examples/l3fwd/main.c

   1 /*-
   2  *   BSD LICENSE
   3  *
   4  *   Copyright(c) 2010-2013 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 #ifndef IPv6_BYTES
  96 #define IPv6_BYTES_FMT "%02x%02x:%02x%02x:%02x%02x:%02x%02x:"\
  97                        "%02x%02x:%02x%02x:%02x%02x:%02x%02x"
  98 #define IPv6_BYTES(addr) \
  99         addr[0],  addr[1], addr[2],  addr[3], \
 100         addr[4],  addr[5], addr[6],  addr[7], \
 101         addr[8],  addr[9], addr[10], addr[11],\
 102         addr[12], addr[13],addr[14], addr[15]
 103 #endif
 104
 105
 106 #define RTE_LOGTYPE_L3FWD RTE_LOGTYPE_USER1
 107
 108 #define MAX_PORTS 32
 109
 110 #define MAX_JUMBO_PKT_LEN  9600
 111
 112 #define IPV6_ADDR_LEN 16
 113
 114 #define MEMPOOL_CACHE_SIZE 256
 115
 116 #define MBUF_SIZE (2048 + sizeof(struct rte_mbuf) + RTE_PKTMBUF_HEADROOM)
 117
 118 /*
 119  * This expression is used to calculate the number of mbufs needed depending on user input, taking
 120  *  into account memory for rx and tx hardware rings, cache per lcore and mtable per port per lcore.
 121  *  RTE_MAX is used to ensure that NB_MBUF never goes below a minimum value of 8192
 122  */
 123
 124 #define NB_MBUF RTE_MAX (                                                                                                                                       \
 125                                 (nb_ports*nb_rx_queue*RTE_TEST_RX_DESC_DEFAULT +                                                        \
 126                                 nb_ports*nb_lcores*MAX_PKT_BURST +                                                                                      \
 127                                 nb_ports*n_tx_queue*RTE_TEST_TX_DESC_DEFAULT +                                                          \
 128                                 nb_lcores*MEMPOOL_CACHE_SIZE),                                                                                          \
 129                                 (unsigned)8192)
 130
 131 /*
 132  * RX and TX Prefetch, Host, and Write-back threshold values should be
 133  * carefully set for optimal performance. Consult the network
 134  * controller's datasheet and supporting DPDK documentation for guidance
 135  * on how these parameters should be set.
 136  */
 137 #define RX_PTHRESH 8 /**< Default values of RX prefetch threshold reg. */
 138 #define RX_HTHRESH 8 /**< Default values of RX host threshold reg. */
 139 #define RX_WTHRESH 4 /**< Default values of RX write-back threshold reg. */
 140
 141 /*
 142  * These default values are optimized for use with the Intel(R) 82599 10 GbE
 143  * Controller and the DPDK ixgbe PMD. Consider using other values for other
 144  * network controllers and/or network drivers.
 145  */
 146 #define TX_PTHRESH 36 /**< Default values of TX prefetch threshold reg. */
 147 #define TX_HTHRESH 0  /**< Default values of TX host threshold reg. */
 148 #define TX_WTHRESH 0  /**< Default values of TX write-back threshold reg. */
 149
 150 #define MAX_PKT_BURST 32
 151 #define BURST_TX_DRAIN 200000ULL /* around 100us at 2 Ghz */
 152
 153 #define NB_SOCKETS 8
 154
 155 #define SOCKET0 0
 156
 157 /* Configure how many packets ahead to prefetch, when reading packets */
 158 #define PREFETCH_OFFSET 3
 159
 160 /*
 161  * Configurable number of RX/TX ring descriptors
 162  */
 163 #define RTE_TEST_RX_DESC_DEFAULT 128
 164 #define RTE_TEST_TX_DESC_DEFAULT 512
 165 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
 166 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
 167
 168 /* ethernet addresses of ports */
 169 static struct ether_addr ports_eth_addr[MAX_PORTS];
 170
 171 /* mask of enabled ports */
 172 static uint32_t enabled_port_mask = 0;
 173 static int promiscuous_on = 0; /**< Ports set in promiscuous mode off by default. */
 174 static int numa_on = 1; /**< NUMA is enabled by default. */
 175
 176 struct mbuf_table {
 177         uint16_t len;
 178         struct rte_mbuf *m_table[MAX_PKT_BURST];
 179 };
 180
 181 struct lcore_rx_queue {
 182         uint8_t port_id;
 183         uint8_t queue_id;
 184 } __rte_cache_aligned;
 185
 186 #define MAX_RX_QUEUE_PER_LCORE 16
 187 #define MAX_TX_QUEUE_PER_PORT MAX_PORTS
 188 #define MAX_RX_QUEUE_PER_PORT 128
 189
 190 #define MAX_LCORE_PARAMS 1024
 191 struct lcore_params {
 192         uint8_t port_id;
 193         uint8_t queue_id;
 194         uint8_t lcore_id;
 195 } __rte_cache_aligned;
 196
 197 static struct lcore_params lcore_params_array[MAX_LCORE_PARAMS];
 198 static struct lcore_params lcore_params_array_default[] = {
 199         {0, 0, 2},
 200         {0, 1, 2},
 201         {0, 2, 2},
 202         {1, 0, 2},
 203         {1, 1, 2},
 204         {1, 2, 2},
 205         {2, 0, 2},
 206         {3, 0, 3},
 207         {3, 1, 3},
 208 };
 209
 210 static struct lcore_params * lcore_params = lcore_params_array_default;
 211 static uint16_t nb_lcore_params = sizeof(lcore_params_array_default) /
 212                                 sizeof(lcore_params_array_default[0]);
 213
 214 static struct rte_eth_conf port_conf = {
 215         .rxmode = {
 216                 .max_rx_pkt_len = ETHER_MAX_LEN,
 217                 .split_hdr_size = 0,
 218                 .header_split   = 0, /**< Header Split disabled */
 219                 .hw_ip_checksum = 1, /**< IP checksum offload enabled */
 220                 .hw_vlan_filter = 0, /**< VLAN filtering disabled */
 221                 .jumbo_frame    = 0, /**< Jumbo Frame Support disabled */
 222                 .hw_strip_crc   = 0, /**< CRC stripped by hardware */
 223         },
 224         .rx_adv_conf = {
 225                 .rss_conf = {
 226                         .rss_key = NULL,
 227                         .rss_hf = ETH_RSS_IPV4 | ETH_RSS_IPV6,
 228                 },
 229         },
 230         .txmode = {
 231                 .mq_mode = ETH_DCB_NONE,
 232         },
 233 };
 234
 235 static const struct rte_eth_rxconf rx_conf = {
 236         .rx_thresh = {
 237                 .pthresh = RX_PTHRESH,
 238                 .hthresh = RX_HTHRESH,
 239                 .wthresh = RX_WTHRESH,
 240         },
 241         .rx_free_thresh = 32,
 242 };
 243
 244 static const struct rte_eth_txconf tx_conf = {
 245         .tx_thresh = {
 246                 .pthresh = TX_PTHRESH,
 247                 .hthresh = TX_HTHRESH,
 248                 .wthresh = TX_WTHRESH,
 249         },
 250         .tx_free_thresh = 0, /* Use PMD default values */
