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