examples/tep_termination/main.c

   1 /*-
   2  *   BSD LICENSE
   3  *
   4  *   Copyright(c) 2010-2015 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 <arpa/inet.h>
  35 #include <getopt.h>
  36 #include <linux/if_ether.h>
  37 #include <linux/if_vlan.h>
  38 #include <linux/virtio_net.h>
  39 #include <linux/virtio_ring.h>
  40 #include <signal.h>
  41 #include <stdint.h>
  42 #include <sys/eventfd.h>
  43 #include <sys/param.h>
  44 #include <unistd.h>
  45
  46 #include <rte_atomic.h>
  47 #include <rte_cycles.h>
  48 #include <rte_ethdev.h>
  49 #include <rte_log.h>
  50 #include <rte_string_fns.h>
  51 #include <rte_malloc.h>
  52 #include <rte_virtio_net.h>
  53
  54 #include "main.h"
  55 #include "vxlan.h"
  56 #include "vxlan_setup.h"
  57
  58 /* the maximum number of external ports supported */
  59 #define MAX_SUP_PORTS 1
  60
  61 /**
  62  * Calculate the number of buffers needed per port
  63  */
  64 #define NUM_MBUFS_PER_PORT ((MAX_QUEUES * RTE_TEST_RX_DESC_DEFAULT) +\
  65                                 (nb_switching_cores * MAX_PKT_BURST) +\
  66                                 (nb_switching_cores * \
  67                                 RTE_TEST_TX_DESC_DEFAULT) +\
  68                                 (nb_switching_cores * MBUF_CACHE_SIZE))
  69
  70 #define MBUF_CACHE_SIZE 128
  71 #define MBUF_SIZE (2048 + sizeof(struct rte_mbuf) + RTE_PKTMBUF_HEADROOM)
  72
  73 #define MAX_PKT_BURST 32        /* Max burst size for RX/TX */
  74 #define BURST_TX_DRAIN_US 100   /* TX drain every ~100us */
  75
  76 /* Defines how long we wait between retries on RX */
  77 #define BURST_RX_WAIT_US 15
  78
  79 #define BURST_RX_RETRIES 4      /* Number of retries on RX. */
  80
  81 #define JUMBO_FRAME_MAX_SIZE    0x2600
  82
  83 /* State of virtio device. */
  84 #define DEVICE_MAC_LEARNING 0
  85 #define DEVICE_RX           1
  86 #define DEVICE_SAFE_REMOVE  2
  87
  88 /* Config_core_flag status definitions. */
  89 #define REQUEST_DEV_REMOVAL 1
  90 #define ACK_DEV_REMOVAL     0
  91
  92 /* Configurable number of RX/TX ring descriptors */
  93 #define RTE_TEST_RX_DESC_DEFAULT 1024
  94 #define RTE_TEST_TX_DESC_DEFAULT 512
  95
  96 /* Get first 4 bytes in mbuf headroom. */
  97 #define MBUF_HEADROOM_UINT32(mbuf) (*(uint32_t *)((uint8_t *)(mbuf) \
  98                 + sizeof(struct rte_mbuf)))
  99
 100 #define INVALID_PORT_ID 0xFF
 101
 102 /* Size of buffers used for snprintfs. */
 103 #define MAX_PRINT_BUFF 6072
 104
 105 /* Maximum character device basename size. */
 106 #define MAX_BASENAME_SZ 20
 107
 108 /* Maximum long option length for option parsing. */
 109 #define MAX_LONG_OPT_SZ 64
 110
 111 /* Used to compare MAC addresses. */
 112 #define MAC_ADDR_CMP 0xFFFFFFFFFFFFULL
 113
 114 #define CMD_LINE_OPT_NB_DEVICES "nb-devices"
 115 #define CMD_LINE_OPT_UDP_PORT "udp-port"
 116 #define CMD_LINE_OPT_TX_CHECKSUM "tx-checksum"
 117 #define CMD_LINE_OPT_TSO_SEGSZ "tso-segsz"
 118 #define CMD_LINE_OPT_FILTER_TYPE "filter-type"
 119 #define CMD_LINE_OPT_RX_RETRY "rx-retry"
 120 #define CMD_LINE_OPT_RX_RETRY_DELAY "rx-retry-delay"
 121 #define CMD_LINE_OPT_RX_RETRY_NUM "rx-retry-num"
 122 #define CMD_LINE_OPT_STATS "stats"
 123 #define CMD_LINE_OPT_DEV_BASENAME "dev-basename"
 124
 125 /* mask of enabled ports */
 126 static uint32_t enabled_port_mask;
 127
 128 /*Number of switching cores enabled*/
 129 static uint32_t nb_switching_cores;
 130
 131 /* number of devices/queues to support*/
 132 uint16_t nb_devices = 2;
 133
 134 /* max ring descriptor, ixgbe, i40e, e1000 all are 4096. */
 135 #define MAX_RING_DESC 4096
 136
 137 struct vpool {
 138         struct rte_mempool *pool;
 139         struct rte_ring *ring;
 140         uint32_t buf_size;
 141 } vpool_array[MAX_QUEUES+MAX_QUEUES];
 142
 143 /* UDP tunneling port */
 144 uint16_t udp_port = 4789;
 145
 146 /* enable/disable inner TX checksum */
 147 uint8_t tx_checksum = 0;
 148
 149 /* TCP segment size */
 150 uint16_t tso_segsz = 0;
 151
 152 /* RX filter type for tunneling packet */
 153 uint8_t filter_idx = 1;
 154
 155 /* overlay packet operation */
 156 struct ol_switch_ops overlay_options = {
 157         .port_configure = vxlan_port_init,
 158         .tunnel_setup = vxlan_link,
 159         .tunnel_destroy = vxlan_unlink,
 160         .tx_handle = vxlan_tx_pkts,
 161         .rx_handle = vxlan_rx_pkts,
 162         .param_handle = NULL,
 163 };
 164
 165 /* Enable stats. */
 166 uint32_t enable_stats = 0;
 167 /* Enable retries on RX. */
 168 static uint32_t enable_retry = 1;
 169 /* Specify timeout (in useconds) between retries on RX. */
 170 static uint32_t burst_rx_delay_time = BURST_RX_WAIT_US;
 171 /* Specify the number of retries on RX. */
 172 static uint32_t burst_rx_retry_num = BURST_RX_RETRIES;
 173
 174 /* Character device basename. Can be set by user. */
 175 static char dev_basename[MAX_BASENAME_SZ] = "vhost-net";
 176
 177 static unsigned lcore_ids[RTE_MAX_LCORE];
 178 uint8_t ports[RTE_MAX_ETHPORTS];
 179
 180 static unsigned nb_ports; /**< The number of ports specified in command line */
 181
 182 /* ethernet addresses of ports */
 183 struct ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
 184
 185 /* heads for the main used and free linked lists for the data path. */
 186 static struct virtio_net_data_ll *ll_root_used;
 187 static struct virtio_net_data_ll *ll_root_free;
 188
 189 /**
 190  * Array of data core structures containing information on
 191  * individual core linked lists.
