617
 618         for (i = 0; i < nb_lcore_params; ++i) {
 619                 portid = lcore_params[i].port_id;
 620                 if ((enabled_port_mask & (1 << portid)) == 0) {
 621                         printf("port %u is not enabled in port mask\n", portid);
 622                         return -1;
 623                 }
 624                 if (portid >= nb_ports) {
 625                         printf("port %u is not present on the board\n", portid);
 626                         return -1;
 627                 }
 628         }
 629         return 0;
 630 }
 631
 632 static uint8_t
 633 get_port_n_rx_queues(const uint8_t port)
 634 {
 635         int queue = -1;
 636         uint16_t i;
 637
 638         for (i = 0; i < nb_lcore_params; ++i) {
 639                 if (lcore_params[i].port_id == port && lcore_params[i].queue_id > queue)
 640                         queue = lcore_params[i].queue_id;
 641         }
 642         return (uint8_t)(++queue);
 643 }
 644
 645 static int
 646 init_lcore_rx_queues(void)
 647 {
 648         uint16_t i, nb_rx_queue;
 649         uint8_t lcore;
 650
 651         for (i = 0; i < nb_lcore_params; ++i) {
 652                 lcore = lcore_params[i].lcore_id;
 653                 nb_rx_queue = lcore_conf[lcore].n_rx_queue;
 654                 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
 655                         printf("error: too many queues (%u) for lcore: %u\n",
 656                                 (unsigned)nb_rx_queue + 1, (unsigned)lcore);
 657                         return -1;
 658                 } else {
 659                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
 660                                 lcore_params[i].port_id;
 661                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
 662                                 lcore_params[i].queue_id;
 663                         lcore_conf[lcore].n_rx_queue++;
 664                 }
 665         }
 666         return 0;
 667 }
 668
 669 /* display usage */
 670 static void
 671 print_usage(const char *prgname)
 672 {
 673         printf ("%s [EAL options] -- -p PORTMASK -P"
 674                 "  [--config (port,queue,lcore)[,(port,queue,lcore]]\n"
 675                 "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
 676                 "  --config (port,queue,lcore): rx queues configuration\n"
 677                 "  --no-numa: optional, disable numa awareness\n",
 678                 prgname);
 679 }
 680
 681 static int
 682 parse_portmask(const char *portmask)
 683 {
 684         char *end = NULL;
 685         unsigned long pm;
 686
 687         /* parse hexadecimal string */
 688         pm = strtoul(portmask, &end, 16);
 689         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
 690                 return -1;
 691
 692         if (pm == 0)
 693                 return -1;
 694
 695         return pm;
 696 }
 697
 698 static int
 699 parse_config(const char *q_arg)
 700 {
 701         char s[256];
 702         const char *p, *p0 = q_arg;
 703         char *end;
 704         enum fieldnames {
 705                 FLD_PORT = 0,
 706                 FLD_QUEUE,
 707                 FLD_LCORE,
 708                 _NUM_FLD
 709         };
 710         unsigned long int_fld[_NUM_FLD];
 711         char *str_fld[_NUM_FLD];
 712         int i;
 713         unsigned size;
 714
 715         nb_lcore_params = 0;
 716
 717         while ((p = strchr(p0,'(')) != NULL) {
 718                 ++p;
 719                 if((p0 = strchr(p,')')) == NULL)
 720                         return -1;
 721
 722                 size = p0 - p;
 723                 if(size >= sizeof(s))
 724                         return -1;
 725
 726                 rte_snprintf(s, sizeof(s), "%.*s", size, p);
 727                 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') != _NUM_FLD)
 728                         return -1;
 729                 for (i = 0; i < _NUM_FLD; i++){
 730                         errno = 0;
 731                         int_fld[i] = strtoul(str_fld[i], &end, 0);
 732                         if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
 733                                 return -1;
 734                 }
 735                 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
 736                         printf("exceeded max number of lcore params: %hu\n",
 737                                 nb_lcore_params);
 738                         return -1;
 739                 }
 740                 lcore_params_array[nb_lcore_params].port_id = (uint8_t)int_fld[FLD_PORT];
 741                 lcore_params_array[nb_lcore_params].queue_id = (uint8_t)int_fld[FLD_QUEUE];
 742                 lcore_params_array[nb_lcore_params].lcore_id = (uint8_t)int_fld[FLD_LCORE];
 743                 ++nb_lcore_params;
 744         }
 745         lcore_params = lcore_params_array;
 746         return 0;
 747 }
 748
 749 /* Parse the argument given in the command line of the application */
 750 static int
 751 parse_args(int argc, char **argv)
 752 {
 753         int opt, ret;
 754         char **argvopt;
