examples/l3fwd-vf/main.c

   1 /*-
   2  *   BSD LICENSE
   3  *
   4  *   Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
   5  *   All rights reserved.
   6  *
   7  *   Redistribution and use in source and binary forms, with or without
   8  *   modification, are permitted provided that the following conditions
   9  *   are met:
  10  *
  11  *     * Redistributions of source code must retain the above copyright
  12  *       notice, this list of conditions and the following disclaimer.
  13  *     * Redistributions in binary form must reproduce the above copyright
  14  *       notice, this list of conditions and the following disclaimer in
  15  *       the documentation and/or other materials provided with the
  16  *       distribution.
  17  *     * Neither the name of Intel Corporation nor the names of its
  18  *       contributors may be used to endorse or promote products derived
  19  *       from this software without specific prior written permission.
  20  *
  21  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  22  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  23  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
  24  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  25  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  26  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
  27  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  28  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  29  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  30  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
  31  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  32  */
  33
  34 #include <stdio.h>
  35 #include <stdlib.h>
  36 #include <stdint.h>
  37 #include <inttypes.h>
  38 #include <sys/types.h>
  39 #include <string.h>
  40 #include <sys/queue.h>
  41 #include <stdarg.h>
  42 #include <errno.h>
  43 #include <getopt.h>
  44 #include <signal.h>
  45
  46 #include <rte_common.h>
  47 #include <rte_byteorder.h>
  48 #include <rte_log.h>
  49 #include <rte_memory.h>
  50 #include <rte_memcpy.h>
  51 #include <rte_eal.h>
  52 #include <rte_launch.h>
  53 #include <rte_atomic.h>
  54 #include <rte_spinlock.h>
  55 #include <rte_cycles.h>
  56 #include <rte_prefetch.h>
  57 #include <rte_lcore.h>
  58 #include <rte_per_lcore.h>
  59 #include <rte_branch_prediction.h>
  60 #include <rte_interrupts.h>
  61 #include <rte_pci.h>
  62 #include <rte_random.h>
  63 #include <rte_debug.h>
  64 #include <rte_ether.h>
  65 #include <rte_ethdev.h>
  66 #include <rte_mempool.h>
  67 #include <rte_mbuf.h>
  68 #include <rte_ip.h>
  69 #include <rte_tcp.h>
  70 #include <rte_udp.h>
  71 #include <rte_string_fns.h>
  72
  73 #define APP_LOOKUP_EXACT_MATCH          0
  74 #define APP_LOOKUP_LPM                  1
  75 #define DO_RFC_1812_CHECKS
  76
  77 //#define APP_LOOKUP_METHOD             APP_LOOKUP_EXACT_MATCH
  78 #ifndef APP_LOOKUP_METHOD
  79 #define APP_LOOKUP_METHOD             APP_LOOKUP_LPM
  80 #endif
  81
  82 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
  83 #include <rte_hash.h>
  84 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
  85 #include <rte_lpm.h>
  86 #else
  87 #error "APP_LOOKUP_METHOD set to incorrect value"
  88 #endif
  89
  90 #define RTE_LOGTYPE_L3FWD RTE_LOGTYPE_USER1
  91
  92 #define MEMPOOL_CACHE_SIZE 256
  93
  94 /*
  95  * This expression is used to calculate the number of mbufs needed depending on user input, taking
  96  *  into account memory for rx and tx hardware rings, cache per lcore and mtable per port per lcore.
  97  *  RTE_MAX is used to ensure that NB_MBUF never goes below a minimum value of 8192
  98  */
  99
 100 #define NB_MBUF RTE_MAX (                                               \
 101                                 (nb_ports*nb_rx_queue*nb_rxd +          \
 102                                 nb_ports*nb_lcores*MAX_PKT_BURST +      \
 103                                 nb_ports*n_tx_queue*nb_txd +            \
 104                                 nb_lcores*MEMPOOL_CACHE_SIZE),          \
 105                                 (unsigned)8192)
 106
 107 /*
 108  * RX and TX Prefetch, Host, and Write-back threshold values should be
 109  * carefully set for optimal performance. Consult the network
 110  * controller's datasheet and supporting DPDK documentation for guidance
 111  * on how these parameters should be set.
 112  */
 113 #define RX_PTHRESH 8 /**< Default values of RX prefetch threshold reg. */
 114 #define RX_HTHRESH 8 /**< Default values of RX host threshold reg. */
 115 #define RX_WTHRESH 4 /**< Default values of RX write-back threshold reg. */
 116
 117 /*
 118  * These default values are optimized for use with the Intel(R) 82599 10 GbE
 119  * Controller and the DPDK ixgbe PMD. Consider using other values for other
 120  * network controllers and/or network drivers.