 192  */
 193 static struct lcore_info lcore_info[RTE_MAX_LCORE];
 194
 195 /* Used for queueing bursts of TX packets. */
 196 struct mbuf_table {
 197         unsigned len;
 198         unsigned txq_id;
 199         struct rte_mbuf *m_table[MAX_PKT_BURST];
 200 };
 201
 202 /* TX queue for each data core. */
 203 struct mbuf_table lcore_tx_queue[RTE_MAX_LCORE];
 204
 205 struct device_statistics dev_statistics[MAX_DEVICES];
 206
 207 /**
 208  * Set character device basename.
 209  */
 210 static int
 211 us_vhost_parse_basename(const char *q_arg)
 212 {
 213         /* parse number string */
 214         if (strlen(q_arg) >= MAX_BASENAME_SZ)
 215                 return -1;
 216         else
 217                 snprintf((char *)&dev_basename, MAX_BASENAME_SZ, "%s", q_arg);
 218
 219         return 0;
 220 }
 221
 222 /**
 223  * Parse the portmask provided at run time.
 224  */
 225 static int
 226 parse_portmask(const char *portmask)
 227 {
 228         char *end = NULL;
 229         unsigned long pm;
 230
 231         /* parse hexadecimal string */
 232         pm = strtoul(portmask, &end, 16);
 233         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
 234                 return -1;
 235
 236         if (pm == 0)
 237                 return -1;
 238
 239         return pm;
 240 }
 241
 242 /**
 243  * Parse num options at run time.
 244  */
 245 static int
 246 parse_num_opt(const char *q_arg, uint32_t max_valid_value)
 247 {
 248         char *end = NULL;
 249         unsigned long num;
 250
 251         /* parse unsigned int string */
 252         num = strtoul(q_arg, &end, 10);
 253         if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
 254                 return -1;
 255
 256         if (num > max_valid_value)
 257                 return -1;
 258
 259         return num;
 260 }
 261
 262 /**
 263  * Display usage
 264  */
 265 static void
 266 tep_termination_usage(const char *prgname)
 267 {
 268         RTE_LOG(INFO, VHOST_CONFIG, "%s [EAL options] -- -p PORTMASK\n"
 269         "               --udp-port: UDP destination port for VXLAN packet\n"
 270         "               --nb-devices[1-64]: The number of virtIO device\n"
 271         "               --tx-checksum [0|1]: inner Tx checksum offload\n"
 272         "               --tso-segsz [0-N]: TCP segment size\n"
 273         "               --filter-type[1-3]: filter type for tunneling packet\n"
 274         "                   1: Inner MAC and tenent ID\n"
 275         "                   2: Inner MAC and VLAN, and tenent ID\n"
 276         "                   3: Outer MAC, Inner MAC and tenent ID\n"
 277         "               -p PORTMASK: Set mask for ports to be used by application\n"
 278         "               --rx-retry [0|1]: disable/enable(default) retries on rx."
 279         "                Enable retry if destintation queue is full\n"
 280         "               --rx-retry-delay [0-N]: timeout(in usecond) between retries on RX."
 281         "                This makes effect only if retries on rx enabled\n"
 282         "               --rx-retry-num [0-N]: the number of retries on rx."
 283         "                This makes effect only if retries on rx enabled\n"
 284         "               --stats [0-N]: 0: Disable stats, N: Time in seconds to print stats\n"
 285         "               --dev-basename: The basename to be used for the character device.\n",
 286                prgname);
 287 }
 288
 289 /**
 290  * Parse the arguments given in the command line of the application.
