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_memzone.h>
  52 #include <rte_eal.h>
  53 #include <rte_per_lcore.h>
  54 #include <rte_launch.h>
  55 #include <rte_atomic.h>
  56 #include <rte_spinlock.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 #define APP_LOOKUP_EXACT_MATCH          0
  77 #define APP_LOOKUP_LPM                  1
  78 #define DO_RFC_1812_CHECKS
  79
  80 //#define APP_LOOKUP_METHOD             APP_LOOKUP_EXACT_MATCH
  81 #ifndef APP_LOOKUP_METHOD
  82 #define APP_LOOKUP_METHOD             APP_LOOKUP_LPM
  83 #endif
  84
  85 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
  86 #include <rte_hash.h>
  87 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
  88 #include <rte_lpm.h>
  89 #else
  90 #error "APP_LOOKUP_METHOD set to incorrect value"
  91 #endif
  92
  93 #define RTE_LOGTYPE_L3FWD RTE_LOGTYPE_USER1
  94
  95 #define MEMPOOL_CACHE_SIZE 256
  96
  97 #define MBUF_DATA_SIZE (2048 + RTE_PKTMBUF_HEADROOM)
  98
  99 /*
 100  * This expression is used to calculate the number of mbufs needed depending on user input, taking
 101  *  into account memory for rx and tx hardware rings, cache per lcore and mtable per port per lcore.
 102  *  RTE_MAX is used to ensure that NB_MBUF never goes below a minimum value of 8192
 103  */
 104
 105 #define NB_MBUF RTE_MAX (                                                                                                                                       \
 106                                 (nb_ports*nb_rx_queue*RTE_TEST_RX_DESC_DEFAULT +                                                        \
 107                                 nb_ports*nb_lcores*MAX_PKT_BURST +                                                                                      \
 108                                 nb_ports*n_tx_queue*RTE_TEST_TX_DESC_DEFAULT +                                                          \
 109                                 nb_lcores*MEMPOOL_CACHE_SIZE),                                                                                          \
 110                                 (unsigned)8192)
 111
 112 /*
 113  * RX and TX Prefetch, Host, and Write-back threshold values should be
 114  * carefully set for optimal performance. Consult the network
 115  * controller's datasheet and supporting DPDK documentation for guidance
 116  * on how these parameters should be set.
 117  */
 118 #define RX_PTHRESH 8 /**< Default values of RX prefetch threshold reg. */
 119 #define RX_HTHRESH 8 /**< Default values of RX host threshold reg. */
 120 #define RX_WTHRESH 4 /**< Default values of RX write-back threshold reg. */
 121
 122 /*
 123  * These default values are optimized for use with the Intel(R) 82599 10 GbE
 124  * Controller and the DPDK ixgbe PMD. Consider using other values for other
 125  * network controllers and/or network drivers.
 126  */
 127 #define TX_PTHRESH 36 /**< Default values of TX prefetch threshold reg. */
 128 #define TX_HTHRESH 0  /**< Default values of TX host threshold reg. */
 129 #define TX_WTHRESH 0  /**< Default values of TX write-back threshold reg. */
 130
 131 #define MAX_PKT_BURST 32
 132 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
 133
 134 #define NB_SOCKETS 8
 135
 136 #define SOCKET0 0
 137
 138 /* Configure how many packets ahead to prefetch, when reading packets */
 139 #define PREFETCH_OFFSET 3
 140
 141 /*
 142  * Configurable number of RX/TX ring descriptors
 143  */
 144 #define RTE_TEST_RX_DESC_DEFAULT 128
 145 #define RTE_TEST_TX_DESC_DEFAULT 512
 146 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
 147 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
 148
 149 /* ethernet addresses of ports */
 150 static struct ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
 151
 152 /* mask of enabled ports */
 153 static uint32_t enabled_port_mask = 0;
 154 static int numa_on = 1; /**< NUMA is enabled by default. */
 155
 156 struct mbuf_table {
 157         uint16_t len;
 158         struct rte_mbuf *m_table[MAX_PKT_BURST];
 159 };
 160
 161 struct lcore_rx_queue {
 162         uint8_t port_id;
 163         uint8_t queue_id;
 164 } __rte_cache_aligned;
 165
 166 #define MAX_RX_QUEUE_PER_LCORE 16
 167 #define MAX_TX_QUEUE_PER_PORT 1
 168 #define MAX_RX_QUEUE_PER_PORT 1
 169
 170 #define MAX_LCORE_PARAMS 1024
 171 struct lcore_params {
 172         uint8_t port_id;
 173         uint8_t queue_id;
 174         uint8_t lcore_id;
 175 } __rte_cache_aligned;
 176
 177 static struct lcore_params lcore_params_array[MAX_LCORE_PARAMS];
 178 static struct lcore_params lcore_params_array_default[] = {
 179         {0, 0, 2},
 180         {0, 1, 2},
 181         {0, 2, 2},
 182         {1, 0, 2},
 183         {1, 1, 2},
 184         {1, 2, 2},
 185         {2, 0, 2},