 251         .tx_rs_thresh = 0, /* Use PMD default values */
 252         .txq_flags = 0x0,
 253 };
 254
 255 static struct rte_mempool * pktmbuf_pool[NB_SOCKETS];
 256
 257
 258 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 259
 260 #ifdef RTE_MACHINE_CPUFLAG_SSE4_2
 261 #include <rte_hash_crc.h>
 262 #define DEFAULT_HASH_FUNC       rte_hash_crc
 263 #else
 264 #include <rte_jhash.h>
 265 #define DEFAULT_HASH_FUNC       rte_jhash
 266 #endif
 267
 268 struct ipv4_5tuple {
 269         uint32_t ip_dst;
 270         uint32_t ip_src;
 271         uint16_t port_dst;
 272         uint16_t port_src;
 273         uint8_t  proto;
 274 } __attribute__((__packed__));
 275
 276 struct ipv6_5tuple {
 277         uint8_t  ip_dst[IPV6_ADDR_LEN];
 278         uint8_t  ip_src[IPV6_ADDR_LEN];
 279         uint16_t port_dst;
 280         uint16_t port_src;
 281         uint8_t  proto;
 282 } __attribute__((__packed__));
 283
 284 struct ipv4_l3fwd_route {
 285         struct ipv4_5tuple key;
 286         uint8_t if_out;
 287 };
 288
 289 struct ipv6_l3fwd_route {
 290         struct ipv6_5tuple key;
 291         uint8_t if_out;
 292 };
 293
 294 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
 295         {{IPv4(100,10,0,1), IPv4(200,10,0,1), 101, 11, IPPROTO_TCP}, 0},
 296         {{IPv4(100,20,0,2), IPv4(200,20,0,2), 102, 12, IPPROTO_TCP}, 1},
 297         {{IPv4(100,30,0,3), IPv4(200,30,0,3), 103, 13, IPPROTO_TCP}, 2},
 298         {{IPv4(100,40,0,4), IPv4(200,40,0,4), 104, 14, IPPROTO_TCP}, 3},
 299 };
 300
 301 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
 302         {
 303                 {
 304                         {0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
 305                          0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
 306                         {0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
 307                          0x02, 0x1e, 0x67, 0xff, 0xfe, 0x0d, 0xb6, 0x0a},
 308                          1, 10, IPPROTO_UDP
 309                 }, 4
 310         },
 311 };
 312
 313 typedef struct rte_hash lookup_struct_t;
 314 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
 315 static lookup_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
 316
 317 #define L3FWD_HASH_ENTRIES      1024
 318
 319 struct rte_hash_parameters ipv4_l3fwd_hash_params = {
 320         .name = "ipv4_l3fwd_hash_0",
 321         .entries = L3FWD_HASH_ENTRIES,
 322         .bucket_entries = 4,
 323         .key_len = sizeof(struct ipv4_5tuple),
 324         .hash_func = DEFAULT_HASH_FUNC,
 325         .hash_func_init_val = 0,
 326         .socket_id = SOCKET0,
 327 };
 328
 329 struct rte_hash_parameters ipv6_l3fwd_hash_params = {
 330         .name = "ipv6_l3fwd_hash_0",
 331         .entries = L3FWD_HASH_ENTRIES,
 332         .bucket_entries = 4,
 333         .key_len = sizeof(struct ipv6_5tuple),
 334         .hash_func = DEFAULT_HASH_FUNC,
 335         .hash_func_init_val = 0,
 336         .socket_id = SOCKET0,
 337 };
 338
 339 #define IPV4_L3FWD_NUM_ROUTES \
 340         (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
 341
 342 #define IPV6_L3FWD_NUM_ROUTES \
 343         (sizeof(ipv6_l3fwd_route_array) / sizeof(ipv6_l3fwd_route_array[0]))
 344
 345 static uint8_t ipv4_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 346 static uint8_t ipv6_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 347 #endif
 348
 349 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 350 struct ipv4_l3fwd_route {
 351         uint32_t ip;
 352         uint8_t  depth;
 353         uint8_t  if_out;
 354 };
 355
 356 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
 357         {IPv4(1,1,1,0), 24, 0},
 358         {IPv4(2,1,1,0), 24, 1},
 359         {IPv4(3,1,1,0), 24, 2},
 360         {IPv4(4,1,1,0), 24, 3},
 361         {IPv4(5,1,1,0), 24, 4},
 362         {IPv4(6,1,1,0), 24, 5},
 363         {IPv4(7,1,1,0), 24, 6},
 364         {IPv4(8,1,1,0), 24, 7},
 365 };
 366
 367 #define IPV4_L3FWD_NUM_ROUTES \
 368         (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
 369
 370 #define IPV4_L3FWD_LPM_MAX_RULES     1024
 371
 372 typedef struct rte_lpm lookup_struct_t;
 373 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
 374 #endif
 375
 376 struct lcore_conf {
 377         uint16_t n_rx_queue;
 378         struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
 379         uint16_t tx_queue_id[MAX_PORTS];
 380         struct mbuf_table tx_mbufs[MAX_PORTS];
 381         lookup_struct_t * ipv4_lookup_struct;
 382         lookup_struct_t * ipv6_lookup_struct;
 383 } __rte_cache_aligned;
 384
 385 static struct lcore_conf lcore_conf[RTE_MAX_LCORE];
 386
 387 /* Send burst of packets on an output interface */
 388 static inline int
 389 send_burst(struct lcore_conf *qconf, uint16_t n, uint8_t port)
 390 {
 391         struct rte_mbuf **m_table;
 392         int ret;
 393         uint16_t queueid;
 394
 395         queueid = qconf->tx_queue_id[port];
 396         m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
 397
 398         ret = rte_eth_tx_burst(port, queueid, m_table, n);
 399         if (unlikely(ret < n)) {
 400                 do {
 401                         rte_pktmbuf_free(m_table[ret]);
 402                 } while (++ret < n);
 403         }
 404
 405         return 0;
 406 }
 407
 408 /* Enqueue a single packet, and send burst if queue is filled */
 409 static inline int
 410 send_single_packet(struct rte_mbuf *m, uint8_t port)
 411 {
 412         uint32_t lcore_id;
 413         uint16_t len;
 414         struct lcore_conf *qconf;
 415
 416         lcore_id = rte_lcore_id();
 417
 418         qconf = &lcore_conf[lcore_id];
 419         len = qconf->tx_mbufs[port].len;
 420         qconf->tx_mbufs[port].m_table[len] = m;
 421         len++;
 422
 423         /* enough pkts to be sent */
 424         if (unlikely(len == MAX_PKT_BURST)) {
 425                 send_burst(qconf, MAX_PKT_BURST, port);
 426                 len = 0;
 427         }
 428
 429         qconf->tx_mbufs[port].len = len;
 430         return 0;
 431 }
 432
 433 #ifdef DO_RFC_1812_CHECKS
 434 static inline int
 435 is_valid_ipv4_pkt(struct ipv4_hdr *pkt, uint32_t link_len)
 436 {
 437         /* From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2 */
 438         /*
 439          * 1. The packet length reported by the Link Layer must be large
 440          * enough to hold the minimum length legal IP datagram (20 bytes).