 291  */
 292 static int
 293 tep_termination_parse_args(int argc, char **argv)
 294 {
 295         int opt, ret;
 296         int option_index;
 297         unsigned i;
 298         const char *prgname = argv[0];
 299         static struct option long_option[] = {
 300                 {CMD_LINE_OPT_NB_DEVICES, required_argument, NULL, 0},
 301                 {CMD_LINE_OPT_UDP_PORT, required_argument, NULL, 0},
 302                 {CMD_LINE_OPT_TX_CHECKSUM, required_argument, NULL, 0},
 303                 {CMD_LINE_OPT_TSO_SEGSZ, required_argument, NULL, 0},
 304                 {CMD_LINE_OPT_FILTER_TYPE, required_argument, NULL, 0},
 305                 {CMD_LINE_OPT_RX_RETRY, required_argument, NULL, 0},
 306                 {CMD_LINE_OPT_RX_RETRY_DELAY, required_argument, NULL, 0},
 307                 {CMD_LINE_OPT_RX_RETRY_NUM, required_argument, NULL, 0},
 308                 {CMD_LINE_OPT_STATS, required_argument, NULL, 0},
 309                 {CMD_LINE_OPT_DEV_BASENAME, required_argument, NULL, 0},
 310                 {NULL, 0, 0, 0},
 311         };
 312
 313         /* Parse command line */
 314         while ((opt = getopt_long(argc, argv, "p:",
 315                         long_option, &option_index)) != EOF) {
 316                 switch (opt) {
 317                 /* Portmask */
 318                 case 'p':
 319                         enabled_port_mask = parse_portmask(optarg);
 320                         if (enabled_port_mask == 0) {
 321                                 RTE_LOG(INFO, VHOST_CONFIG,
 322                                         "Invalid portmask\n");
 323                                 tep_termination_usage(prgname);
 324                                 return -1;
 325                         }
 326                         break;
 327                 case 0:
 328                         if (!strncmp(long_option[option_index].name,
 329                                 CMD_LINE_OPT_NB_DEVICES,
 330                                 sizeof(CMD_LINE_OPT_NB_DEVICES))) {
 331                                 ret = parse_num_opt(optarg, MAX_DEVICES);
 332                                 if (ret == -1) {
 333                                         RTE_LOG(INFO, VHOST_CONFIG,
 334                                         "Invalid argument for nb-devices [0-%d]\n",
 335                                         MAX_DEVICES);
 336                                         tep_termination_usage(prgname);
 337                                         return -1;
 338                                 } else
 339                                         nb_devices = ret;
 340                         }
 341
 342                         /* Enable/disable retries on RX. */
 343                         if (!strncmp(long_option[option_index].name,
 344                                 CMD_LINE_OPT_RX_RETRY,
 345                                 sizeof(CMD_LINE_OPT_RX_RETRY))) {
 346                                 ret = parse_num_opt(optarg, 1);
 347                                 if (ret == -1) {
 348                                         RTE_LOG(INFO, VHOST_CONFIG,
 349                                                 "Invalid argument for rx-retry [0|1]\n");
 350                                         tep_termination_usage(prgname);
 351                                         return -1;
 352                                 } else
 353                                         enable_retry = ret;
 354                         }
 355
 356                         if (!strncmp(long_option[option_index].name,
 357                                 CMD_LINE_OPT_TSO_SEGSZ,
 358                                 sizeof(CMD_LINE_OPT_TSO_SEGSZ))) {
 359                                 ret = parse_num_opt(optarg, INT16_MAX);
 360                                 if (ret == -1) {
 361                                         RTE_LOG(INFO, VHOST_CONFIG,
 362                                                 "Invalid argument for TCP segment size [0-N]\n");
 363                                         tep_termination_usage(prgname);
 364                                         return -1;
 365                                 } else
 366                                         tso_segsz = ret;
 367                         }
 368
 369                         if (!strncmp(long_option[option_index].name,
 370                                         CMD_LINE_OPT_UDP_PORT,
 371                                         sizeof(CMD_LINE_OPT_UDP_PORT))) {
 372                                 ret = parse_num_opt(optarg, INT16_MAX);
 373                                 if (ret == -1) {
 374                                         RTE_LOG(INFO, VHOST_CONFIG,
 375                                                 "Invalid argument for UDP port [0-N]\n");
 376                                         tep_termination_usage(prgname);
 377                                         return -1;
 378                                 } else
 379                                         udp_port = ret;
 380                         }
 381
 382                         /* Specify the retries delay time (in useconds) on RX.*/
 383                         if (!strncmp(long_option[option_index].name,
 384                                 CMD_LINE_OPT_RX_RETRY_DELAY,
 385                                 sizeof(CMD_LINE_OPT_RX_RETRY_DELAY))) {
 386                                 ret = parse_num_opt(optarg, INT32_MAX);
 387                                 if (ret == -1) {
 388                                         RTE_LOG(INFO, VHOST_CONFIG,
 389                                                 "Invalid argument for rx-retry-delay [0-N]\n");
 390                                         tep_termination_usage(prgname);
 391                                         return -1;
 392                                 } else
 393                                         burst_rx_delay_time = ret;
 394                         }
 395
 396                         /* Specify the retries number on RX. */
 397                         if (!strncmp(long_option[option_index].name,
 398                                 CMD_LINE_OPT_RX_RETRY_NUM,
 399                                 sizeof(CMD_LINE_OPT_RX_RETRY_NUM))) {
 400                                 ret = parse_num_opt(optarg, INT32_MAX);
 401                                 if (ret == -1) {
 402                                         RTE_LOG(INFO, VHOST_CONFIG,
 403                                                 "Invalid argument for rx-retry-num [0-N]\n");
 404                                         tep_termination_usage(prgname);
 405                                         return -1;
 406                                 } else
 407                                         burst_rx_retry_num = ret;
 408                         }
 409
 410                         if (!strncmp(long_option[option_index].name,
 411                                 CMD_LINE_OPT_TX_CHECKSUM,
 412                                 sizeof(CMD_LINE_OPT_TX_CHECKSUM))) {
 413                                 ret = parse_num_opt(optarg, 1);
 414                                 if (ret == -1) {
 415                                         RTE_LOG(INFO, VHOST_CONFIG,
 416                                                 "Invalid argument for tx-checksum [0|1]\n");
 417                                         tep_termination_usage(prgname);
 418                                         return -1;
 419                                 } else
 420                                         tx_checksum = ret;
 421                         }
 422
 423                         if (!strncmp(long_option[option_index].name,
 424                                         CMD_LINE_OPT_FILTER_TYPE,
 425                                         sizeof(CMD_LINE_OPT_FILTER_TYPE))) {
 426                                 ret = parse_num_opt(optarg, 3);
 427                                 if ((ret == -1) || (ret == 0)) {
 428                                         RTE_LOG(INFO, VHOST_CONFIG,
 429                                                 "Invalid argument for filter type [1-3]\n");
 430                                         tep_termination_usage(prgname);
 431                                         return -1;
 432                                 } else
 433                                         filter_idx = ret - 1;
 434                         }
 435
 436                         /* Enable/disable stats. */
 437                         if (!strncmp(long_option[option_index].name,