 186         {3, 0, 3},
 187         {3, 1, 3},
 188 };
 189
 190 static struct lcore_params * lcore_params = lcore_params_array_default;
 191 static uint16_t nb_lcore_params = sizeof(lcore_params_array_default) /
 192                                 sizeof(lcore_params_array_default[0]);
 193
 194 static struct rte_eth_conf port_conf = {
 195         .rxmode = {
 196                 .mq_mode        = ETH_MQ_RX_RSS,
 197                 .max_rx_pkt_len = ETHER_MAX_LEN,
 198                 .split_hdr_size = 0,
 199                 .header_split   = 0, /**< Header Split disabled */
 200                 .hw_ip_checksum = 1, /**< IP checksum offload enabled */
 201                 .hw_vlan_filter = 0, /**< VLAN filtering disabled */
 202                 .jumbo_frame    = 0, /**< Jumbo Frame Support disabled */
 203                 .hw_strip_crc   = 0, /**< CRC stripped by hardware */
 204         },
 205         .rx_adv_conf = {
 206                 .rss_conf = {
 207                         .rss_key = NULL,
 208                         .rss_hf = ETH_RSS_IP,
 209                 },
 210         },
 211         .txmode = {
 212                 .mq_mode = ETH_MQ_TX_NONE,
 213         },
 214 };
 215
 216 static struct rte_mempool * pktmbuf_pool[NB_SOCKETS];
 217
 218
 219 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 220
 221 #ifdef RTE_MACHINE_CPUFLAG_SSE4_2
 222 #include <rte_hash_crc.h>
 223 #define DEFAULT_HASH_FUNC       rte_hash_crc
 224 #else
 225 #include <rte_jhash.h>
 226 #define DEFAULT_HASH_FUNC       rte_jhash
 227 #endif
 228
 229 struct ipv4_5tuple {
 230         uint32_t ip_dst;
 231         uint32_t ip_src;
 232         uint16_t port_dst;
 233         uint16_t port_src;
 234         uint8_t proto;
 235 } __attribute__((__packed__));
 236
 237 struct l3fwd_route {
 238         struct ipv4_5tuple key;
 239         uint8_t if_out;
 240 };
 241
 242 static struct l3fwd_route l3fwd_route_array[] = {
 243         {{IPv4(100,10,0,1), IPv4(200,10,0,1), 101, 11, IPPROTO_TCP}, 0},
 244         {{IPv4(100,20,0,2), IPv4(200,20,0,2), 102, 12, IPPROTO_TCP}, 1},
 245         {{IPv4(100,30,0,3), IPv4(200,30,0,3), 103, 13, IPPROTO_TCP}, 2},
 246         {{IPv4(100,40,0,4), IPv4(200,40,0,4), 104, 14, IPPROTO_TCP}, 3},
 247 };
 248
 249 typedef struct rte_hash lookup_struct_t;
 250 static lookup_struct_t *l3fwd_lookup_struct[NB_SOCKETS];
 251
 252 #define L3FWD_HASH_ENTRIES      1024
 253 struct rte_hash_parameters l3fwd_hash_params = {
 254         .name = "l3fwd_hash_0",
 255         .entries = L3FWD_HASH_ENTRIES,
 256         .bucket_entries = 4,
 257         .key_len = sizeof(struct ipv4_5tuple),
 258         .hash_func = DEFAULT_HASH_FUNC,
 259         .hash_func_init_val = 0,
 260         .socket_id = SOCKET0,
 261 };
 262
 263 #define L3FWD_NUM_ROUTES \
 264         (sizeof(l3fwd_route_array) / sizeof(l3fwd_route_array[0]))
 265
 266 static uint8_t l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 267 #endif
 268
 269 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 270 struct l3fwd_route {
 271         uint32_t ip;
 272         uint8_t  depth;
 273         uint8_t  if_out;
 274 };
 275
 276 static struct l3fwd_route l3fwd_route_array[] = {
 277         {IPv4(1,1,1,0), 24, 0},
 278         {IPv4(2,1,1,0), 24, 1},
 279         {IPv4(3,1,1,0), 24, 2},
 280         {IPv4(4,1,1,0), 24, 3},
 281         {IPv4(5,1,1,0), 24, 4},
 282         {IPv4(6,1,1,0), 24, 5},
 283         {IPv4(7,1,1,0), 24, 6},
 284         {IPv4(8,1,1,0), 24, 7},
 285 };
 286
 287 #define L3FWD_NUM_ROUTES \
 288         (sizeof(l3fwd_route_array) / sizeof(l3fwd_route_array[0]))
 289
 290 #define L3FWD_LPM_MAX_RULES     1024
 291
 292 typedef struct rte_lpm lookup_struct_t;
 293 static lookup_struct_t *l3fwd_lookup_struct[NB_SOCKETS];
 294 #endif
 295
 296 struct lcore_conf {
 297         uint16_t n_rx_queue;
 298         struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
 299         uint16_t tx_queue_id;
 300         struct mbuf_table tx_mbufs[RTE_MAX_ETHPORTS];
 301         lookup_struct_t * lookup_struct;
 302 } __rte_cache_aligned;
 303
 304 static struct lcore_conf lcore_conf[RTE_MAX_LCORE];
 305 static rte_spinlock_t spinlock_conf[RTE_MAX_ETHPORTS] = {RTE_SPINLOCK_INITIALIZER};
 306 /* Send burst of packets on an output interface */
 307 static inline int
 308 send_burst(struct lcore_conf *qconf, uint16_t n, uint8_t port)
 309 {
 310         struct rte_mbuf **m_table;
 311         int ret;
 312         uint16_t queueid;
 313
 314         queueid = qconf->tx_queue_id;
 315         m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
 316
 317         rte_spinlock_lock(&spinlock_conf[port]);
 318         ret = rte_eth_tx_burst(port, queueid, m_table, n);
 319         rte_spinlock_unlock(&spinlock_conf[port]);
