examples/l2fwd-ivshmem/host/host.c

   1 /*-
   2  *   BSD LICENSE
   3  *
   4  *   Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
   5  *   All rights reserved.
   6  *
   7  *   Redistribution and use in source and binary forms, with or without
   8  *   modification, are permitted provided that the following conditions
   9  *   are met:
  10  *
  11  *     * Redistributions of source code must retain the above copyright
  12  *       notice, this list of conditions and the following disclaimer.
  13  *     * Redistributions in binary form must reproduce the above copyright
  14  *       notice, this list of conditions and the following disclaimer in
  15  *       the documentation and/or other materials provided with the
  16  *       distribution.
  17  *     * Neither the name of Intel Corporation nor the names of its
  18  *       contributors may be used to endorse or promote products derived
  19  *       from this software without specific prior written permission.
  20  *
  21  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  22  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  23  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
  24  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  25  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  26  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
  27  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  28  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  29  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  30  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
  31  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  32  */
  33
  34 #include <unistd.h>
  35 #include <stdlib.h>
  36 #include <stdio.h>
  37 #include <string.h>
  38 #include <limits.h>
  39 #include <inttypes.h>
  40 #include <getopt.h>
  41 #include <signal.h>
  42
  43 #include <rte_eal.h>
  44 #include <rte_config.h>
  45 #include <rte_cycles.h>
  46 #include <rte_eal_memconfig.h>
  47 #include <rte_debug.h>
  48 #include <rte_ether.h>
  49 #include <rte_ethdev.h>
  50 #include <rte_string_fns.h>
  51 #include <rte_ivshmem.h>
  52 #include <rte_ring.h>
  53 #include <rte_mempool.h>
  54 #include <rte_mbuf.h>
  55
  56 #include "../include/common.h"
  57
  58 /*
  59  * Configurable number of RX/TX ring descriptors
  60  */
  61 #define RTE_TEST_RX_DESC_DEFAULT 128
  62 #define RTE_TEST_TX_DESC_DEFAULT 512
  63 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
  64 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
  65
  66 /*
  67  * RX and TX Prefetch, Host, and Write-back threshold values should be
  68  * carefully set for optimal performance. Consult the network
  69  * controller's datasheet and supporting DPDK documentation for guidance
  70  * on how these parameters should be set.
  71  */
  72 #define RX_PTHRESH 8 /**< Default values of RX prefetch threshold reg. */
  73 #define RX_HTHRESH 8 /**< Default values of RX host threshold reg. */
  74 #define RX_WTHRESH 4 /**< Default values of RX write-back threshold reg. */
  75
  76 /*
  77  * These default values are optimized for use with the Intel(R) 82599 10 GbE
  78  * Controller and the DPDK ixgbe PMD. Consider using other values for other
  79  * network controllers and/or network drivers.
  80  */
  81 #define TX_PTHRESH 36 /**< Default values of TX prefetch threshold reg. */
  82 #define TX_HTHRESH 0  /**< Default values of TX host threshold reg. */
  83 #define TX_WTHRESH 0  /**< Default values of TX write-back threshold reg. */
  84
  85 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
  86
  87 /* mask of enabled ports */
  88 static uint32_t l2fwd_ivshmem_enabled_port_mask = 0;
  89
  90 static struct ether_addr l2fwd_ivshmem_ports_eth_addr[RTE_MAX_ETHPORTS];
  91
  92 #define NB_MBUF   8192
  93 #define MBUF_SIZE (2048 + sizeof(struct rte_mbuf) + RTE_PKTMBUF_HEADROOM)
  94
  95 #define MAX_RX_QUEUE_PER_LCORE 16
  96 #define MAX_TX_QUEUE_PER_PORT 16
  97 struct lcore_queue_conf {
  98         unsigned n_rx_port;
  99         unsigned rx_port_list[MAX_RX_QUEUE_PER_LCORE];
 100         struct vm_port_param * port_param[MAX_RX_QUEUE_PER_LCORE];
 101         struct mbuf_table tx_mbufs[RTE_MAX_ETHPORTS];
 102         struct mbuf_table rx_mbufs[RTE_MAX_ETHPORTS];
 103 } __rte_cache_aligned;
 104 static struct lcore_queue_conf lcore_queue_conf[RTE_MAX_LCORE];
 105
 106 static const struct rte_eth_conf port_conf = {
 107         .rxmode = {
 108                 .split_hdr_size = 0,
 109                 .header_split   = 0, /**< Header Split disabled */
 110                 .hw_ip_checksum = 0, /**< IP checksum offload disabled */
 111                 .hw_vlan_filter = 0, /**< VLAN filtering disabled */
 112                 .jumbo_frame    = 0, /**< Jumbo Frame Support disabled */
 113                 .hw_strip_crc   = 0, /**< CRC stripped by hardware */
 114         },
 115         .txmode = {
 116                 .mq_mode = ETH_MQ_TX_NONE,