 441          */
 442         if (link_len < sizeof(struct ipv4_hdr))
 443                 return -1;
 444
 445         /* 2. The IP checksum must be correct. */
 446         /* this is checked in H/W */
 447
 448         /*
 449          * 3. The IP version number must be 4. If the version number is not 4
 450          * then the packet may be another version of IP, such as IPng or
 451          * ST-II.
 452          */
 453         if (((pkt->version_ihl) >> 4) != 4)
 454                 return -3;
 455         /*
 456          * 4. The IP header length field must be large enough to hold the
 457          * minimum length legal IP datagram (20 bytes = 5 words).
 458          */
 459         if ((pkt->version_ihl & 0xf) < 5)
 460                 return -4;
 461
 462         /*
 463          * 5. The IP total length field must be large enough to hold the IP
 464          * datagram header, whose length is specified in the IP header length
 465          * field.
 466          */
 467         if (rte_cpu_to_be_16(pkt->total_length) < sizeof(struct ipv4_hdr))
 468                 return -5;
 469
 470         return 0;
 471 }
 472 #endif
 473
 474 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 475 static void
 476 print_ipv4_key(struct ipv4_5tuple key)
 477 {
 478         printf("IP dst = %08x, IP src = %08x, port dst = %d, port src = %d, proto = %d\n",
 479                         (unsigned)key.ip_dst, (unsigned)key.ip_src, key.port_dst, key.port_src, key.proto);
 480 }
 481 static void
 482 print_ipv6_key(struct ipv6_5tuple key)
 483 {
 484         printf( "IP dst = " IPv6_BYTES_FMT ", IP src = " IPv6_BYTES_FMT ", "
 485                 "port dst = %d, port src = %d, proto = %d\n",
 486                 IPv6_BYTES(key.ip_dst), IPv6_BYTES(key.ip_src),
 487                 key.port_dst, key.port_src, key.proto);
 488 }
 489
 490 static inline uint8_t
 491 get_ipv4_dst_port(struct ipv4_hdr *ipv4_hdr,  uint8_t portid, lookup_struct_t * ipv4_l3fwd_lookup_struct)
 492 {
 493         struct ipv4_5tuple key;
 494         struct tcp_hdr *tcp;
 495         struct udp_hdr *udp;
 496         int ret = 0;
 497
 498         key.ip_dst = rte_be_to_cpu_32(ipv4_hdr->dst_addr);
 499         key.ip_src = rte_be_to_cpu_32(ipv4_hdr->src_addr);
 500         key.proto = ipv4_hdr->next_proto_id;
 501
 502         switch (ipv4_hdr->next_proto_id) {
 503         case IPPROTO_TCP:
 504                 tcp = (struct tcp_hdr *)((unsigned char *) ipv4_hdr +
 505                                         sizeof(struct ipv4_hdr));
 506                 key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
 507                 key.port_src = rte_be_to_cpu_16(tcp->src_port);
 508                 break;
 509
 510         case IPPROTO_UDP:
 511                 udp = (struct udp_hdr *)((unsigned char *) ipv4_hdr +
 512                                         sizeof(struct ipv4_hdr));
 513                 key.port_dst = rte_be_to_cpu_16(udp->dst_port);
 514                 key.port_src = rte_be_to_cpu_16(udp->src_port);
 515                 break;
 516
 517         default:
 518                 key.port_dst = 0;
 519                 key.port_src = 0;
 520                 break;
 521         }
 522
 523         /* Find destination port */
 524         ret = rte_hash_lookup(ipv4_l3fwd_lookup_struct, (const void *)&key);
 525         return (uint8_t)((ret < 0)? portid : ipv4_l3fwd_out_if[ret]);
 526 }
 527
 528 static inline uint8_t
 529 get_ipv6_dst_port(struct ipv6_hdr *ipv6_hdr,  uint8_t portid, lookup_struct_t * ipv6_l3fwd_lookup_struct)
 530 {
 531         struct ipv6_5tuple key;
 532         struct tcp_hdr *tcp;
 533         struct udp_hdr *udp;
 534         int ret = 0;
 535
 536         memcpy(key.ip_dst, ipv6_hdr->dst_addr, IPV6_ADDR_LEN);
 537         memcpy(key.ip_src, ipv6_hdr->src_addr, IPV6_ADDR_LEN);
 538
 539         key.proto = ipv6_hdr->proto;
 540
 541         switch (ipv6_hdr->proto) {
 542         case IPPROTO_TCP:
 543                 tcp = (struct tcp_hdr *)((unsigned char *) ipv6_hdr +
 544                                         sizeof(struct ipv6_hdr));
 545                 key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
 546                 key.port_src = rte_be_to_cpu_16(tcp->src_port);
 547                 break;
 548
 549         case IPPROTO_UDP:
 550                 udp = (struct udp_hdr *)((unsigned char *) ipv6_hdr +
 551                                         sizeof(struct ipv6_hdr));
 552                 key.port_dst = rte_be_to_cpu_16(udp->dst_port);
 553                 key.port_src = rte_be_to_cpu_16(udp->src_port);
 554                 break;
 555
 556         default:
 557                 key.port_dst = 0;
 558                 key.port_src = 0;
 559                 break;
 560         }
 561
 562         /* Find destination port */
 563         ret = rte_hash_lookup(ipv6_l3fwd_lookup_struct, (const void *)&key);
 564         return (uint8_t)((ret < 0)? portid : ipv6_l3fwd_out_if[ret]);
 565 }
 566 #endif
 567
 568 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 569 static inline uint8_t
 570 get_ipv4_dst_port(struct ipv4_hdr *ipv4_hdr,  uint8_t portid, lookup_struct_t * ipv4_l3fwd_lookup_struct)
 571 {
 572         uint8_t next_hop;
 573
 574         return (uint8_t) ((rte_lpm_lookup(ipv4_l3fwd_lookup_struct,
 575                         rte_be_to_cpu_32(ipv4_hdr->dst_addr), &next_hop) == 0)?