 320
 321         if (unlikely(ret < n)) {
 322                 do {
 323                         rte_pktmbuf_free(m_table[ret]);
 324                 } while (++ret < n);
 325         }
 326
 327         return 0;
 328 }
 329
 330 /* Enqueue a single packet, and send burst if queue is filled */
 331 static inline int
 332 send_single_packet(struct rte_mbuf *m, uint8_t port)
 333 {
 334         uint32_t lcore_id;
 335         uint16_t len;
 336         struct lcore_conf *qconf;
 337
 338         lcore_id = rte_lcore_id();
 339
 340         qconf = &lcore_conf[lcore_id];
 341         len = qconf->tx_mbufs[port].len;
 342         qconf->tx_mbufs[port].m_table[len] = m;
 343         len++;
 344
 345         /* enough pkts to be sent */
 346         if (unlikely(len == MAX_PKT_BURST)) {
 347                 send_burst(qconf, MAX_PKT_BURST, port);
 348                 len = 0;
 349         }
 350
 351         qconf->tx_mbufs[port].len = len;
 352         return 0;
 353 }
 354
 355 #ifdef DO_RFC_1812_CHECKS
 356 static inline int
 357 is_valid_ipv4_pkt(struct ipv4_hdr *pkt, uint32_t link_len)
 358 {
 359         /* From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2 */
 360         /*
 361          * 1. The packet length reported by the Link Layer must be large
 362          * enough to hold the minimum length legal IP datagram (20 bytes).
 363          */
 364         if (link_len < sizeof(struct ipv4_hdr))
 365                 return -1;
 366
 367         /* 2. The IP checksum must be correct. */
 368         /* this is checked in H/W */
 369
 370         /*
 371          * 3. The IP version number must be 4. If the version number is not 4
 372          * then the packet may be another version of IP, such as IPng or
 373          * ST-II.
 374          */
 375         if (((pkt->version_ihl) >> 4) != 4)
 376                 return -3;
 377         /*
 378          * 4. The IP header length field must be large enough to hold the
 379          * minimum length legal IP datagram (20 bytes = 5 words).
 380          */
 381         if ((pkt->version_ihl & 0xf) < 5)
 382                 return -4;
 383
 384         /*
 385          * 5. The IP total length field must be large enough to hold the IP
 386          * datagram header, whose length is specified in the IP header length
 387          * field.
 388          */
 389         if (rte_cpu_to_be_16(pkt->total_length) < sizeof(struct ipv4_hdr))
 390                 return -5;
 391
 392         return 0;
 393 }
 394 #endif
 395
 396 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 397 static void
 398 print_key(struct ipv4_5tuple key)
 399 {
 400         printf("IP dst = %08x, IP src = %08x, port dst = %d, port src = %d, proto = %d\n",
 401                (unsigned)key.ip_dst, (unsigned)key.ip_src, key.port_dst, key.port_src, key.proto);
 402 }
 403
 404 static inline uint8_t
 405 get_dst_port(struct ipv4_hdr *ipv4_hdr,  uint8_t portid, lookup_struct_t * l3fwd_lookup_struct)
 406 {
 407         struct ipv4_5tuple key;
 408         struct tcp_hdr *tcp;
 409         struct udp_hdr *udp;
 410         int ret = 0;
 411
 412         key.ip_dst = rte_be_to_cpu_32(ipv4_hdr->dst_addr);
 413         key.ip_src = rte_be_to_cpu_32(ipv4_hdr->src_addr);
 414         key.proto = ipv4_hdr->next_proto_id;
 415
 416         switch (ipv4_hdr->next_proto_id) {
 417         case IPPROTO_TCP:
 418                 tcp = (struct tcp_hdr *)((unsigned char *) ipv4_hdr +
 419                                         sizeof(struct ipv4_hdr));
 420                 key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
 421                 key.port_src = rte_be_to_cpu_16(tcp->src_port);
 422                 break;
 423
 424         case IPPROTO_UDP:
 425                 udp = (struct udp_hdr *)((unsigned char *) ipv4_hdr +
 426                                         sizeof(struct ipv4_hdr));
 427                 key.port_dst = rte_be_to_cpu_16(udp->dst_port);
 428                 key.port_src = rte_be_to_cpu_16(udp->src_port);
 429                 break;
 430
 431         default:
 432                 key.port_dst = 0;
 433                 key.port_src = 0;
 434         }
 435
 436         /* Find destination port */
 437         ret = rte_hash_lookup(l3fwd_lookup_struct, (const void *)&key);
 438         return (uint8_t)((ret < 0)? portid : l3fwd_out_if[ret]);
 439 }
 440 #endif
 441
 442 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 443 static inline uint8_t
 444 get_dst_port(struct ipv4_hdr *ipv4_hdr,  uint8_t portid, lookup_struct_t * l3fwd_lookup_struct)
 445 {
 446         uint8_t next_hop;
 447
 448         return (uint8_t) ((rte_lpm_lookup(l3fwd_lookup_struct,
 449                         rte_be_to_cpu_32(ipv4_hdr->dst_addr), &next_hop) == 0)?
