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_tailq.h>
  53 #include <rte_eal.h>
  54 #include <rte_per_lcore.h>
  55 #include <rte_launch.h>
  56 #include <rte_atomic.h>
  57 #include <rte_cycles.h>
  58 #include <rte_prefetch.h>
  59 #include <rte_lcore.h>
  60 #include <rte_per_lcore.h>
  61 #include <rte_branch_prediction.h>
  62 #include <rte_interrupts.h>
  63 #include <rte_pci.h>
  64 #include <rte_random.h>
  65 #include <rte_debug.h>
  66 #include <rte_ether.h>
  67 #include <rte_ethdev.h>
  68 #include <rte_ring.h>
  69 #include <rte_mempool.h>
  70 #include <rte_mbuf.h>
  71 #include <rte_ip.h>
  72 #include <rte_tcp.h>
  73 #include <rte_udp.h>
  74 #include <rte_string_fns.h>
  75
  76 #include "main.h"
  77
  78 #define APP_LOOKUP_EXACT_MATCH          0
  79 #define APP_LOOKUP_LPM                  1
  80 #define DO_RFC_1812_CHECKS
  81
  82 //#define APP_LOOKUP_METHOD             APP_LOOKUP_EXACT_MATCH
  83 #ifndef APP_LOOKUP_METHOD
  84 #define APP_LOOKUP_METHOD             APP_LOOKUP_LPM
  85 #endif
  86
  87 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
  88 #include <rte_hash.h>
  89 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
  90 #include <rte_lpm.h>
  91 #else
  92 #error "APP_LOOKUP_METHOD set to incorrect value"
  93 #endif
  94
  95 #define RTE_LOGTYPE_L3FWD RTE_LOGTYPE_USER1
  96
  97 #define MEMPOOL_CACHE_SIZE 256
  98
  99 #define MBUF_SIZE (2048 + sizeof(struct rte_mbuf) + RTE_PKTMBUF_HEADROOM)
 100
 101 /*
 102  * This expression is used to calculate the number of mbufs needed depending on user input, taking
 103  *  into account memory for rx and tx hardware rings, cache per lcore and mtable per port per lcore.
 104  *  RTE_MAX is used to ensure that NB_MBUF never goes below a minimum value of 8192
 105  */
 106
 107 #define NB_MBUF RTE_MAX (                                                                                                                                       \
 108                                 (nb_ports*nb_rx_queue*RTE_TEST_RX_DESC_DEFAULT +                                                        \
 109                                 nb_ports*nb_lcores*MAX_PKT_BURST +                                                                                      \
 110                                 nb_ports*n_tx_queue*RTE_TEST_TX_DESC_DEFAULT +                                                          \
 111                                 nb_lcores*MEMPOOL_CACHE_SIZE),                                                                                          \
 112                                 (unsigned)8192)
 113
 114 /*
 115  * RX and TX Prefetch, Host, and Write-back threshold values should be
 116  * carefully set for optimal performance. Consult the network
 117  * controller's datasheet and supporting DPDK documentation for guidance
 118  * on how these parameters should be set.
 119  */
 120 #define RX_PTHRESH 8 /**< Default values of RX prefetch threshold reg. */
 121 #define RX_HTHRESH 8 /**< Default values of RX host threshold reg. */
 122 #define RX_WTHRESH 4 /**< Default values of RX write-back threshold reg. */
 123
 124 /*
 125  * These default values are optimized for use with the Intel(R) 82599 10 GbE
 126  * Controller and the DPDK ixgbe PMD. Consider using other values for other
 127  * network controllers and/or network drivers.
 128  */
 129 #define TX_PTHRESH 36 /**< Default values of TX prefetch threshold reg. */
 130 #define TX_HTHRESH 0  /**< Default values of TX host threshold reg. */
 131 #define TX_WTHRESH 0  /**< Default values of TX write-back threshold reg. */
 132
 133 #define MAX_PKT_BURST 32
 134 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
 135
 136 #define NB_SOCKETS 8
 137
 138 #define SOCKET0 0
 139
 140 /* Configure how many packets ahead to prefetch, when reading packets */
 141 #define PREFETCH_OFFSET 3
 142
 143 /*
 144  * Configurable number of RX/TX ring descriptors
 145  */
 146 #define RTE_TEST_RX_DESC_DEFAULT 128
 147 #define RTE_TEST_TX_DESC_DEFAULT 512
 148 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
 149 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
 150
 151 /* ethernet addresses of ports */
 152 static struct ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
 153
 154 /* mask of enabled ports */
 155 static uint32_t enabled_port_mask = 0;