 438                                 CMD_LINE_OPT_STATS,
 439                                 sizeof(CMD_LINE_OPT_STATS))) {
 440                                 ret = parse_num_opt(optarg, INT32_MAX);
 441                                 if (ret == -1) {
 442                                         RTE_LOG(INFO, VHOST_CONFIG,
 443                                                         "Invalid argument for stats [0..N]\n");
 444                                         tep_termination_usage(prgname);
 445                                         return -1;
 446                                 } else
 447                                         enable_stats = ret;
 448                         }
 449
 450                         /* Set character device basename. */
 451                         if (!strncmp(long_option[option_index].name,
 452                                 CMD_LINE_OPT_DEV_BASENAME,
 453                                 sizeof(CMD_LINE_OPT_DEV_BASENAME))) {
 454                                 if (us_vhost_parse_basename(optarg) == -1) {
 455                                         RTE_LOG(INFO, VHOST_CONFIG,
 456                                                 "Invalid argument for character "
 457                                                 "device basename (Max %d characters)\n",
 458                                                 MAX_BASENAME_SZ);
 459                                         tep_termination_usage(prgname);
 460                                         return -1;
 461                                 }
 462                         }
 463
 464                         break;
 465
 466                         /* Invalid option - print options. */
 467                 default:
 468                         tep_termination_usage(prgname);
 469                         return -1;
 470                 }
 471         }
 472
 473         for (i = 0; i < RTE_MAX_ETHPORTS; i++) {
 474                 if (enabled_port_mask & (1 << i))
 475                         ports[nb_ports++] = (uint8_t)i;
 476         }
 477
 478         if ((nb_ports ==  0) || (nb_ports > MAX_SUP_PORTS)) {
 479                 RTE_LOG(INFO, VHOST_PORT, "Current enabled port number is %u,"
 480                         "but only %u port can be enabled\n", nb_ports,
 481                         MAX_SUP_PORTS);
 482                 return -1;
 483         }
 484
 485         return 0;
 486 }
 487
 488 /**
 489  * Update the global var NB_PORTS and array PORTS
 490  * according to system ports number and return valid ports number
 491  */
 492 static unsigned
 493 check_ports_num(unsigned max_nb_ports)
 494 {
 495         unsigned valid_nb_ports = nb_ports;
 496         unsigned portid;
 497
 498         if (nb_ports > max_nb_ports) {
 499                 RTE_LOG(INFO, VHOST_PORT, "\nSpecified port number(%u) "
 500                         " exceeds total system port number(%u)\n",
 501                         nb_ports, max_nb_ports);
 502                 nb_ports = max_nb_ports;
 503         }
 504
 505         for (portid = 0; portid < nb_ports; portid++) {
 506                 if (ports[portid] >= max_nb_ports) {
 507                         RTE_LOG(INFO, VHOST_PORT,
 508                                 "\nSpecified port ID(%u) exceeds max "
 509                                 " system port ID(%u)\n",
 510                                 ports[portid], (max_nb_ports - 1));
 511                         ports[portid] = INVALID_PORT_ID;
 512                         valid_nb_ports--;
 513                 }
 514         }
 515         return valid_nb_ports;
 516 }
 517
 518 /**
 519  * This function routes the TX packet to the correct interface. This may be a local device
 520  * or the physical port.
 521  */
 522 static inline void __attribute__((always_inline))
 523 virtio_tx_route(struct vhost_dev *vdev, struct rte_mbuf *m)
 524 {
 525         struct mbuf_table *tx_q;
 526         struct rte_mbuf **m_table;
 527         unsigned len, ret = 0;
 528         const uint16_t lcore_id = rte_lcore_id();
 529         struct virtio_net *dev = vdev->dev;
 530
 531         LOG_DEBUG(VHOST_DATA, "(%"PRIu64") TX: MAC address is external\n",
 532                 dev->device_fh);
 533
 534         /* Add packet to the port tx queue */
 535         tx_q = &lcore_tx_queue[lcore_id];
 536         len = tx_q->len;
 537
 538         tx_q->m_table[len] = m;
 539         len++;
 540         if (enable_stats) {
 541                 dev_statistics[dev->device_fh].tx_total++;
 542                 dev_statistics[dev->device_fh].tx++;
 543         }
 544
 545         if (unlikely(len == MAX_PKT_BURST)) {
 546                 m_table = (struct rte_mbuf **)tx_q->m_table;
 547                 ret = overlay_options.tx_handle(ports[0],
 548                         (uint16_t)tx_q->txq_id, m_table,
 549                         (uint16_t)tx_q->len);
 550
 551                 /* Free any buffers not handled by TX and update
 552                  * the port stats.
 553                  */
 554                 if (unlikely(ret < len)) {
 555                         do {
 556                                 rte_pktmbuf_free(m_table[ret]);
 557                         } while (++ret < len);
 558                 }
 559
 560                 len = 0;
 561         }
 562
 563         tx_q->len = len;
 564         return;
 565 }
 566
 567 /**
 568  * This function is called by each data core. It handles all
 569  * RX/TX registered with the core. For TX the specific lcore
 570  * linked list is used. For RX, MAC addresses are compared
 571  * with all devices in the main linked list.
 572  */
 573 static int
 574 switch_worker(__rte_unused void *arg)
 575 {
 576         struct rte_mempool *mbuf_pool = arg;
 577         struct virtio_net *dev = NULL;
 578         struct vhost_dev *vdev = NULL;
 579         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
 580         struct virtio_net_data_ll *dev_ll;
 581         struct mbuf_table *tx_q;
 582         volatile struct lcore_ll_info *lcore_ll;
 583         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1)
 584                                         / US_PER_S * BURST_TX_DRAIN_US;
 585         uint64_t prev_tsc, diff_tsc, cur_tsc, ret_count = 0;
 586         unsigned i, ret = 0;
 587         const uint16_t lcore_id = rte_lcore_id();
 588         const uint16_t num_cores = (uint16_t)rte_lcore_count();
 589         uint16_t rx_count = 0;
 590         uint16_t tx_count;
 591         uint32_t retry = 0;
 592
 593         RTE_LOG(INFO, VHOST_DATA, "Procesing on Core %u started\n", lcore_id);
 594         lcore_ll = lcore_info[lcore_id].lcore_ll;
 595         prev_tsc = 0;
 596
 597         tx_q = &lcore_tx_queue[lcore_id];
 598         for (i = 0; i < num_cores; i++) {
 599                 if (lcore_ids[i] == lcore_id) {
 600                         tx_q->txq_id = i;
 601                         break;
 602                 }
 603         }
 604
 605         while (1) {
 606                 cur_tsc = rte_rdtsc();
 607                 /*
 608                  * TX burst queue drain
 609                  */
 610                 diff_tsc = cur_tsc - prev_tsc;
 611                 if (unlikely(diff_tsc > drain_tsc)) {
 612
 613                         if (tx_q->len) {
 614                                 LOG_DEBUG(VHOST_DATA, "TX queue drained after "
 615                                         "timeout with burst size %u\n",
 616                                         tx_q->len);
 617                                 ret = overlay_options.tx_handle(ports[0],
 618                                         (uint16_t)tx_q->txq_id,
 619                                         (struct rte_mbuf **)tx_q->m_table,
 620                                         (uint16_t)tx_q->len);
 621                                 if (unlikely(ret < tx_q->len)) {
 622                                         do {
 623                                                 rte_pktmbuf_free(tx_q->m_table[ret]);
 624                                         } while (++ret < tx_q->len);
 625                                 }
 626
 627                                 tx_q->len = 0;
 628                         }
 629
 630                         prev_tsc = cur_tsc;
 631
 632                 }
 633
 634                 rte_prefetch0(lcore_ll->ll_root_used);
 635
 636                 /**
 637                  * Inform the configuration core that we have exited
 638                  * the linked list and that no devices are
 639                  * in use if requested.