 117         },
 118 };
 119
 120 static const struct rte_eth_rxconf rx_conf = {
 121         .rx_thresh = {
 122                 .pthresh = RX_PTHRESH,
 123                 .hthresh = RX_HTHRESH,
 124                 .wthresh = RX_WTHRESH,
 125         },
 126 };
 127
 128 static const struct rte_eth_txconf tx_conf = {
 129         .tx_thresh = {
 130                 .pthresh = TX_PTHRESH,
 131                 .hthresh = TX_HTHRESH,
 132                 .wthresh = TX_WTHRESH,
 133         },
 134         .tx_free_thresh = 0, /* Use PMD default values */
 135         .tx_rs_thresh = 0, /* Use PMD default values */
 136 };
 137
 138 #define METADATA_NAME "l2fwd_ivshmem"
 139 #define CMDLINE_OPT_FWD_CONF "fwd-conf"
 140
 141 #define QEMU_CMD_FMT "/tmp/ivshmem_qemu_cmdline_%s"
 142
 143 struct port_statistics port_statistics[RTE_MAX_ETHPORTS];
 144
 145 struct rte_mempool * l2fwd_ivshmem_pktmbuf_pool = NULL;
 146
 147 /* Print out statistics on packets dropped */
 148 static void
 149 print_stats(void)
 150 {
 151         uint64_t total_packets_dropped, total_packets_tx, total_packets_rx;
 152         uint64_t total_vm_packets_dropped, total_vm_packets_tx, total_vm_packets_rx;
 153         unsigned portid;
 154
 155         total_packets_dropped = 0;
 156         total_packets_tx = 0;
 157         total_packets_rx = 0;
 158         total_vm_packets_tx = 0;
 159         total_vm_packets_rx = 0;
 160
 161         const char clr[] = { 27, '[', '2', 'J', '\0' };
 162         const char topLeft[] = { 27, '[', '1', ';', '1', 'H','\0' };
 163
 164                 /* Clear screen and move to top left */
 165         printf("%s%s", clr, topLeft);
 166
 167         printf("\nPort statistics ====================================");
 168
 169         for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
 170                 /* skip disabled ports */
 171                 if ((l2fwd_ivshmem_enabled_port_mask & (1 << portid)) == 0)
 172                         continue;
 173                 printf("\nStatistics for port %u ------------------------------"
 174                            "\nPackets sent: %24"PRIu64
 175                            "\nPackets received: %20"PRIu64
 176                            "\nPackets dropped: %21"PRIu64,
 177                            portid,
 178                            port_statistics[portid].tx,
 179                            port_statistics[portid].rx,
 180                            port_statistics[portid].dropped);
 181
 182                 total_packets_dropped += port_statistics[portid].dropped;
 183                 total_packets_tx += port_statistics[portid].tx;
 184                 total_packets_rx += port_statistics[portid].rx;
 185         }
 186
 187         printf("\nVM statistics ======================================");
 188         for (portid = 0; portid < ctrl->nb_ports; portid++) {
 189                 printf("\nStatistics for port %u ------------------------------"
 190                            "\nPackets sent: %24"PRIu64
 191                            "\nPackets received: %20"PRIu64,
 192                            portid,
 193                            ctrl->vm_ports[portid].stats.tx,
 194                            ctrl->vm_ports[portid].stats.rx);
 195
 196                 total_vm_packets_dropped += ctrl->vm_ports[portid].stats.dropped;
 197                 total_vm_packets_tx += ctrl->vm_ports[portid].stats.tx;
 198                 total_vm_packets_rx += ctrl->vm_ports[portid].stats.rx;
 199         }
 200         printf("\nAggregate statistics ==============================="
 201                            "\nTotal packets sent: %18"PRIu64
 202                            "\nTotal packets received: %14"PRIu64
 203                            "\nTotal packets dropped: %15"PRIu64
 204                            "\nTotal VM packets sent: %15"PRIu64
 205                            "\nTotal VM packets received: %11"PRIu64,
 206                            total_packets_tx,
 207                            total_packets_rx,
 208                            total_packets_dropped,
 209                            total_vm_packets_tx,
 210                            total_vm_packets_rx);
 211         printf("\n====================================================\n");
 212 }
 213
 214 static int
 215 print_to_file(const char *cmdline, const char *config_name)
 216 {
 217         FILE *file;
 218         char path[PATH_MAX];
 219
 220         snprintf(path, sizeof(path), QEMU_CMD_FMT, config_name);
 221         file = fopen(path, "w");
 222         if (file == NULL) {
 223                 RTE_LOG(ERR, L2FWD_IVSHMEM, "Could not open '%s' \n", path);
 224                 return -1;
 225         }
 226
 227         RTE_LOG(DEBUG, L2FWD_IVSHMEM, "QEMU command line for config '%s': %s \n",
 228                         config_name, cmdline);
 229
 230         fprintf(file, "%s\n", cmdline);
 231         fclose(file);
 232         return 0;
 233 }
 234
 235 static int