 450                         next_hop : portid);
 451 }
 452 #endif
 453
 454 static inline void
 455 l3fwd_simple_forward(struct rte_mbuf *m, uint8_t portid, lookup_struct_t * l3fwd_lookup_struct)
 456 {
 457         struct ether_hdr *eth_hdr;
 458         struct ipv4_hdr *ipv4_hdr;
 459         void *tmp;
 460         uint8_t dst_port;
 461
 462         eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
 463
 464         ipv4_hdr = (struct ipv4_hdr *)(rte_pktmbuf_mtod(m, unsigned char *) +
 465                                 sizeof(struct ether_hdr));
 466
 467 #ifdef DO_RFC_1812_CHECKS
 468         /* Check to make sure the packet is valid (RFC1812) */
 469         if (is_valid_ipv4_pkt(ipv4_hdr, m->pkt_len) < 0) {
 470                 rte_pktmbuf_free(m);
 471                 return;
 472         }
 473 #endif
 474
 475         dst_port = get_dst_port(ipv4_hdr, portid, l3fwd_lookup_struct);
 476         if (dst_port >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port) == 0)
 477                 dst_port = portid;
 478
 479         /* 02:00:00:00:00:xx */
 480         tmp = &eth_hdr->d_addr.addr_bytes[0];
 481         *((uint64_t *)tmp) = 0x000000000002 + ((uint64_t)dst_port << 40);
 482
 483 #ifdef DO_RFC_1812_CHECKS
 484         /* Update time to live and header checksum */
 485         --(ipv4_hdr->time_to_live);
 486         ++(ipv4_hdr->hdr_checksum);
 487 #endif
 488
 489         /* src addr */
 490         ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
 491
 492         send_single_packet(m, dst_port);
 493
 494 }
 495
 496 /* main processing loop */
 497 static int
 498 main_loop(__attribute__((unused)) void *dummy)
 499 {
 500         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
 501         unsigned lcore_id;
 502         uint64_t prev_tsc, diff_tsc, cur_tsc;
 503         int i, j, nb_rx;
 504         uint8_t portid, queueid;
 505         struct lcore_conf *qconf;
 506         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
 507
 508         prev_tsc = 0;
 509
 510         lcore_id = rte_lcore_id();
 511         qconf = &lcore_conf[lcore_id];
 512
 513         if (qconf->n_rx_queue == 0) {
 514                 RTE_LOG(INFO, L3FWD, "lcore %u has nothing to do\n", lcore_id);
 515                 return 0;
 516         }
 517
 518         RTE_LOG(INFO, L3FWD, "entering main loop on lcore %u\n", lcore_id);
 519
 520         for (i = 0; i < qconf->n_rx_queue; i++) {
 521
 522                 portid = qconf->rx_queue_list[i].port_id;
 523                 queueid = qconf->rx_queue_list[i].queue_id;
 524                 RTE_LOG(INFO, L3FWD, " -- lcoreid=%u portid=%hhu rxqueueid=%hhu\n", lcore_id,
 525                         portid, queueid);
 526         }
 527
 528         while (1) {
 529
 530                 cur_tsc = rte_rdtsc();
 531
 532                 /*
 533                  * TX burst queue drain
 534                  */
 535                 diff_tsc = cur_tsc - prev_tsc;
 536                 if (unlikely(diff_tsc > drain_tsc)) {
 537
 538                         /*
 539                          * This could be optimized (use queueid instead of
 540                          * portid), but it is not called so often
 541                          */
 542                         for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
 543                                 if (qconf->tx_mbufs[portid].len == 0)
 544                                         continue;
 545                                 send_burst(&lcore_conf[lcore_id],
 546                                         qconf->tx_mbufs[portid].len,
 547                                         portid);
 548                                 qconf->tx_mbufs[portid].len = 0;
 549                         }
 550
 551                         prev_tsc = cur_tsc;
 552                 }
 553
 554                 /*
 555                  * Read packet from RX queues
 556                  */
 557                 for (i = 0; i < qconf->n_rx_queue; ++i) {
 558
 559                         portid = qconf->rx_queue_list[i].port_id;
 560                         queueid = qconf->rx_queue_list[i].queue_id;
 561                         nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst, MAX_PKT_BURST);
 562
 563                         /* Prefetch first packets */
 564                         for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
 565                                 rte_prefetch0(rte_pktmbuf_mtod(
 566                                                 pkts_burst[j], void *));
 567                         }
 568
 569                         /* Prefetch and forward already prefetched packets */
 570                         for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
 571                                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
 572                                                 j + PREFETCH_OFFSET], void *));
 573                                 l3fwd_simple_forward(pkts_burst[j], portid, qconf->lookup_struct);
 574                         }
 575
 576                         /* Forward remaining prefetched packets */