 121  */
 122 #define TX_PTHRESH 36 /**< Default values of TX prefetch threshold reg. */
 123 #define TX_HTHRESH 0  /**< Default values of TX host threshold reg. */
 124 #define TX_WTHRESH 0  /**< Default values of TX write-back threshold reg. */
 125
 126 #define MAX_PKT_BURST 32
 127 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
 128
 129 #define NB_SOCKETS 8
 130
 131 #define SOCKET0 0
 132
 133 /* Configure how many packets ahead to prefetch, when reading packets */
 134 #define PREFETCH_OFFSET 3
 135
 136 /*
 137  * Configurable number of RX/TX ring descriptors
 138  */
 139 #define RTE_TEST_RX_DESC_DEFAULT 128
 140 #define RTE_TEST_TX_DESC_DEFAULT 512
 141 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
 142 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
 143
 144 /* ethernet addresses of ports */
 145 static struct ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
 146
 147 /* mask of enabled ports */
 148 static uint32_t enabled_port_mask = 0;
 149 static int numa_on = 1; /**< NUMA is enabled by default. */
 150
 151 struct mbuf_table {
 152         uint16_t len;
 153         struct rte_mbuf *m_table[MAX_PKT_BURST];
 154 };
 155
 156 struct lcore_rx_queue {
 157         uint16_t port_id;
 158         uint8_t queue_id;
 159 } __rte_cache_aligned;
 160
 161 #define MAX_RX_QUEUE_PER_LCORE 16
 162 #define MAX_TX_QUEUE_PER_PORT 1
 163 #define MAX_RX_QUEUE_PER_PORT 1
 164
 165 #define MAX_LCORE_PARAMS 1024
 166 struct lcore_params {
 167         uint16_t port_id;
 168         uint8_t queue_id;
 169         uint8_t lcore_id;
 170 } __rte_cache_aligned;
 171
 172 static struct lcore_params lcore_params_array[MAX_LCORE_PARAMS];
 173 static struct lcore_params lcore_params_array_default[] = {
 174         {0, 0, 2},
 175         {0, 1, 2},
 176         {0, 2, 2},
 177         {1, 0, 2},
 178         {1, 1, 2},
 179         {1, 2, 2},
 180         {2, 0, 2},
 181         {3, 0, 3},
 182         {3, 1, 3},
 183 };
 184
 185 static struct lcore_params * lcore_params = lcore_params_array_default;
 186 static uint16_t nb_lcore_params = sizeof(lcore_params_array_default) /
 187                                 sizeof(lcore_params_array_default[0]);
 188
 189 static struct rte_eth_conf port_conf = {
 190         .rxmode = {
 191                 .mq_mode        = ETH_MQ_RX_RSS,
 192                 .max_rx_pkt_len = ETHER_MAX_LEN,
 193                 .split_hdr_size = 0,
 194                 .header_split   = 0, /**< Header Split disabled */
 195                 .hw_ip_checksum = 1, /**< IP checksum offload enabled */
 196                 .hw_vlan_filter = 0, /**< VLAN filtering disabled */
 197                 .jumbo_frame    = 0, /**< Jumbo Frame Support disabled */
 198                 .hw_strip_crc   = 1, /**< CRC stripped by hardware */
 199         },
 200         .rx_adv_conf = {
 201                 .rss_conf = {
 202                         .rss_key = NULL,
 203                         .rss_hf = ETH_RSS_IP,
 204                 },
 205         },
 206         .txmode = {
 207                 .mq_mode = ETH_MQ_TX_NONE,
 208         },
 209 };
 210
 211 static struct rte_mempool * pktmbuf_pool[NB_SOCKETS];
 212
 213
 214 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 215
 216 #ifdef RTE_ARCH_X86
 217 #include <rte_hash_crc.h>
 218 #define DEFAULT_HASH_FUNC       rte_hash_crc
 219 #else
 220 #include <rte_jhash.h>
 221 #define DEFAULT_HASH_FUNC       rte_jhash
 222 #endif
 223
 224 struct ipv4_5tuple {
 225         uint32_t ip_dst;
 226         uint32_t ip_src;
 227         uint16_t port_dst;
 228         uint16_t port_src;
 229         uint8_t proto;
 230 } __attribute__((__packed__));
 231
 232 struct l3fwd_route {
 233         struct ipv4_5tuple key;
 234         uint8_t if_out;
 235 };
 236
 237 static struct l3fwd_route l3fwd_route_array[] = {
 238         {{IPv4(100,10,0,1), IPv4(200,10,0,1), 101, 11, IPPROTO_TCP}, 0},
 239         {{IPv4(100,20,0,2), IPv4(200,20,0,2), 102, 12, IPPROTO_TCP}, 1},
 240         {{IPv4(100,30,0,3), IPv4(200,30,0,3), 103, 13, IPPROTO_TCP}, 2},
 241         {{IPv4(100,40,0,4), IPv4(200,40,0,4), 104, 14, IPPROTO_TCP}, 3},
 242 };
 243
 244 typedef struct rte_hash lookup_struct_t;
 245 static lookup_struct_t *l3fwd_lookup_struct[NB_SOCKETS];
 246
 247 #define L3FWD_HASH_ENTRIES      1024
 248 struct rte_hash_parameters l3fwd_hash_params = {
 249         .name = "l3fwd_hash_0",
 250         .entries = L3FWD_HASH_ENTRIES,
 251         .key_len = sizeof(struct ipv4_5tuple),