 236 generate_ivshmem_cmdline(const char *config_name)
 237 {
 238         char cmdline[PATH_MAX];
 239         if (rte_ivshmem_metadata_cmdline_generate(cmdline, sizeof(cmdline),
 240                         config_name) < 0)
 241                 return -1;
 242
 243         if (print_to_file(cmdline, config_name) < 0)
 244                 return -1;
 245
 246         rte_ivshmem_metadata_dump(stdout, config_name);
 247         return 0;
 248 }
 249
 250 /* display usage */
 251 static void
 252 l2fwd_ivshmem_usage(const char *prgname)
 253 {
 254         printf("%s [EAL options] -- -p PORTMASK [-q NQ -T PERIOD]\n"
 255                    "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
 256                    "  -q NQ: number of queue (=ports) per lcore (default is 1)\n"
 257                    "  -T PERIOD: statistics will be refreshed each PERIOD seconds "
 258                        "(0 to disable, 10 default, 86400 maximum)\n",
 259                prgname);
 260 }
 261
 262 static unsigned int
 263 l2fwd_ivshmem_parse_nqueue(const char *q_arg)
 264 {
 265         char *end = NULL;
 266         unsigned long n;
 267
 268         /* parse hexadecimal string */
 269         n = strtoul(q_arg, &end, 10);
 270         if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
 271                 return 0;
 272         if (n == 0)
 273                 return 0;
 274         if (n >= MAX_RX_QUEUE_PER_LCORE)
 275                 return 0;
 276
 277         return n;
 278 }
 279
 280 static int
 281 l2fwd_ivshmem_parse_portmask(const char *portmask)
 282 {
 283         char *end = NULL;
 284         unsigned long pm;
 285
 286         /* parse hexadecimal string */
 287         pm = strtoul(portmask, &end, 16);
 288         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
 289                 return -1;
 290
 291         if (pm == 0)
 292                 return -1;
 293
 294         return pm;
 295 }
 296
 297 static int
 298 l2fwd_ivshmem_parse_timer_period(const char *q_arg)
 299 {
 300         char *end = NULL;
 301         int n;
 302
 303         /* parse number string */
 304         n = strtol(q_arg, &end, 10);
 305         if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
 306                 return -1;
 307         if (n >= MAX_TIMER_PERIOD)
 308                 return -1;
 309
 310         return n;
 311 }
 312
 313 /* Parse the argument given in the command line of the application */
 314 static int
 315 l2fwd_ivshmem_parse_args(int argc, char **argv)
 316 {
 317         int opt, ret;
 318         char **argvopt;
 319         int option_index;
 320         char *prgname = argv[0];
 321         static struct option lgopts[] = {
 322                         {CMDLINE_OPT_FWD_CONF, 1, 0, 0},
 323                 {NULL, 0, 0, 0}
 324         };
 325
 326         argvopt = argv;
 327
 328         while ((opt = getopt_long(argc, argvopt, "q:p:T:",
 329                                   lgopts, &option_index)) != EOF) {
 330
 331                 switch (opt) {
 332                 /* portmask */
 333                 case 'p':
 334                         l2fwd_ivshmem_enabled_port_mask = l2fwd_ivshmem_parse_portmask(optarg);
 335                         if (l2fwd_ivshmem_enabled_port_mask == 0) {
 336                                 printf("invalid portmask\n");
 337                                 l2fwd_ivshmem_usage(prgname);
 338                                 return -1;
 339                         }
 340                         break;
 341
 342                 /* nqueue */
 343                 case 'q':
 344                         l2fwd_ivshmem_rx_queue_per_lcore = l2fwd_ivshmem_parse_nqueue(optarg);
 345                         if (l2fwd_ivshmem_rx_queue_per_lcore == 0) {
 346                                 printf("invalid queue number\n");
 347                                 l2fwd_ivshmem_usage(prgname);
 348                                 return -1;
 349                         }
 350                         break;
 351
 352                 /* timer period */
 353                 case 'T':
 354                         timer_period = l2fwd_ivshmem_parse_timer_period(optarg) * 1000 * TIMER_MILLISECOND;
 355                         if (timer_period < 0) {
 356                                 printf("invalid timer period\n");
 357                                 l2fwd_ivshmem_usage(prgname);
 358                                 return -1;
 359                         }
 360                         break;
 361
 362                 /* long options */
 363                 case 0:
 364                         l2fwd_ivshmem_usage(prgname);
 365                         return -1;
 366
 367                 default:
 368                         l2fwd_ivshmem_usage(prgname);
 369                         return -1;
 370                 }
 371         }
 372
 373         if (optind >= 0)
 374                 argv[optind-1] = prgname;
 375
 376         ret = optind-1;