 577                         for (; j < nb_rx; j++) {
 578                                 l3fwd_simple_forward(pkts_burst[j], portid, qconf->lookup_struct);
 579                         }
 580                 }
 581         }
 582 }
 583
 584 static int
 585 check_lcore_params(void)
 586 {
 587         uint8_t queue, lcore;
 588         uint16_t i;
 589         int socketid;
 590
 591         for (i = 0; i < nb_lcore_params; ++i) {
 592                 queue = lcore_params[i].queue_id;
 593                 if (queue >= MAX_RX_QUEUE_PER_PORT) {
 594                         printf("invalid queue number: %hhu\n", queue);
 595                         return -1;
 596                 }
 597                 lcore = lcore_params[i].lcore_id;
 598                 if (!rte_lcore_is_enabled(lcore)) {
 599                         printf("error: lcore %hhu is not enabled in lcore mask\n", lcore);
 600                         return -1;
 601                 }
 602                 if ((socketid = rte_lcore_to_socket_id(lcore) != 0) &&
 603                         (numa_on == 0)) {
 604                         printf("warning: lcore %hhu is on socket %d with numa off \n",
 605                                 lcore, socketid);
 606                 }
 607         }
 608         return 0;
 609 }
 610
 611 static int
 612 check_port_config(const unsigned nb_ports)
 613 {
 614         unsigned portid;
 615         uint16_t i;
 616
 617         for (i = 0; i < nb_lcore_params; ++i) {
 618                 portid = lcore_params[i].port_id;
 619                 if ((enabled_port_mask & (1 << portid)) == 0) {
 620                         printf("port %u is not enabled in port mask\n", portid);
 621                         return -1;
 622                 }
 623                 if (portid >= nb_ports) {
 624                         printf("port %u is not present on the board\n", portid);
 625                         return -1;
 626                 }
 627         }
 628         return 0;
 629 }
 630
 631 static uint8_t
 632 get_port_n_rx_queues(const uint8_t port)
 633 {
 634         int queue = -1;
 635         uint16_t i;
 636
 637         for (i = 0; i < nb_lcore_params; ++i) {
 638                 if (lcore_params[i].port_id == port && lcore_params[i].queue_id > queue)
 639                         queue = lcore_params[i].queue_id;
 640         }
 641         return (uint8_t)(++queue);
 642 }
 643
 644 static int
 645 init_lcore_rx_queues(void)
 646 {
 647         uint16_t i, nb_rx_queue;
 648         uint8_t lcore;
 649
 650         for (i = 0; i < nb_lcore_params; ++i) {
 651                 lcore = lcore_params[i].lcore_id;
 652                 nb_rx_queue = lcore_conf[lcore].n_rx_queue;
 653                 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
 654                         printf("error: too many queues (%u) for lcore: %u\n",
 655                                 (unsigned)nb_rx_queue + 1, (unsigned)lcore);
 656                         return -1;
 657                 } else {
 658                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
 659                                 lcore_params[i].port_id;
 660                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
 661                                 lcore_params[i].queue_id;
 662                         lcore_conf[lcore].n_rx_queue++;
 663                 }
 664         }
 665         return 0;
 666 }
 667
 668 /* display usage */
 669 static void
 670 print_usage(const char *prgname)
 671 {
 672         printf ("%s [EAL options] -- -p PORTMASK"
 673                 "  [--config (port,queue,lcore)[,(port,queue,lcore]]\n"
 674                 "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
 675                 "  --config (port,queue,lcore): rx queues configuration\n"
 676                 "  --no-numa: optional, disable numa awareness\n",
 677                 prgname);
 678 }
 679
 680 /* Custom handling of signals to handle process terminal */
 681 static void
 682 signal_handler(int signum)
 683 {
 684         uint8_t portid;
 685         uint8_t nb_ports = rte_eth_dev_count();
 686
 687         /* When we receive a SIGINT signal */
 688         if (signum == SIGINT) {
 689                 for (portid = 0; portid < nb_ports; portid++) {
 690                         /* skip ports that are not enabled */
 691                         if ((enabled_port_mask & (1 << portid)) == 0)
 692                                 continue;
 693                         rte_eth_dev_close(portid);
 694                 }
 695         }
 696         rte_exit(EXIT_SUCCESS, "\n User forced exit\n");
 697 }
 698 static int
 699 parse_portmask(const char *portmask)
 700 {
 701         char *end = NULL;
 702         unsigned long pm;
 703
 704         /* parse hexadecimal string */
 705         pm = strtoul(portmask, &end, 16);
 706         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
 707                 return -1;
 708
 709         if (pm == 0)
 710                 return -1;
 711
 712         return pm;
 713 }
 714
 715 static int
 716 parse_config(const char *q_arg)
 717 {
 718         char s[256];
