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 1024
 110 #define RTE_TEST_TX_DESC_DEFAULT 1024
 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 rte_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 = RTE_ETHER_MAX_LEN,
 163                 .split_hdr_size = 0,
 164                 .offloads = DEV_RX_OFFLOAD_CHECKSUM,
 165         },
 166         .rx_adv_conf = {
 167                 .rss_conf = {
 168                         .rss_key = NULL,
 169                         .rss_hf = ETH_RSS_IP,
 170                 },
 171         },
 172         .txmode = {
 173                 .mq_mode = ETH_MQ_TX_NONE,
 174         },
 175 };
 176
 177 static struct rte_mempool * pktmbuf_pool[NB_SOCKETS];
 178
 179
 180 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 181
 182 #ifdef RTE_ARCH_X86
 183 #include <rte_hash_crc.h>
 184 #define DEFAULT_HASH_FUNC       rte_hash_crc
 185 #else
 186 #include <rte_jhash.h>
 187 #define DEFAULT_HASH_FUNC       rte_jhash
 188 #endif
 189
 190 struct ipv4_5tuple {
 191         uint32_t ip_dst;
 192         uint32_t ip_src;
 193         uint16_t port_dst;
 194         uint16_t port_src;
 195         uint8_t proto;
 196 } __attribute__((__packed__));
 197
 198 struct l3fwd_route {
 199         struct ipv4_5tuple key;
 200         uint8_t if_out;
 201 };
 202
 203 static struct l3fwd_route l3fwd_route_array[] = {
 204         {{RTE_IPv4(100,10,0,1), RTE_IPv4(200,10,0,1), 101, 11, IPPROTO_TCP}, 0},
 205         {{RTE_IPv4(100,20,0,2), RTE_IPv4(200,20,0,2), 102, 12, IPPROTO_TCP}, 1},
 206         {{RTE_IPv4(100,30,0,3), RTE_IPv4(200,30,0,3), 103, 13, IPPROTO_TCP}, 2},
 207         {{RTE_IPv4(100,40,0,4), RTE_IPv4(200,40,0,4), 104, 14, IPPROTO_TCP}, 3},
 208 };
 209
 210 typedef struct rte_hash lookup_struct_t;
 211 static lookup_struct_t *l3fwd_lookup_struct[NB_SOCKETS];
 212
 213 #define L3FWD_HASH_ENTRIES      1024
 214 struct rte_hash_parameters l3fwd_hash_params = {
 215         .name = "l3fwd_hash_0",
 216         .entries = L3FWD_HASH_ENTRIES,
 217         .key_len = sizeof(struct ipv4_5tuple),
 218         .hash_func = DEFAULT_HASH_FUNC,
 219         .hash_func_init_val = 0,
 220         .socket_id = SOCKET0,
 221 };
 222
 223 #define L3FWD_NUM_ROUTES \
 224         (sizeof(l3fwd_route_array) / sizeof(l3fwd_route_array[0]))
 225
 226 static uint8_t l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 227 #endif
 228
 229 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 230 struct l3fwd_route {
 231         uint32_t ip;
 232         uint8_t  depth;
 233         uint8_t  if_out;
 234 };
 235
 236 static struct l3fwd_route l3fwd_route_array[] = {
 237         {RTE_IPv4(1,1,1,0), 24, 0},
 238         {RTE_IPv4(2,1,1,0), 24, 1},
 239         {RTE_IPv4(3,1,1,0), 24, 2},
 240         {RTE_IPv4(4,1,1,0), 24, 3},
 241         {RTE_IPv4(5,1,1,0), 24, 4},
 242         {RTE_IPv4(6,1,1,0), 24, 5},
 243         {RTE_IPv4(7,1,1,0), 24, 6},
 244         {RTE_IPv4(8,1,1,0), 24, 7},
 245 };
 246
 247 #define L3FWD_NUM_ROUTES \
 248         (sizeof(l3fwd_route_array) / sizeof(l3fwd_route_array[0]))
 249
 250 #define L3FWD_LPM_MAX_RULES     1024
 251
 252 typedef struct rte_lpm lookup_struct_t;
 253 static lookup_struct_t *l3fwd_lookup_struct[NB_SOCKETS];
 254 #endif
 255
 256 struct lcore_conf {