 576                         next_hop : portid);
 577 }
 578 #endif
 579
 580 static inline void
 581 l3fwd_simple_forward(struct rte_mbuf *m, uint8_t portid, struct lcore_conf *qconf)
 582 {
 583         struct ether_hdr *eth_hdr;
 584         struct ipv4_hdr *ipv4_hdr;
 585         void *d_addr_bytes;
 586         uint8_t dst_port;
 587
 588         eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
 589
 590         if (m->ol_flags & PKT_RX_IPV4_HDR) {
 591                 /* Handle IPv4 headers.*/
 592                 ipv4_hdr = (struct ipv4_hdr *)(rte_pktmbuf_mtod(m, unsigned char *) +
 593                                 sizeof(struct ether_hdr));
 594
 595 #ifdef DO_RFC_1812_CHECKS
 596                 /* Check to make sure the packet is valid (RFC1812) */
 597                 if (is_valid_ipv4_pkt(ipv4_hdr, m->pkt.pkt_len) < 0) {
 598                         rte_pktmbuf_free(m);
 599                         return;
 600                 }
 601 #endif
 602
 603                 dst_port = get_ipv4_dst_port(ipv4_hdr, portid, qconf->ipv4_lookup_struct);
 604                 if (dst_port >= MAX_PORTS || (enabled_port_mask & 1 << dst_port) == 0)
 605                         dst_port = portid;
 606
 607                 /* 02:00:00:00:00:xx */
 608                 d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
 609                 *((uint64_t *)d_addr_bytes) = 0x000000000002 + ((uint64_t)dst_port << 40);
 610
 611 #ifdef DO_RFC_1812_CHECKS
 612                 /* Update time to live and header checksum */
 613                 --(ipv4_hdr->time_to_live);
 614                 ++(ipv4_hdr->hdr_checksum);
 615 #endif
 616
 617                 /* src addr */
 618                 ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
 619
 620                 send_single_packet(m, dst_port);
 621         }
 622         else {
 623                 /* Handle IPv6 headers.*/
 624 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 625                 struct ipv6_hdr *ipv6_hdr;
 626
 627                 ipv6_hdr = (struct ipv6_hdr *)(rte_pktmbuf_mtod(m, unsigned char *) +
 628                                 sizeof(struct ether_hdr));
 629
 630                 dst_port = get_ipv6_dst_port(ipv6_hdr, portid, qconf->ipv6_lookup_struct);
 631
 632                 if (dst_port >= MAX_PORTS || (enabled_port_mask & 1 << dst_port) == 0)
 633                         dst_port = portid;
 634
 635                 /* 02:00:00:00:00:xx */
 636                 d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
 637                 *((uint64_t *)d_addr_bytes) = 0x000000000002 + ((uint64_t)dst_port << 40);
 638
 639                 /* src addr */
 640                 ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
 641
 642                 send_single_packet(m, dst_port);
 643 #else
 644                 /* We don't currently handle IPv6 packets in LPM mode. */
 645                 rte_pktmbuf_free(m);
 646 #endif
 647         }
 648
 649 }
 650
 651 /* main processing loop */
 652 static __attribute__((noreturn)) int
 653 main_loop(__attribute__((unused)) void *dummy)
 654 {
 655         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
 656         unsigned lcore_id;
 657         uint64_t prev_tsc = 0;
 658         uint64_t diff_tsc, cur_tsc;
 659         int i, j, nb_rx;
 660         uint8_t portid, queueid;
 661         struct lcore_conf *qconf;
 662
 663         lcore_id = rte_lcore_id();
 664         qconf = &lcore_conf[lcore_id];
 665
 666         if (qconf->n_rx_queue == 0) {
 667                 RTE_LOG(INFO, L3FWD, "lcore %u has nothing to do\n", lcore_id);
 668                 while(1);
 669         }
 670
 671         RTE_LOG(INFO, L3FWD, "entering main loop on lcore %u\n", lcore_id);
 672
 673         for (i = 0; i < qconf->n_rx_queue; i++) {
 674
 675                 portid = qconf->rx_queue_list[i].port_id;
 676                 queueid = qconf->rx_queue_list[i].queue_id;
 677                 RTE_LOG(INFO, L3FWD, " -- lcoreid=%u portid=%hhu rxqueueid=%hhu\n", lcore_id,
 678                         portid, queueid);
 679         }
 680
 681         while (1) {
 682
 683                 cur_tsc = rte_rdtsc();
 684
 685                 /*
 686                  * TX burst queue drain
 687                  */
 688                 diff_tsc = cur_tsc - prev_tsc;
 689                 if (unlikely(diff_tsc > BURST_TX_DRAIN)) {
 690
 691                         /*
 692                          * This could be optimized (use queueid instead of
 693                          * portid), but it is not called so often
 694                          */
 695                         for (portid = 0; portid < MAX_PORTS; portid++) {
 696                                 if (qconf->tx_mbufs[portid].len == 0)
 697                                         continue;
 698                                 send_burst(&lcore_conf[lcore_id],
 699                                         qconf->tx_mbufs[portid].len,
 700                                         portid);
 701                                 qconf->tx_mbufs[portid].len = 0;
 702                         }
 703
 704                         prev_tsc = cur_tsc;
 705                 }
 706
 707                 /*
 708                  * Read packet from RX queues
 709                  */
 710                 for (i = 0; i < qconf->n_rx_queue; ++i) {
 711
 712                         portid = qconf->rx_queue_list[i].port_id;
 713                         queueid = qconf->rx_queue_list[i].queue_id;
 714                         nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst, MAX_PKT_BURST);
 715
 716                         /* Prefetch first packets */
 717                         for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
 718                                 rte_prefetch0(rte_pktmbuf_mtod(
 719                                                 pkts_burst[j], void *));
 720                         }
 721
 722                         /* Prefetch and forward already prefetched packets */
 723                         for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
 724                                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
 725                                                 j + PREFETCH_OFFSET], void *));
 726                                 l3fwd_simple_forward(pkts_burst[j], portid, qconf);
 727                         }
 728
 729                         /* Forward remaining prefetched packets */
 730                         for (; j < nb_rx; j++) {
 731                                 l3fwd_simple_forward(pkts_burst[j], portid, qconf);
 732                         }
 733                 }
 734         }
 735 }
 736
 737 static int
 738 check_lcore_params(void)
 739 {
 740         uint8_t queue, lcore;
 741         uint16_t i;
 742         int socketid;
 743