 156 static int numa_on = 1; /**< NUMA is enabled by default. */
 157
 158 struct mbuf_table {
 159         uint16_t len;
 160         struct rte_mbuf *m_table[MAX_PKT_BURST];
 161 };
 162
 163 struct lcore_rx_queue {
 164         uint8_t port_id;
 165         uint8_t queue_id;
 166 } __rte_cache_aligned;
 167
 168 #define MAX_RX_QUEUE_PER_LCORE 16
 169 #define MAX_TX_QUEUE_PER_PORT 1
 170 #define MAX_RX_QUEUE_PER_PORT 1
 171
 172 #define MAX_LCORE_PARAMS 1024
 173 struct lcore_params {
 174         uint8_t port_id;
 175         uint8_t queue_id;
 176         uint8_t lcore_id;
 177 } __rte_cache_aligned;
 178
 179 static struct lcore_params lcore_params_array[MAX_LCORE_PARAMS];
 180 static struct lcore_params lcore_params_array_default[] = {
 181         {0, 0, 2},
 182         {0, 1, 2},
 183         {0, 2, 2},
 184         {1, 0, 2},
 185         {1, 1, 2},
 186         {1, 2, 2},
 187         {2, 0, 2},
 188         {3, 0, 3},
 189         {3, 1, 3},
 190 };
 191
 192 static struct lcore_params * lcore_params = lcore_params_array_default;
 193 static uint16_t nb_lcore_params = sizeof(lcore_params_array_default) /
 194                                 sizeof(lcore_params_array_default[0]);
 195
 196 static struct rte_eth_conf port_conf = {
 197         .rxmode = {
 198                 .max_rx_pkt_len = ETHER_MAX_LEN,
 199                 .split_hdr_size = 0,
 200                 .header_split   = 0, /**< Header Split disabled */
 201                 .hw_ip_checksum = 1, /**< IP checksum offload enabled */
 202                 .hw_vlan_filter = 0, /**< VLAN filtering disabled */
 203                 .jumbo_frame    = 0, /**< Jumbo Frame Support disabled */
 204                 .hw_strip_crc   = 0, /**< CRC stripped by hardware */
 205         },
 206         .rx_adv_conf = {
 207                 .rss_conf = {
 208                         .rss_key = NULL,
 209                         .rss_hf = ETH_RSS_IPV4 | ETH_RSS_IPV6,
 210                 },
 211         },
 212         .txmode = {
 213                 .mq_mode = ETH_MQ_TX_NONE,
 214         },
 215 };
 216
 217 static const struct rte_eth_rxconf rx_conf = {
 218         .rx_thresh = {
 219                 .pthresh = RX_PTHRESH,
 220                 .hthresh = RX_HTHRESH,
 221                 .wthresh = RX_WTHRESH,
 222         },
 223         .rx_free_thresh = 32,
 224 };
 225
 226 static const struct rte_eth_txconf tx_conf = {
 227         .tx_thresh = {
 228                 .pthresh = TX_PTHRESH,
 229                 .hthresh = TX_HTHRESH,
 230                 .wthresh = TX_WTHRESH,
 231         },
 232         .tx_free_thresh = 0, /* Use PMD default values */
 233         .tx_rs_thresh = 0, /* Use PMD default values */
 234         .txq_flags = (ETH_TXQ_FLAGS_NOMULTSEGS |
 235                       ETH_TXQ_FLAGS_NOVLANOFFL |
 236                       ETH_TXQ_FLAGS_NOXSUMSCTP |
 237                       ETH_TXQ_FLAGS_NOXSUMUDP |
 238                       ETH_TXQ_FLAGS_NOXSUMTCP)
 239 };
 240
 241 static struct rte_mempool * pktmbuf_pool[NB_SOCKETS];
 242
 243
 244 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 245
 246 #ifdef RTE_MACHINE_CPUFLAG_SSE4_2
 247 #include <rte_hash_crc.h>
 248 #define DEFAULT_HASH_FUNC       rte_hash_crc
 249 #else
 250 #include <rte_jhash.h>
 251 #define DEFAULT_HASH_FUNC       rte_jhash
 252 #endif
 253
 254 struct ipv4_5tuple {
 255         uint32_t ip_dst;
 256         uint32_t ip_src;
 257         uint16_t port_dst;
 258         uint16_t port_src;
 259         uint8_t proto;
 260 } __attribute__((__packed__));
 261
 262 struct l3fwd_route {
 263         struct ipv4_5tuple key;
 264         uint8_t if_out;
 265 };
 266
 267 static struct l3fwd_route l3fwd_route_array[] = {
 268         {{IPv4(100,10,0,1), IPv4(200,10,0,1), 101, 11, IPPROTO_TCP}, 0},
 269         {{IPv4(100,20,0,2), IPv4(200,20,0,2), 102, 12, IPPROTO_TCP}, 1},
 270         {{IPv4(100,30,0,3), IPv4(200,30,0,3), 103, 13, IPPROTO_TCP}, 2},
 271         {{IPv4(100,40,0,4), IPv4(200,40,0,4), 104, 14, IPPROTO_TCP}, 3},
 272 };
 273
 274 typedef struct rte_hash lookup_struct_t;
 275 static lookup_struct_t *l3fwd_lookup_struct[NB_SOCKETS];
 276
 277 #define L3FWD_HASH_ENTRIES      1024
 278 struct rte_hash_parameters l3fwd_hash_params = {
 279         .name = "l3fwd_hash_0",
 280         .entries = L3FWD_HASH_ENTRIES,