 252         .hash_func = DEFAULT_HASH_FUNC,
 253         .hash_func_init_val = 0,
 254         .socket_id = SOCKET0,
 255 };
 256
 257 #define L3FWD_NUM_ROUTES \
 258         (sizeof(l3fwd_route_array) / sizeof(l3fwd_route_array[0]))
 259
 260 static uint8_t l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 261 #endif
 262
 263 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 264 struct l3fwd_route {
 265         uint32_t ip;
 266         uint8_t  depth;
 267         uint8_t  if_out;
 268 };
 269
 270 static struct l3fwd_route l3fwd_route_array[] = {
 271         {IPv4(1,1,1,0), 24, 0},
 272         {IPv4(2,1,1,0), 24, 1},
 273         {IPv4(3,1,1,0), 24, 2},
 274         {IPv4(4,1,1,0), 24, 3},
 275         {IPv4(5,1,1,0), 24, 4},
 276         {IPv4(6,1,1,0), 24, 5},
 277         {IPv4(7,1,1,0), 24, 6},
 278         {IPv4(8,1,1,0), 24, 7},
 279 };
 280
 281 #define L3FWD_NUM_ROUTES \
 282         (sizeof(l3fwd_route_array) / sizeof(l3fwd_route_array[0]))
 283
 284 #define L3FWD_LPM_MAX_RULES     1024
 285
 286 typedef struct rte_lpm lookup_struct_t;
 287 static lookup_struct_t *l3fwd_lookup_struct[NB_SOCKETS];
 288 #endif
 289
 290 struct lcore_conf {
 291         uint16_t n_rx_queue;
 292         struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
 293         uint16_t tx_queue_id;
 294         struct mbuf_table tx_mbufs[RTE_MAX_ETHPORTS];
 295         lookup_struct_t * lookup_struct;
 296 } __rte_cache_aligned;
 297
 298 static struct lcore_conf lcore_conf[RTE_MAX_LCORE];
 299 static rte_spinlock_t spinlock_conf[RTE_MAX_ETHPORTS] = {RTE_SPINLOCK_INITIALIZER};
 300 /* Send burst of packets on an output interface */
 301 static inline int
 302 send_burst(struct lcore_conf *qconf, uint16_t n, uint16_t port)
 303 {
 304         struct rte_mbuf **m_table;
 305         int ret;
 306         uint16_t queueid;
 307
 308         queueid = qconf->tx_queue_id;
 309         m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
 310
 311         rte_spinlock_lock(&spinlock_conf[port]);
 312         ret = rte_eth_tx_burst(port, queueid, m_table, n);
 313         rte_spinlock_unlock(&spinlock_conf[port]);
 314
 315         if (unlikely(ret < n)) {
 316                 do {
 317                         rte_pktmbuf_free(m_table[ret]);
 318                 } while (++ret < n);
 319         }
 320
 321         return 0;
 322 }
 323
 324 /* Enqueue a single packet, and send burst if queue is filled */
 325 static inline int
 326 send_single_packet(struct rte_mbuf *m, uint16_t port)
 327 {
 328         uint32_t lcore_id;
 329         uint16_t len;
 330         struct lcore_conf *qconf;
 331
 332         lcore_id = rte_lcore_id();
 333
 334         qconf = &lcore_conf[lcore_id];
 335         len = qconf->tx_mbufs[port].len;
 336         qconf->tx_mbufs[port].m_table[len] = m;
 337         len++;
 338
 339         /* enough pkts to be sent */
 340         if (unlikely(len == MAX_PKT_BURST)) {
 341                 send_burst(qconf, MAX_PKT_BURST, port);
 342                 len = 0;
 343         }
 344
 345         qconf->tx_mbufs[port].len = len;
 346         return 0;
 347 }
 348
 349 #ifdef DO_RFC_1812_CHECKS
 350 static inline int
 351 is_valid_ipv4_pkt(struct ipv4_hdr *pkt, uint32_t link_len)
 352 {
 353         /* From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2 */
 354         /*
 355          * 1. The packet length reported by the Link Layer must be large
 356          * enough to hold the minimum length legal IP datagram (20 bytes).
 357          */
 358         if (link_len < sizeof(struct ipv4_hdr))
 359                 return -1;
 360
 361         /* 2. The IP checksum must be correct. */
 362         /* this is checked in H/W */
 363
 364         /*
 365          * 3. The IP version number must be 4. If the version number is not 4
 366          * then the packet may be another version of IP, such as IPng or
 367          * ST-II.
 368          */
 369         if (((pkt->version_ihl) >> 4) != 4)
 370                 return -3;
 371         /*
 372          * 4. The IP header length field must be large enough to hold the
 373          * minimum length legal IP datagram (20 bytes = 5 words).
 374          */
 375         if ((pkt->version_ihl & 0xf) < 5)
 376                 return -4;
 377
 378         /*
 379          * 5. The IP total length field must be large enough to hold the IP
 380          * datagram header, whose length is specified in the IP header length
 381          * field.