 744         for (i = 0; i < nb_lcore_params; ++i) {
 745                 queue = lcore_params[i].queue_id;
 746                 if (queue >= MAX_RX_QUEUE_PER_PORT) {
 747                         printf("invalid queue number: %hhu\n", queue);
 748                         return -1;
 749                 }
 750                 lcore = lcore_params[i].lcore_id;
 751                 if (!rte_lcore_is_enabled(lcore)) {
 752                         printf("error: lcore %hhu is not enabled in lcore mask\n", lcore);
 753                         return -1;
 754                 }
 755                 if ((socketid = rte_lcore_to_socket_id(lcore) != 0) &&
 756                         (numa_on == 0)) {
 757                         printf("warning: lcore %hhu is on socket %d with numa off \n",
 758                                 lcore, socketid);
 759                 }
 760         }
 761         return 0;
 762 }
 763
 764 static int
 765 check_port_config(const unsigned nb_ports)
 766 {
 767         unsigned portid;
 768         uint16_t i;
 769
 770         for (i = 0; i < nb_lcore_params; ++i) {
 771                 portid = lcore_params[i].port_id;
 772                 if ((enabled_port_mask & (1 << portid)) == 0) {
 773                         printf("port %u is not enabled in port mask\n", portid);
 774                         return -1;
 775                 }
 776                 if (portid >= nb_ports) {
 777                         printf("port %u is not present on the board\n", portid);
 778                         return -1;
 779                 }
 780         }
 781         return 0;
 782 }
 783
 784 static uint8_t
 785 get_port_n_rx_queues(const uint8_t port)
 786 {
 787         int queue = -1;
 788         uint16_t i;
 789
 790         for (i = 0; i < nb_lcore_params; ++i) {
 791                 if (lcore_params[i].port_id == port && lcore_params[i].queue_id > queue)
 792                         queue = lcore_params[i].queue_id;
 793         }
 794         return (uint8_t)(++queue);
 795 }
 796
 797 static int
 798 init_lcore_rx_queues(void)
 799 {
 800         uint16_t i, nb_rx_queue;
 801         uint8_t lcore;
 802
 803         for (i = 0; i < nb_lcore_params; ++i) {
 804                 lcore = lcore_params[i].lcore_id;
 805                 nb_rx_queue = lcore_conf[lcore].n_rx_queue;
 806                 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
 807                         printf("error: too many queues (%u) for lcore: %u\n",
 808                                 (unsigned)nb_rx_queue + 1, (unsigned)lcore);
 809                         return -1;
 810                 } else {
 811                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
 812                                 lcore_params[i].port_id;
 813                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
 814                                 lcore_params[i].queue_id;
 815                         lcore_conf[lcore].n_rx_queue++;
 816                 }
 817         }
 818         return 0;
 819 }
 820
 821 /* display usage */
 822 static void
 823 print_usage(const char *prgname)
 824 {
 825         printf ("%s [EAL options] -- -p PORTMASK -P"
 826                 "  [--config (port,queue,lcore)[,(port,queue,lcore]]"
 827                 "  [--enable-jumbo [--max-pkt-len PKTLEN]]\n"
 828                 "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
 829                 "  -P : enable promiscuous mode\n"
 830                 "  --config (port,queue,lcore): rx queues configuration\n"
 831                 "  --no-numa: optional, disable numa awareness\n"
 832                 "  --enable-jumbo: enable jumbo frame"
 833                 " which max packet len is PKTLEN in decimal (64-9600)\n",
 834                 prgname);
 835 }
 836
 837 static int parse_max_pkt_len(const char *pktlen)
 838 {
 839         char *end = NULL;
 840         unsigned long len;
 841
 842         /* parse decimal string */
 843         len = strtoul(pktlen, &end, 10);
 844         if ((pktlen[0] == '\0') || (end == NULL) || (*end != '\0'))
 845                 return -1;
 846
 847         if (len == 0)
 848                 return -1;
 849
 850         return len;
 851 }
 852
 853 static int
 854 parse_portmask(const char *portmask)
 855 {
 856         char *end = NULL;
 857         unsigned long pm;
 858
 859         /* parse hexadecimal string */
 860         pm = strtoul(portmask, &end, 16);
 861         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
 862                 return -1;
 863
 864         if (pm == 0)
 865                 return -1;
 866
 867         return pm;
 868 }
 869
 870 static int
 871 parse_config(const char *q_arg)
 872 {
 873         char s[256];
 874         const char *p, *p0 = q_arg;
 875         char *end;
 876         enum fieldnames {
 877                 FLD_PORT = 0,
 878                 FLD_QUEUE,
 879                 FLD_LCORE,
 880                 _NUM_FLD
 881         };
 882         unsigned long int_fld[_NUM_FLD];
 883         char *str_fld[_NUM_FLD];
 884         int i;
 885         unsigned size;
 886
 887         nb_lcore_params = 0;
 888
 889         while ((p = strchr(p0,'(')) != NULL) {
 890                 ++p;
 891                 if((p0 = strchr(p,')')) == NULL)
 892                         return -1;
 893
 894                 size = p0 - p;
 895                 if(size >= sizeof(s))
 896                         return -1;
 897
 898                 rte_snprintf(s, sizeof(s), "%.*s", size, p);
 899                 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') != _NUM_FLD)
 900                         return -1;
 901                 for (i = 0; i < _NUM_FLD; i++){
 902                         errno = 0;
 903                         int_fld[i] = strtoul(str_fld[i], &end, 0);
 904                         if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
 905                                 return -1;
 906                 }
 907                 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
 908                         printf("exceeded max number of lcore params: %hu\n",
 909                                 nb_lcore_params);
 910                         return -1;
 911                 }
 912                 lcore_params_array[nb_lcore_params].port_id = (uint8_t)int_fld[FLD_PORT];
 913                 lcore_params_array[nb_lcore_params].queue_id = (uint8_t)int_fld[FLD_QUEUE];
 914                 lcore_params_array[nb_lcore_params].lcore_id = (uint8_t)int_fld[FLD_LCORE];
 915                 ++nb_lcore_params;
 916         }
 917         lcore_params = lcore_params_array;
 918         return 0;
 919 }
 920
 921 /* Parse the argument given in the command line of the application */
 922 static int
 923 parse_args(int argc, char **argv)
 924 {
 925         int opt, ret;
 926         char **argvopt;