 281         .bucket_entries = 4,
 282         .key_len = sizeof(struct ipv4_5tuple),
 283         .hash_func = DEFAULT_HASH_FUNC,
 284         .hash_func_init_val = 0,
 285         .socket_id = SOCKET0,
 286 };
 287
 288 #define L3FWD_NUM_ROUTES \
 289         (sizeof(l3fwd_route_array) / sizeof(l3fwd_route_array[0]))
 290
 291 static uint8_t l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 292 #endif
 293
 294 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 295 struct l3fwd_route {
 296         uint32_t ip;
 297         uint8_t  depth;
 298         uint8_t  if_out;
 299 };
 300
 301 static struct l3fwd_route l3fwd_route_array[] = {
 302         {IPv4(1,1,1,0), 24, 0},
 303         {IPv4(2,1,1,0), 24, 1},
 304         {IPv4(3,1,1,0), 24, 2},
 305         {IPv4(4,1,1,0), 24, 3},
 306         {IPv4(5,1,1,0), 24, 4},
 307         {IPv4(6,1,1,0), 24, 5},
 308         {IPv4(7,1,1,0), 24, 6},
 309         {IPv4(8,1,1,0), 24, 7},
 310 };
 311
 312 #define L3FWD_NUM_ROUTES \
 313         (sizeof(l3fwd_route_array) / sizeof(l3fwd_route_array[0]))
 314
 315 #define L3FWD_LPM_MAX_RULES     1024
 316
 317 typedef struct rte_lpm lookup_struct_t;
 318 static lookup_struct_t *l3fwd_lookup_struct[NB_SOCKETS];
 319 #endif
 320
 321 struct lcore_conf {
 322         uint16_t n_rx_queue;
 323         struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
 324         uint16_t tx_queue_id;
 325         struct mbuf_table tx_mbufs[RTE_MAX_ETHPORTS];
 326         lookup_struct_t * lookup_struct;
 327 } __rte_cache_aligned;
 328
 329 static struct lcore_conf lcore_conf[RTE_MAX_LCORE];
 330
 331 /* Send burst of packets on an output interface */
 332 static inline int
 333 send_burst(struct lcore_conf *qconf, uint16_t n, uint8_t port)
 334 {
 335         struct rte_mbuf **m_table;
 336         int ret;
 337         uint16_t queueid;
 338
 339         queueid = qconf->tx_queue_id;
 340         m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
 341
 342         ret = rte_eth_tx_burst(port, queueid, m_table, n);
 343         if (unlikely(ret < n)) {
 344                 do {
 345                         rte_pktmbuf_free(m_table[ret]);
 346                 } while (++ret < n);
 347         }
 348
 349         return 0;
 350 }
 351
 352 /* Enqueue a single packet, and send burst if queue is filled */
 353 static inline int
 354 send_single_packet(struct rte_mbuf *m, uint8_t port)
 355 {
 356         uint32_t lcore_id;
 357         uint16_t len;
 358         struct lcore_conf *qconf;
 359
 360         lcore_id = rte_lcore_id();
 361
 362         qconf = &lcore_conf[lcore_id];
 363         len = qconf->tx_mbufs[port].len;
 364         qconf->tx_mbufs[port].m_table[len] = m;
 365         len++;
 366
 367         /* enough pkts to be sent */
 368         if (unlikely(len == MAX_PKT_BURST)) {
 369                 send_burst(qconf, MAX_PKT_BURST, port);
 370                 len = 0;
 371         }
 372
 373         qconf->tx_mbufs[port].len = len;
 374         return 0;
 375 }
 376
 377 #ifdef DO_RFC_1812_CHECKS
 378 static inline int
 379 is_valid_ipv4_pkt(struct ipv4_hdr *pkt, uint32_t link_len)
 380 {
 381         /* From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2 */
 382         /*
 383          * 1. The packet length reported by the Link Layer must be large
 384          * enough to hold the minimum length legal IP datagram (20 bytes).
 385          */
 386         if (link_len < sizeof(struct ipv4_hdr))
 387                 return -1;
 388
 389         /* 2. The IP checksum must be correct. */
 390         /* this is checked in H/W */
 391
 392         /*
 393          * 3. The IP version number must be 4. If the version number is not 4
 394          * then the packet may be another version of IP, such as IPng or
 395          * ST-II.
 396          */
 397         if (((pkt->version_ihl) >> 4) != 4)
 398                 return -3;
 399         /*
 400          * 4. The IP header length field must be large enough to hold the
 401          * minimum length legal IP datagram (20 bytes = 5 words).
 402          */
 403         if ((pkt->version_ihl & 0xf) < 5)
 404                 return -4;
 405
 406         /*
 407          * 5. The IP total length field must be large enough to hold the IP
 408          * datagram header, whose length is specified in the IP header length
 409          * field.