 719         const char *p, *p0 = q_arg;
 720         char *end;
 721         enum fieldnames {
 722                 FLD_PORT = 0,
 723                 FLD_QUEUE,
 724                 FLD_LCORE,
 725                 _NUM_FLD
 726         };
 727         unsigned long int_fld[_NUM_FLD];
 728         char *str_fld[_NUM_FLD];
 729         int i;
 730         unsigned size;
 731
 732         nb_lcore_params = 0;
 733
 734         while ((p = strchr(p0,'(')) != NULL) {
 735                 ++p;
 736                 if((p0 = strchr(p,')')) == NULL)
 737                         return -1;
 738
 739                 size = p0 - p;
 740                 if(size >= sizeof(s))
 741                         return -1;
 742
 743                 snprintf(s, sizeof(s), "%.*s", size, p);
 744                 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') != _NUM_FLD)
 745                         return -1;
 746                 for (i = 0; i < _NUM_FLD; i++){
 747                         errno = 0;
 748                         int_fld[i] = strtoul(str_fld[i], &end, 0);
 749                         if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
 750                                 return -1;
 751                 }
 752                 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
 753                         printf("exceeded max number of lcore params: %hu\n",
 754                                 nb_lcore_params);
 755                         return -1;
 756                 }
 757                 lcore_params_array[nb_lcore_params].port_id = (uint8_t)int_fld[FLD_PORT];
 758                 lcore_params_array[nb_lcore_params].queue_id = (uint8_t)int_fld[FLD_QUEUE];
 759                 lcore_params_array[nb_lcore_params].lcore_id = (uint8_t)int_fld[FLD_LCORE];
 760                 ++nb_lcore_params;
 761         }
 762         lcore_params = lcore_params_array;
 763         return 0;
 764 }
 765
 766 /* Parse the argument given in the command line of the application */
 767 static int
 768 parse_args(int argc, char **argv)
 769 {
 770         int opt, ret;
 771         char **argvopt;
 772         int option_index;
 773         char *prgname = argv[0];
 774         static struct option lgopts[] = {
 775                 {"config", 1, 0, 0},
 776                 {"no-numa", 0, 0, 0},
 777                 {NULL, 0, 0, 0}
 778         };
 779
 780         argvopt = argv;
 781
 782         while ((opt = getopt_long(argc, argvopt, "p:",
 783                                 lgopts, &option_index)) != EOF) {
 784
 785                 switch (opt) {
 786                 /* portmask */
 787                 case 'p':
 788                         enabled_port_mask = parse_portmask(optarg);
 789                         if (enabled_port_mask == 0) {
 790                                 printf("invalid portmask\n");
 791                                 print_usage(prgname);
 792                                 return -1;
 793                         }
 794                         break;
 795
 796                 /* long options */
 797                 case 0:
 798                         if (!strcmp(lgopts[option_index].name, "config")) {
 799                                 ret = parse_config(optarg);
 800                                 if (ret) {
 801                                         printf("invalid config\n");
 802                                         print_usage(prgname);
 803                                         return -1;
 804                                 }
 805                         }
 806
 807                         if (!strcmp(lgopts[option_index].name, "no-numa")) {
 808                                 printf("numa is disabled \n");
 809                                 numa_on = 0;
 810                         }
 811                         break;
 812
 813                 default:
 814                         print_usage(prgname);
 815                         return -1;
 816                 }
 817         }
 818
 819         if (optind >= 0)
 820                 argv[optind-1] = prgname;
 821
 822         ret = optind-1;
 823         optind = 0; /* reset getopt lib */
 824         return ret;
 825 }
 826
 827 static void
 828 print_ethaddr(const char *name, const struct ether_addr *eth_addr)
 829 {
 830         char buf[ETHER_ADDR_FMT_SIZE];
 831         ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr);
 832         printf("%s%s", name, buf);
 833 }
 834
 835 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 836 static void
 837 setup_hash(int socketid)
 838 {
 839         unsigned i;
 840         int ret;
 841         char s[64];
 842
 843         /* create  hashes */
 844         snprintf(s, sizeof(s), "l3fwd_hash_%d", socketid);
 845         l3fwd_hash_params.name = s;
 846         l3fwd_hash_params.socket_id = socketid;
 847         l3fwd_lookup_struct[socketid] = rte_hash_create(&l3fwd_hash_params);
 848         if (l3fwd_lookup_struct[socketid] == NULL)
 849                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
 850                                 "socket %d\n", socketid);
 851