 640                  */
 641                 if (lcore_ll->dev_removal_flag == REQUEST_DEV_REMOVAL)
 642                         lcore_ll->dev_removal_flag = ACK_DEV_REMOVAL;
 643
 644                 /*
 645                  * Process devices
 646                  */
 647                 dev_ll = lcore_ll->ll_root_used;
 648
 649                 while (dev_ll != NULL) {
 650                         vdev = dev_ll->vdev;
 651                         dev = vdev->dev;
 652
 653                         if (unlikely(vdev->remove)) {
 654                                 dev_ll = dev_ll->next;
 655                                 overlay_options.tunnel_destroy(vdev);
 656                                 vdev->ready = DEVICE_SAFE_REMOVE;
 657                                 continue;
 658                         }
 659                         if (likely(vdev->ready == DEVICE_RX)) {
 660                                 /* Handle guest RX */
 661                                 rx_count = rte_eth_rx_burst(ports[0],
 662                                         vdev->rx_q, pkts_burst, MAX_PKT_BURST);
 663
 664                                 if (rx_count) {
 665                                         /*
 666                                         * Retry is enabled and the queue is
 667                                         * full then we wait and retry to
 668                                         * avoid packet loss. Here MAX_PKT_BURST
 669                                         * must be less than virtio queue size
 670                                         */
 671                                         if (enable_retry && unlikely(rx_count >
 672                                                 rte_vring_available_entries(dev, VIRTIO_RXQ))) {
 673                                                 for (retry = 0; retry < burst_rx_retry_num;
 674                                                         retry++) {
 675                                                         rte_delay_us(burst_rx_delay_time);
 676                                                         if (rx_count <= rte_vring_available_entries(dev, VIRTIO_RXQ))
 677                                                                 break;
 678                                                 }
 679                                         }
 680
 681                                         ret_count = overlay_options.rx_handle(dev, pkts_burst, rx_count);
 682                                         if (enable_stats) {
 683                                                 rte_atomic64_add(
 684                                                 &dev_statistics[dev->device_fh].rx_total_atomic,
 685                                                 rx_count);
 686                                                 rte_atomic64_add(
 687                                                 &dev_statistics[dev->device_fh].rx_atomic, ret_count);
 688                                         }
 689                                         while (likely(rx_count)) {
 690                                                 rx_count--;
 691                                                 rte_pktmbuf_free(pkts_burst[rx_count]);
 692                                         }
 693
 694                                 }
 695                         }
 696
 697                         if (likely(!vdev->remove)) {
 698                                 /* Handle guest TX*/
 699                                 tx_count = rte_vhost_dequeue_burst(dev,
 700                                                 VIRTIO_TXQ, mbuf_pool,
 701                                                 pkts_burst, MAX_PKT_BURST);
 702                                 /* If this is the first received packet we need to learn the MAC */
 703                                 if (unlikely(vdev->ready == DEVICE_MAC_LEARNING) && tx_count) {
 704                                         if (vdev->remove ||
 705                                                 (overlay_options.tunnel_setup(vdev, pkts_burst[0]) == -1)) {
 706                                                 while (tx_count)
 707                                                         rte_pktmbuf_free(pkts_burst[--tx_count]);
 708                                         }
 709                                 }
 710                                 while (tx_count)
 711                                         virtio_tx_route(vdev, pkts_burst[--tx_count]);
 712                         }
 713
 714                         /* move to the next device in the list */
 715                         dev_ll = dev_ll->next;
 716                 }
 717         }
 718
 719         return 0;
 720 }
 721
 722 /**
 723  * Add an entry to a used linked list. A free entry must first be found
 724  * in the free linked list using get_data_ll_free_entry();
 725  */
 726 static void
 727 add_data_ll_entry(struct virtio_net_data_ll **ll_root_addr,
 728         struct virtio_net_data_ll *ll_dev)
 729 {
 730         struct virtio_net_data_ll *ll = *ll_root_addr;
 731
 732         /* Set next as NULL and use a compiler barrier to avoid reordering. */
 733         ll_dev->next = NULL;
 734         rte_compiler_barrier();
 735
 736         /* If ll == NULL then this is the first device. */
 737         if (ll) {
 738                 /* Increment to the tail of the linked list. */
 739                 while (ll->next != NULL)
 740                         ll = ll->next;
 741
 742                 ll->next = ll_dev;
 743         } else {
 744                 *ll_root_addr = ll_dev;
 745         }
 746 }
 747
 748 /**
 749  * Remove an entry from a used linked list. The entry must then be added to
 750  * the free linked list using put_data_ll_free_entry().