 755         int option_index;
 756         char *prgname = argv[0];
 757         static struct option lgopts[] = {
 758                 {"config", 1, 0, 0},
 759                 {"no-numa", 0, 0, 0},
 760                 {NULL, 0, 0, 0}
 761         };
 762
 763         argvopt = argv;
 764
 765         while ((opt = getopt_long(argc, argvopt, "p:P",
 766                                 lgopts, &option_index)) != EOF) {
 767
 768                 switch (opt) {
 769                 /* portmask */
 770                 case 'p':
 771                         enabled_port_mask = parse_portmask(optarg);
 772                         if (enabled_port_mask == 0) {
 773                                 printf("invalid portmask\n");
 774                                 print_usage(prgname);
 775                                 return -1;
 776                         }
 777                         break;
 778                 case 'P':
 779                         printf("Promiscuous mode selected\n");
 780                         promiscuous_on = 1;
 781                         break;
 782
 783                 /* long options */
 784                 case 0:
 785                         if (!strcmp(lgopts[option_index].name, "config")) {
 786                                 ret = parse_config(optarg);
 787                                 if (ret) {
 788                                         printf("invalid config\n");
 789                                         print_usage(prgname);
 790                                         return -1;
 791                                 }
 792                         }
 793
 794                         if (!strcmp(lgopts[option_index].name, "no-numa")) {
 795                                 printf("numa is disabled \n");
 796                                 numa_on = 0;
 797                         }
 798                         break;
 799
 800                 default:
 801                         print_usage(prgname);
 802                         return -1;
 803                 }
 804         }
 805
 806         if (optind >= 0)
 807                 argv[optind-1] = prgname;
 808
 809         ret = optind-1;
 810         optind = 0; /* reset getopt lib */
 811         return ret;
 812 }
 813
 814 static void
 815 print_ethaddr(const char *name, const struct ether_addr *eth_addr)
 816 {
 817         printf ("%s%02X:%02X:%02X:%02X:%02X:%02X", name,
 818                 eth_addr->addr_bytes[0],
 819                 eth_addr->addr_bytes[1],
 820                 eth_addr->addr_bytes[2],
 821                 eth_addr->addr_bytes[3],
 822                 eth_addr->addr_bytes[4],
 823                 eth_addr->addr_bytes[5]);
 824 }
 825
 826 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 827 static void
 828 setup_hash(int socketid)
 829 {
 830         unsigned i;
 831         int ret;
 832         char s[64];
 833
 834         /* create  hashes */
 835         rte_snprintf(s, sizeof(s), "l3fwd_hash_%d", socketid);
 836         l3fwd_hash_params.name = s;
 837         l3fwd_hash_params.socket_id = socketid;
 838         l3fwd_lookup_struct[socketid] = rte_hash_create(&l3fwd_hash_params);
 839         if (l3fwd_lookup_struct[socketid] == NULL)
 840                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
 841                                 "socket %d\n", socketid);
 842
 843         /* populate the hash */
 844         for (i = 0; i < L3FWD_NUM_ROUTES; i++) {
 845                 ret = rte_hash_add_key (l3fwd_lookup_struct[socketid],
 846                                 (void *) &l3fwd_route_array[i].key);
 847                 if (ret < 0) {
 848                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
 849                                 "l3fwd hash on socket %d\n", i, socketid);
 850                 }
 851                 l3fwd_out_if[ret] = l3fwd_route_array[i].if_out;
 852                 printf("Hash: Adding key\n");
 853                 print_key(l3fwd_route_array[i].key);
 854         }
 855 }
 856 #endif
 857
 858 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 859 static void
 860 setup_lpm(int socketid)
 861 {
 862         unsigned i;
 863         int ret;
 864         char s[64];
 865
 866         /* create the LPM table */
 867         rte_snprintf(s, sizeof(s), "L3FWD_LPM_%d", socketid);
 868         l3fwd_lookup_struct[socketid] = rte_lpm_create(s, socketid,
 869                                 L3FWD_LPM_MAX_RULES, RTE_LPM_MEMZONE);
 870         if (l3fwd_lookup_struct[socketid] == NULL)
 871                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
 872                                 " on socket %d\n", socketid);
 873
 874         /* populate the LPM table */
 875         for (i = 0; i < L3FWD_NUM_ROUTES; i++) {
 876                 ret = rte_lpm_add(l3fwd_lookup_struct[socketid],
 877                         l3fwd_route_array[i].ip,
 878                         l3fwd_route_array[i].depth,
 879                         l3fwd_route_array[i].if_out);
 880
 881                 if (ret < 0) {
 882                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