 257         uint16_t n_rx_queue;
 258         struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
 259         uint16_t tx_queue_id;
 260         struct mbuf_table tx_mbufs[RTE_MAX_ETHPORTS];
 261         lookup_struct_t * lookup_struct;
 262 } __rte_cache_aligned;
 263
 264 static struct lcore_conf lcore_conf[RTE_MAX_LCORE];
 265 static rte_spinlock_t spinlock_conf[RTE_MAX_ETHPORTS] = {RTE_SPINLOCK_INITIALIZER};
 266 /* Send burst of packets on an output interface */
 267 static inline int
 268 send_burst(struct lcore_conf *qconf, uint16_t n, uint16_t port)
 269 {
 270         struct rte_mbuf **m_table;
 271         int ret;
 272         uint16_t queueid;
 273
 274         queueid = qconf->tx_queue_id;
 275         m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
 276
 277         rte_spinlock_lock(&spinlock_conf[port]);
 278         ret = rte_eth_tx_burst(port, queueid, m_table, n);
 279         rte_spinlock_unlock(&spinlock_conf[port]);
 280
 281         if (unlikely(ret < n)) {
 282                 do {
 283                         rte_pktmbuf_free(m_table[ret]);
 284                 } while (++ret < n);
 285         }
 286
 287         return 0;
 288 }
 289
 290 /* Enqueue a single packet, and send burst if queue is filled */
 291 static inline int
 292 send_single_packet(struct rte_mbuf *m, uint16_t port)
 293 {
 294         uint32_t lcore_id;
 295         uint16_t len;
 296         struct lcore_conf *qconf;
 297
 298         lcore_id = rte_lcore_id();
 299
 300         qconf = &lcore_conf[lcore_id];
 301         len = qconf->tx_mbufs[port].len;
 302         qconf->tx_mbufs[port].m_table[len] = m;
 303         len++;
 304
 305         /* enough pkts to be sent */
 306         if (unlikely(len == MAX_PKT_BURST)) {
 307                 send_burst(qconf, MAX_PKT_BURST, port);
 308                 len = 0;
 309         }
 310
 311         qconf->tx_mbufs[port].len = len;
 312         return 0;
 313 }
 314
 315 #ifdef DO_RFC_1812_CHECKS
 316 static inline int
 317 is_valid_ipv4_pkt(struct rte_ipv4_hdr *pkt, uint32_t link_len)
 318 {
 319         /* From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2 */
 320         /*
 321          * 1. The packet length reported by the Link Layer must be large
 322          * enough to hold the minimum length legal IP datagram (20 bytes).
 323          */
 324         if (link_len < sizeof(struct rte_ipv4_hdr))
 325                 return -1;
 326
 327         /* 2. The IP checksum must be correct. */
 328         /* this is checked in H/W */
 329
 330         /*
 331          * 3. The IP version number must be 4. If the version number is not 4
 332          * then the packet may be another version of IP, such as IPng or
 333          * ST-II.
 334          */
 335         if (((pkt->version_ihl) >> 4) != 4)
 336                 return -3;
 337         /*
 338          * 4. The IP header length field must be large enough to hold the
 339          * minimum length legal IP datagram (20 bytes = 5 words).
 340          */
 341         if ((pkt->version_ihl & 0xf) < 5)
 342                 return -4;
 343
 344         /*
 345          * 5. The IP total length field must be large enough to hold the IP
 346          * datagram header, whose length is specified in the IP header length
 347          * field.
