examples/l3fwd-vf/main.c

   1 /*-
   2  *   BSD LICENSE
   3  *
   4  *   Copyright(c) 2010-2013 Intel Corporation. All rights reserved.
   5  *   All rights reserved.
   6  *
   7  *   Redistribution and use in source and binary forms, with or without
   8  *   modification, are permitted provided that the following conditions
   9  *   are met:
  10  *
  11  *     * Redistributions of source code must retain the above copyright
  12  *       notice, this list of conditions and the following disclaimer.
  13  *     * Redistributions in binary form must reproduce the above copyright
  14  *       notice, this list of conditions and the following disclaimer in
  15  *       the documentation and/or other materials provided with the
  16  *       distribution.
  17  *     * Neither the name of Intel Corporation nor the names of its
  18  *       contributors may be used to endorse or promote products derived
  19  *       from this software without specific prior written permission.
  20  *
  21  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  22  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  23  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
  24  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  25  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  26  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
  27  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  28  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  29  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  30  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
  31  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  32  *
  33  */
  34
  35 #include <stdio.h>
  36 #include <stdlib.h>
  37 #include <stdint.h>
  38 #include <inttypes.h>
  39 #include <sys/types.h>
  40 #include <string.h>
  41 #include <sys/queue.h>
  42 #include <stdarg.h>
  43 #include <errno.h>
  44 #include <getopt.h>
  45 #include <signal.h>
  46
  47 #include <rte_common.h>
  48 #include <rte_byteorder.h>
  49 #include <rte_log.h>
  50 #include <rte_memory.h>
  51 #include <rte_memcpy.h>
  52 #include <rte_memzone.h>
  53 #include <rte_tailq.h>
  54 #include <rte_eal.h>
  55 #include <rte_per_lcore.h>
  56 #include <rte_launch.h>
  57 #include <rte_atomic.h>
  58 #include <rte_cycles.h>
  59 #include <rte_prefetch.h>
  60 #include <rte_lcore.h>
  61 #include <rte_per_lcore.h>
  62 #include <rte_branch_prediction.h>
  63 #include <rte_interrupts.h>
  64 #include <rte_pci.h>
  65 #include <rte_random.h>
  66 #include <rte_debug.h>
  67 #include <rte_ether.h>
  68 #include <rte_ethdev.h>
  69 #include <rte_ring.h>
  70 #include <rte_mempool.h>
  71 #include <rte_mbuf.h>
  72 #include <rte_ip.h>
  73 #include <rte_tcp.h>
  74 #include <rte_udp.h>
  75 #include <rte_string_fns.h>
  76
  77 #include "main.h"
  78
  79 #define APP_LOOKUP_EXACT_MATCH          0
  80 #define APP_LOOKUP_LPM                  1
  81 #define DO_RFC_1812_CHECKS
  82
  83 //#define APP_LOOKUP_METHOD             APP_LOOKUP_EXACT_MATCH
  84 #ifndef APP_LOOKUP_METHOD
  85 #define APP_LOOKUP_METHOD             APP_LOOKUP_LPM
  86 #endif
  87
  88 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
  89 #include <rte_hash.h>
  90 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
  91 #include <rte_lpm.h>
  92 #else
  93 #error "APP_LOOKUP_METHOD set to incorrect value"
  94 #endif
  95
  96 #define RTE_LOGTYPE_L3FWD RTE_LOGTYPE_USER1
  97
  98 #define MBUF_SIZE (2048 + sizeof(struct rte_mbuf) + RTE_PKTMBUF_HEADROOM)
  99 #define NB_MBUF   8192
 100
 101 /*
 102  * RX and TX Prefetch, Host, and Write-back threshold values should be
 103  * carefully set for optimal performance. Consult the network
 104  * controller's datasheet and supporting DPDK documentation for guidance
 105  * on how these parameters should be set.
 106  */
 107 #define RX_PTHRESH 8 /**< Default values of RX prefetch threshold reg. */
 108 #define RX_HTHRESH 8 /**< Default values of RX host threshold reg. */
 109 #define RX_WTHRESH 4 /**< Default values of RX write-back threshold reg. */
 110
 111 /*
 112  * These default values are optimized for use with the Intel(R) 82599 10 GbE
 113  * Controller and the DPDK ixgbe PMD. Consider using other values for other
 114  * network controllers and/or network drivers.