 883                                 "l3fwd LPM table on socket %d\n",
 884                                 i, socketid);
 885                 }
 886
 887                 printf("LPM: Adding route 0x%08x / %d (%d)\n",
 888                         (unsigned)l3fwd_route_array[i].ip,
 889                         l3fwd_route_array[i].depth,
 890                         l3fwd_route_array[i].if_out);
 891         }
 892 }
 893 #endif
 894
 895 static int
 896 init_mem(void)
 897 {
 898         struct lcore_conf *qconf;
 899         int socketid;
 900         unsigned lcore_id;
 901         char s[64];
 902
 903         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
 904                 if (rte_lcore_is_enabled(lcore_id) == 0)
 905                         continue;
 906
 907                 if (numa_on)
 908                         socketid = rte_lcore_to_socket_id(lcore_id);
 909                 else
 910                         socketid = 0;
 911
 912                 if (socketid >= NB_SOCKETS) {
 913                         rte_exit(EXIT_FAILURE, "Socket %d of lcore %u is out of range %d\n",
 914                                 socketid, lcore_id, NB_SOCKETS);
 915                 }
 916                 if (pktmbuf_pool[socketid] == NULL) {
 917                         rte_snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
 918                         pktmbuf_pool[socketid] =
 919                                 rte_mempool_create(s, NB_MBUF, MBUF_SIZE, 32,
 920                                         sizeof(struct rte_pktmbuf_pool_private),
 921                                         rte_pktmbuf_pool_init, NULL,
 922                                         rte_pktmbuf_init, NULL,
 923                                         socketid, 0);
 924                         if (pktmbuf_pool[socketid] == NULL)
 925                                 rte_exit(EXIT_FAILURE,
 926                                                 "Cannot init mbuf pool on socket %d\n", socketid);
 927                         else
 928                                 printf("Allocated mbuf pool on socket %d\n", socketid);
 929
 930 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 931                         setup_lpm(socketid);
 932 #else
 933                         setup_hash(socketid);
 934 #endif
 935                 }
 936                 qconf = &lcore_conf[lcore_id];
 937                 qconf->lookup_struct = l3fwd_lookup_struct[socketid];
 938         }
 939         return 0;
 940 }
 941
 942 int
 943 MAIN(int argc, char **argv)
 944 {
 945         struct lcore_conf *qconf;
 946         struct rte_eth_link link;
 947         int ret;
 948         unsigned nb_ports;
 949         uint16_t queueid;
 950         unsigned lcore_id;
 951         uint32_t n_tx_queue, nb_lcores;
 952         uint8_t portid, nb_rx_queue, queue, socketid;
 953
 954         /* init EAL */
 955         ret = rte_eal_init(argc, argv);
 956         if (ret < 0)
 957                 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
 958         argc -= ret;
 959         argv += ret;
 960
 961         /* parse application arguments (after the EAL ones) */
 962         ret = parse_args(argc, argv);
 963         if (ret < 0)
 964                 rte_exit(EXIT_FAILURE, "Invalid L3FWD parameters\n");
 965
 966         if (check_lcore_params() < 0)
 967                 rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
 968
 969         ret = init_lcore_rx_queues();
 970         if (ret < 0)
 971                 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
 972
 973         ret = init_mem();
 974         if (ret < 0)
 975                 rte_exit(EXIT_FAILURE, "init_mem failed\n");
 976
 977         /* init driver */
 978 #ifdef RTE_LIBRTE_IGB_PMD
 979         if (rte_igb_pmd_init() < 0)
 980                 rte_exit(EXIT_FAILURE, "Cannot init igb pmd\n");
 981 #endif
 982 #ifdef RTE_LIBRTE_IXGBE_PMD
 983         if (rte_ixgbe_pmd_init() < 0)
 984                 rte_exit(EXIT_FAILURE, "Cannot init ixgbe pmd\n");
 985 #endif
 986
 987         if (rte_eal_pci_probe() < 0)
 988                 rte_exit(EXIT_FAILURE, "Cannot probe PCI\n");
 989
 990         nb_ports = rte_eth_dev_count();
 991         if (nb_ports > MAX_PORTS)
 992                 nb_ports = MAX_PORTS;
 993
 994         if (check_port_config(nb_ports) < 0)
 995                 rte_exit(EXIT_FAILURE, "check_port_config failed\n");
 996
 997         nb_lcores = rte_lcore_count();
 998
 999         /* initialize all ports */
1000         for (portid = 0; portid < nb_ports; portid++) {
1001                 /* skip ports that are not enabled */
1002                 if ((enabled_port_mask & (1 << portid)) == 0) {
1003                         printf("\nSkipping disabled port %d\n", portid);
1004                         continue;
1005                 }
1006
1007                 /* init port */
1008                 printf("Initializing port %d ... ", portid );
1009                 fflush(stdout);
1010
1011                 nb_rx_queue = get_port_n_rx_queues(portid);
1012                 n_tx_queue = nb_lcores;