 348          */
 349         if (rte_cpu_to_be_16(pkt->total_length) < sizeof(struct rte_ipv4_hdr))
 350                 return -5;
 351
 352         return 0;
 353 }
 354 #endif
 355
 356 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 357 static void
 358 print_key(struct ipv4_5tuple key)
 359 {
 360         printf("IP dst = %08x, IP src = %08x, port dst = %d, port src = %d, proto = %d\n",
 361                (unsigned)key.ip_dst, (unsigned)key.ip_src, key.port_dst, key.port_src, key.proto);
 362 }
 363
 364 static inline uint16_t
 365 get_dst_port(struct rte_ipv4_hdr *ipv4_hdr, uint16_t portid,
 366               lookup_struct_t *l3fwd_lookup_struct)
 367 {
 368         struct ipv4_5tuple key;
 369         struct tcp_hdr *tcp;
 370         struct udp_hdr *udp;
 371         int ret = 0;
 372
 373         key.ip_dst = rte_be_to_cpu_32(ipv4_hdr->dst_addr);
 374         key.ip_src = rte_be_to_cpu_32(ipv4_hdr->src_addr);
 375         key.proto = ipv4_hdr->next_proto_id;
 376
 377         switch (ipv4_hdr->next_proto_id) {
 378         case IPPROTO_TCP:
 379                 tcp = (struct tcp_hdr *)((unsigned char *) ipv4_hdr +
 380                                         sizeof(struct rte_ipv4_hdr));
 381                 key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
 382                 key.port_src = rte_be_to_cpu_16(tcp->src_port);
 383                 break;
 384
 385         case IPPROTO_UDP:
 386                 udp = (struct udp_hdr *)((unsigned char *) ipv4_hdr +
 387                                         sizeof(struct rte_ipv4_hdr));
 388                 key.port_dst = rte_be_to_cpu_16(udp->dst_port);
 389                 key.port_src = rte_be_to_cpu_16(udp->src_port);
 390                 break;
 391
 392         default:
 393                 key.port_dst = 0;
 394                 key.port_src = 0;
 395         }
 396
 397         /* Find destination port */
 398         ret = rte_hash_lookup(l3fwd_lookup_struct, (const void *)&key);
 399         return ((ret < 0) ? portid : l3fwd_out_if[ret]);
 400 }
 401 #endif
 402
 403 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 404 static inline uint32_t
 405 get_dst_port(struct rte_ipv4_hdr *ipv4_hdr, uint16_t portid,
 406               lookup_struct_t *l3fwd_lookup_struct)
 407 {
 408         uint32_t next_hop;
 409
 410         return ((rte_lpm_lookup(l3fwd_lookup_struct,
 411                 rte_be_to_cpu_32(ipv4_hdr->dst_addr), &next_hop) == 0) ?
 412                 next_hop : portid);
 413 }
 414 #endif
 415
 416 static inline void
 417 l3fwd_simple_forward(struct rte_mbuf *m, uint16_t portid,
 418                       lookup_struct_t *l3fwd_lookup_struct)
 419 {
 420         struct rte_ether_hdr *eth_hdr;
 421         struct rte_ipv4_hdr *ipv4_hdr;
 422         void *tmp;
 423         uint16_t dst_port;
 424
 425         eth_hdr = rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
 426
 427         ipv4_hdr = rte_pktmbuf_mtod_offset(m, struct rte_ipv4_hdr *,
 428                                            sizeof(struct rte_ether_hdr));
 429
 430 #ifdef DO_RFC_1812_CHECKS
 431         /* Check to make sure the packet is valid (RFC1812) */
 432         if (is_valid_ipv4_pkt(ipv4_hdr, m->pkt_len) < 0) {
 433                 rte_pktmbuf_free(m);
 434                 return;
 435         }
 436 #endif
 437
 438         dst_port = get_dst_port(ipv4_hdr, portid, l3fwd_lookup_struct);
 439         if (dst_port >= RTE_MAX_ETHPORTS || (enabled_port_mask & 1 << dst_port) == 0)
 440                 dst_port = portid;
 441
 442         /* 02:00:00:00:00:xx */
 443         tmp = &eth_hdr->d_addr.addr_bytes[0];
 444         *((uint64_t *)tmp) = 0x000000000002 + ((uint64_t)dst_port << 40);
 445
 446 #ifdef DO_RFC_1812_CHECKS
 447         /* Update time to live and header checksum */
 448         --(ipv4_hdr->time_to_live);
 449         ++(ipv4_hdr->hdr_checksum);
 450 #endif
 451
 452         /* src addr */
 453         rte_ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
 454
 455         send_single_packet(m, dst_port);
 456
 457 }
 458
 459 /* main processing loop */
 460 static int
 461 main_loop(__attribute__((unused)) void *dummy)
 462 {
 463         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
 464         unsigned lcore_id;
 465         uint64_t prev_tsc, diff_tsc, cur_tsc;
 466         int i, j, nb_rx;
 467         uint8_t queueid;
 468         uint16_t portid;
 469         struct lcore_conf *qconf;
 470         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
 471
 472         prev_tsc = 0;
 473
 474         lcore_id = rte_lcore_id();
 475         qconf = &lcore_conf[lcore_id];
 476
 477         if (qconf->n_rx_queue == 0) {
 478                 RTE_LOG(INFO, L3FWD, "lcore %u has nothing to do\n", lcore_id);
 479                 return 0;
 480         }
 481
 482         RTE_LOG(INFO, L3FWD, "entering main loop on lcore %u\n", lcore_id);
 483
 484         for (i = 0; i < qconf->n_rx_queue; i++) {
 485
 486                 portid = qconf->rx_queue_list[i].port_id;