 115  */
 116 #define TX_PTHRESH 36 /**< Default values of TX prefetch threshold reg. */
 117 #define TX_HTHRESH 0  /**< Default values of TX host threshold reg. */
 118 #define TX_WTHRESH 0  /**< Default values of TX write-back threshold reg. */
 119
 120 #define MAX_PKT_BURST 32
 121 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
 122
 123 #define NB_SOCKETS 8
 124
 125 #define SOCKET0 0
 126
 127 /* Configure how many packets ahead to prefetch, when reading packets */
 128 #define PREFETCH_OFFSET 3
 129
 130 /*
 131  * Configurable number of RX/TX ring descriptors
 132  */
 133 #define RTE_TEST_RX_DESC_DEFAULT 128
 134 #define RTE_TEST_TX_DESC_DEFAULT 512
 135 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
 136 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
 137
 138 /* ethernet addresses of ports */
 139 static struct ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
 140
 141 /* mask of enabled ports */
 142 static uint32_t enabled_port_mask = 0;
 143 static int numa_on = 1; /**< NUMA is enabled by default. */
 144
 145 struct mbuf_table {
 146         uint16_t len;
 147         struct rte_mbuf *m_table[MAX_PKT_BURST];
 148 };
 149
 150 struct lcore_rx_queue {
 151         uint8_t port_id;
 152         uint8_t queue_id;
 153 } __rte_cache_aligned;
 154
 155 #define MAX_RX_QUEUE_PER_LCORE 16
 156 #define MAX_TX_QUEUE_PER_PORT 1
 157 #define MAX_RX_QUEUE_PER_PORT 1
 158
 159 #define MAX_LCORE_PARAMS 1024
 160 struct lcore_params {
 161         uint8_t port_id;
 162         uint8_t queue_id;
 163         uint8_t lcore_id;
 164 } __rte_cache_aligned;
 165
 166 static struct lcore_params lcore_params_array[MAX_LCORE_PARAMS];
 167 static struct lcore_params lcore_params_array_default[] = {
 168         {0, 0, 2},
 169         {0, 1, 2},
 170         {0, 2, 2},
 171         {1, 0, 2},
 172         {1, 1, 2},
 173         {1, 2, 2},
 174         {2, 0, 2},
 175         {3, 0, 3},
 176         {3, 1, 3},
 177 };
 178
 179 static struct lcore_params * lcore_params = lcore_params_array_default;
 180 static uint16_t nb_lcore_params = sizeof(lcore_params_array_default) /
 181                                 sizeof(lcore_params_array_default[0]);
 182
 183 static struct rte_eth_conf port_conf = {
 184         .rxmode = {
 185                 .split_hdr_size = 0,
 186                 .header_split   = 0, /**< Header Split disabled */
 187                 .hw_ip_checksum = 1, /**< IP checksum offload enabled */
 188                 .hw_vlan_filter = 0, /**< VLAN filtering disabled */
 189                 .jumbo_frame    = 0, /**< Jumbo Frame Support disabled */
 190                 .hw_strip_crc   = 1, /**< CRC stripped by hardware */
 191         },
 192         .rx_adv_conf = {
 193                 .rss_conf = {
 194                         .rss_key = NULL,
 195                         .rss_hf = ETH_RSS_IPV4 | ETH_RSS_IPV6,
 196                 },
 197         },
 198         .txmode = {
 199                 .mq_mode = ETH_MQ_TX_NONE,
 200         },
 201 };
 202
 203 static const struct rte_eth_rxconf rx_conf = {
 204         .rx_thresh = {
 205                 .pthresh = RX_PTHRESH,
 206                 .hthresh = RX_HTHRESH,
 207                 .wthresh = RX_WTHRESH,
 208         },
 209 };
 210
 211 static const struct rte_eth_txconf tx_conf = {
 212         .tx_thresh = {
 213                 .pthresh = TX_PTHRESH,
 214                 .hthresh = TX_HTHRESH,
 215                 .wthresh = TX_WTHRESH,
 216         },
 217         .tx_free_thresh = 0, /* Use PMD default values */
 218         .tx_rs_thresh = 0, /* Use PMD default values */
 219 };
 220
 221 static struct rte_mempool * pktmbuf_pool[NB_SOCKETS];
 222
 223
 224 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 225
 226 #ifdef RTE_MACHINE_CPUFLAG_SSE4_2
 227 #include <rte_hash_crc.h>
 228 #define DEFAULT_HASH_FUNC       rte_hash_crc
 229 #else
 230 #include <rte_jhash.h>
 231 #define DEFAULT_HASH_FUNC       rte_jhash
 232 #endif
 233
 234 struct ipv4_5tuple {
 235         uint32_t ip_dst;
 236         uint32_t ip_src;
 237         uint16_t port_dst;
 238         uint16_t port_src;
 239         uint8_t proto;
 240 } __attribute__((__packed__));
 241
 242 struct l3fwd_route {
 243         struct ipv4_5tuple key;
 244         uint8_t if_out;
 245 };
 246
 247 static struct l3fwd_route l3fwd_route_array[] = {
 248         {{IPv4(100,10,0,1), IPv4(200,10,0,1), 101, 11, IPPROTO_TCP}, 0},
 249         {{IPv4(100,20,0,2), IPv4(200,20,0,2), 102, 12, IPPROTO_TCP}, 1},
 250         {{IPv4(100,30,0,3), IPv4(200,30,0,3), 103, 13, IPPROTO_TCP}, 2},
 251         {{IPv4(100,40,0,4), IPv4(200,40,0,4), 104, 14, IPPROTO_TCP}, 3},