 751  */
 752 static void
 753 rm_data_ll_entry(struct virtio_net_data_ll **ll_root_addr,
 754         struct virtio_net_data_ll *ll_dev,
 755         struct virtio_net_data_ll *ll_dev_last)
 756 {
 757         struct virtio_net_data_ll *ll = *ll_root_addr;
 758
 759         if (unlikely((ll == NULL) || (ll_dev == NULL)))
 760                 return;
 761
 762         if (ll_dev == ll)
 763                 *ll_root_addr = ll_dev->next;
 764         else
 765                 if (likely(ll_dev_last != NULL))
 766                         ll_dev_last->next = ll_dev->next;
 767                 else
 768                         RTE_LOG(ERR, VHOST_CONFIG,
 769                                 "Remove entry form ll failed.\n");
 770 }
 771
 772 /**
 773  * Find and return an entry from the free linked list.
 774  */
 775 static struct virtio_net_data_ll *
 776 get_data_ll_free_entry(struct virtio_net_data_ll **ll_root_addr)
 777 {
 778         struct virtio_net_data_ll *ll_free = *ll_root_addr;
 779         struct virtio_net_data_ll *ll_dev;
 780
 781         if (ll_free == NULL)
 782                 return NULL;
 783
 784         ll_dev = ll_free;
 785         *ll_root_addr = ll_free->next;
 786
 787         return ll_dev;
 788 }
 789
 790 /**
 791  * Place an entry back on to the free linked list.
 792  */
 793 static void
 794 put_data_ll_free_entry(struct virtio_net_data_ll **ll_root_addr,
 795         struct virtio_net_data_ll *ll_dev)
 796 {
 797         struct virtio_net_data_ll *ll_free = *ll_root_addr;
 798
 799         if (ll_dev == NULL)
 800                 return;
 801
 802         ll_dev->next = ll_free;
 803         *ll_root_addr = ll_dev;
 804 }
 805
 806 /**
 807  * Creates a linked list of a given size.
 808  */
 809 static struct virtio_net_data_ll *
 810 alloc_data_ll(uint32_t size)
 811 {
 812         struct virtio_net_data_ll *ll_new;
 813         uint32_t i;
 814
 815         /* Malloc and then chain the linked list. */
 816         ll_new = malloc(size * sizeof(struct virtio_net_data_ll));
 817         if (ll_new == NULL) {
 818                 RTE_LOG(ERR, VHOST_CONFIG,
 819                         "Failed to allocate memory for ll_new.\n");
 820                 return NULL;
 821         }
 822
 823         for (i = 0; i < size - 1; i++) {
 824                 ll_new[i].vdev = NULL;
 825                 ll_new[i].next = &ll_new[i+1];
 826         }
 827         ll_new[i].next = NULL;
 828
 829         return ll_new;
 830 }
 831
 832 /**
 833  * Create the main linked list along with each individual cores
 834  * linked list. A used and a free list are created to manage entries.
 835  */
 836 static int
 837 init_data_ll(void)
 838 {
 839         int lcore;
 840
 841         RTE_LCORE_FOREACH_SLAVE(lcore) {
 842                 lcore_info[lcore].lcore_ll =
 843                         malloc(sizeof(struct lcore_ll_info));
 844                 if (lcore_info[lcore].lcore_ll == NULL) {
 845                         RTE_LOG(ERR, VHOST_CONFIG,
 846                                 "Failed to allocate memory for lcore_ll.\n");
 847                         return -1;
 848                 }
 849
 850                 lcore_info[lcore].lcore_ll->device_num = 0;
 851                 lcore_info[lcore].lcore_ll->dev_removal_flag = ACK_DEV_REMOVAL;
 852                 lcore_info[lcore].lcore_ll->ll_root_used = NULL;
 853                 if (nb_devices % nb_switching_cores)
 854                         lcore_info[lcore].lcore_ll->ll_root_free =
 855                                 alloc_data_ll((nb_devices / nb_switching_cores)
 856                                                 + 1);
 857                 else
 858                         lcore_info[lcore].lcore_ll->ll_root_free =
 859                                 alloc_data_ll(nb_devices / nb_switching_cores);
 860         }
 861
 862         /* Allocate devices up to a maximum of MAX_DEVICES. */
 863         ll_root_free = alloc_data_ll(MIN((nb_devices), MAX_DEVICES));
 864
 865         return 0;
 866 }
 867
 868 /**
 869  * Remove a device from the specific data core linked list and
 870  * from the main linked list. Synchonization occurs through the use
 871  * of the lcore dev_removal_flag. Device is made volatile here
 872  * to avoid re-ordering of dev->remove=1 which can cause an infinite
 873  * loop in the rte_pause loop.
 874  */
 875 static void
 876 destroy_device(volatile struct virtio_net *dev)
 877 {
 878         struct virtio_net_data_ll *ll_lcore_dev_cur;
 879         struct virtio_net_data_ll *ll_main_dev_cur;
 880         struct virtio_net_data_ll *ll_lcore_dev_last = NULL;
 881         struct virtio_net_data_ll *ll_main_dev_last = NULL;
 882         struct vhost_dev *vdev;
 883         int lcore;
 884
 885         dev->flags &= ~VIRTIO_DEV_RUNNING;
 886
 887         vdev = (struct vhost_dev *)dev->priv;
 888
 889         /* set the remove flag. */
 890         vdev->remove = 1;
 891         while (vdev->ready != DEVICE_SAFE_REMOVE)
 892                 rte_pause();
 893
 894         /* Search for entry to be removed from lcore ll */
 895         ll_lcore_dev_cur = lcore_info[vdev->coreid].lcore_ll->ll_root_used;
 896         while (ll_lcore_dev_cur != NULL) {
 897                 if (ll_lcore_dev_cur->vdev == vdev) {
 898                         break;
 899                 } else {
 900                         ll_lcore_dev_last = ll_lcore_dev_cur;
 901                         ll_lcore_dev_cur = ll_lcore_dev_cur->next;
 902                 }
 903         }
 904
 905         if (ll_lcore_dev_cur == NULL) {
 906                 RTE_LOG(ERR, VHOST_CONFIG,
 907                         "(%"PRIu64") Failed to find the dev to be destroy.\n",
 908                         dev->device_fh);
 909                 return;
 910         }
 911
 912         /* Search for entry to be removed from main ll */
 913         ll_main_dev_cur = ll_root_used;
 914         ll_main_dev_last = NULL;
 915         while (ll_main_dev_cur != NULL) {
 916                 if (ll_main_dev_cur->vdev == vdev) {
 917                         break;
 918                 } else {
 919                         ll_main_dev_last = ll_main_dev_cur;
 920                         ll_main_dev_cur = ll_main_dev_cur->next;
 921                 }
 922         }
 923
 924         /* Remove entries from the lcore and main ll. */
 925         rm_data_ll_entry(&lcore_info[vdev->coreid].lcore_ll->ll_root_used,
 926                         ll_lcore_dev_cur, ll_lcore_dev_last);
 927         rm_data_ll_entry(&ll_root_used, ll_main_dev_cur, ll_main_dev_last);
 928
 929         /* Set the dev_removal_flag on each lcore. */
 930         RTE_LCORE_FOREACH_SLAVE(lcore) {
 931                 lcore_info[lcore].lcore_ll->dev_removal_flag =
 932                         REQUEST_DEV_REMOVAL;
 933         }
 934
 935         /*
 936          * Once each core has set the dev_removal_flag to
 937          * ACK_DEV_REMOVAL we can be sure that they can no longer access
 938          * the device removed from the linked lists and that the devices
 939          * are no longer in use.