 927         int option_index;
 928         char *prgname = argv[0];
 929         static struct option lgopts[] = {
 930                 {"config", 1, 0, 0},
 931                 {"no-numa", 0, 0, 0},
 932                 {"enable-jumbo", 0, 0, 0},
 933                 {NULL, 0, 0, 0}
 934         };
 935
 936         argvopt = argv;
 937
 938         while ((opt = getopt_long(argc, argvopt, "p:P",
 939                                 lgopts, &option_index)) != EOF) {
 940
 941                 switch (opt) {
 942                 /* portmask */
 943                 case 'p':
 944                         enabled_port_mask = parse_portmask(optarg);
 945                         if (enabled_port_mask == 0) {
 946                                 printf("invalid portmask\n");
 947                                 print_usage(prgname);
 948                                 return -1;
 949                         }
 950                         break;
 951                 case 'P':
 952                         printf("Promiscuous mode selected\n");
 953                         promiscuous_on = 1;
 954                         break;
 955
 956                 /* long options */
 957                 case 0:
 958                         if (!strncmp(lgopts[option_index].name, "config", 6)) {
 959                                 ret = parse_config(optarg);
 960                                 if (ret) {
 961                                         printf("invalid config\n");
 962                                         print_usage(prgname);
 963                                         return -1;
 964                                 }
 965                         }
 966
 967                         if (!strncmp(lgopts[option_index].name, "no-numa", 7)) {
 968                                 printf("numa is disabled \n");
 969                                 numa_on = 0;
 970                         }
 971
 972                         if (!strncmp(lgopts[option_index].name, "enable-jumbo", 12)) {
 973                                 struct option lenopts = {"max-pkt-len", required_argument, 0, 0};
 974
 975                                 printf("jumbo frame is enabled \n");
 976                                 port_conf.rxmode.jumbo_frame = 1;
 977
 978                                 /* if no max-pkt-len set, use the default value ETHER_MAX_LEN */
 979                                 if (0 == getopt_long(argc, argvopt, "", &lenopts, &option_index)) {
 980                                         ret = parse_max_pkt_len(optarg);
 981                                         if ((ret < 64) || (ret > MAX_JUMBO_PKT_LEN)){
 982                                                 printf("invalid packet length\n");
 983                                                 print_usage(prgname);
 984                                                 return -1;
 985                                         }
 986                                         port_conf.rxmode.max_rx_pkt_len = ret;
 987                                 }
 988                                 printf("set jumbo frame max packet length to %u\n",
 989                                                 (unsigned int)port_conf.rxmode.max_rx_pkt_len);
 990                         }
 991
 992                         break;
 993
 994                 default:
 995                         print_usage(prgname);
 996                         return -1;
 997                 }
 998         }
 999
1000         if (optind >= 0)
1001                 argv[optind-1] = prgname;
1002
1003         ret = optind-1;
1004         optind = 0; /* reset getopt lib */
1005         return ret;
1006 }
1007
1008 static void
1009 print_ethaddr(const char *name, const struct ether_addr *eth_addr)
1010 {
1011         printf ("%s%02X:%02X:%02X:%02X:%02X:%02X", name,
1012                 eth_addr->addr_bytes[0],
1013                 eth_addr->addr_bytes[1],
1014                 eth_addr->addr_bytes[2],
1015                 eth_addr->addr_bytes[3],
1016                 eth_addr->addr_bytes[4],
1017                 eth_addr->addr_bytes[5]);
1018 }
1019
1020 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1021 static void
1022 setup_hash(int socketid)
1023 {
1024         unsigned i;
1025         int ret;
1026         char s[64];
1027
1028         /* create ipv4 hash */
1029         rte_snprintf(s, sizeof(s), "ipv4_l3fwd_hash_%d", socketid);
1030         ipv4_l3fwd_hash_params.name = s;
1031         ipv4_l3fwd_hash_params.socket_id = socketid;
1032         ipv4_l3fwd_lookup_struct[socketid] = rte_hash_create(&ipv4_l3fwd_hash_params);
1033         if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
1034                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
1035                                 "socket %d\n", socketid);
1036
1037         /* create ipv6 hash */
1038         rte_snprintf(s, sizeof(s), "ipv6_l3fwd_hash_%d", socketid);
1039         ipv6_l3fwd_hash_params.name = s;
1040         ipv6_l3fwd_hash_params.socket_id = socketid;
1041         ipv6_l3fwd_lookup_struct[socketid] = rte_hash_create(&ipv6_l3fwd_hash_params);
1042         if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
1043                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
1044                                 "socket %d\n", socketid);
1045
1046
1047         /* populate the ipv4 hash */
1048         for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
1049                 ret = rte_hash_add_key (ipv4_l3fwd_lookup_struct[socketid],
1050                                 (void *) &ipv4_l3fwd_route_array[i].key);
1051                 if (ret < 0) {
1052                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
1053                                 "l3fwd hash on socket %d\n", i, socketid);
1054                 }
1055                 ipv4_l3fwd_out_if[ret] = ipv4_l3fwd_route_array[i].if_out;
1056                 printf("Hash: Adding key\n");
1057                 print_ipv4_key(ipv4_l3fwd_route_array[i].key);
1058         }
1059
1060         /* populate the ipv6 hash */
1061         for (i = 0; i < IPV6_L3FWD_NUM_ROUTES; i++) {
1062                 ret = rte_hash_add_key (ipv6_l3fwd_lookup_struct[socketid],
1063                                 (void *) &ipv6_l3fwd_route_array[i].key);
1064                 if (ret < 0) {
1065                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
1066                                 "l3fwd hash on socket %d\n", i, socketid);
1067                 }
1068                 ipv6_l3fwd_out_if[ret] = ipv6_l3fwd_route_array[i].if_out;
1069                 printf("Hash: Adding key\n");
1070                 print_ipv6_key(ipv6_l3fwd_route_array[i].key);
1071         }
1072 }
1073 #endif
1074
1075 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1076 static void
1077 setup_lpm(int socketid)
1078 {
1079         unsigned i;