 252 };
 253
 254 typedef struct rte_hash lookup_struct_t;
 255 static lookup_struct_t *l3fwd_lookup_struct[NB_SOCKETS];
 256
 257 #define L3FWD_HASH_ENTRIES      1024
 258 struct rte_hash_parameters l3fwd_hash_params = {
 259         .name = "l3fwd_hash_0",
 260         .entries = L3FWD_HASH_ENTRIES,
 261         .bucket_entries = 4,
 262         .key_len = sizeof(struct ipv4_5tuple),
 263         .hash_func = DEFAULT_HASH_FUNC,
 264         .hash_func_init_val = 0,
 265         .socket_id = SOCKET0,
 266 };
 267
 268 #define L3FWD_NUM_ROUTES \
 269         (sizeof(l3fwd_route_array) / sizeof(l3fwd_route_array[0]))
 270
 271 static uint8_t l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 272 #endif
 273
 274 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 275 struct l3fwd_route {
 276         uint32_t ip;
 277         uint8_t  depth;
 278         uint8_t  if_out;
 279 };
 280
 281 static struct l3fwd_route l3fwd_route_array[] = {
 282         {IPv4(1,1,1,0), 24, 0},
 283         {IPv4(2,1,1,0), 24, 1},
 284         {IPv4(3,1,1,0), 24, 2},
 285         {IPv4(4,1,1,0), 24, 3},
 286         {IPv4(5,1,1,0), 24, 4},
 287         {IPv4(6,1,1,0), 24, 5},
 288         {IPv4(7,1,1,0), 24, 6},
 289         {IPv4(8,1,1,0), 24, 7},
 290 };
 291
 292 #define L3FWD_NUM_ROUTES \
 293         (sizeof(l3fwd_route_array) / sizeof(l3fwd_route_array[0]))
 294
 295 #define L3FWD_LPM_MAX_RULES     1024
 296
 297 typedef struct rte_lpm lookup_struct_t;
 298 static lookup_struct_t *l3fwd_lookup_struct[NB_SOCKETS];
 299 #endif
 300
 301 struct lcore_conf {
 302         uint16_t n_rx_queue;
 303         struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
 304         uint16_t tx_queue_id;
 305         struct mbuf_table tx_mbufs[RTE_MAX_ETHPORTS];
 306         lookup_struct_t * lookup_struct;
 307 } __rte_cache_aligned;
 308
 309 static struct lcore_conf lcore_conf[RTE_MAX_LCORE];
 310
 311 /* Send burst of packets on an output interface */
 312 static inline int
 313 send_burst(struct lcore_conf *qconf, uint16_t n, uint8_t port)
 314 {
 315         struct rte_mbuf **m_table;
 316         int ret;
 317         uint16_t queueid;
 318
 319         queueid = qconf->tx_queue_id;
 320         m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
 321
 322         ret = rte_eth_tx_burst(port, queueid, m_table, n);
 323         if (unlikely(ret < n)) {
 324                 do {
 325                         rte_pktmbuf_free(m_table[ret]);
 326                 } while (++ret < n);
 327         }
 328
 329         return 0;
 330 }
 331
 332 /* Enqueue a single packet, and send burst if queue is filled */
 333 static inline int
 334 send_single_packet(struct rte_mbuf *m, uint8_t port)
 335 {
 336         uint32_t lcore_id;
 337         uint16_t len;
 338         struct lcore_conf *qconf;
 339
 340         lcore_id = rte_lcore_id();
 341
 342         qconf = &lcore_conf[lcore_id];
 343         len = qconf->tx_mbufs[port].len;
 344         qconf->tx_mbufs[port].m_table[len] = m;
 345         len++;
 346
 347         /* enough pkts to be sent */
 348         if (unlikely(len == MAX_PKT_BURST)) {
 349                 send_burst(qconf, MAX_PKT_BURST, port);
 350                 len = 0;
 351         }
 352
 353         qconf->tx_mbufs[port].len = len;
 354         return 0;
 355 }
 356
 357 #ifdef DO_RFC_1812_CHECKS
 358 static inline int
 359 is_valid_ipv4_pkt(struct ipv4_hdr *pkt, uint32_t link_len)
 360 {
 361         /* From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2 */
 362         /*
 363          * 1. The packet length reported by the Link Layer must be large
 364          * enough to hold the minimum length legal IP datagram (20 bytes).
 365          */
 366         if (link_len < sizeof(struct ipv4_hdr))
 367                 return -1;
 368
 369         /* 2. The IP checksum must be correct. */
 370         /* this is checked in H/W */
 371
 372         /*
 373          * 3. The IP version number must be 4. If the version number is not 4
 374          * then the packet may be another version of IP, such as IPng or
 375          * ST-II.
 376          */
 377         if (((pkt->version_ihl) >> 4) != 4)
 378                 return -3;
 379         /*
 380          * 4. The IP header length field must be large enough to hold the
 381          * minimum length legal IP datagram (20 bytes = 5 words).
 382          */
 383         if ((pkt->version_ihl & 0xf) < 5)
 384                 return -4;
 385
 386         /*
 387          * 5. The IP total length field must be large enough to hold the IP
 388          * datagram header, whose length is specified in the IP header length
 389          * field.