 382          */
 383         if (rte_cpu_to_be_16(pkt->total_length) < sizeof(struct ipv4_hdr))
 384                 return -5;
 385
 386         return 0;
 387 }
 388 #endif
 389
 390 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 391 static void
 392 print_key(struct ipv4_5tuple key)
 393 {
 394         printf("IP dst = %08x, IP src = %08x, port dst = %d, port src = %d, proto = %d\n",
 395                (unsigned)key.ip_dst, (unsigned)key.ip_src, key.port_dst, key.port_src, key.proto);
 396 }
 397
 398 static inline uint16_t
 399 get_dst_port(struct ipv4_hdr *ipv4_hdr, uint16_t portid,
 400               lookup_struct_t *l3fwd_lookup_struct)
 401 {
 402         struct ipv4_5tuple key;
 403         struct tcp_hdr *tcp;
 404         struct udp_hdr *udp;
 405         int ret = 0;
 406
 407         key.ip_dst = rte_be_to_cpu_32(ipv4_hdr->dst_addr);
 408         key.ip_src = rte_be_to_cpu_32(ipv4_hdr->src_addr);
 409         key.proto = ipv4_hdr->next_proto_id;
 410
 411         switch (ipv4_hdr->next_proto_id) {
 412         case IPPROTO_TCP:
 413                 tcp = (struct tcp_hdr *)((unsigned char *) ipv4_hdr +
 414                                         sizeof(struct ipv4_hdr));
 415                 key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
 416                 key.port_src = rte_be_to_cpu_16(tcp->src_port);
 417                 break;
 418
 419         case IPPROTO_UDP:
 420                 udp = (struct udp_hdr *)((unsigned char *) ipv4_hdr +
 421                                         sizeof(struct ipv4_hdr));
 422                 key.port_dst = rte_be_to_cpu_16(udp->dst_port);
 423                 key.port_src = rte_be_to_cpu_16(udp->src_port);
 424                 break;
 425
 426         default:
 427                 key.port_dst = 0;
 428                 key.port_src = 0;
 429         }
 430
 431         /* Find destination port */
 432         ret = rte_hash_lookup(l3fwd_lookup_struct, (const void *)&key);
 433         return ((ret < 0) ? portid : l3fwd_out_if[ret]);
 434 }
 435 #endif
 436
 437 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 438 static inline uint32_t
 439 get_dst_port(struct ipv4_hdr *ipv4_hdr, uint16_t portid,
 440               lookup_struct_t *l3fwd_lookup_struct)
 441 {
 442         uint32_t next_hop;
 443
 444         return ((rte_lpm_lookup(l3fwd_lookup_struct,
 445                 rte_be_to_cpu_32(ipv4_hdr->dst_addr), &next_hop) == 0) ?
 446                 next_hop : portid);
 447 }
 448 #endif
 449
 450 static inline void
 451 l3fwd_simple_forward(struct rte_mbuf *m, uint16_t portid,
 452                       lookup_struct_t *l3fwd_lookup_struct)
 453 {
 454         struct ether_hdr *eth_hdr;
 455         struct ipv4_hdr *ipv4_hdr;
 456         void *tmp;
 457         uint16_t dst_port;
 458
 459         eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
 460
 461         ipv4_hdr = rte_pktmbuf_mtod_offset(m, struct ipv4_hdr *,
 462                                            sizeof(struct ether_hdr));
 463
 464 #ifdef DO_RFC_1812_CHECKS
 465         /* Check to make sure the packet is valid (RFC1812) */
 466         if (is_valid_ipv4_pkt(ipv4_hdr, m->pkt_len) < 0) {
 467                 rte_pktmbuf_free(m);
 468                 return;
 469         }
 470 #endif
 471
 472         dst_port = get_dst_port(ipv4_hdr, portid, l3fwd_lookup_struct);
 473         if (dst_port >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port) == 0)
 474                 dst_port = portid;
 475
 476         /* 02:00:00:00:00:xx */
 477         tmp = &eth_hdr->d_addr.addr_bytes[0];
 478         *((uint64_t *)tmp) = 0x000000000002 + ((uint64_t)dst_port << 40);
 479
 480 #ifdef DO_RFC_1812_CHECKS
 481         /* Update time to live and header checksum */
 482         --(ipv4_hdr->time_to_live);
 483         ++(ipv4_hdr->hdr_checksum);
 484 #endif
 485
 486         /* src addr */
 487         ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
 488
 489         send_single_packet(m, dst_port);
 490
 491 }
 492
 493 /* main processing loop */
 494 static int
 495 main_loop(__attribute__((unused)) void *dummy)
 496 {
 497         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
 498         unsigned lcore_id;
 499         uint64_t prev_tsc, diff_tsc, cur_tsc;
 500         int i, j, nb_rx;
 501         uint8_t queueid;
 502         uint16_t portid;
 503         struct lcore_conf *qconf;
 504         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
 505
 506         prev_tsc = 0;
 507
 508         lcore_id = rte_lcore_id();
 509         qconf = &lcore_conf[lcore_id];
 510
 511         if (qconf->n_rx_queue == 0) {
 512                 RTE_LOG(INFO, L3FWD, "lcore %u has nothing to do\n", lcore_id);
 513                 return 0;
 514         }
 515
 516         RTE_LOG(INFO, L3FWD, "entering main loop on lcore %u\n", lcore_id);
 517
 518         for (i = 0; i < qconf->n_rx_queue; i++) {
 519
 520                 portid = qconf->rx_queue_list[i].port_id;
 521                 queueid = qconf->rx_queue_list[i].queue_id;
 522                 RTE_LOG(INFO, L3FWD, " --lcoreid=%u portid=%u rxqueueid=%hhu\n",
 523                 lcore_id, portid, queueid);
 524         }
 525
 526         while (1) {
 527
 528                 cur_tsc = rte_rdtsc();
 529
 530                 /*
 531                  * TX burst queue drain
 532                  */
 533                 diff_tsc = cur_tsc - prev_tsc;
 534                 if (unlikely(diff_tsc > drain_tsc)) {
 535
 536                         /*
 537                          * This could be optimized (use queueid instead of
 538                          * portid), but it is not called so often
 539                          */