 410          */
 411         if (rte_cpu_to_be_16(pkt->total_length) < sizeof(struct ipv4_hdr))
 412                 return -5;
 413
 414         return 0;
 415 }
 416 #endif
 417
 418 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 419 static void
 420 print_key(struct ipv4_5tuple key)
 421 {
 422         printf("IP dst = %08x, IP src = %08x, port dst = %d, port src = %d, proto = %d\n",
 423                (unsigned)key.ip_dst, (unsigned)key.ip_src, key.port_dst, key.port_src, key.proto);
 424 }
 425
 426 static inline uint8_t
 427 get_dst_port(struct ipv4_hdr *ipv4_hdr,  uint8_t portid, lookup_struct_t * l3fwd_lookup_struct)
 428 {
 429         struct ipv4_5tuple key;
 430         struct tcp_hdr *tcp;
 431         struct udp_hdr *udp;
 432         int ret = 0;
 433
 434         key.ip_dst = rte_be_to_cpu_32(ipv4_hdr->dst_addr);
 435         key.ip_src = rte_be_to_cpu_32(ipv4_hdr->src_addr);
 436         key.proto = ipv4_hdr->next_proto_id;
 437
 438         switch (ipv4_hdr->next_proto_id) {
 439         case IPPROTO_TCP:
 440                 tcp = (struct tcp_hdr *)((unsigned char *) ipv4_hdr +
 441                                         sizeof(struct ipv4_hdr));
 442                 key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
 443                 key.port_src = rte_be_to_cpu_16(tcp->src_port);
 444                 break;
 445
 446         case IPPROTO_UDP:
 447                 udp = (struct udp_hdr *)((unsigned char *) ipv4_hdr +
 448                                         sizeof(struct ipv4_hdr));
 449                 key.port_dst = rte_be_to_cpu_16(udp->dst_port);
 450                 key.port_src = rte_be_to_cpu_16(udp->src_port);
 451                 break;
 452
 453         default:
 454                 key.port_dst = 0;
 455                 key.port_src = 0;
 456         }
 457
 458         /* Find destination port */
 459         ret = rte_hash_lookup(l3fwd_lookup_struct, (const void *)&key);
 460         return (uint8_t)((ret < 0)? portid : l3fwd_out_if[ret]);
 461 }
 462 #endif
 463
 464 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 465 static inline uint8_t
 466 get_dst_port(struct ipv4_hdr *ipv4_hdr,  uint8_t portid, lookup_struct_t * l3fwd_lookup_struct)
 467 {
 468         uint8_t next_hop;
 469
 470         return (uint8_t) ((rte_lpm_lookup(l3fwd_lookup_struct,
 471                         rte_be_to_cpu_32(ipv4_hdr->dst_addr), &next_hop) == 0)?
 472                         next_hop : portid);
 473 }
 474 #endif
 475
 476 static inline void
 477 l3fwd_simple_forward(struct rte_mbuf *m, uint8_t portid, lookup_struct_t * l3fwd_lookup_struct)
 478 {
 479         struct ether_hdr *eth_hdr;
 480         struct ipv4_hdr *ipv4_hdr;
 481         void *tmp;
 482         uint8_t dst_port;
 483
 484         eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
 485
 486         ipv4_hdr = (struct ipv4_hdr *)(rte_pktmbuf_mtod(m, unsigned char *) +
 487                                 sizeof(struct ether_hdr));
 488
 489 #ifdef DO_RFC_1812_CHECKS
 490         /* Check to make sure the packet is valid (RFC1812) */
 491         if (is_valid_ipv4_pkt(ipv4_hdr, m->pkt.pkt_len) < 0) {
 492                 rte_pktmbuf_free(m);
 493                 return;
 494         }
 495 #endif
 496
 497         dst_port = get_dst_port(ipv4_hdr, portid, l3fwd_lookup_struct);
 498         if (dst_port >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port) == 0)
 499                 dst_port = portid;
 500
 501         /* 02:00:00:00:00:xx */
 502         tmp = &eth_hdr->d_addr.addr_bytes[0];
 503         *((uint64_t *)tmp) = 0x000000000002 + ((uint64_t)dst_port << 40);
 504
 505 #ifdef DO_RFC_1812_CHECKS
 506         /* Update time to live and header checksum */
 507         --(ipv4_hdr->time_to_live);
 508         ++(ipv4_hdr->hdr_checksum);
 509 #endif
 510
 511         /* src addr */
 512         ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
 513
 514         send_single_packet(m, dst_port);
 515
 516 }
 517
 518 /* main processing loop */
 519 static int
 520 main_loop(__attribute__((unused)) void *dummy)
 521 {
 522         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
 523         unsigned lcore_id;
 524         uint64_t prev_tsc, diff_tsc, cur_tsc;
 525         int i, j, nb_rx;
 526         uint8_t portid, queueid;
 527         struct lcore_conf *qconf;
 528         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
 529
 530         prev_tsc = 0;
 531
 532         lcore_id = rte_lcore_id();
 533         qconf = &lcore_conf[lcore_id];
 534
 535         if (qconf->n_rx_queue == 0) {
 536                 RTE_LOG(INFO, L3FWD, "lcore %u has nothing to do\n", lcore_id);
 537                 return 0;
 538         }
 539
 540         RTE_LOG(INFO, L3FWD, "entering main loop on lcore %u\n", lcore_id);
 541
 542         for (i = 0; i < qconf->n_rx_queue; i++) {
 543
 544                 portid = qconf->rx_queue_list[i].port_id;
 545                 queueid = qconf->rx_queue_list[i].queue_id;
 546                 RTE_LOG(INFO, L3FWD, " -- lcoreid=%u portid=%hhu rxqueueid=%hhu\n", lcore_id,
 547                         portid, queueid);
 548         }
 549
 550         while (1) {
 551
 552                 cur_tsc = rte_rdtsc();
 553
 554                 /*
 555                  * TX burst queue drain
 556                  */
 557                 diff_tsc = cur_tsc - prev_tsc;
 558                 if (unlikely(diff_tsc > drain_tsc)) {
 559
 560                         /*
 561                          * This could be optimized (use queueid instead of
 562                          * portid), but it is not called so often
 563                          */