 377         optind = 0; /* reset getopt lib */
 378         return ret;
 379 }
 380
 381 /* Check the link status of all ports in up to 9s, and print them finally */
 382 static void
 383 check_all_ports_link_status(uint8_t port_num, uint32_t port_mask)
 384 {
 385 #define CHECK_INTERVAL 100 /* 100ms */
 386 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
 387         uint8_t portid, count, all_ports_up, print_flag = 0;
 388         struct rte_eth_link link;
 389
 390         printf("\nChecking link status");
 391         fflush(stdout);
 392         for (count = 0; count <= MAX_CHECK_TIME; count++) {
 393                 all_ports_up = 1;
 394                 for (portid = 0; portid < port_num; portid++) {
 395                         if ((port_mask & (1 << portid)) == 0)
 396                                 continue;
 397                         memset(&link, 0, sizeof(link));
 398                         rte_eth_link_get_nowait(portid, &link);
 399                         /* print link status if flag set */
 400                         if (print_flag == 1) {
 401                                 if (link.link_status)
 402                                         printf("Port %d Link Up - speed %u "
 403                                                 "Mbps - %s\n", (uint8_t)portid,
 404                                                 (unsigned)link.link_speed,
 405                                 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
 406                                         ("full-duplex") : ("half-duplex\n"));
 407                                 else
 408                                         printf("Port %d Link Down\n",
 409                                                 (uint8_t)portid);
 410                                 continue;
 411                         }
 412                         /* clear all_ports_up flag if any link down */
 413                         if (link.link_status == 0) {
 414                                 all_ports_up = 0;
 415                                 break;
 416                         }
 417                 }
 418                 /* after finally printing all link status, get out */
 419                 if (print_flag == 1)
 420                         break;
 421
 422                 if (all_ports_up == 0) {
 423                         printf(".");
 424                         fflush(stdout);
 425                         rte_delay_ms(CHECK_INTERVAL);
 426                 }
 427
 428                 /* set the print_flag if all ports up or timeout */
 429                 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
 430                         print_flag = 1;
 431                         printf("done\n");
 432                 }
 433         }
 434 }
 435
 436 /* Send the burst of packets on an output interface */
 437 static int
 438 l2fwd_ivshmem_send_burst(struct lcore_queue_conf *qconf, unsigned n, uint8_t port)
 439 {
 440         struct rte_mbuf **m_table;
 441         unsigned ret;
 442         unsigned queueid =0;
 443
 444         m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
 445
 446         ret = rte_eth_tx_burst(port, (uint16_t) queueid, m_table, (uint16_t) n);
 447         port_statistics[port].tx += ret;
 448         if (unlikely(ret < n)) {
 449                 port_statistics[port].dropped += (n - ret);
 450                 do {
 451                         rte_pktmbuf_free(m_table[ret]);
 452                 } while (++ret < n);
 453         }
 454
 455         return 0;
 456 }
 457
 458 /* Enqueue packets for TX and prepare them to be sent on the network */
 459 static int
 460 l2fwd_ivshmem_send_packet(struct rte_mbuf *m, uint8_t port)
 461 {
 462         unsigned lcore_id, len;
 463         struct lcore_queue_conf *qconf;
 464
 465         lcore_id = rte_lcore_id();
 466
 467         qconf = &lcore_queue_conf[lcore_id];
 468         len = qconf->tx_mbufs[port].len;
 469         qconf->tx_mbufs[port].m_table[len] = m;
 470         len++;
 471
 472         /* enough pkts to be sent */
 473         if (unlikely(len == MAX_PKT_BURST)) {
 474                 l2fwd_ivshmem_send_burst(qconf, MAX_PKT_BURST, port);
 475                 len = 0;
 476         }
 477
 478         qconf->tx_mbufs[port].len = len;
 479         return 0;
 480 }
 481
 482 static int
 483 l2fwd_ivshmem_receive_burst(struct lcore_queue_conf *qconf, unsigned portid,
 484                 unsigned vm_port)
 485 {
 486         struct rte_mbuf ** m;
 487         struct rte_ring * rx;
 488         unsigned len, pkt_idx;
 489
 490         m = qconf->rx_mbufs[portid].m_table;
 491         len = qconf->rx_mbufs[portid].len;
 492         rx = qconf->port_param[vm_port]->rx_ring;
 493
 494         /* if enqueueing failed, ring is probably full, so drop the packets */
 495         if (rte_ring_enqueue_bulk(rx, (void**) m, len) < 0) {
 496                 port_statistics[portid].dropped += len;
 497
 498                 pkt_idx = 0;
 499                 do {