 540                         for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
 541                                 if (qconf->tx_mbufs[portid].len == 0)
 542                                         continue;
 543                                 send_burst(&lcore_conf[lcore_id],
 544                                         qconf->tx_mbufs[portid].len,
 545                                         portid);
 546                                 qconf->tx_mbufs[portid].len = 0;
 547                         }
 548
 549                         prev_tsc = cur_tsc;
 550                 }
 551
 552                 /*
 553                  * Read packet from RX queues
 554                  */
 555                 for (i = 0; i < qconf->n_rx_queue; ++i) {
 556
 557                         portid = qconf->rx_queue_list[i].port_id;
 558                         queueid = qconf->rx_queue_list[i].queue_id;
 559                         nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst, MAX_PKT_BURST);
 560
 561                         /* Prefetch first packets */
 562                         for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
 563                                 rte_prefetch0(rte_pktmbuf_mtod(
 564                                                 pkts_burst[j], void *));
 565                         }
 566
 567                         /* Prefetch and forward already prefetched packets */
 568                         for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
 569                                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
 570                                                 j + PREFETCH_OFFSET], void *));
 571                                 l3fwd_simple_forward(pkts_burst[j], portid, qconf->lookup_struct);
 572                         }
 573
 574                         /* Forward remaining prefetched packets */
 575                         for (; j < nb_rx; j++) {
 576                                 l3fwd_simple_forward(pkts_burst[j], portid, qconf->lookup_struct);
 577                         }
 578                 }
 579         }
 580 }
 581
 582 static int
 583 check_lcore_params(void)
 584 {
 585         uint8_t queue, lcore;
 586         uint16_t i;
 587         int socketid;
 588
 589         for (i = 0; i < nb_lcore_params; ++i) {
 590                 queue = lcore_params[i].queue_id;
 591                 if (queue >= MAX_RX_QUEUE_PER_PORT) {
 592                         printf("invalid queue number: %hhu\n", queue);
 593                         return -1;
 594                 }
 595                 lcore = lcore_params[i].lcore_id;
 596                 if (!rte_lcore_is_enabled(lcore)) {
 597                         printf("error: lcore %hhu is not enabled in lcore mask\n", lcore);
 598                         return -1;
 599                 }
 600                 if ((socketid = rte_lcore_to_socket_id(lcore) != 0) &&
 601                         (numa_on == 0)) {
 602                         printf("warning: lcore %hhu is on socket %d with numa off \n",
 603                                 lcore, socketid);
 604                 }
 605         }
 606         return 0;
 607 }
 608
 609 static int
 610 check_port_config(const unsigned nb_ports)
 611 {
 612         unsigned portid;
 613         uint16_t i;
 614
 615         for (i = 0; i < nb_lcore_params; ++i) {
 616                 portid = lcore_params[i].port_id;
 617                 if ((enabled_port_mask & (1 << portid)) == 0) {
 618                         printf("port %u is not enabled in port mask\n", portid);
 619                         return -1;
 620                 }
 621                 if (portid >= nb_ports) {
 622                         printf("port %u is not present on the board\n", portid);
 623                         return -1;
 624                 }
 625         }
 626         return 0;
 627 }
 628
 629 static uint8_t
 630 get_port_n_rx_queues(const uint16_t port)
 631 {
 632         int queue = -1;
 633         uint16_t i;
 634
 635         for (i = 0; i < nb_lcore_params; ++i) {
 636                 if (lcore_params[i].port_id == port && lcore_params[i].queue_id > queue)
 637                         queue = lcore_params[i].queue_id;
 638         }
 639         return (uint8_t)(++queue);
 640 }
 641
 642 static int
 643 init_lcore_rx_queues(void)
 644 {
 645         uint16_t i, nb_rx_queue;
 646         uint8_t lcore;
 647
 648         for (i = 0; i < nb_lcore_params; ++i) {
 649                 lcore = lcore_params[i].lcore_id;
 650                 nb_rx_queue = lcore_conf[lcore].n_rx_queue;
 651                 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
 652                         printf("error: too many queues (%u) for lcore: %u\n",
 653                                 (unsigned)nb_rx_queue + 1, (unsigned)lcore);
 654                         return -1;
 655                 } else {
 656                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
 657                                 lcore_params[i].port_id;
 658                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
 659                                 lcore_params[i].queue_id;
 660                         lcore_conf[lcore].n_rx_queue++;
 661                 }
 662         }
 663         return 0;
 664 }
 665
 666 /* display usage */
 667 static void
 668 print_usage(const char *prgname)
 669 {
 670         printf ("%s [EAL options] -- -p PORTMASK"
 671                 "  [--config (port,queue,lcore)[,(port,queue,lcore]]\n"
 672                 "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
 673                 "  --config (port,queue,lcore): rx queues configuration\n"
 674                 "  --no-numa: optional, disable numa awareness\n",