 487                 queueid = qconf->rx_queue_list[i].queue_id;
 488                 RTE_LOG(INFO, L3FWD, " --lcoreid=%u portid=%u rxqueueid=%hhu\n",
 489                 lcore_id, portid, queueid);
 490         }
 491
 492         while (1) {
 493
 494                 cur_tsc = rte_rdtsc();
 495
 496                 /*
 497                  * TX burst queue drain
 498                  */
 499                 diff_tsc = cur_tsc - prev_tsc;
 500                 if (unlikely(diff_tsc > drain_tsc)) {
 501
 502                         /*
 503                          * This could be optimized (use queueid instead of
 504                          * portid), but it is not called so often
 505                          */
 506                         for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
 507                                 if (qconf->tx_mbufs[portid].len == 0)
 508                                         continue;
 509                                 send_burst(&lcore_conf[lcore_id],
 510                                         qconf->tx_mbufs[portid].len,
 511                                         portid);
 512                                 qconf->tx_mbufs[portid].len = 0;
 513                         }
 514
 515                         prev_tsc = cur_tsc;
 516                 }
 517
 518                 /*
 519                  * Read packet from RX queues
 520                  */
 521                 for (i = 0; i < qconf->n_rx_queue; ++i) {
 522
 523                         portid = qconf->rx_queue_list[i].port_id;
 524                         queueid = qconf->rx_queue_list[i].queue_id;
 525                         nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst, MAX_PKT_BURST);
 526
 527                         /* Prefetch first packets */
 528                         for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
 529                                 rte_prefetch0(rte_pktmbuf_mtod(
 530                                                 pkts_burst[j], void *));
 531                         }
 532
 533                         /* Prefetch and forward already prefetched packets */
 534                         for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
 535                                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
 536                                                 j + PREFETCH_OFFSET], void *));
 537                                 l3fwd_simple_forward(pkts_burst[j], portid, qconf->lookup_struct);
 538                         }
 539
 540                         /* Forward remaining prefetched packets */
 541                         for (; j < nb_rx; j++) {
 542                                 l3fwd_simple_forward(pkts_burst[j], portid, qconf->lookup_struct);
 543                         }
 544                 }
 545         }
 546 }
 547
 548 static int
 549 check_lcore_params(void)
 550 {
 551         uint8_t queue, lcore;
 552         uint16_t i;
 553         int socketid;
 554
 555         for (i = 0; i < nb_lcore_params; ++i) {
 556                 queue = lcore_params[i].queue_id;
 557                 if (queue >= MAX_RX_QUEUE_PER_PORT) {
 558                         printf("invalid queue number: %hhu\n", queue);
 559                         return -1;
 560                 }
 561                 lcore = lcore_params[i].lcore_id;
 562                 if (!rte_lcore_is_enabled(lcore)) {
 563                         printf("error: lcore %hhu is not enabled in lcore mask\n", lcore);
 564                         return -1;
 565                 }
 566                 if ((socketid = rte_lcore_to_socket_id(lcore) != 0) &&
 567                         (numa_on == 0)) {
 568                         printf("warning: lcore %hhu is on socket %d with numa off \n",
 569                                 lcore, socketid);
 570                 }
 571         }
 572         return 0;
 573 }
 574
 575 static int
 576 check_port_config(void)
 577 {
 578         unsigned portid;
 579         uint16_t i;
 580
 581         for (i = 0; i < nb_lcore_params; ++i) {
 582                 portid = lcore_params[i].port_id;
 583                 if ((enabled_port_mask & (1 << portid)) == 0) {
 584                         printf("port %u is not enabled in port mask\n", portid);
 585                         return -1;
 586                 }
 587                 if (!rte_eth_dev_is_valid_port(portid)) {
 588                         printf("port %u is not present on the board\n", portid);
 589                         return -1;
 590                 }
 591         }
 592         return 0;
 593 }
 594
 595 static uint8_t
 596 get_port_n_rx_queues(const uint16_t port)
 597 {
 598         int queue = -1;
 599         uint16_t i;
 600
 601         for (i = 0; i < nb_lcore_params; ++i) {
 602                 if (lcore_params[i].port_id == port && lcore_params[i].queue_id > queue)
 603                         queue = lcore_params[i].queue_id;
 604         }