 852         /* populate the hash */
 853         for (i = 0; i < L3FWD_NUM_ROUTES; i++) {
 854                 ret = rte_hash_add_key (l3fwd_lookup_struct[socketid],
 855                                 (void *) &l3fwd_route_array[i].key);
 856                 if (ret < 0) {
 857                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
 858                                 "l3fwd hash on socket %d\n", i, socketid);
 859                 }
 860                 l3fwd_out_if[ret] = l3fwd_route_array[i].if_out;
 861                 printf("Hash: Adding key\n");
 862                 print_key(l3fwd_route_array[i].key);
 863         }
 864 }
 865 #endif
 866
 867 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 868 static void
 869 setup_lpm(int socketid)
 870 {
 871         unsigned i;
 872         int ret;
 873         char s[64];
 874
 875         /* create the LPM table */
 876         snprintf(s, sizeof(s), "L3FWD_LPM_%d", socketid);
 877         l3fwd_lookup_struct[socketid] = rte_lpm_create(s, socketid,
 878                                 L3FWD_LPM_MAX_RULES, 0);
 879         if (l3fwd_lookup_struct[socketid] == NULL)
 880                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
 881                                 " on socket %d\n", socketid);
 882
 883         /* populate the LPM table */
 884         for (i = 0; i < L3FWD_NUM_ROUTES; i++) {
 885                 ret = rte_lpm_add(l3fwd_lookup_struct[socketid],
 886                         l3fwd_route_array[i].ip,
 887                         l3fwd_route_array[i].depth,
 888                         l3fwd_route_array[i].if_out);
 889
 890                 if (ret < 0) {
 891                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
 892                                 "l3fwd LPM table on socket %d\n",
 893                                 i, socketid);
 894                 }
 895
 896                 printf("LPM: Adding route 0x%08x / %d (%d)\n",
 897                         (unsigned)l3fwd_route_array[i].ip,
 898                         l3fwd_route_array[i].depth,
 899                         l3fwd_route_array[i].if_out);
 900         }
 901 }
 902 #endif
 903
 904 static int
 905 init_mem(unsigned nb_mbuf)
 906 {
 907         struct lcore_conf *qconf;
 908         int socketid;
 909         unsigned lcore_id;
 910         char s[64];
 911
 912         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
 913                 if (rte_lcore_is_enabled(lcore_id) == 0)
 914                         continue;
 915
 916                 if (numa_on)
 917                         socketid = rte_lcore_to_socket_id(lcore_id);
 918                 else
 919                         socketid = 0;
 920
 921                 if (socketid >= NB_SOCKETS) {
 922                         rte_exit(EXIT_FAILURE, "Socket %d of lcore %u is out of range %d\n",
 923                                 socketid, lcore_id, NB_SOCKETS);
 924                 }
 925                 if (pktmbuf_pool[socketid] == NULL) {
 926                         snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
 927                         pktmbuf_pool[socketid] = rte_pktmbuf_pool_create(s,
 928                                 nb_mbuf, MEMPOOL_CACHE_SIZE, 0, MBUF_DATA_SIZE,
 929                                 socketid);
 930                         if (pktmbuf_pool[socketid] == NULL)
 931                                 rte_exit(EXIT_FAILURE, "Cannot init mbuf pool on socket %d\n", socketid);
 932                         else
 933                                 printf("Allocated mbuf pool on socket %d\n", socketid);
 934
 935 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 936                         setup_lpm(socketid);
 937 #else
 938                         setup_hash(socketid);
 939 #endif
 940                 }
 941                 qconf = &lcore_conf[lcore_id];
 942                 qconf->lookup_struct = l3fwd_lookup_struct[socketid];
 943         }
 944         return 0;
 945 }
 946
 947 int
 948 main(int argc, char **argv)
 949 {
 950         struct lcore_conf *qconf;
 951         struct rte_eth_dev_info dev_info;
 952         struct rte_eth_txconf *txconf;
 953         int ret;
 954         unsigned nb_ports;
 955         uint16_t queueid;
 956         unsigned lcore_id;
 957         uint32_t nb_lcores;
 958         uint16_t n_tx_queue;
 959         uint8_t portid, nb_rx_queue, queue, socketid;
 960
 961         signal(SIGINT, signal_handler);
 962         /* init EAL */
 963         ret = rte_eal_init(argc, argv);
 964         if (ret < 0)
 965                 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
 966         argc -= ret;
 967         argv += ret;
 968
 969         /* parse application arguments (after the EAL ones) */
 970         ret = parse_args(argc, argv);
 971         if (ret < 0)
 972                 rte_exit(EXIT_FAILURE, "Invalid L3FWD-VF parameters\n");
 973
 974         if (check_lcore_params() < 0)
 975                 rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
 976
 977         ret = init_lcore_rx_queues();
 978         if (ret < 0)