 564                         for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
 565                                 if (qconf->tx_mbufs[portid].len == 0)
 566                                         continue;
 567                                 send_burst(&lcore_conf[lcore_id],
 568                                         qconf->tx_mbufs[portid].len,
 569                                         portid);
 570                                 qconf->tx_mbufs[portid].len = 0;
 571                         }
 572
 573                         prev_tsc = cur_tsc;
 574                 }
 575
 576                 /*
 577                  * Read packet from RX queues
 578                  */
 579                 for (i = 0; i < qconf->n_rx_queue; ++i) {
 580
 581                         portid = qconf->rx_queue_list[i].port_id;
 582                         queueid = qconf->rx_queue_list[i].queue_id;
 583                         nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst, MAX_PKT_BURST);
 584
 585                         /* Prefetch first packets */
 586                         for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
 587                                 rte_prefetch0(rte_pktmbuf_mtod(
 588                                                 pkts_burst[j], void *));
 589                         }
 590
 591                         /* Prefetch and forward already prefetched packets */
 592                         for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
 593                                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
 594                                                 j + PREFETCH_OFFSET], void *));
 595                                 l3fwd_simple_forward(pkts_burst[j], portid, qconf->lookup_struct);
 596                         }
 597
 598                         /* Forward remaining prefetched packets */
 599                         for (; j < nb_rx; j++) {
 600                                 l3fwd_simple_forward(pkts_burst[j], portid, qconf->lookup_struct);
 601                         }
 602                 }
 603         }
 604 }
 605
 606 static int
 607 check_lcore_params(void)
 608 {
 609         uint8_t queue, lcore;
 610         uint16_t i;
 611         int socketid;
 612
 613         for (i = 0; i < nb_lcore_params; ++i) {
 614                 queue = lcore_params[i].queue_id;
 615                 if (queue >= MAX_RX_QUEUE_PER_PORT) {
 616                         printf("invalid queue number: %hhu\n", queue);
 617                         return -1;
 618                 }
 619                 lcore = lcore_params[i].lcore_id;
 620                 if (!rte_lcore_is_enabled(lcore)) {
 621                         printf("error: lcore %hhu is not enabled in lcore mask\n", lcore);
 622                         return -1;
 623                 }
 624                 if ((socketid = rte_lcore_to_socket_id(lcore) != 0) &&
 625                         (numa_on == 0)) {
 626                         printf("warning: lcore %hhu is on socket %d with numa off \n",
 627                                 lcore, socketid);
 628                 }
 629         }
 630         return 0;
 631 }
 632
 633 static int
 634 check_port_config(const unsigned nb_ports)
 635 {
 636         unsigned portid;
 637         uint16_t i;
 638
 639         for (i = 0; i < nb_lcore_params; ++i) {
 640                 portid = lcore_params[i].port_id;
 641                 if ((enabled_port_mask & (1 << portid)) == 0) {
 642                         printf("port %u is not enabled in port mask\n", portid);
 643                         return -1;
 644                 }
 645                 if (portid >= nb_ports) {
 646                         printf("port %u is not present on the board\n", portid);
 647                         return -1;
 648                 }
 649         }
 650         return 0;
 651 }
 652
 653 static uint8_t
 654 get_port_n_rx_queues(const uint8_t port)
 655 {
 656         int queue = -1;
 657         uint16_t i;
 658
 659         for (i = 0; i < nb_lcore_params; ++i) {
 660                 if (lcore_params[i].port_id == port && lcore_params[i].queue_id > queue)
 661                         queue = lcore_params[i].queue_id;
 662         }
 663         return (uint8_t)(++queue);
 664 }
 665
 666 static int
 667 init_lcore_rx_queues(void)
 668 {
 669         uint16_t i, nb_rx_queue;
 670         uint8_t lcore;
 671
 672         for (i = 0; i < nb_lcore_params; ++i) {
 673                 lcore = lcore_params[i].lcore_id;
 674                 nb_rx_queue = lcore_conf[lcore].n_rx_queue;
 675                 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
 676                         printf("error: too many queues (%u) for lcore: %u\n",
 677                                 (unsigned)nb_rx_queue + 1, (unsigned)lcore);
 678                         return -1;
 679                 } else {
 680                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
 681                                 lcore_params[i].port_id;
 682                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
 683                                 lcore_params[i].queue_id;
 684                         lcore_conf[lcore].n_rx_queue++;
 685                 }
 686         }
 687         return 0;
 688 }
 689
 690 /* display usage */
 691 static void
 692 print_usage(const char *prgname)
 693 {
 694         printf ("%s [EAL options] -- -p PORTMASK"
 695                 "  [--config (port,queue,lcore)[,(port,queue,lcore]]\n"
 696                 "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
 697                 "  --config (port,queue,lcore): rx queues configuration\n"