 605         return (uint8_t)(++queue);
 606 }
 607
 608 static int
 609 init_lcore_rx_queues(void)
 610 {
 611         uint16_t i, nb_rx_queue;
 612         uint8_t lcore;
 613
 614         for (i = 0; i < nb_lcore_params; ++i) {
 615                 lcore = lcore_params[i].lcore_id;
 616                 nb_rx_queue = lcore_conf[lcore].n_rx_queue;
 617                 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
 618                         printf("error: too many queues (%u) for lcore: %u\n",
 619                                 (unsigned)nb_rx_queue + 1, (unsigned)lcore);
 620                         return -1;
 621                 } else {
 622                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
 623                                 lcore_params[i].port_id;
 624                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
 625                                 lcore_params[i].queue_id;
 626                         lcore_conf[lcore].n_rx_queue++;
 627                 }
 628         }
 629         return 0;
 630 }
 631
 632 /* display usage */
 633 static void
 634 print_usage(const char *prgname)
 635 {
 636         printf ("%s [EAL options] -- -p PORTMASK"
 637                 "  [--config (port,queue,lcore)[,(port,queue,lcore]]\n"
 638                 "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
 639                 "  --config (port,queue,lcore): rx queues configuration\n"
 640                 "  --no-numa: optional, disable numa awareness\n",
 641                 prgname);
 642 }
 643
 644 /* Custom handling of signals to handle process terminal */
 645 static void
 646 signal_handler(int signum)
 647 {
 648         uint16_t portid;
 649
 650         /* When we receive a SIGINT signal */
 651         if (signum == SIGINT) {
 652                 RTE_ETH_FOREACH_DEV(portid) {
 653                         /* skip ports that are not enabled */
 654                         if ((enabled_port_mask & (1 << portid)) == 0)
 655                                 continue;
 656                         rte_eth_dev_close(portid);
 657                 }
 658         }
 659         rte_exit(EXIT_SUCCESS, "\n User forced exit\n");
 660 }
 661 static int
 662 parse_portmask(const char *portmask)
 663 {
 664         char *end = NULL;
 665         unsigned long pm;
 666
 667         /* parse hexadecimal string */
 668         pm = strtoul(portmask, &end, 16);
 669         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
 670                 return -1;
 671
 672         if (pm == 0)
 673                 return -1;
 674
 675         return pm;
 676 }
 677
 678 static int
 679 parse_config(const char *q_arg)
 680 {
 681         char s[256];
 682         const char *p, *p0 = q_arg;
 683         char *end;
 684         enum fieldnames {
 685                 FLD_PORT = 0,
 686                 FLD_QUEUE,
 687                 FLD_LCORE,
 688                 _NUM_FLD
 689         };
 690         unsigned long int_fld[_NUM_FLD];
 691         char *str_fld[_NUM_FLD];
 692         int i;
 693         unsigned size;
 694
 695         nb_lcore_params = 0;
 696
 697         while ((p = strchr(p0,'(')) != NULL) {
 698                 ++p;
 699                 if((p0 = strchr(p,')')) == NULL)
 700                         return -1;
 701
 702                 size = p0 - p;
 703                 if(size >= sizeof(s))
 704                         return -1;
 705
 706                 snprintf(s, sizeof(s), "%.*s", size, p);
 707                 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') != _NUM_FLD)
 708                         return -1;
 709                 for (i = 0; i < _NUM_FLD; i++){
 710                         errno = 0;
 711                         int_fld[i] = strtoul(str_fld[i], &end, 0);
 712                         if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
 713                                 return -1;
 714                 }
 715                 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
 716                         printf("exceeded max number of lcore params: %hu\n",
 717                                 nb_lcore_params);
 718                         return -1;
 719                 }
 720                 lcore_params_array[nb_lcore_params].port_id = int_fld[FLD_PORT];
 721                 lcore_params_array[nb_lcore_params].queue_id = (uint8_t)int_fld[FLD_QUEUE];
 722                 lcore_params_array[nb_lcore_params].lcore_id = (uint8_t)int_fld[FLD_LCORE];
 723                 ++nb_lcore_params;
 724         }
 725         lcore_params = lcore_params_array;
 726         return 0;
 727 }
 728
 729 /* Parse the argument given in the command line of the application */
 730 static int
 731 parse_args(int argc, char **argv)
 732 {
 733         int opt, ret;
 734         char **argvopt;
 735         int option_index;
 736         char *prgname = argv[0];
 737         static struct option lgopts[] = {
 738                 {"config", 1, 0, 0},
 739                 {"no-numa", 0, 0, 0},