 940          */
 941         RTE_LCORE_FOREACH_SLAVE(lcore) {
 942                 while (lcore_info[lcore].lcore_ll->dev_removal_flag
 943                         != ACK_DEV_REMOVAL)
 944                         rte_pause();
 945         }
 946
 947         /* Add the entries back to the lcore and main free ll.*/
 948         put_data_ll_free_entry(&lcore_info[vdev->coreid].lcore_ll->ll_root_free,
 949                                 ll_lcore_dev_cur);
 950         put_data_ll_free_entry(&ll_root_free, ll_main_dev_cur);
 951
 952         /* Decrement number of device on the lcore. */
 953         lcore_info[vdev->coreid].lcore_ll->device_num--;
 954
 955         RTE_LOG(INFO, VHOST_DATA, "(%"PRIu64") Device has been removed "
 956                 "from data core\n", dev->device_fh);
 957
 958         rte_free(vdev);
 959
 960 }
 961
 962 /**
 963  * A new device is added to a data core. First the device is added
 964  * to the main linked list and the allocated to a specific data core.
 965  */
 966 static int
 967 new_device(struct virtio_net *dev)
 968 {
 969         struct virtio_net_data_ll *ll_dev;
 970         int lcore, core_add = 0;
 971         uint32_t device_num_min = nb_devices;
 972         struct vhost_dev *vdev;
 973
 974         vdev = rte_zmalloc("vhost device", sizeof(*vdev), RTE_CACHE_LINE_SIZE);
 975         if (vdev == NULL) {
 976                 RTE_LOG(INFO, VHOST_DATA,
 977                         "(%"PRIu64") Couldn't allocate memory for vhost dev\n",
 978                         dev->device_fh);
 979                 return -1;
 980         }
 981         vdev->dev = dev;
 982         dev->priv = vdev;
 983         /* Add device to main ll */
 984         ll_dev = get_data_ll_free_entry(&ll_root_free);
 985         if (ll_dev == NULL) {
 986                 RTE_LOG(INFO, VHOST_DATA, "(%"PRIu64") No free entry found in"
 987                         " linked list Device limit of %d devices per core"
 988                         " has been reached\n", dev->device_fh, nb_devices);
 989                 if (vdev->regions_hpa)
 990                         rte_free(vdev->regions_hpa);
 991                 rte_free(vdev);
 992                 return -1;
 993         }
 994         ll_dev->vdev = vdev;
 995         add_data_ll_entry(&ll_root_used, ll_dev);
 996         vdev->rx_q = dev->device_fh;
 997
 998         /* reset ready flag */
 999         vdev->ready = DEVICE_MAC_LEARNING;
1000         vdev->remove = 0;
1001
1002         /* Find a suitable lcore to add the device. */
1003         RTE_LCORE_FOREACH_SLAVE(lcore) {
1004                 if (lcore_info[lcore].lcore_ll->device_num < device_num_min) {
1005                         device_num_min = lcore_info[lcore].lcore_ll->device_num;
1006                         core_add = lcore;
1007                 }
1008         }
1009         /* Add device to lcore ll */
1010         ll_dev = get_data_ll_free_entry(&lcore_info[core_add].lcore_ll->ll_root_free);
1011         if (ll_dev == NULL) {
1012                 RTE_LOG(INFO, VHOST_DATA,
1013                         "(%"PRIu64") Failed to add device to data core\n",
1014                         dev->device_fh);
1015                 vdev->ready = DEVICE_SAFE_REMOVE;
1016                 destroy_device(dev);
1017                 rte_free(vdev->regions_hpa);
1018                 rte_free(vdev);
1019                 return -1;
1020         }
1021         ll_dev->vdev = vdev;
1022         vdev->coreid = core_add;
1023
1024         add_data_ll_entry(&lcore_info[vdev->coreid].lcore_ll->ll_root_used,
1025                         ll_dev);
1026
1027         /* Initialize device stats */
1028         memset(&dev_statistics[dev->device_fh], 0,
1029                 sizeof(struct device_statistics));
1030
1031         /* Disable notifications. */
1032         rte_vhost_enable_guest_notification(dev, VIRTIO_RXQ, 0);
1033         rte_vhost_enable_guest_notification(dev, VIRTIO_TXQ, 0);
1034         lcore_info[vdev->coreid].lcore_ll->device_num++;
1035         dev->flags |= VIRTIO_DEV_RUNNING;
1036
1037         RTE_LOG(INFO, VHOST_DATA, "(%"PRIu64") Device has been added to data core %d\n",
1038                 dev->device_fh, vdev->coreid);
1039
1040         return 0;
1041 }
1042
1043 /**
1044  * These callback allow devices to be added to the data core when configuration
1045  * has been fully complete.