 698                 "  --no-numa: optional, disable numa awareness\n",
 699                 prgname);
 700 }
 701
 702 /* Custom handling of signals to handle process terminal */
 703 static void
 704 signal_handler(int signum)
 705 {
 706         uint8_t portid;
 707         uint8_t nb_ports = rte_eth_dev_count();
 708
 709         /* When we receive a SIGINT signal */
 710         if (signum == SIGINT) {
 711                 for (portid = 0; portid < nb_ports; portid++) {
 712                         /* skip ports that are not enabled */
 713                         if ((enabled_port_mask & (1 << portid)) == 0)
 714                                 continue;
 715                         rte_eth_dev_close(portid);
 716                 }
 717         }
 718         rte_exit(EXIT_SUCCESS, "\n User forced exit\n");
 719 }
 720 static int
 721 parse_portmask(const char *portmask)
 722 {
 723         char *end = NULL;
 724         unsigned long pm;
 725
 726         /* parse hexadecimal string */
 727         pm = strtoul(portmask, &end, 16);
 728         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
 729                 return -1;
 730
 731         if (pm == 0)
 732                 return -1;
 733
 734         return pm;
 735 }
 736
 737 static int
 738 parse_config(const char *q_arg)
 739 {
 740         char s[256];
 741         const char *p, *p0 = q_arg;
 742         char *end;
 743         enum fieldnames {
 744                 FLD_PORT = 0,
 745                 FLD_QUEUE,
 746                 FLD_LCORE,
 747                 _NUM_FLD
 748         };
 749         unsigned long int_fld[_NUM_FLD];
 750         char *str_fld[_NUM_FLD];
 751         int i;
 752         unsigned size;
 753
 754         nb_lcore_params = 0;
 755
 756         while ((p = strchr(p0,'(')) != NULL) {
 757                 ++p;
 758                 if((p0 = strchr(p,')')) == NULL)
 759                         return -1;
 760
 761                 size = p0 - p;
 762                 if(size >= sizeof(s))
 763                         return -1;
 764
 765                 rte_snprintf(s, sizeof(s), "%.*s", size, p);
 766                 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') != _NUM_FLD)
 767                         return -1;
 768                 for (i = 0; i < _NUM_FLD; i++){
 769                         errno = 0;
 770                         int_fld[i] = strtoul(str_fld[i], &end, 0);
 771                         if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
 772                                 return -1;
 773                 }
 774                 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
 775                         printf("exceeded max number of lcore params: %hu\n",
 776                                 nb_lcore_params);
 777                         return -1;
 778                 }
 779                 lcore_params_array[nb_lcore_params].port_id = (uint8_t)int_fld[FLD_PORT];
 780                 lcore_params_array[nb_lcore_params].queue_id = (uint8_t)int_fld[FLD_QUEUE];
 781                 lcore_params_array[nb_lcore_params].lcore_id = (uint8_t)int_fld[FLD_LCORE];
 782                 ++nb_lcore_params;
 783         }
 784         lcore_params = lcore_params_array;
 785         return 0;
 786 }
 787
 788 /* Parse the argument given in the command line of the application */
 789 static int
 790 parse_args(int argc, char **argv)
 791 {
 792         int opt, ret;
 793         char **argvopt;
 794         int option_index;
 795         char *prgname = argv[0];
 796         static struct option lgopts[] = {
 797                 {"config", 1, 0, 0},
 798                 {"no-numa", 0, 0, 0},
 799                 {NULL, 0, 0, 0}
 800         };
 801
 802         argvopt = argv;
 803
 804         while ((opt = getopt_long(argc, argvopt, "p:",
 805                                 lgopts, &option_index)) != EOF) {
 806
 807                 switch (opt) {
 808                 /* portmask */
 809                 case 'p':
 810                         enabled_port_mask = parse_portmask(optarg);
 811                         if (enabled_port_mask == 0) {
 812                                 printf("invalid portmask\n");
 813                                 print_usage(prgname);
 814                                 return -1;
 815                         }
 816                         break;
 817
 818                 /* long options */
 819                 case 0:
 820                         if (!strcmp(lgopts[option_index].name, "config")) {
 821                                 ret = parse_config(optarg);
 822                                 if (ret) {
 823                                         printf("invalid config\n");
 824                                         print_usage(prgname);
 825                                         return -1;
 826                                 }
 827                         }
 828
 829                         if (!strcmp(lgopts[option_index].name, "no-numa")) {
 830                                 printf("numa is disabled \n");
 831                                 numa_on = 0;
 832                         }
 833                         break;
 834
 835                 default:
 836                         print_usage(prgname);
 837                         return -1;
 838                 }
 839         }
 840
 841         if (optind >= 0)
 842                 argv[optind-1] = prgname;
 843
 844         ret = optind-1;
 845         optind = 0; /* reset getopt lib */
 846         return ret;
 847 }
 848
 849 static void
 850 print_ethaddr(const char *name, const struct ether_addr *eth_addr)
 851 {
 852         printf ("%s%02X:%02X:%02X:%02X:%02X:%02X", name,
 853                 eth_addr->addr_bytes[0],