 740                 {NULL, 0, 0, 0}
 741         };
 742
 743         argvopt = argv;
 744
 745         while ((opt = getopt_long(argc, argvopt, "p:",
 746                                 lgopts, &option_index)) != EOF) {
 747
 748                 switch (opt) {
 749                 /* portmask */
 750                 case 'p':
 751                         enabled_port_mask = parse_portmask(optarg);
 752                         if (enabled_port_mask == 0) {
 753                                 printf("invalid portmask\n");
 754                                 print_usage(prgname);
 755                                 return -1;
 756                         }
 757                         break;
 758
 759                 /* long options */
 760                 case 0:
 761                         if (!strcmp(lgopts[option_index].name, "config")) {
 762                                 ret = parse_config(optarg);
 763                                 if (ret) {
 764                                         printf("invalid config\n");
 765                                         print_usage(prgname);
 766                                         return -1;
 767                                 }
 768                         }
 769
 770                         if (!strcmp(lgopts[option_index].name, "no-numa")) {
 771                                 printf("numa is disabled \n");
 772                                 numa_on = 0;
 773                         }
 774                         break;
 775
 776                 default:
 777                         print_usage(prgname);
 778                         return -1;
 779                 }
 780         }
 781
 782         if (optind >= 0)
 783                 argv[optind-1] = prgname;
 784
 785         ret = optind-1;
 786         optind = 1; /* reset getopt lib */
 787         return ret;
 788 }
 789
 790 static void
 791 print_ethaddr(const char *name, const struct rte_ether_addr *eth_addr)
 792 {
 793         char buf[RTE_ETHER_ADDR_FMT_SIZE];
 794         rte_ether_format_addr(buf, RTE_ETHER_ADDR_FMT_SIZE, eth_addr);
 795         printf("%s%s", name, buf);
 796 }
 797
 798 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 799 static void
 800 setup_hash(int socketid)
 801 {
 802         unsigned i;
 803         int ret;
 804         char s[64];
 805
 806         /* create  hashes */
 807         snprintf(s, sizeof(s), "l3fwd_hash_%d", socketid);
 808         l3fwd_hash_params.name = s;
 809         l3fwd_hash_params.socket_id = socketid;
 810         l3fwd_lookup_struct[socketid] = rte_hash_create(&l3fwd_hash_params);
 811         if (l3fwd_lookup_struct[socketid] == NULL)
 812                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
 813                                 "socket %d\n", socketid);
 814
 815         /* populate the hash */
 816         for (i = 0; i < L3FWD_NUM_ROUTES; i++) {
 817                 ret = rte_hash_add_key (l3fwd_lookup_struct[socketid],
 818                                 (void *) &l3fwd_route_array[i].key);
 819                 if (ret < 0) {
 820                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
 821                                 "l3fwd hash on socket %d\n", i, socketid);
 822                 }
 823                 l3fwd_out_if[ret] = l3fwd_route_array[i].if_out;
 824                 printf("Hash: Adding key\n");
 825                 print_key(l3fwd_route_array[i].key);
 826         }
 827 }
 828 #endif
 829
 830 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 831 static void
 832 setup_lpm(int socketid)
 833 {
 834         unsigned i;
 835         int ret;
 836         char s[64];
 837
 838         struct rte_lpm_config lpm_ipv4_config;
 839
 840         lpm_ipv4_config.max_rules = L3FWD_LPM_MAX_RULES;
 841         lpm_ipv4_config.number_tbl8s = 256;
 842         lpm_ipv4_config.flags = 0;
 843
 844         /* create the LPM table */
 845         snprintf(s, sizeof(s), "L3FWD_LPM_%d", socketid);
 846         l3fwd_lookup_struct[socketid] =
 847                         rte_lpm_create(s, socketid, &lpm_ipv4_config);
 848         if (l3fwd_lookup_struct[socketid] == NULL)
 849                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
 850                                 " on socket %d\n", socketid);
 851
 852         /* populate the LPM table */
 853         for (i = 0; i < L3FWD_NUM_ROUTES; i++) {
 854                 ret = rte_lpm_add(l3fwd_lookup_struct[socketid],
 855                         l3fwd_route_array[i].ip,
 856                         l3fwd_route_array[i].depth,
 857                         l3fwd_route_array[i].if_out);
 858
 859                 if (ret < 0) {
 860                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
 861                                 "l3fwd LPM table on socket %d\n",
 862                                 i, socketid);
 863                 }
 864
 865                 printf("LPM: Adding route 0x%08x / %d (%d)\n",
 866                         (unsigned)l3fwd_route_array[i].ip,