 390          */
 391         if (rte_cpu_to_be_16(pkt->total_length) < sizeof(struct ipv4_hdr))
 392                 return -5;
 393
 394         return 0;
 395 }
 396 #endif
 397
 398 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 399 static void
 400 print_key(struct ipv4_5tuple key)
 401 {
 402         printf("IP dst = %08x, IP src = %08x, port dst = %d, port src = %d, proto = %d\n",
 403                (unsigned)key.ip_dst, (unsigned)key.ip_src, key.port_dst, key.port_src, key.proto);
 404 }
 405
 406 static inline uint8_t
 407 get_dst_port(struct ipv4_hdr *ipv4_hdr,  uint8_t portid, lookup_struct_t * l3fwd_lookup_struct)
 408 {
 409         struct ipv4_5tuple key;
 410         struct tcp_hdr *tcp;
 411         struct udp_hdr *udp;
 412         int ret = 0;
 413
 414         key.ip_dst = rte_be_to_cpu_32(ipv4_hdr->dst_addr);
 415         key.ip_src = rte_be_to_cpu_32(ipv4_hdr->src_addr);
 416         key.proto = ipv4_hdr->next_proto_id;
 417
 418         switch (ipv4_hdr->next_proto_id) {
 419         case IPPROTO_TCP:
 420                 tcp = (struct tcp_hdr *)((unsigned char *) ipv4_hdr +
 421                                         sizeof(struct ipv4_hdr));
 422                 key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
 423                 key.port_src = rte_be_to_cpu_16(tcp->src_port);
 424                 break;
 425
 426         case IPPROTO_UDP:
 427                 udp = (struct udp_hdr *)((unsigned char *) ipv4_hdr +
 428                                         sizeof(struct ipv4_hdr));
 429                 key.port_dst = rte_be_to_cpu_16(udp->dst_port);
 430                 key.port_src = rte_be_to_cpu_16(udp->src_port);
 431                 break;
 432
 433         default:
 434                 key.port_dst = 0;
 435                 key.port_src = 0;
 436         }
 437
 438         /* Find destination port */
 439         ret = rte_hash_lookup(l3fwd_lookup_struct, (const void *)&key);
 440         return (uint8_t)((ret < 0)? portid : l3fwd_out_if[ret]);
 441 }
 442 #endif
 443
 444 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 445 static inline uint8_t
 446 get_dst_port(struct ipv4_hdr *ipv4_hdr,  uint8_t portid, lookup_struct_t * l3fwd_lookup_struct)
 447 {
 448         uint8_t next_hop;
 449
 450         return (uint8_t) ((rte_lpm_lookup(l3fwd_lookup_struct,
 451                         rte_be_to_cpu_32(ipv4_hdr->dst_addr), &next_hop) == 0)?
 452                         next_hop : portid);
 453 }
 454 #endif
 455
 456 static inline void
 457 l3fwd_simple_forward(struct rte_mbuf *m, uint8_t portid, lookup_struct_t * l3fwd_lookup_struct)
 458 {
 459         struct ether_hdr *eth_hdr;
 460         struct ipv4_hdr *ipv4_hdr;
 461         void *tmp;
 462         uint8_t dst_port;
 463
 464         eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
 465
 466         ipv4_hdr = (struct ipv4_hdr *)(rte_pktmbuf_mtod(m, unsigned char *) +
 467                                 sizeof(struct ether_hdr));
 468
 469 #ifdef DO_RFC_1812_CHECKS
 470         /* Check to make sure the packet is valid (RFC1812) */
 471         if (is_valid_ipv4_pkt(ipv4_hdr, m->pkt.pkt_len) < 0) {
 472                 rte_pktmbuf_free(m);
 473                 return;
 474         }
 475 #endif
 476
 477         dst_port = get_dst_port(ipv4_hdr, portid, l3fwd_lookup_struct);
 478         if (dst_port >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port) == 0)
 479                 dst_port = portid;
 480
 481         /* 02:00:00:00:00:xx */
 482         tmp = &eth_hdr->d_addr.addr_bytes[0];
 483         *((uint64_t *)tmp) = 0x000000000002 + ((uint64_t)dst_port << 40);
 484
 485 #ifdef DO_RFC_1812_CHECKS
 486         /* Update time to live and header checksum */
 487         --(ipv4_hdr->time_to_live);
 488         ++(ipv4_hdr->hdr_checksum);
 489 #endif
 490
 491         /* src addr */
 492         ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
 493
 494         send_single_packet(m, dst_port);
 495
 496 }
 497
 498 /* main processing loop */
 499 static int
 500 main_loop(__attribute__((unused)) void *dummy)
 501 {
 502         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
 503         unsigned lcore_id;
 504         uint64_t prev_tsc, diff_tsc, cur_tsc;
 505         int i, j, nb_rx;
 506         uint8_t portid, queueid;
 507         struct lcore_conf *qconf;
 508         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
 509
 510         prev_tsc = 0;
 511
 512         lcore_id = rte_lcore_id();
 513         qconf = &lcore_conf[lcore_id];
 514
 515         if (qconf->n_rx_queue == 0) {
 516                 RTE_LOG(INFO, L3FWD, "lcore %u has nothing to do\n", lcore_id);
 517                 return 0;
 518         }
 519
 520         RTE_LOG(INFO, L3FWD, "entering main loop on lcore %u\n", lcore_id);
 521
 522         for (i = 0; i < qconf->n_rx_queue; i++) {
 523
 524                 portid = qconf->rx_queue_list[i].port_id;
 525                 queueid = qconf->rx_queue_list[i].queue_id;
 526                 RTE_LOG(INFO, L3FWD, " -- lcoreid=%u portid=%hhu rxqueueid=%hhu\n", lcore_id,
 527                         portid, queueid);
 528         }
 529
 530         while (1) {
 531
 532                 cur_tsc = rte_rdtsc();
 533
 534                 /*
 535                  * TX burst queue drain
 536                  */
 537                 diff_tsc = cur_tsc - prev_tsc;
 538                 if (unlikely(diff_tsc > drain_tsc)) {
 539
 540                         /*
 541                          * This could be optimized (use queueid instead of
 542                          * portid), but it is not called so often
