examples/l3fwd-vf/main.c

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