 854                 eth_addr->addr_bytes[1],
 855                 eth_addr->addr_bytes[2],
 856                 eth_addr->addr_bytes[3],
 857                 eth_addr->addr_bytes[4],
 858                 eth_addr->addr_bytes[5]);
 859 }
 860
 861 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 862 static void
 863 setup_hash(int socketid)
 864 {
 865         unsigned i;
 866         int ret;
 867         char s[64];
 868
 869         /* create  hashes */
 870         rte_snprintf(s, sizeof(s), "l3fwd_hash_%d", socketid);
 871         l3fwd_hash_params.name = s;
 872         l3fwd_hash_params.socket_id = socketid;
 873         l3fwd_lookup_struct[socketid] = rte_hash_create(&l3fwd_hash_params);
 874         if (l3fwd_lookup_struct[socketid] == NULL)
 875                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
 876                                 "socket %d\n", socketid);
 877
 878         /* populate the hash */
 879         for (i = 0; i < L3FWD_NUM_ROUTES; i++) {
 880                 ret = rte_hash_add_key (l3fwd_lookup_struct[socketid],
 881                                 (void *) &l3fwd_route_array[i].key);
 882                 if (ret < 0) {
 883                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
 884                                 "l3fwd hash on socket %d\n", i, socketid);
 885                 }
 886                 l3fwd_out_if[ret] = l3fwd_route_array[i].if_out;
 887                 printf("Hash: Adding key\n");
 888                 print_key(l3fwd_route_array[i].key);
 889         }
 890 }
 891 #endif
 892
 893 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 894 static void
 895 setup_lpm(int socketid)
 896 {
 897         unsigned i;
 898         int ret;
 899         char s[64];
 900
 901         /* create the LPM table */
 902         rte_snprintf(s, sizeof(s), "L3FWD_LPM_%d", socketid);
 903         l3fwd_lookup_struct[socketid] = rte_lpm_create(s, socketid,
 904                                 L3FWD_LPM_MAX_RULES, 0);
 905         if (l3fwd_lookup_struct[socketid] == NULL)
 906                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
 907                                 " on socket %d\n", socketid);
 908
 909         /* populate the LPM table */
 910         for (i = 0; i < L3FWD_NUM_ROUTES; i++) {
 911                 ret = rte_lpm_add(l3fwd_lookup_struct[socketid],
 912                         l3fwd_route_array[i].ip,
 913                         l3fwd_route_array[i].depth,
 914                         l3fwd_route_array[i].if_out);
 915
 916                 if (ret < 0) {
 917                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
 918                                 "l3fwd LPM table on socket %d\n",
 919                                 i, socketid);
 920                 }
 921
 922                 printf("LPM: Adding route 0x%08x / %d (%d)\n",
 923                         (unsigned)l3fwd_route_array[i].ip,
 924                         l3fwd_route_array[i].depth,
 925                         l3fwd_route_array[i].if_out);
 926         }
 927 }
 928 #endif
 929
 930 static int
 931 init_mem(unsigned nb_mbuf)
 932 {
 933         struct lcore_conf *qconf;
 934         int socketid;
 935         unsigned lcore_id;
 936         char s[64];
 937
 938         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
 939                 if (rte_lcore_is_enabled(lcore_id) == 0)
 940                         continue;
 941
 942                 if (numa_on)
 943                         socketid = rte_lcore_to_socket_id(lcore_id);
 944                 else
 945                         socketid = 0;
 946
 947                 if (socketid >= NB_SOCKETS) {
 948                         rte_exit(EXIT_FAILURE, "Socket %d of lcore %u is out of range %d\n",
 949                                 socketid, lcore_id, NB_SOCKETS);
 950                 }
 951                 if (pktmbuf_pool[socketid] == NULL) {
 952                         rte_snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
 953                         pktmbuf_pool[socketid] =
 954                                 rte_mempool_create(s, nb_mbuf, MBUF_SIZE,
 955                                                    MEMPOOL_CACHE_SIZE,
 956                                         sizeof(struct rte_pktmbuf_pool_private),
 957                                         rte_pktmbuf_pool_init, NULL,
 958                                         rte_pktmbuf_init, NULL,
 959                                         socketid, 0);
 960                         if (pktmbuf_pool[socketid] == NULL)
 961                                 rte_exit(EXIT_FAILURE, "Cannot init mbuf pool on socket %d\n", socketid);
 962                         else
 963                                 printf("Allocated mbuf pool on socket %d\n", socketid);
 964
 965 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 966                         setup_lpm(socketid);
 967 #else
 968                         setup_hash(socketid);
 969 #endif
 970                 }
 971                 qconf = &lcore_conf[lcore_id];
 972                 qconf->lookup_struct = l3fwd_lookup_struct[socketid];
 973         }
 974         return 0;
 975 }
 976
 977 int
 978 MAIN(int argc, char **argv)
 979 {
 980         struct lcore_conf *qconf;
 981         int ret;
 982         unsigned nb_ports;
 983         uint16_t queueid;
 984         unsigned lcore_id;
 985         uint32_t nb_lcores;
 986         uint16_t n_tx_queue;
 987         uint8_t portid, nb_rx_queue, queue, socketid;
 988
 989         signal(SIGINT, signal_handler);
 990         /* init EAL */
 991         ret = rte_eal_init(argc, argv);
 992         if (ret < 0)