 543                          */
 544                         for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
 545                                 if (qconf->tx_mbufs[portid].len == 0)
 546                                         continue;
 547                                 send_burst(&lcore_conf[lcore_id],
 548                                         qconf->tx_mbufs[portid].len,
 549                                         portid);
 550                                 qconf->tx_mbufs[portid].len = 0;
 551                         }
 552
 553                         prev_tsc = cur_tsc;
 554                 }
 555
 556                 /*
 557                  * Read packet from RX queues
 558                  */
 559                 for (i = 0; i < qconf->n_rx_queue; ++i) {
 560
 561                         portid = qconf->rx_queue_list[i].port_id;
 562                         queueid = qconf->rx_queue_list[i].queue_id;
 563                         nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst, MAX_PKT_BURST);
 564
 565                         /* Prefetch first packets */
 566                         for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
 567                                 rte_prefetch0(rte_pktmbuf_mtod(
 568                                                 pkts_burst[j], void *));
 569                         }
 570
 571                         /* Prefetch and forward already prefetched packets */
 572                         for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
 573                                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
 574                                                 j + PREFETCH_OFFSET], void *));
 575                                 l3fwd_simple_forward(pkts_burst[j], portid, qconf->lookup_struct);
 576                         }
 577
 578                         /* Forward remaining prefetched packets */
 579                         for (; j < nb_rx; j++) {
 580                                 l3fwd_simple_forward(pkts_burst[j], portid, qconf->lookup_struct);
 581                         }
 582                 }
 583         }
 584 }
 585
 586 static int
 587 check_lcore_params(void)
 588 {
 589         uint8_t queue, lcore;
 590         uint16_t i;
 591         int socketid;
 592
 593         for (i = 0; i < nb_lcore_params; ++i) {
 594                 queue = lcore_params[i].queue_id;
 595                 if (queue >= MAX_RX_QUEUE_PER_PORT) {
 596                         printf("invalid queue number: %hhu\n", queue);
 597                         return -1;
 598                 }
 599                 lcore = lcore_params[i].lcore_id;
 600                 if (!rte_lcore_is_enabled(lcore)) {
 601                         printf("error: lcore %hhu is not enabled in lcore mask\n", lcore);
 602                         return -1;
 603                 }
 604                 if ((socketid = rte_lcore_to_socket_id(lcore) != 0) &&
 605                         (numa_on == 0)) {
 606                         printf("warning: lcore %hhu is on socket %d with numa off \n",
 607                                 lcore, socketid);
 608                 }
 609         }
 610         return 0;
 611 }
 612
 613 static int
 614 check_port_config(const unsigned nb_ports)
 615 {
 616         unsigned portid;
 617         uint16_t i;
 618
 619         for (i = 0; i < nb_lcore_params; ++i) {
 620                 portid = lcore_params[i].port_id;
 621                 if ((enabled_port_mask & (1 << portid)) == 0) {
 622                         printf("port %u is not enabled in port mask\n", portid);
 623                         return -1;
 624                 }
 625                 if (portid >= nb_ports) {
 626                         printf("port %u is not present on the board\n", portid);
 627                         return -1;
 628                 }
 629         }
 630         return 0;
 631 }
 632
 633 static uint8_t
 634 get_port_n_rx_queues(const uint8_t port)
 635 {
 636         int queue = -1;
 637         uint16_t i;
 638
 639         for (i = 0; i < nb_lcore_params; ++i) {
 640                 if (lcore_params[i].port_id == port && lcore_params[i].queue_id > queue)
 641                         queue = lcore_params[i].queue_id;
 642         }
 643         return (uint8_t)(++queue);
 644 }
 645
 646 static int
 647 init_lcore_rx_queues(void)
 648 {
 649         uint16_t i, nb_rx_queue;
 650         uint8_t lcore;
 651
 652         for (i = 0; i < nb_lcore_params; ++i) {
 653                 lcore = lcore_params[i].lcore_id;
 654                 nb_rx_queue = lcore_conf[lcore].n_rx_queue;
 655                 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
 656                         printf("error: too many queues (%u) for lcore: %u\n",
 657                                 (unsigned)nb_rx_queue + 1, (unsigned)lcore);
 658                         return -1;
 659                 } else {
 660                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
 661                                 lcore_params[i].port_id;
 662                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
 663                                 lcore_params[i].queue_id;
 664                         lcore_conf[lcore].n_rx_queue++;
 665                 }
 666         }
 667         return 0;
 668 }
 669
 670 /* display usage */
 671 static void
 672 print_usage(const char *prgname)
 673 {
 674         printf ("%s [EAL options] -- -p PORTMASK"
 675                 "  [--config (port,queue,lcore)[,(port,queue,lcore]]\n"