 867                         l3fwd_route_array[i].depth,
 868                         l3fwd_route_array[i].if_out);
 869         }
 870 }
 871 #endif
 872
 873 static int
 874 init_mem(unsigned nb_mbuf)
 875 {
 876         struct lcore_conf *qconf;
 877         int socketid;
 878         unsigned lcore_id;
 879         char s[64];
 880
 881         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
 882                 if (rte_lcore_is_enabled(lcore_id) == 0)
 883                         continue;
 884
 885                 if (numa_on)
 886                         socketid = rte_lcore_to_socket_id(lcore_id);
 887                 else
 888                         socketid = 0;
 889
 890                 if (socketid >= NB_SOCKETS) {
 891                         rte_exit(EXIT_FAILURE, "Socket %d of lcore %u is out of range %d\n",
 892                                 socketid, lcore_id, NB_SOCKETS);
 893                 }
 894                 if (pktmbuf_pool[socketid] == NULL) {
 895                         snprintf(s, sizeof(s), "mbuf_pool_%d", socketid);
 896                         pktmbuf_pool[socketid] = rte_pktmbuf_pool_create(s,
 897                                 nb_mbuf, MEMPOOL_CACHE_SIZE, 0,
 898                                 RTE_MBUF_DEFAULT_BUF_SIZE, socketid);
 899                         if (pktmbuf_pool[socketid] == NULL)
 900                                 rte_exit(EXIT_FAILURE, "Cannot init mbuf pool on socket %d\n", socketid);
 901                         else
 902                                 printf("Allocated mbuf pool on socket %d\n", socketid);
 903
 904 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 905                         setup_lpm(socketid);
 906 #else
 907                         setup_hash(socketid);
 908 #endif
 909                 }
 910                 qconf = &lcore_conf[lcore_id];
 911                 qconf->lookup_struct = l3fwd_lookup_struct[socketid];
 912         }
 913         return 0;
 914 }
 915
 916 int
 917 main(int argc, char **argv)
 918 {
 919         struct lcore_conf *qconf;
 920         struct rte_eth_dev_info dev_info;
 921         struct rte_eth_txconf *txconf;
 922         int ret;
 923         unsigned nb_ports;
 924         uint16_t queueid, portid;
 925         unsigned lcore_id;
 926         uint32_t nb_lcores;
 927         uint16_t n_tx_queue;
 928         uint8_t nb_rx_queue, queue, socketid;
 929
 930         signal(SIGINT, signal_handler);
 931         /* init EAL */
 932         ret = rte_eal_init(argc, argv);
 933         if (ret < 0)
 934                 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
 935         argc -= ret;
 936         argv += ret;
 937
 938         /* parse application arguments (after the EAL ones) */
 939         ret = parse_args(argc, argv);
 940         if (ret < 0)
 941                 rte_exit(EXIT_FAILURE, "Invalid L3FWD-VF parameters\n");
 942
 943         if (check_lcore_params() < 0)
 944                 rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
 945
 946         ret = init_lcore_rx_queues();
 947         if (ret < 0)
 948                 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
 949
 950         nb_ports = rte_eth_dev_count_avail();
 951
 952         if (check_port_config() < 0)
 953                 rte_exit(EXIT_FAILURE, "check_port_config failed\n");
 954
 955         nb_lcores = rte_lcore_count();
 956
 957         /* initialize all ports */
 958         RTE_ETH_FOREACH_DEV(portid) {
 959                 struct rte_eth_conf local_port_conf = port_conf;
 960
 961                 /* skip ports that are not enabled */
 962                 if ((enabled_port_mask & (1 << portid)) == 0) {
 963                         printf("\nSkipping disabled port %d\n", portid);
 964                         continue;
 965                 }
 966
 967                 /* init port */
 968                 printf("Initializing port %d ... ", portid );
 969                 fflush(stdout);
 970
 971                 /* must always equal(=1) */
 972                 nb_rx_queue = get_port_n_rx_queues(portid);
 973                 n_tx_queue = MAX_TX_QUEUE_PER_PORT;
 974
 975                 printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
 976                         nb_rx_queue, (unsigned)1 );
 977
 978                 rte_eth_dev_info_get(portid, &dev_info);
 979                 if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
 980                         local_port_conf.txmode.offloads |=
 981                                 DEV_TX_OFFLOAD_MBUF_FAST_FREE;
 982
 983                 local_port_conf.rx_adv_conf.rss_conf.rss_hf &=
 984                         dev_info.flow_type_rss_offloads;
 985                 if (local_port_conf.rx_adv_conf.rss_conf.rss_hf !=
 986                                 port_conf.rx_adv_conf.rss_conf.rss_hf) {
 987                         printf("Port %u modified RSS hash function based on hardware support,"
 988                                 "requested:%#"PRIx64" configured:%#"PRIx64"\n",