 675                 prgname);
 676 }
 677
 678 /* Custom handling of signals to handle process terminal */
 679 static void
 680 signal_handler(int signum)
 681 {
 682         uint16_t portid;
 683         uint16_t nb_ports = rte_eth_dev_count();
 684
 685         /* When we receive a SIGINT signal */
 686         if (signum == SIGINT) {
 687                 for (portid = 0; portid < nb_ports; portid++) {
 688                         /* skip ports that are not enabled */
 689                         if ((enabled_port_mask & (1 << portid)) == 0)
 690                                 continue;
 691                         rte_eth_dev_close(portid);
 692                 }
 693         }
 694         rte_exit(EXIT_SUCCESS, "\n User forced exit\n");
 695 }
 696 static int
 697 parse_portmask(const char *portmask)
 698 {
 699         char *end = NULL;
 700         unsigned long pm;
 701
 702         /* parse hexadecimal string */
 703         pm = strtoul(portmask, &end, 16);
 704         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
 705                 return -1;
 706
 707         if (pm == 0)
 708                 return -1;
 709
 710         return pm;
 711 }
 712
 713 static int
 714 parse_config(const char *q_arg)
 715 {
 716         char s[256];
 717         const char *p, *p0 = q_arg;
 718         char *end;
 719         enum fieldnames {
 720                 FLD_PORT = 0,
 721                 FLD_QUEUE,
 722                 FLD_LCORE,
 723                 _NUM_FLD
 724         };
 725         unsigned long int_fld[_NUM_FLD];
 726         char *str_fld[_NUM_FLD];
 727         int i;
 728         unsigned size;
 729
 730         nb_lcore_params = 0;
 731
 732         while ((p = strchr(p0,'(')) != NULL) {
 733                 ++p;
 734                 if((p0 = strchr(p,')')) == NULL)
 735                         return -1;
 736
 737                 size = p0 - p;
 738                 if(size >= sizeof(s))
 739                         return -1;
 740
 741                 snprintf(s, sizeof(s), "%.*s", size, p);
 742                 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') != _NUM_FLD)
 743                         return -1;
 744                 for (i = 0; i < _NUM_FLD; i++){
 745                         errno = 0;
 746                         int_fld[i] = strtoul(str_fld[i], &end, 0);
 747                         if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
 748                                 return -1;
 749                 }
 750                 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
 751                         printf("exceeded max number of lcore params: %hu\n",
 752                                 nb_lcore_params);
 753                         return -1;
 754                 }
 755                 lcore_params_array[nb_lcore_params].port_id = int_fld[FLD_PORT];
 756                 lcore_params_array[nb_lcore_params].queue_id = (uint8_t)int_fld[FLD_QUEUE];
 757                 lcore_params_array[nb_lcore_params].lcore_id = (uint8_t)int_fld[FLD_LCORE];
 758                 ++nb_lcore_params;
 759         }
 760         lcore_params = lcore_params_array;
 761         return 0;
 762 }
 763
 764 /* Parse the argument given in the command line of the application */
 765 static int
 766 parse_args(int argc, char **argv)
 767 {
 768         int opt, ret;
 769         char **argvopt;
 770         int option_index;
 771         char *prgname = argv[0];
 772         static struct option lgopts[] = {
 773                 {"config", 1, 0, 0},
 774                 {"no-numa", 0, 0, 0},
 775                 {NULL, 0, 0, 0}
 776         };
 777
 778         argvopt = argv;
 779
 780         while ((opt = getopt_long(argc, argvopt, "p:",
 781                                 lgopts, &option_index)) != EOF) {
 782
 783                 switch (opt) {
 784                 /* portmask */
 785                 case 'p':
 786                         enabled_port_mask = parse_portmask(optarg);
 787                         if (enabled_port_mask == 0) {
 788                                 printf("invalid portmask\n");
 789                                 print_usage(prgname);
 790                                 return -1;
 791                         }
 792                         break;
 793
 794                 /* long options */
 795                 case 0:
 796                         if (!strcmp(lgopts[option_index].name, "config")) {
 797                                 ret = parse_config(optarg);
 798                                 if (ret) {
 799                                         printf("invalid config\n");
 800                                         print_usage(prgname);
 801                                         return -1;
 802                                 }
 803                         }
 804
 805                         if (!strcmp(lgopts[option_index].name, "no-numa")) {
 806                                 printf("numa is disabled \n");
 807                                 numa_on = 0;
 808                         }
 809                         break;
 810
 811                 default:
 812                         print_usage(prgname);
 813                         return -1;
 814                 }
 815         }
 816
 817         if (optind >= 0)
 818                 argv[optind-1] = prgname;
 819
 820         ret = optind-1;
 821         optind = 1; /* reset getopt lib */
 822         return ret;
 823 }
 824
 825 static void
 826 print_ethaddr(const char *name, const struct ether_addr *eth_addr)
 827 {
 828         char buf[ETHER_ADDR_FMT_SIZE];
 829         ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr);
 830         printf("%s%s", name, buf);