 676                 "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
 677                 "  --config (port,queue,lcore): rx queues configuration\n"
 678                 "  --no-numa: optional, disable numa awareness\n",
 679                 prgname);
 680 }
 681
 682 /* Custom handling of signals to handle process terminal */
 683 static void
 684 signal_handler(int signum)
 685 {
 686         uint8_t portid;
 687         uint8_t nb_ports = rte_eth_dev_count();
 688
 689         /* When we receive a SIGINT signal */
 690         if (signum == SIGINT) {
 691                 for (portid = 0; portid < nb_ports; portid++) {
 692                         /* skip ports that are not enabled */
 693                         if ((enabled_port_mask & (1 << portid)) == 0)
 694                                 continue;
 695                         rte_eth_dev_close(portid);
 696                 }
 697         }
 698         rte_exit(EXIT_SUCCESS, "\n User forced exit\n");
 699 }
 700 static int
 701 parse_portmask(const char *portmask)
 702 {
 703         char *end = NULL;
 704         unsigned long pm;
 705
 706         /* parse hexadecimal string */
 707         pm = strtoul(portmask, &end, 16);
 708         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
 709                 return -1;
 710
 711         if (pm == 0)
 712                 return -1;
 713
 714         return pm;
 715 }
 716
 717 static int
 718 parse_config(const char *q_arg)
 719 {
 720         char s[256];
 721         const char *p, *p0 = q_arg;
 722         char *end;
 723         enum fieldnames {
 724                 FLD_PORT = 0,
 725                 FLD_QUEUE,
 726                 FLD_LCORE,
 727                 _NUM_FLD
 728         };
 729         unsigned long int_fld[_NUM_FLD];
 730         char *str_fld[_NUM_FLD];
 731         int i;
 732         unsigned size;
 733
 734         nb_lcore_params = 0;
 735
 736         while ((p = strchr(p0,'(')) != NULL) {
 737                 ++p;
 738                 if((p0 = strchr(p,')')) == NULL)
 739                         return -1;
 740
 741                 size = p0 - p;
 742                 if(size >= sizeof(s))
 743                         return -1;
 744
 745                 rte_snprintf(s, sizeof(s), "%.*s", size, p);
 746                 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') != _NUM_FLD)
 747                         return -1;
 748                 for (i = 0; i < _NUM_FLD; i++){
 749                         errno = 0;
 750                         int_fld[i] = strtoul(str_fld[i], &end, 0);
 751                         if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
 752                                 return -1;
 753                 }
 754                 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
 755                         printf("exceeded max number of lcore params: %hu\n",
 756                                 nb_lcore_params);
 757                         return -1;
 758                 }
 759                 lcore_params_array[nb_lcore_params].port_id = (uint8_t)int_fld[FLD_PORT];
 760                 lcore_params_array[nb_lcore_params].queue_id = (uint8_t)int_fld[FLD_QUEUE];
 761                 lcore_params_array[nb_lcore_params].lcore_id = (uint8_t)int_fld[FLD_LCORE];
 762                 ++nb_lcore_params;
 763         }
 764         lcore_params = lcore_params_array;
 765         return 0;
 766 }
 767
 768 /* Parse the argument given in the command line of the application */
 769 static int
 770 parse_args(int argc, char **argv)
 771 {
 772         int opt, ret;
 773         char **argvopt;
 774         int option_index;
 775         char *prgname = argv[0];
 776         static struct option lgopts[] = {
 777                 {"config", 1, 0, 0},
 778                 {"no-numa", 0, 0, 0},
 779                 {NULL, 0, 0, 0}
 780         };
 781
 782         argvopt = argv;
 783
 784         while ((opt = getopt_long(argc, argvopt, "p:",
 785                                 lgopts, &option_index)) != EOF) {
 786
 787                 switch (opt) {
 788                 /* portmask */
 789                 case 'p':
 790                         enabled_port_mask = parse_portmask(optarg);
 791                         if (enabled_port_mask == 0) {
 792                                 printf("invalid portmask\n");
 793                                 print_usage(prgname);
 794                                 return -1;
 795                         }
 796                         break;
 797
 798                 /* long options */
 799                 case 0:
 800                         if (!strcmp(lgopts[option_index].name, "config")) {
 801                                 ret = parse_config(optarg);
 802                                 if (ret) {
 803                                         printf("invalid config\n");
 804                                         print_usage(prgname);
 805                                         return -1;
 806                                 }
 807                         }
 808
 809                         if (!strcmp(lgopts[option_index].name, "no-numa")) {
 810                                 printf("numa is disabled \n");
 811                                 numa_on = 0;
 812                         }
 813                         break;
 814
 815                 default:
 816                         print_usage(prgname);
 817                         return -1;
 818                 }
 819         }
 820
 821         if (optind >= 0)
 822                 argv[optind-1] = prgname;
 823
 824         ret = optind-1;
 825         optind = 0; /* reset getopt lib */