1013                 if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
1014                         n_tx_queue = MAX_TX_QUEUE_PER_PORT;
1015                 printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
1016                         nb_rx_queue, (unsigned)n_tx_queue );
1017                 ret = rte_eth_dev_configure(portid, nb_rx_queue,
1018                                         (uint16_t)n_tx_queue, &port_conf);
1019                 if (ret < 0)
1020                         rte_exit(EXIT_FAILURE, "Cannot configure device: 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
1028                 /* init one TX queue per couple (lcore,port) */
1029                 queueid = 0;
1030                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1031                         if (rte_lcore_is_enabled(lcore_id) == 0)
1032                                 continue;
1033
1034                         if (numa_on)
1035                                 socketid = (uint8_t)rte_lcore_to_socket_id(lcore_id);
1036                         else
1037                                 socketid = 0;
1038
1039                         printf("txq=%u,%d,%d ", lcore_id, queueid, socketid);
1040                         fflush(stdout);
1041                         ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
1042                                                      socketid, &tx_conf);
1043                         if (ret < 0)
1044                                 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: err=%d, "
1045                                         "port=%d\n", ret, portid);
1046
1047                         qconf = &lcore_conf[lcore_id];
1048                         qconf->tx_queue_id[portid] = queueid;
1049                         queueid++;
1050                 }
1051                 printf("\n");
1052         }
1053
1054         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1055                 if (rte_lcore_is_enabled(lcore_id) == 0)
1056                         continue;
1057                 qconf = &lcore_conf[lcore_id];
1058                 printf("\nInitializing rx queues on lcore %u ... ", lcore_id );
1059                 fflush(stdout);
1060                 /* init RX queues */
1061                 for(queue = 0; queue < qconf->n_rx_queue; ++queue) {
1062                         portid = qconf->rx_queue_list[queue].port_id;
1063                         queueid = qconf->rx_queue_list[queue].queue_id;
1064
1065                         if (numa_on)
1066                                 socketid = (uint8_t)rte_lcore_to_socket_id(lcore_id);
1067                         else
1068                                 socketid = 0;
1069
1070                         printf("rxq=%d,%d,%d ", portid, queueid, socketid);
1071                         fflush(stdout);
1072
1073                         ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
1074                                         socketid, &rx_conf, pktmbuf_pool[socketid]);
1075                         if (ret < 0)
1076                                 rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: err=%d,"
1077                                                 "port=%d\n", ret, portid);
1078                 }
1079         }
1080
1081         printf("\n");
1082
1083         /* start ports */
1084         for (portid = 0; portid < nb_ports; portid++) {
1085                 if ((enabled_port_mask & (1 << portid)) == 0) {
1086                         continue;
1087                 }
1088                 /* Start device */
1089                 ret = rte_eth_dev_start(portid);
1090                 if (ret < 0)
1091                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%d\n",
1092                                 ret, portid);
1093
1094                 printf("done: Port %d ", portid);
1095
1096                 /* get link status */
1097                 rte_eth_link_get(portid, &link);
1098                 if (link.link_status) {
1099                         printf(" Link Up - speed %u Mbps - %s\n",
1100                                (unsigned) link.link_speed,
1101                                (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
1102                                ("full-duplex") : ("half-duplex\n"));
1103                 } else {
1104                         printf(" Link Down\n");
1105                 }
1106                 /*
1107                  * If enabled, put device in promiscuous mode.
1108                  * This allows IO forwarding mode to forward packets
1109                  * to itself through 2 cross-connected  ports of the
1110                  * target machine.
1111                  */
1112                 if (promiscuous_on)
1113                         rte_eth_promiscuous_enable(portid);
1114         }
1115
1116         /* launch per-lcore init on every lcore */
1117         rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
1118         RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1119                 if (rte_eal_wait_lcore(lcore_id) < 0)
1120                         return -1;
1121         }
1122
1123         return 0;
1124 }