 993                 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
 994         argc -= ret;
 995         argv += ret;
 996
 997         /* parse application arguments (after the EAL ones) */
 998         ret = parse_args(argc, argv);
 999         if (ret < 0)
1000                 rte_exit(EXIT_FAILURE, "Invalid L3FWD-VF parameters\n");
1001
1002         if (check_lcore_params() < 0)
1003                 rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
1004
1005         ret = init_lcore_rx_queues();
1006         if (ret < 0)
1007                 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
1008
1009         /* init driver */
1010         if (rte_pmd_init_all() < 0)
1011                 rte_exit(EXIT_FAILURE, "Cannot init pmd\n");
1012
1013         if (rte_eal_pci_probe() < 0)
1014                 rte_exit(EXIT_FAILURE, "Cannot probe PCI\n");
1015
1016         nb_ports = rte_eth_dev_count();
1017         if (nb_ports > RTE_MAX_ETHPORTS)
1018                 nb_ports = RTE_MAX_ETHPORTS;
1019
1020         if (check_port_config(nb_ports) < 0)
1021                 rte_exit(EXIT_FAILURE, "check_port_config failed\n");
1022
1023         nb_lcores = rte_lcore_count();
1024
1025         /* initialize all ports */
1026         for (portid = 0; portid < nb_ports; portid++) {
1027                 /* skip ports that are not enabled */
1028                 if ((enabled_port_mask & (1 << portid)) == 0) {
1029                         printf("\nSkipping disabled port %d\n", portid);
1030                         continue;
1031                 }
1032
1033                 /* init port */
1034                 printf("Initializing port %d ... ", portid );
1035                 fflush(stdout);
1036
1037                 /* must always equal(=1) */
1038                 nb_rx_queue = get_port_n_rx_queues(portid);
1039                 n_tx_queue = MAX_TX_QUEUE_PER_PORT;
1040
1041                 printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
1042                         nb_rx_queue, (unsigned)1 );
1043                 ret = rte_eth_dev_configure(portid, nb_rx_queue, n_tx_queue, &port_conf);
1044                 if (ret < 0)
1045                         rte_exit(EXIT_FAILURE, "Cannot configure device: err=%d, port=%d\n",
1046                                 ret, portid);
1047
1048                 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
1049                 print_ethaddr(" Address:", &ports_eth_addr[portid]);
1050                 printf(", ");
1051
1052                 ret = init_mem(NB_MBUF);
1053                 if (ret < 0)
1054                         rte_exit(EXIT_FAILURE, "init_mem failed\n");
1055
1056                 /* init one TX queue */
1057                 socketid = (uint8_t)rte_lcore_to_socket_id(rte_get_master_lcore());
1058
1059                 printf("txq=%d,%d,%d ", portid, 0, socketid);
1060                 fflush(stdout);
1061                 ret = rte_eth_tx_queue_setup(portid, 0, nb_txd,
1062                                                  socketid, &tx_conf);
1063                 if (ret < 0)
1064                         rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: err=%d, "
1065                                 "port=%d\n", ret, portid);
1066
1067                 printf("\n");
1068         }
1069
1070         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1071                 if (rte_lcore_is_enabled(lcore_id) == 0)
1072                         continue;
1073                 qconf = &lcore_conf[lcore_id];
1074                 qconf->tx_queue_id = 0;
1075
1076                 printf("\nInitializing rx queues on lcore %u ... ", lcore_id );
1077                 fflush(stdout);
1078                 /* init RX queues */
1079                 for(queue = 0; queue < qconf->n_rx_queue; ++queue) {
1080                         portid = qconf->rx_queue_list[queue].port_id;
1081                         queueid = qconf->rx_queue_list[queue].queue_id;
1082
1083                         if (numa_on)
1084                                 socketid = (uint8_t)rte_lcore_to_socket_id(lcore_id);
1085                         else
1086                                 socketid = 0;
1087
1088                         printf("rxq=%d,%d,%d ", portid, queueid, socketid);
1089                         fflush(stdout);
1090
1091                         ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
1092                                                 socketid, &rx_conf, pktmbuf_pool[socketid]);
1093                         if (ret < 0)
1094                                 rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: err=%d,"
1095                                                 "port=%d\n", ret, portid);
1096                 }
1097         }
1098         printf("\n");
1099
1100         /* start ports */
1101         for (portid = 0; portid < nb_ports; portid++) {
1102                 if ((enabled_port_mask & (1 << portid)) == 0) {
1103                         continue;
1104                 }
1105                 /* Start device */
1106                 ret = rte_eth_dev_start(portid);
1107                 if (ret < 0)
1108                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%d\n",
1109                                 ret, portid);
1110
1111                 printf("done: Port %d\n", portid);
1112
1113         }
1114
1115         /* launch per-lcore init on every lcore */
1116         rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
1117         RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1118                 if (rte_eal_wait_lcore(lcore_id) < 0)
1119                         return -1;
1120         }
1121
1122         return 0;
1123 }