 826         return ret;
 827 }
 828
 829 static void
 830 print_ethaddr(const char *name, const struct ether_addr *eth_addr)
 831 {
 832         printf ("%s%02X:%02X:%02X:%02X:%02X:%02X", name,
 833                 eth_addr->addr_bytes[0],
 834                 eth_addr->addr_bytes[1],
 835                 eth_addr->addr_bytes[2],
 836                 eth_addr->addr_bytes[3],
 837                 eth_addr->addr_bytes[4],
 838                 eth_addr->addr_bytes[5]);
 839 }
 840
 841 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 842 static void
 843 setup_hash(int socketid)
 844 {
 845         unsigned i;
 846         int ret;
 847         char s[64];
 848
 849         /* create  hashes */
 850         rte_snprintf(s, sizeof(s), "l3fwd_hash_%d", socketid);
 851         l3fwd_hash_params.name = s;
 852         l3fwd_hash_params.socket_id = socketid;
 853         l3fwd_lookup_struct[socketid] = rte_hash_create(&l3fwd_hash_params);
 854         if (l3fwd_lookup_struct[socketid] == NULL)
 855                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
 856                                 "socket %d\n", socketid);
 857
 858         /* populate the hash */
 859         for (i = 0; i < L3FWD_NUM_ROUTES; i++) {
 860                 ret = rte_hash_add_key (l3fwd_lookup_struct[socketid],
 861                                 (void *) &l3fwd_route_array[i].key);
 862                 if (ret < 0) {
 863                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
 864                                 "l3fwd hash on socket %d\n", i, socketid);
 865                 }
 866                 l3fwd_out_if[ret] = l3fwd_route_array[i].if_out;
 867                 printf("Hash: Adding key\n");
 868                 print_key(l3fwd_route_array[i].key);
 869         }
 870 }
 871 #endif
 872
 873 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 874 static void
 875 setup_lpm(int socketid)
 876 {
 877         unsigned i;
 878         int ret;
 879         char s[64];
 880
 881         /* create the LPM table */
 882         rte_snprintf(s, sizeof(s), "L3FWD_LPM_%d", socketid);
 883         l3fwd_lookup_struct[socketid] = rte_lpm_create(s, socketid,
 884                                 L3FWD_LPM_MAX_RULES, 0);
 885         if (l3fwd_lookup_struct[socketid] == NULL)
 886                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
 887                                 " on socket %d\n", socketid);
 888
 889         /* populate the LPM table */
 890         for (i = 0; i < L3FWD_NUM_ROUTES; i++) {
 891                 ret = rte_lpm_add(l3fwd_lookup_struct[socketid],
 892                         l3fwd_route_array[i].ip,
 893                         l3fwd_route_array[i].depth,
 894                         l3fwd_route_array[i].if_out);
 895
 896                 if (ret < 0) {
 897                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
 898                                 "l3fwd LPM table on socket %d\n",
 899                                 i, socketid);
 900                 }
 901
 902                 printf("LPM: Adding route 0x%08x / %d (%d)\n",
 903                         (unsigned)l3fwd_route_array[i].ip,
 904                         l3fwd_route_array[i].depth,
 905                         l3fwd_route_array[i].if_out);
 906         }
 907 }
 908 #endif
 909
 910 static int
 911 init_mem(void)
 912 {
 913         struct lcore_conf *qconf;
 914         int socketid;
 915         unsigned lcore_id;
 916         char s[64];
 917
 918         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
 919                 if (rte_lcore_is_enabled(lcore_id) == 0)
 920                         continue;
 921
 922                 if (numa_on)
 923                         socketid = rte_lcore_to_socket_id(lcore_id);
 924                 else
 925                         socketid = 0;
 926
 927                 if (socketid >= NB_SOCKETS) {
 928                         rte_exit(EXIT_FAILURE, "Socket %d of lcore %u is out of range %d\n",
 929                                 socketid, lcore_id, NB_SOCKETS);
 930                 }
 931                 if (pktmbuf_pool[socketid] == NULL) {
 932                         rte_snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
 933                         pktmbuf_pool[socketid] =
 934                                 rte_mempool_create(s, NB_MBUF, MBUF_SIZE, 32,
 935                                         sizeof(struct rte_pktmbuf_pool_private),
 936                                         rte_pktmbuf_pool_init, NULL,
 937                                         rte_pktmbuf_init, NULL,
 938                                         socketid, 0);
 939                         if (pktmbuf_pool[socketid] == NULL)
 940                                 rte_exit(EXIT_FAILURE, "Cannot init mbuf pool on socket %d\n", socketid);
 941                         else
 942                                 printf("Allocated mbuf pool on socket %d\n", socketid);
 943
 944 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 945                         setup_lpm(socketid);
 946 #else
 947                         setup_hash(socketid);
 948 #endif
 949                 }
 950                 qconf = &lcore_conf[lcore_id];
 951                 qconf->lookup_struct = l3fwd_lookup_struct[socketid];
 952         }
 953         return 0;
 954 }
 955
 956 int
 957 MAIN(int argc, char **argv)
 958 {
 959         struct lcore_conf *qconf;
 960         int ret;
 961         unsigned nb_ports;
 962         uint16_t queueid;
 963         unsigned lcore_id;
 964         uint8_t portid, nb_rx_queue, queue, socketid;
 965