 500                         rte_pktmbuf_free(m[pkt_idx]);
 501                 } while (++pkt_idx < len);
 502         }
 503         else
 504                 /* increment rx stats by however many packets we managed to receive */
 505                 port_statistics[portid].rx += len;
 506
 507         return 0;
 508 }
 509
 510 /* Enqueue packets for RX and prepare them to be sent to VM */
 511 static int
 512 l2fwd_ivshmem_receive_packets(struct rte_mbuf ** m, unsigned n, unsigned portid,
 513                 unsigned vm_port)
 514 {
 515         unsigned lcore_id, len, pkt_idx;
 516         struct lcore_queue_conf *qconf;
 517
 518         lcore_id = rte_lcore_id();
 519
 520         qconf = &lcore_queue_conf[lcore_id];
 521
 522         len = qconf->rx_mbufs[portid].len;
 523         pkt_idx = 0;
 524
 525         /* enqueue packets */
 526         while (pkt_idx < n && len < MAX_PKT_BURST * 2) {
 527                 qconf->rx_mbufs[portid].m_table[len++] = m[pkt_idx++];
 528         }
 529
 530         /* increment queue len by however many packets we managed to receive */
 531         qconf->rx_mbufs[portid].len += pkt_idx;
 532
 533         /* drop the unreceived packets */
 534         if (unlikely(pkt_idx < n)) {
 535                 port_statistics[portid].dropped += n - pkt_idx;
 536                 do {
 537                         rte_pktmbuf_free(m[pkt_idx]);
 538                 } while (++pkt_idx < n);
 539         }
 540
 541         /* drain the queue halfway through the maximum capacity */
 542         if (unlikely(qconf->rx_mbufs[portid].len >= MAX_PKT_BURST))
 543                 l2fwd_ivshmem_receive_burst(qconf, portid, vm_port);
 544
 545         return 0;
 546 }
 547
 548 /* loop for host forwarding mode.
 549  * the data flow is as follows:
 550  *  1) get packets from TX queue and send it out from a given port
 551  *  2) RX packets from given port and enqueue them on RX ring
 552  *  3) dequeue packets from TX ring and put them on TX queue for a given port
 553  */
 554 static void
 555 fwd_loop(void)
 556 {
 557         struct rte_mbuf *pkts_burst[MAX_PKT_BURST * 2];
 558         struct rte_mbuf *m;
 559         unsigned lcore_id;
 560         uint64_t prev_tsc, diff_tsc, cur_tsc, timer_tsc;
 561         unsigned i, j, portid, nb_rx;
 562         struct lcore_queue_conf *qconf;
 563         struct rte_ring *tx;
 564         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
 565
 566         prev_tsc = 0;
 567         timer_tsc = 0;
 568
 569         lcore_id = rte_lcore_id();
 570         qconf = &lcore_queue_conf[lcore_id];
 571
 572         if (qconf->n_rx_port == 0) {
 573                 RTE_LOG(INFO, L2FWD_IVSHMEM, "lcore %u has nothing to do\n", lcore_id);
 574                 return;
 575         }
 576
 577         RTE_LOG(INFO, L2FWD_IVSHMEM, "entering main loop on lcore %u\n", lcore_id);
 578
 579         for (i = 0; i < qconf->n_rx_port; i++) {
 580
 581                 portid = qconf->rx_port_list[i];
 582                 RTE_LOG(INFO, L2FWD_IVSHMEM, " -- lcoreid=%u portid=%u\n", lcore_id,
 583                         portid);
 584         }
 585
 586         while (ctrl->state == STATE_FWD) {
 587
 588                 cur_tsc = rte_rdtsc();
 589
 590                 /*
 591                  * Burst queue drain
 592                  */
 593                 diff_tsc = cur_tsc - prev_tsc;
 594                 if (unlikely(diff_tsc > drain_tsc)) {
 595
 596                         /*
 597                          * TX
 598                          */
 599                         for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
 600                                 if (qconf->tx_mbufs[portid].len == 0)
 601                                         continue;
 602                                 l2fwd_ivshmem_send_burst(qconf,
 603                                                  qconf->tx_mbufs[portid].len,
 604                                                  (uint8_t) portid);
 605                                 qconf->tx_mbufs[portid].len = 0;
 606                         }
 607
 608                         /*
 609                          * RX
 610                          */
 611                         for (i = 0; i < qconf->n_rx_port; i++) {
 612                                 portid = qconf->rx_port_list[i];
 613                                 if (qconf->rx_mbufs[portid].len == 0)
 614                                         continue;
 615                                 l2fwd_ivshmem_receive_burst(qconf, portid, i);
 616                                 qconf->rx_mbufs[portid].len = 0;
 617                         }
 618
 619                         /* if timer is enabled */
 620                         if (timer_period > 0) {
 621
 622                                 /* advance the timer */
 623                                 timer_tsc += diff_tsc;
 624
 625                                 /* if timer has reached its timeout */