1080         int ret;
1081         char s[64];
1082
1083         /* create the LPM table */
1084         rte_snprintf(s, sizeof(s), "IPV4_L3FWD_LPM_%d", socketid);
1085         ipv4_l3fwd_lookup_struct[socketid] = rte_lpm_create(s, socketid,
1086                                 IPV4_L3FWD_LPM_MAX_RULES, 0);
1087         if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
1088                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
1089                                 " on socket %d\n", socketid);
1090
1091         /* populate the LPM table */
1092         for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
1093                 ret = rte_lpm_add(ipv4_l3fwd_lookup_struct[socketid],
1094                         ipv4_l3fwd_route_array[i].ip,
1095                         ipv4_l3fwd_route_array[i].depth,
1096                         ipv4_l3fwd_route_array[i].if_out);
1097
1098                 if (ret < 0) {
1099                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
1100                                 "l3fwd LPM table on socket %d\n",
1101                                 i, socketid);
1102                 }
1103
1104                 printf("LPM: Adding route 0x%08x / %d (%d)\n",
1105                         (unsigned)ipv4_l3fwd_route_array[i].ip,
1106                         ipv4_l3fwd_route_array[i].depth,
1107                         ipv4_l3fwd_route_array[i].if_out);
1108         }
1109 }
1110 #endif
1111
1112 static int
1113 init_mem(unsigned nb_mbuf)
1114 {
1115         struct lcore_conf *qconf;
1116         int socketid;
1117         unsigned lcore_id;
1118         char s[64];
1119
1120         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1121                 if (rte_lcore_is_enabled(lcore_id) == 0)
1122                         continue;
1123
1124                 if (numa_on)
1125                         socketid = rte_lcore_to_socket_id(lcore_id);
1126                 else
1127                         socketid = 0;
1128
1129                 if (socketid >= NB_SOCKETS) {
1130                         rte_exit(EXIT_FAILURE, "Socket %d of lcore %u is out of range %d\n",
1131                                 socketid, lcore_id, NB_SOCKETS);
1132                 }
1133                 if (pktmbuf_pool[socketid] == NULL) {
1134                         rte_snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
1135                         pktmbuf_pool[socketid] =
1136                                 rte_mempool_create(s, nb_mbuf, MBUF_SIZE, MEMPOOL_CACHE_SIZE,
1137                                         sizeof(struct rte_pktmbuf_pool_private),
1138                                         rte_pktmbuf_pool_init, NULL,
1139                                         rte_pktmbuf_init, NULL,
1140                                         socketid, 0);
1141                         if (pktmbuf_pool[socketid] == NULL)
1142                                 rte_exit(EXIT_FAILURE,
1143                                                 "Cannot init mbuf pool on socket %d\n", socketid);
1144                         else
1145                                 printf("Allocated mbuf pool on socket %d\n", socketid);
1146
1147 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1148                         setup_lpm(socketid);
1149 #else
1150                         setup_hash(socketid);
1151 #endif
1152                 }
1153                 qconf = &lcore_conf[lcore_id];
1154                 qconf->ipv4_lookup_struct = ipv4_l3fwd_lookup_struct[socketid];
1155 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1156                 qconf->ipv6_lookup_struct = ipv6_l3fwd_lookup_struct[socketid];
1157 #endif
1158         }
1159         return 0;
1160 }
1161
1162 /* Check the link status of all ports in up to 9s, and print them finally */
1163 static void
1164 check_all_ports_link_status(uint8_t port_num, uint32_t port_mask)
1165 {
1166 #define CHECK_INTERVAL 100 /* 100ms */
1167 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
1168         uint8_t portid, count, all_ports_up, print_flag = 0;
1169         struct rte_eth_link link;
1170
1171         printf("\nChecking link status");
1172         fflush(stdout);
1173         for (count = 0; count <= MAX_CHECK_TIME; count++) {
1174                 all_ports_up = 1;
1175                 for (portid = 0; portid < port_num; portid++) {
1176                         if ((port_mask & (1 << portid)) == 0)
1177                                 continue;
1178                         memset(&link, 0, sizeof(link));
1179                         rte_eth_link_get_nowait(portid, &link);
1180                         /* print link status if flag set */
1181                         if (print_flag == 1) {
1182                                 if (link.link_status)
1183                                         printf("Port %d Link Up - speed %u "
1184                                                 "Mbps - %s\n", (uint8_t)portid,
1185                                                 (unsigned)link.link_speed,
1186                                 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
1187                                         ("full-duplex") : ("half-duplex\n"));
1188                                 else
1189                                         printf("Port %d Link Down\n",
1190                                                 (uint8_t)portid);
1191                                 continue;
1192                         }
1193                         /* clear all_ports_up flag if any link down */
1194                         if (link.link_status == 0) {
1195                                 all_ports_up = 0;
1196                                 break;
1197                         }
1198                 }
1199                 /* after finally printing all link status, get out */
1200                 if (print_flag == 1)
1201                         break;
1202
1203                 if (all_ports_up == 0) {
1204                         printf(".");
1205                         fflush(stdout);
1206                         rte_delay_ms(CHECK_INTERVAL);
1207                 }
1208
1209                 /* set the print_flag if all ports up or timeout */
1210                 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
1211                         print_flag = 1;
1212                         printf("done\n");
1213                 }
1214         }
1215 }
1216
1217 int
1218 MAIN(int argc, char **argv)
1219 {
1220         struct lcore_conf *qconf;
1221         int ret;
1222         unsigned nb_ports;
1223         uint16_t queueid;
1224         unsigned lcore_id;
1225         uint32_t n_tx_queue, nb_lcores;
1226         uint8_t portid, nb_rx_queue, queue, socketid;
1227
1228         /* init EAL */
1229         ret = rte_eal_init(argc, argv);
1230         if (ret < 0)