 831 }
 832
 833 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 834 static void
 835 setup_hash(int socketid)
 836 {
 837         unsigned i;
 838         int ret;
 839         char s[64];
 840
 841         /* create  hashes */
 842         snprintf(s, sizeof(s), "l3fwd_hash_%d", socketid);
 843         l3fwd_hash_params.name = s;
 844         l3fwd_hash_params.socket_id = socketid;
 845         l3fwd_lookup_struct[socketid] = rte_hash_create(&l3fwd_hash_params);
 846         if (l3fwd_lookup_struct[socketid] == NULL)
 847                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
 848                                 "socket %d\n", socketid);
 849
 850         /* populate the hash */
 851         for (i = 0; i < L3FWD_NUM_ROUTES; i++) {
 852                 ret = rte_hash_add_key (l3fwd_lookup_struct[socketid],
 853                                 (void *) &l3fwd_route_array[i].key);
 854                 if (ret < 0) {
 855                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
 856                                 "l3fwd hash on socket %d\n", i, socketid);
 857                 }
 858                 l3fwd_out_if[ret] = l3fwd_route_array[i].if_out;
 859                 printf("Hash: Adding key\n");
 860                 print_key(l3fwd_route_array[i].key);
 861         }
 862 }
 863 #endif
 864
 865 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 866 static void
 867 setup_lpm(int socketid)
 868 {
 869         unsigned i;
 870         int ret;
 871         char s[64];
 872
 873         struct rte_lpm_config lpm_ipv4_config;
 874
 875         lpm_ipv4_config.max_rules = L3FWD_LPM_MAX_RULES;
 876         lpm_ipv4_config.number_tbl8s = 256;
 877         lpm_ipv4_config.flags = 0;
 878
 879         /* create the LPM table */
 880         snprintf(s, sizeof(s), "L3FWD_LPM_%d", socketid);
 881         l3fwd_lookup_struct[socketid] =
 882                         rte_lpm_create(s, socketid, &lpm_ipv4_config);
 883         if (l3fwd_lookup_struct[socketid] == NULL)
 884                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
 885                                 " on socket %d\n", socketid);
 886
 887         /* populate the LPM table */
 888         for (i = 0; i < L3FWD_NUM_ROUTES; i++) {
 889                 ret = rte_lpm_add(l3fwd_lookup_struct[socketid],
 890                         l3fwd_route_array[i].ip,
 891                         l3fwd_route_array[i].depth,
 892                         l3fwd_route_array[i].if_out);
 893
 894                 if (ret < 0) {
 895                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
 896                                 "l3fwd LPM table on socket %d\n",
 897                                 i, socketid);
 898                 }
 899
 900                 printf("LPM: Adding route 0x%08x / %d (%d)\n",
 901                         (unsigned)l3fwd_route_array[i].ip,
 902                         l3fwd_route_array[i].depth,
 903                         l3fwd_route_array[i].if_out);
 904         }
 905 }
 906 #endif
 907
 908 static int
 909 init_mem(unsigned nb_mbuf)
 910 {
 911         struct lcore_conf *qconf;
 912         int socketid;
 913         unsigned lcore_id;
 914         char s[64];
 915
 916         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
 917                 if (rte_lcore_is_enabled(lcore_id) == 0)
 918                         continue;
 919
 920                 if (numa_on)
 921                         socketid = rte_lcore_to_socket_id(lcore_id);
 922                 else
 923                         socketid = 0;
 924
 925                 if (socketid >= NB_SOCKETS) {
 926                         rte_exit(EXIT_FAILURE, "Socket %d of lcore %u is out of range %d\n",
 927                                 socketid, lcore_id, NB_SOCKETS);
 928                 }
 929                 if (pktmbuf_pool[socketid] == NULL) {
 930                         snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
 931                         pktmbuf_pool[socketid] = rte_pktmbuf_pool_create(s,
 932                                 nb_mbuf, MEMPOOL_CACHE_SIZE, 0,
 933                                 RTE_MBUF_DEFAULT_BUF_SIZE, socketid);
 934                         if (pktmbuf_pool[socketid] == NULL)
 935                                 rte_exit(EXIT_FAILURE, "Cannot init mbuf pool on socket %d\n", socketid);
 936                         else
 937                                 printf("Allocated mbuf pool on socket %d\n", socketid);
 938
 939 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 940                         setup_lpm(socketid);
 941 #else
 942                         setup_hash(socketid);
 943 #endif
 944                 }
 945                 qconf = &lcore_conf[lcore_id];
 946                 qconf->lookup_struct = l3fwd_lookup_struct[socketid];
 947         }
 948         return 0;
 949 }
 950
 951 int
 952 main(int argc, char **argv)
 953 {
 954         struct lcore_conf *qconf;
 955         struct rte_eth_dev_info dev_info;
 956         struct rte_eth_txconf *txconf;
 957         int ret;
 958         unsigned nb_ports;
 959         uint16_t queueid, portid;
 960         unsigned lcore_id;
 961         uint32_t nb_lcores;
 962         uint16_t n_tx_queue;
 963         uint8_t nb_rx_queue, queue, socketid;
 964
 965         signal(SIGINT, signal_handler);
 966         /* init EAL */
 967         ret = rte_eal_init(argc, argv);
 968         if (ret < 0)
 969                 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
 970         argc -= ret;
 971         argv += ret;
 972
 973         /* parse application arguments (after the EAL ones) */
 974         ret = parse_args(argc, argv);
 975         if (ret < 0)