 626                                 if (unlikely(timer_tsc >= (uint64_t) timer_period)) {
 627
 628                                         /* do this only on master core */
 629                                         if (lcore_id == rte_get_master_lcore()) {
 630                                                 print_stats();
 631                                                 /* reset the timer */
 632                                                 timer_tsc = 0;
 633                                         }
 634                                 }
 635                         }
 636
 637                         prev_tsc = cur_tsc;
 638                 }
 639
 640                 /*
 641                  * packet RX and forwarding
 642                  */
 643                 for (i = 0; i < qconf->n_rx_port; i++) {
 644
 645                         /* RX packets from port and put them on RX ring */
 646                         portid = qconf->rx_port_list[i];
 647                         nb_rx = rte_eth_rx_burst((uint8_t) portid, 0,
 648                                                  pkts_burst, MAX_PKT_BURST);
 649
 650                         if (nb_rx != 0)
 651                                 l2fwd_ivshmem_receive_packets(pkts_burst, nb_rx, portid, i);
 652
 653                         /* dequeue packets from TX ring and send them to TX queue */
 654                         tx = qconf->port_param[i]->tx_ring;
 655
 656                         nb_rx = rte_ring_count(tx);
 657
 658                         nb_rx = RTE_MIN(nb_rx, (unsigned) MAX_PKT_BURST);
 659
 660                         if (nb_rx == 0)
 661                                 continue;
 662
 663                         /* should not happen */
 664                         if (unlikely(rte_ring_dequeue_bulk(tx, (void**) pkts_burst, nb_rx) < 0)) {
 665                                 ctrl->state = STATE_FAIL;
 666                                 return;
 667                         }
 668
 669                         for (j = 0; j < nb_rx; j++) {
 670                                 m = pkts_burst[j];
 671                                 l2fwd_ivshmem_send_packet(m, portid);
 672                         }
 673                 }
 674         }
 675 }
 676
 677 static int
 678 l2fwd_ivshmem_launch_one_lcore(__attribute__((unused)) void *dummy)
 679 {
 680         fwd_loop();
 681         return 0;
 682 }
 683
 684 int main(int argc, char **argv)
 685 {
 686         char name[RTE_RING_NAMESIZE];
 687         struct rte_ring *r;
 688         struct lcore_queue_conf *qconf;
 689         struct rte_eth_dev_info dev_info;
 690         uint8_t portid, port_nr;
 691         uint8_t nb_ports, nb_ports_available;
 692         uint8_t nb_ports_in_mask;
 693         int ret;
 694         unsigned lcore_id, rx_lcore_id;
 695
 696         /* init EAL */
 697         ret = rte_eal_init(argc, argv);
 698         if (ret < 0)
 699                 rte_exit(EXIT_FAILURE, "Invalid EAL arguments\n");
 700         argc -= ret;
 701         argv += ret;
 702
 703         /* parse application arguments (after the EAL ones) */
 704         ret = l2fwd_ivshmem_parse_args(argc, argv);
 705         if (ret < 0)
 706                 rte_exit(EXIT_FAILURE, "Invalid l2fwd-ivshmem arguments\n");
 707
 708         /* create a shared mbuf pool */
 709         l2fwd_ivshmem_pktmbuf_pool =
 710                 rte_mempool_create(MBUF_MP_NAME, NB_MBUF,
 711                                    MBUF_SIZE, 32,
 712                                    sizeof(struct rte_pktmbuf_pool_private),
 713                                    rte_pktmbuf_pool_init, NULL,
 714                                    rte_pktmbuf_init, NULL,
 715                                    rte_socket_id(), 0);
 716         if (l2fwd_ivshmem_pktmbuf_pool == NULL)
 717                 rte_exit(EXIT_FAILURE, "Cannot init mbuf pool\n");
 718
 719         if (rte_eal_pci_probe() < 0)
 720                 rte_exit(EXIT_FAILURE, "Cannot probe PCI\n");
 721
 722         nb_ports = rte_eth_dev_count();
 723         if (nb_ports == 0)
 724                 rte_exit(EXIT_FAILURE, "No Ethernet ports - bye\n");
 725
 726         if (nb_ports > RTE_MAX_ETHPORTS)
 727                 nb_ports = RTE_MAX_ETHPORTS;
 728
 729         /*
 730          * reserve memzone to communicate with VMs - we cannot use rte_malloc here
 731          * because while it is technically possible, it is a very bad idea to share
 732          * the heap between two primary processes.
 733          */
 734         ctrl_mz = rte_memzone_reserve(CTRL_MZ_NAME, sizeof(struct ivshmem_ctrl),
 735                         SOCKET_ID_ANY, 0);
 736         if (ctrl_mz == NULL)
 737                 rte_exit(EXIT_FAILURE, "Cannot reserve control memzone\n");
 738         ctrl = (struct ivshmem_ctrl*) ctrl_mz->addr;
 739
 740         memset(ctrl, 0, sizeof(struct ivshmem_ctrl));
 741
 742         /*
 743          * Each port is assigned an output port.