1231                 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
1232         argc -= ret;
1233         argv += ret;
1234
1235         /* parse application arguments (after the EAL ones) */
1236         ret = parse_args(argc, argv);
1237         if (ret < 0)
1238                 rte_exit(EXIT_FAILURE, "Invalid L3FWD parameters\n");
1239
1240         if (check_lcore_params() < 0)
1241                 rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
1242
1243         ret = init_lcore_rx_queues();
1244         if (ret < 0)
1245                 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
1246
1247
1248         /* init driver(s) */
1249         if (rte_pmd_init_all() < 0)
1250                 rte_exit(EXIT_FAILURE, "Cannot init pmd\n");
1251
1252         if (rte_eal_pci_probe() < 0)
1253                 rte_exit(EXIT_FAILURE, "Cannot probe PCI\n");
1254
1255         nb_ports = rte_eth_dev_count();
1256         if (nb_ports > MAX_PORTS)
1257                 nb_ports = MAX_PORTS;
1258
1259         if (check_port_config(nb_ports) < 0)
1260                 rte_exit(EXIT_FAILURE, "check_port_config failed\n");
1261
1262         nb_lcores = rte_lcore_count();
1263
1264         /* initialize all ports */
1265         for (portid = 0; portid < nb_ports; portid++) {
1266                 /* skip ports that are not enabled */
1267                 if ((enabled_port_mask & (1 << portid)) == 0) {
1268                         printf("\nSkipping disabled port %d\n", portid);
1269                         continue;
1270                 }
1271
1272                 /* init port */
1273                 printf("Initializing port %d ... ", portid );
1274                 fflush(stdout);
1275
1276                 nb_rx_queue = get_port_n_rx_queues(portid);
1277                 n_tx_queue = nb_lcores;
1278                 if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
1279                         n_tx_queue = MAX_TX_QUEUE_PER_PORT;
1280                 printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
1281                         nb_rx_queue, (unsigned)n_tx_queue );
1282                 ret = rte_eth_dev_configure(portid, nb_rx_queue,
1283                                         (uint16_t)n_tx_queue, &port_conf);
1284                 if (ret < 0)
1285                         rte_exit(EXIT_FAILURE, "Cannot configure device: err=%d, port=%d\n",
1286                                 ret, portid);
1287
1288                 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
1289                 print_ethaddr(" Address:", &ports_eth_addr[portid]);
1290                 printf(", ");
1291
1292                 /* init memory */
1293                 ret = init_mem(NB_MBUF);
1294                 if (ret < 0)
1295                         rte_exit(EXIT_FAILURE, "init_mem failed\n");
1296
1297                 /* init one TX queue per couple (lcore,port) */
1298                 queueid = 0;
1299                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1300                         if (rte_lcore_is_enabled(lcore_id) == 0)
1301                                 continue;
1302
1303                         if (numa_on)
1304                                 socketid = (uint8_t)rte_lcore_to_socket_id(lcore_id);
1305                         else
1306                                 socketid = 0;
1307
1308                         printf("txq=%u,%d,%d ", lcore_id, queueid, socketid);
1309                         fflush(stdout);
1310                         ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
1311                                                      socketid, &tx_conf);
1312                         if (ret < 0)
1313                                 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: err=%d, "
1314                                         "port=%d\n", ret, portid);
1315
1316                         qconf = &lcore_conf[lcore_id];
1317                         qconf->tx_queue_id[portid] = queueid;
1318                         queueid++;
1319                 }
1320                 printf("\n");
1321         }
1322
1323         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1324                 if (rte_lcore_is_enabled(lcore_id) == 0)
1325                         continue;
1326                 qconf = &lcore_conf[lcore_id];
1327                 printf("\nInitializing rx queues on lcore %u ... ", lcore_id );
1328                 fflush(stdout);
1329                 /* init RX queues */
1330                 for(queue = 0; queue < qconf->n_rx_queue; ++queue) {
1331                         portid = qconf->rx_queue_list[queue].port_id;
1332                         queueid = qconf->rx_queue_list[queue].queue_id;
1333
1334                         if (numa_on)
1335                                 socketid = (uint8_t)rte_lcore_to_socket_id(lcore_id);
1336                         else
1337                                 socketid = 0;
1338
1339                         printf("rxq=%d,%d,%d ", portid, queueid, socketid);
1340                         fflush(stdout);
1341
1342                         ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
1343                                         socketid, &rx_conf, pktmbuf_pool[socketid]);
1344                         if (ret < 0)
1345                                 rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: err=%d,"
1346                                                 "port=%d\n", ret, portid);
1347                 }
1348         }
1349
1350         printf("\n");
1351
1352         /* start ports */
1353         for (portid = 0; portid < nb_ports; portid++) {
1354                 if ((enabled_port_mask & (1 << portid)) == 0) {
1355                         continue;
1356                 }
1357                 /* Start device */
1358                 ret = rte_eth_dev_start(portid);
1359                 if (ret < 0)
1360                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%d\n",
1361                                 ret, portid);
1362
1363                 /*
1364                  * If enabled, put device in promiscuous mode.
1365                  * This allows IO forwarding mode to forward packets
1366                  * to itself through 2 cross-connected  ports of the
1367                  * target machine.
1368                  */
1369                 if (promiscuous_on)
1370                         rte_eth_promiscuous_enable(portid);
1371         }
1372
1373         check_all_ports_link_status((uint8_t)nb_ports, enabled_port_mask);
1374
1375         /* launch per-lcore init on every lcore */
1376         rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
1377         RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1378                 if (rte_eal_wait_lcore(lcore_id) < 0)
1379                         return -1;
1380         }
1381
1382         return 0;
1383 }