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