 744          */
 745         nb_ports_in_mask = 0;
 746         for (portid = 0; portid < nb_ports; portid++) {
 747                 /* skip ports that are not enabled */
 748                 if ((l2fwd_ivshmem_enabled_port_mask & (1 << portid)) == 0)
 749                         continue;
 750                 if (portid % 2) {
 751                         ctrl->vm_ports[nb_ports_in_mask].dst = &ctrl->vm_ports[nb_ports_in_mask-1];
 752                         ctrl->vm_ports[nb_ports_in_mask-1].dst = &ctrl->vm_ports[nb_ports_in_mask];
 753                 }
 754
 755                 nb_ports_in_mask++;
 756
 757                 rte_eth_dev_info_get(portid, &dev_info);
 758         }
 759         if (nb_ports_in_mask % 2) {
 760                 printf("Notice: odd number of ports in portmask.\n");
 761                 ctrl->vm_ports[nb_ports_in_mask-1].dst =
 762                                 &ctrl->vm_ports[nb_ports_in_mask-1];
 763         }
 764
 765         rx_lcore_id = 0;
 766         qconf = NULL;
 767
 768         printf("Initializing ports configuration...\n");
 769
 770         nb_ports_available = nb_ports;
 771
 772         /* Initialise each port */
 773         for (portid = 0; portid < nb_ports; portid++) {
 774
 775                 /* skip ports that are not enabled */
 776                 if ((l2fwd_ivshmem_enabled_port_mask & (1 << portid)) == 0) {
 777                         printf("Skipping disabled port %u\n", (unsigned) portid);
 778                         nb_ports_available--;
 779                         continue;
 780                 }
 781
 782                 /* init port */
 783                 printf("Initializing port %u... ", (unsigned) portid);
 784                 fflush(stdout);
 785                 ret = rte_eth_dev_configure(portid, 1, 1, &port_conf);
 786                 if (ret < 0)
 787                         rte_exit(EXIT_FAILURE, "Cannot configure device: err=%d, port=%u\n",
 788                                   ret, (unsigned) portid);
 789
 790                 rte_eth_macaddr_get(portid,&l2fwd_ivshmem_ports_eth_addr[portid]);
 791
 792                 /* init one RX queue */
 793                 fflush(stdout);
 794                 ret = rte_eth_rx_queue_setup(portid, 0, nb_rxd,
 795                                                  rte_eth_dev_socket_id(portid), &rx_conf,
 796                                                  l2fwd_ivshmem_pktmbuf_pool);
 797                 if (ret < 0)
 798                         rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup:err=%d, port=%u\n",
 799                                   ret, (unsigned) portid);
 800
 801                 /* init one TX queue on each port */
 802                 fflush(stdout);
 803                 ret = rte_eth_tx_queue_setup(portid, 0, nb_txd,
 804                                 rte_eth_dev_socket_id(portid), &tx_conf);
 805                 if (ret < 0)
 806                         rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup:err=%d, port=%u\n",
 807                                 ret, (unsigned) portid);
 808
 809                 /* Start device */
 810                 ret = rte_eth_dev_start(portid);
 811                 if (ret < 0)
 812                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start:err=%d, port=%u\n",
 813                                   ret, (unsigned) portid);
 814
 815                 printf("done: \n");
 816
 817                 rte_eth_promiscuous_enable(portid);
 818
 819                 printf("Port %u, MAC address: %02X:%02X:%02X:%02X:%02X:%02X\n\n",
 820                                 (unsigned) portid,
 821                                 l2fwd_ivshmem_ports_eth_addr[portid].addr_bytes[0],
 822                                 l2fwd_ivshmem_ports_eth_addr[portid].addr_bytes[1],
 823                                 l2fwd_ivshmem_ports_eth_addr[portid].addr_bytes[2],
 824                                 l2fwd_ivshmem_ports_eth_addr[portid].addr_bytes[3],
 825                                 l2fwd_ivshmem_ports_eth_addr[portid].addr_bytes[4],
 826                                 l2fwd_ivshmem_ports_eth_addr[portid].addr_bytes[5]);
 827
 828                 /* initialize port stats */
 829                 memset(&port_statistics, 0, sizeof(port_statistics));
 830         }
 831
 832         if (!nb_ports_available) {
 833                 rte_exit(EXIT_FAILURE,
 834                         "All available ports are disabled. Please set portmask.\n");
 835         }
 836         port_nr = 0;
 837
 838         /* Initialize the port/queue configuration of each logical core */
 839         for (portid = 0; portid < nb_ports; portid++) {
 840                 if ((l2fwd_ivshmem_enabled_port_mask & (1 << portid)) == 0)
 841                         continue;
 842
 843                 /* get the lcore_id for this port */
 844                 while (rte_lcore_is_enabled(rx_lcore_id) == 0 ||
 845                            lcore_queue_conf[rx_lcore_id].n_rx_port ==
 846                                            l2fwd_ivshmem_rx_queue_per_lcore) {
 847                         rx_lcore_id++;
 848                         if (rx_lcore_id >= RTE_MAX_LCORE)
 849                                 rte_exit(EXIT_FAILURE, "Not enough cores\n");
 850                 }
 851
 852                 if (qconf != &lcore_queue_conf[rx_lcore_id])
 853                         /* Assigned a new logical core in the loop above. */
 854                         qconf = &lcore_queue_conf[rx_lcore_id];
 855
 856
 857                 rte_eth_macaddr_get(portid, &ctrl->vm_ports[port_nr].ethaddr);
 858
 859                 qconf->rx_port_list[qconf->n_rx_port] = portid;
 860                 qconf->port_param[qconf->n_rx_port] = &ctrl->vm_ports[port_nr];
 861                 qconf->n_rx_port++;
 862                 port_nr++;
 863                 printf("Lcore %u: RX port %u\n", rx_lcore_id, (unsigned) portid);
 864         }
 865
 866         check_all_ports_link_status(nb_ports_available, l2fwd_ivshmem_enabled_port_mask);
 867
 868         /* create rings for each VM port (several ports can be on the same VM).