 976                 rte_exit(EXIT_FAILURE, "Invalid L3FWD-VF parameters\n");
 977
 978         if (check_lcore_params() < 0)
 979                 rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
 980
 981         ret = init_lcore_rx_queues();
 982         if (ret < 0)
 983                 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
 984
 985         nb_ports = rte_eth_dev_count();
 986
 987         if (check_port_config(nb_ports) < 0)
 988                 rte_exit(EXIT_FAILURE, "check_port_config failed\n");
 989
 990         nb_lcores = rte_lcore_count();
 991
 992         /* initialize all ports */
 993         for (portid = 0; portid < nb_ports; portid++) {
 994                 /* skip ports that are not enabled */
 995                 if ((enabled_port_mask & (1 << portid)) == 0) {
 996                         printf("\nSkipping disabled port %d\n", portid);
 997                         continue;
 998                 }
 999
1000                 /* init port */
1001                 printf("Initializing port %d ... ", portid );
1002                 fflush(stdout);
1003
1004                 /* must always equal(=1) */
1005                 nb_rx_queue = get_port_n_rx_queues(portid);
1006                 n_tx_queue = MAX_TX_QUEUE_PER_PORT;
1007
1008                 printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
1009                         nb_rx_queue, (unsigned)1 );
1010                 ret = rte_eth_dev_configure(portid, nb_rx_queue, n_tx_queue, &port_conf);
1011                 if (ret < 0)
1012                         rte_exit(EXIT_FAILURE, "Cannot configure device: err=%d, port=%d\n",
1013                                 ret, portid);
1014
1015                 ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
1016                                                        &nb_txd);
1017                 if (ret < 0)
1018                         rte_exit(EXIT_FAILURE,
1019                                  "Cannot adjust number of descriptors: err=%d, port=%d\n",
1020                                  ret, portid);
1021
1022                 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
1023                 print_ethaddr(" Address:", &ports_eth_addr[portid]);
1024                 printf(", ");
1025
1026                 ret = init_mem(NB_MBUF);
1027                 if (ret < 0)
1028                         rte_exit(EXIT_FAILURE, "init_mem failed\n");
1029
1030                 /* init one TX queue */
1031                 socketid = (uint8_t)rte_lcore_to_socket_id(rte_get_master_lcore());
1032
1033                 printf("txq=%d,%d,%d ", portid, 0, socketid);
1034                 fflush(stdout);
1035
1036                 rte_eth_dev_info_get(portid, &dev_info);
1037                 txconf = &dev_info.default_txconf;
1038                 if (port_conf.rxmode.jumbo_frame)
1039                         txconf->txq_flags = 0;
1040                 ret = rte_eth_tx_queue_setup(portid, 0, nb_txd,
1041                                                  socketid, txconf);
1042                 if (ret < 0)
1043                         rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: err=%d, "
1044                                 "port=%d\n", ret, portid);
1045
1046                 printf("\n");
1047         }
1048
1049         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1050                 if (rte_lcore_is_enabled(lcore_id) == 0)
1051                         continue;
1052                 qconf = &lcore_conf[lcore_id];
1053                 qconf->tx_queue_id = 0;
1054
1055                 printf("\nInitializing rx queues on lcore %u ... ", lcore_id );
1056                 fflush(stdout);
1057                 /* init RX queues */
1058                 for(queue = 0; queue < qconf->n_rx_queue; ++queue) {
1059                         portid = qconf->rx_queue_list[queue].port_id;
1060                         queueid = qconf->rx_queue_list[queue].queue_id;
1061
1062                         if (numa_on)
1063                                 socketid = (uint8_t)rte_lcore_to_socket_id(lcore_id);
1064                         else
1065                                 socketid = 0;
1066
1067                         printf("rxq=%d,%d,%d ", portid, queueid, socketid);
1068                         fflush(stdout);
1069
1070                         ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
1071                                                 socketid, NULL,
1072                                                 pktmbuf_pool[socketid]);
1073                         if (ret < 0)
1074                                 rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: err=%d,"
1075                                                 "port=%d\n", ret, portid);
1076                 }
1077         }
1078         printf("\n");
1079
1080         /* start ports */
1081         for (portid = 0; portid < nb_ports; portid++) {
1082                 if ((enabled_port_mask & (1 << portid)) == 0) {
1083                         continue;
1084                 }
1085                 /* Start device */
1086                 ret = rte_eth_dev_start(portid);
1087                 if (ret < 0)
1088                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%d\n",
1089                                 ret, portid);
1090
1091                 printf("done: Port %d\n", portid);
1092
1093         }
1094
1095         /* launch per-lcore init on every lcore */
1096         rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
1097         RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1098                 if (rte_eal_wait_lcore(lcore_id) < 0)
1099                         return -1;
1100         }
1101
1102         return 0;
1103 }