 979                 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
 980
 981         nb_ports = rte_eth_dev_count();
 982         if (nb_ports > RTE_MAX_ETHPORTS)
 983                 nb_ports = RTE_MAX_ETHPORTS;
 984
 985         if (check_port_config(nb_ports) < 0)
 986                 rte_exit(EXIT_FAILURE, "check_port_config failed\n");
 987
 988         nb_lcores = rte_lcore_count();
 989
 990         /* initialize all ports */
 991         for (portid = 0; portid < nb_ports; portid++) {
 992                 /* skip ports that are not enabled */
 993                 if ((enabled_port_mask & (1 << portid)) == 0) {
 994                         printf("\nSkipping disabled port %d\n", portid);
 995                         continue;
 996                 }
 997
 998                 /* init port */
 999                 printf("Initializing port %d ... ", portid );
1000                 fflush(stdout);
1001
1002                 /* must always equal(=1) */
1003                 nb_rx_queue = get_port_n_rx_queues(portid);
1004                 n_tx_queue = MAX_TX_QUEUE_PER_PORT;
1005
1006                 printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
1007                         nb_rx_queue, (unsigned)1 );
1008                 ret = rte_eth_dev_configure(portid, nb_rx_queue, n_tx_queue, &port_conf);
1009                 if (ret < 0)
1010                         rte_exit(EXIT_FAILURE, "Cannot configure device: err=%d, port=%d\n",
1011                                 ret, portid);
1012
1013                 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
1014                 print_ethaddr(" Address:", &ports_eth_addr[portid]);
1015                 printf(", ");
1016
1017                 ret = init_mem(NB_MBUF);
1018                 if (ret < 0)
1019                         rte_exit(EXIT_FAILURE, "init_mem failed\n");
1020
1021                 /* init one TX queue */
1022                 socketid = (uint8_t)rte_lcore_to_socket_id(rte_get_master_lcore());
1023
1024                 printf("txq=%d,%d,%d ", portid, 0, socketid);
1025                 fflush(stdout);
1026
1027                 rte_eth_dev_info_get(portid, &dev_info);
1028                 txconf = &dev_info.default_txconf;
1029                 if (port_conf.rxmode.jumbo_frame)
1030                         txconf->txq_flags = 0;
1031                 ret = rte_eth_tx_queue_setup(portid, 0, nb_txd,
1032                                                  socketid, txconf);
1033                 if (ret < 0)
1034                         rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: err=%d, "
1035                                 "port=%d\n", ret, portid);
1036
1037                 printf("\n");
1038         }
1039
1040         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1041                 if (rte_lcore_is_enabled(lcore_id) == 0)
1042                         continue;
1043                 qconf = &lcore_conf[lcore_id];
1044                 qconf->tx_queue_id = 0;
1045
1046                 printf("\nInitializing rx queues on lcore %u ... ", lcore_id );
1047                 fflush(stdout);
1048                 /* init RX queues */
1049                 for(queue = 0; queue < qconf->n_rx_queue; ++queue) {
1050                         portid = qconf->rx_queue_list[queue].port_id;
1051                         queueid = qconf->rx_queue_list[queue].queue_id;
1052
1053                         if (numa_on)
1054                                 socketid = (uint8_t)rte_lcore_to_socket_id(lcore_id);
1055                         else
1056                                 socketid = 0;
1057
1058                         printf("rxq=%d,%d,%d ", portid, queueid, socketid);
1059                         fflush(stdout);
1060
1061                         ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
1062                                                 socketid, NULL,
1063                                                 pktmbuf_pool[socketid]);
1064                         if (ret < 0)
1065                                 rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: err=%d,"
1066                                                 "port=%d\n", ret, portid);
1067                 }
1068         }
1069         printf("\n");
1070
1071         /* start ports */
1072         for (portid = 0; portid < nb_ports; portid++) {
1073                 if ((enabled_port_mask & (1 << portid)) == 0) {
1074                         continue;
1075                 }
1076                 /* Start device */
1077                 ret = rte_eth_dev_start(portid);
1078                 if (ret < 0)
1079                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%d\n",
1080                                 ret, portid);
1081
1082                 printf("done: Port %d\n", portid);
1083
1084         }
1085
1086         /* launch per-lcore init on every lcore */
1087         rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
1088         RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1089                 if (rte_eal_wait_lcore(lcore_id) < 0)
1090                         return -1;
1091         }
1092
1093         return 0;
1094 }