 869          * note that we store the pointers in ctrl - that way, they are the same
 870          * and valid across all VMs because ctrl is also in DPDK memory */
 871         for (portid = 0; portid < nb_ports_available; portid++) {
 872
 873                 /* RX ring. SP/SC because it's only used by host and a single VM */
 874                 snprintf(name, sizeof(name), "%s%i", RX_RING_PREFIX, portid);
 875                 r = rte_ring_create(name, NB_MBUF,
 876                                 SOCKET_ID_ANY, RING_F_SP_ENQ | RING_F_SC_DEQ);
 877                 if (r == NULL)
 878                         rte_exit(EXIT_FAILURE, "Cannot create ring %s\n", name);
 879
 880                 ctrl->vm_ports[portid].rx_ring = r;
 881
 882                 /* TX ring. SP/SC because it's only used by host and a single VM */
 883                 snprintf(name, sizeof(name), "%s%i", TX_RING_PREFIX, portid);
 884                 r = rte_ring_create(name, NB_MBUF,
 885                                 SOCKET_ID_ANY, RING_F_SP_ENQ | RING_F_SC_DEQ);
 886                 if (r == NULL)
 887                         rte_exit(EXIT_FAILURE, "Cannot create ring %s\n", name);
 888
 889                 ctrl->vm_ports[portid].tx_ring = r;
 890         }
 891
 892         /* create metadata, output cmdline */
 893         if (rte_ivshmem_metadata_create(METADATA_NAME) < 0)
 894                 rte_exit(EXIT_FAILURE, "Cannot create IVSHMEM metadata\n");
 895
 896         if (rte_ivshmem_metadata_add_memzone(ctrl_mz, METADATA_NAME))
 897                 rte_exit(EXIT_FAILURE, "Cannot add memzone to IVSHMEM metadata\n");
 898
 899         if (rte_ivshmem_metadata_add_mempool(l2fwd_ivshmem_pktmbuf_pool, METADATA_NAME))
 900                 rte_exit(EXIT_FAILURE, "Cannot add mbuf mempool to IVSHMEM metadata\n");
 901
 902         for (portid = 0; portid < nb_ports_available; portid++) {
 903                 if (rte_ivshmem_metadata_add_ring(ctrl->vm_ports[portid].rx_ring,
 904                                 METADATA_NAME) < 0)
 905                         rte_exit(EXIT_FAILURE, "Cannot add ring %s to IVSHMEM metadata\n",
 906                                         ctrl->vm_ports[portid].rx_ring->name);
 907                 if (rte_ivshmem_metadata_add_ring(ctrl->vm_ports[portid].tx_ring,
 908                                 METADATA_NAME) < 0)
 909                         rte_exit(EXIT_FAILURE, "Cannot add ring %s to IVSHMEM metadata\n",
 910                                         ctrl->vm_ports[portid].tx_ring->name);
 911         }
 912         generate_ivshmem_cmdline(METADATA_NAME);
 913
 914         ctrl->nb_ports = nb_ports_available;
 915
 916         printf("Waiting for VM to initialize...\n");
 917
 918         /* wait for VM to initialize */
 919         while (ctrl->state != STATE_FWD) {
 920                 if (ctrl->state == STATE_FAIL)
 921                         rte_exit(EXIT_FAILURE, "VM reported failure\n");
 922
 923                 sleep(1);
 924         }
 925
 926         printf("Done!\n");
 927
 928         sigsetup();
 929
 930         /* launch per-lcore init on every lcore */
 931         rte_eal_mp_remote_launch(l2fwd_ivshmem_launch_one_lcore, NULL, CALL_MASTER);
 932         RTE_LCORE_FOREACH_SLAVE(lcore_id) {
 933                 if (rte_eal_wait_lcore(lcore_id) < 0)
 934                         return -1;
 935         }
 936
 937         if (ctrl->state == STATE_FAIL)
 938                 rte_exit(EXIT_FAILURE, "VM reported failure\n");
 939
 940         return 0;
 941 }