 989                                 portid,
 990                                 port_conf.rx_adv_conf.rss_conf.rss_hf,
 991                                 local_port_conf.rx_adv_conf.rss_conf.rss_hf);
 992                 }
 993
 994                 ret = rte_eth_dev_configure(portid, nb_rx_queue,
 995                                             n_tx_queue, &local_port_conf);
 996                 if (ret < 0)
 997                         rte_exit(EXIT_FAILURE, "Cannot configure device: err=%d, port=%d\n",
 998                                 ret, portid);
 999
1000                 ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
1001                                                        &nb_txd);
1002                 if (ret < 0)
1003                         rte_exit(EXIT_FAILURE,
1004                                  "Cannot adjust number of descriptors: err=%d, port=%d\n",
1005                                  ret, portid);
1006
1007                 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
1008                 print_ethaddr(" Address:", &ports_eth_addr[portid]);
1009                 printf(", ");
1010
1011                 ret = init_mem(NB_MBUF);
1012                 if (ret < 0)
1013                         rte_exit(EXIT_FAILURE, "init_mem failed\n");
1014
1015                 /* init one TX queue */
1016                 socketid = (uint8_t)rte_lcore_to_socket_id(rte_get_master_lcore());
1017
1018                 printf("txq=%d,%d,%d ", portid, 0, socketid);
1019                 fflush(stdout);
1020
1021                 txconf = &dev_info.default_txconf;
1022                 txconf->offloads = local_port_conf.txmode.offloads;
1023                 ret = rte_eth_tx_queue_setup(portid, 0, nb_txd,
1024                                                  socketid, txconf);
1025                 if (ret < 0)
1026                         rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: err=%d, "
1027                                 "port=%d\n", ret, portid);
1028
1029                 printf("\n");
1030         }
1031
1032         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1033                 struct rte_eth_rxconf rxq_conf;
1034
1035                 if (rte_lcore_is_enabled(lcore_id) == 0)
1036                         continue;
1037                 qconf = &lcore_conf[lcore_id];
1038                 qconf->tx_queue_id = 0;
1039
1040                 printf("\nInitializing rx queues on lcore %u ... ", lcore_id );
1041                 fflush(stdout);
1042                 /* init RX queues */
1043                 for(queue = 0; queue < qconf->n_rx_queue; ++queue) {
1044                         struct rte_eth_dev *dev;
1045                         struct rte_eth_conf *conf;
1046
1047                         portid = qconf->rx_queue_list[queue].port_id;
1048                         queueid = qconf->rx_queue_list[queue].queue_id;
1049                         dev = &rte_eth_devices[portid];
1050                         conf = &dev->data->dev_conf;
1051
1052                         if (numa_on)
1053                                 socketid = (uint8_t)rte_lcore_to_socket_id(lcore_id);
1054                         else
1055                                 socketid = 0;
1056
1057                         printf("rxq=%d,%d,%d ", portid, queueid, socketid);
1058                         fflush(stdout);
1059
1060                         rte_eth_dev_info_get(portid, &dev_info);
1061                         rxq_conf = dev_info.default_rxconf;
1062                         rxq_conf.offloads = conf->rxmode.offloads;
1063                         ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
1064                                                 socketid, &rxq_conf,
1065                                                 pktmbuf_pool[socketid]);
1066                         if (ret < 0)
1067                                 rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: err=%d,"
1068                                                 "port=%d\n", ret, portid);
1069                 }
1070         }
1071         printf("\n");
1072
1073         /* start ports */
1074         RTE_ETH_FOREACH_DEV(portid) {
1075                 if ((enabled_port_mask & (1 << portid)) == 0) {
1076                         continue;
1077                 }
1078                 /* Start device */
1079                 ret = rte_eth_dev_start(portid);
1080                 if (ret < 0)
1081                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%d\n",
1082                                 ret, portid);
1083
1084                 printf("done: Port %d\n", portid);
1085
1086         }
1087
1088         /* launch per-lcore init on every lcore */
1089         rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
1090         RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1091                 if (rte_eal_wait_lcore(lcore_id) < 0)
1092                         return -1;
1093         }
1094
1095         return 0;
1096 }