 966         signal(SIGINT, signal_handler);
 967         /* init EAL */
 968         ret = rte_eal_init(argc, argv);
 969         if (ret < 0)
 970                 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
 971         argc -= ret;
 972         argv += ret;
 973
 974         /* parse application arguments (after the EAL ones) */
 975         ret = parse_args(argc, argv);
 976         if (ret < 0)
 977                 rte_exit(EXIT_FAILURE, "Invalid L3FWD-VF parameters\n");
 978
 979         if (check_lcore_params() < 0)
 980                 rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
 981
 982         ret = init_lcore_rx_queues();
 983         if (ret < 0)
 984                 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
 985
 986         ret = init_mem();
 987         if (ret < 0)
 988                 rte_exit(EXIT_FAILURE, "init_mem failed\n");
 989
 990         /* init driver */
 991         if (rte_pmd_init_all() < 0)
 992                 rte_exit(EXIT_FAILURE, "Cannot init pmd\n");
 993
 994         if (rte_eal_pci_probe() < 0)
 995                 rte_exit(EXIT_FAILURE, "Cannot probe PCI\n");
 996
 997         nb_ports = rte_eth_dev_count();
 998         if (nb_ports > RTE_MAX_ETHPORTS)
 999                 nb_ports = RTE_MAX_ETHPORTS;
1000
1001         if (check_port_config(nb_ports) < 0)
1002                 rte_exit(EXIT_FAILURE, "check_port_config failed\n");
1003
1004         /* initialize all ports */
1005         for (portid = 0; portid < nb_ports; portid++) {
1006                 /* skip ports that are not enabled */
1007                 if ((enabled_port_mask & (1 << portid)) == 0) {
1008                         printf("\nSkipping disabled port %d\n", portid);
1009                         continue;
1010                 }
1011
1012                 /* init port */
1013                 printf("Initializing port %d ... ", portid );
1014                 fflush(stdout);
1015
1016                 /* must always equal(=1) */
1017                 nb_rx_queue = get_port_n_rx_queues(portid);
1018
1019                 printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
1020                         nb_rx_queue, (unsigned)1 );
1021                 ret = rte_eth_dev_configure(portid, nb_rx_queue, 1, &port_conf);
1022                 if (ret < 0)
1023                         rte_exit(EXIT_FAILURE, "Cannot configure device: err=%d, port=%d\n",
1024                                 ret, portid);
1025
1026                 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
1027                 print_ethaddr(" Address:", &ports_eth_addr[portid]);
1028                 printf(", ");
1029
1030                 /* init one TX queue */
1031                 socketid = 0;
1032
1033                 printf("txq=%d,%d ", 0, socketid);
1034                 fflush(stdout);
1035                 ret = rte_eth_tx_queue_setup(portid, 0, nb_txd,
1036                                                  socketid, &tx_conf);
1037                 if (ret < 0)
1038                         rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: err=%d, "
1039                                 "port=%d\n", ret, portid);
1040
1041                 printf("\n");
1042         }
1043
1044         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1045                 if (rte_lcore_is_enabled(lcore_id) == 0)
1046                         continue;
1047                 qconf = &lcore_conf[lcore_id];
1048                 qconf->tx_queue_id = 0;
1049
1050                 printf("\nInitializing rx queues on lcore %u ... ", lcore_id );
1051                 fflush(stdout);
1052                 /* init RX queues */
1053                 for(queue = 0; queue < qconf->n_rx_queue; ++queue) {
1054                         portid = qconf->rx_queue_list[queue].port_id;
1055                         queueid = qconf->rx_queue_list[queue].queue_id;
1056
1057                         if (numa_on)
1058                                 socketid = (uint8_t)rte_lcore_to_socket_id(lcore_id);
1059                         else
1060                                 socketid = 0;
1061
1062                         printf("rxq=%d,%d,%d ", portid, queueid, socketid);
1063                         fflush(stdout);
1064
1065                         ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
1066                                                 socketid, &rx_conf, pktmbuf_pool[socketid]);
1067                         if (ret < 0)
1068                                 rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: err=%d,"
1069                                                 "port=%d\n", ret, portid);
1070                 }
1071         }
1072         printf("\n");
1073
1074         /* start ports */
1075         for (portid = 0; portid < nb_ports; portid++) {
1076                 if ((enabled_port_mask & (1 << portid)) == 0) {
1077                         continue;
1078                 }
1079                 /* Start device */
1080                 ret = rte_eth_dev_start(portid);
1081                 if (ret < 0)
1082                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%d\n",
1083                                 ret, portid);
1084
1085                 printf("done: Port %d\n", portid);
1086
1087         }
1088
1089         /* launch per-lcore init on every lcore */
1090         rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
1091         RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1092                 if (rte_eal_wait_lcore(lcore_id) < 0)
1093                         return -1;
1094         }
1095
1096         return 0;
1097 }