1046  */
1047 static const struct virtio_net_device_ops virtio_net_device_ops = {
1048         .new_device =  new_device,
1049         .destroy_device = destroy_device,
1050 };
1051
1052 /**
1053  * This is a thread will wake up after a period to print stats if the user has
1054  * enabled them.
1055  */
1056 static void
1057 print_stats(void)
1058 {
1059         struct virtio_net_data_ll *dev_ll;
1060         uint64_t tx_dropped, rx_dropped;
1061         uint64_t tx, tx_total, rx, rx_total, rx_ip_csum, rx_l4_csum;
1062         uint32_t device_fh;
1063         const char clr[] = { 27, '[', '2', 'J', '\0' };
1064         const char top_left[] = { 27, '[', '1', ';', '1', 'H', '\0' };
1065
1066         while (1) {
1067                 sleep(enable_stats);
1068
1069                 /* Clear screen and move to top left */
1070                 printf("%s%s", clr, top_left);
1071
1072                 printf("\nDevice statistics ================================");
1073
1074                 dev_ll = ll_root_used;
1075                 while (dev_ll != NULL) {
1076                         device_fh = (uint32_t)dev_ll->vdev->dev->device_fh;
1077                         tx_total = dev_statistics[device_fh].tx_total;
1078                         tx = dev_statistics[device_fh].tx;
1079                         tx_dropped = tx_total - tx;
1080
1081                         rx_total = rte_atomic64_read(
1082                                 &dev_statistics[device_fh].rx_total_atomic);
1083                         rx = rte_atomic64_read(
1084                                 &dev_statistics[device_fh].rx_atomic);
1085                         rx_dropped = rx_total - rx;
1086                         rx_ip_csum = rte_atomic64_read(
1087                                 &dev_statistics[device_fh].rx_bad_ip_csum);
1088                         rx_l4_csum = rte_atomic64_read(
1089                                 &dev_statistics[device_fh].rx_bad_l4_csum);
1090
1091                         printf("\nStatistics for device %"PRIu32" ----------"
1092                                         "\nTX total:            %"PRIu64""
1093                                         "\nTX dropped:          %"PRIu64""
1094                                         "\nTX successful:               %"PRIu64""
1095                                         "\nRX total:            %"PRIu64""
1096                                         "\nRX bad IP csum:      %"PRIu64""
1097                                         "\nRX bad L4 csum:      %"PRIu64""
1098                                         "\nRX dropped:          %"PRIu64""
1099                                         "\nRX successful:               %"PRIu64"",
1100                                         device_fh,
1101                                         tx_total,
1102                                         tx_dropped,
1103                                         tx,
1104                                         rx_total,
1105                                         rx_ip_csum,
1106                                         rx_l4_csum,
1107                                         rx_dropped,
1108                                         rx);
1109
1110                         dev_ll = dev_ll->next;
1111                 }
1112                 printf("\n================================================\n");
1113         }
1114 }
1115
1116 /**
1117  * Main function, does initialisation and calls the per-lcore functions. The CUSE
1118  * device is also registered here to handle the IOCTLs.
1119  */
1120 int
1121 main(int argc, char *argv[])
1122 {
1123         struct rte_mempool *mbuf_pool = NULL;
1124         unsigned lcore_id, core_id = 0;
1125         unsigned nb_ports, valid_nb_ports;
1126         int ret;
1127         uint8_t portid;
1128         uint16_t queue_id;
1129         static pthread_t tid;
1130
1131         /* init EAL */
1132         ret = rte_eal_init(argc, argv);
1133         if (ret < 0)
1134                 rte_exit(EXIT_FAILURE, "Error with EAL initialization\n");
1135         argc -= ret;
1136         argv += ret;
1137
1138         /* parse app arguments */
1139         ret = tep_termination_parse_args(argc, argv);
1140         if (ret < 0)
1141                 rte_exit(EXIT_FAILURE, "Invalid argument\n");
1142
1143         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++)
1144                 if (rte_lcore_is_enabled(lcore_id))
1145                         lcore_ids[core_id++] = lcore_id;
1146
1147         /* set the number of swithcing cores available */
1148         nb_switching_cores = rte_lcore_count()-1;
1149
1150         /* Get the number of physical ports. */
1151         nb_ports = rte_eth_dev_count();
1152         if (nb_ports > RTE_MAX_ETHPORTS)
1153                 nb_ports = RTE_MAX_ETHPORTS;
1154
1155         /*
1156          * Update the global var NB_PORTS and global array PORTS
1157          * and get value of var VALID_NB_PORTS according to system ports number
1158          */
1159         valid_nb_ports = check_ports_num(nb_ports);
1160
1161         if ((valid_nb_ports == 0) || (valid_nb_ports > MAX_SUP_PORTS)) {
1162                 rte_exit(EXIT_FAILURE, "Current enabled port number is %u,"
1163                         "but only %u port can be enabled\n", nb_ports,
1164                         MAX_SUP_PORTS);
1165         }
1166         /* Create the mbuf pool. */
1167         mbuf_pool = rte_mempool_create(
1168                         "MBUF_POOL",
1169                         NUM_MBUFS_PER_PORT
1170                         * valid_nb_ports,
1171                         MBUF_SIZE, MBUF_CACHE_SIZE,
1172                         sizeof(struct rte_pktmbuf_pool_private),
1173                         rte_pktmbuf_pool_init, NULL,
1174                         rte_pktmbuf_init, NULL,
1175                         rte_socket_id(), 0);
1176         if (mbuf_pool == NULL)
1177                 rte_exit(EXIT_FAILURE, "Cannot create mbuf pool\n");
1178
1179         for (queue_id = 0; queue_id < MAX_QUEUES + 1; queue_id++)
1180                 vpool_array[queue_id].pool = mbuf_pool;
1181
1182         /* Set log level. */
1183         rte_set_log_level(LOG_LEVEL);
1184
1185         /* initialize all ports */
1186         for (portid = 0; portid < nb_ports; portid++) {
1187                 /* skip ports that are not enabled */
1188                 if ((enabled_port_mask & (1 << portid)) == 0) {
1189                         RTE_LOG(INFO, VHOST_PORT,
1190                                 "Skipping disabled port %d\n", portid);
1191                         continue;
1192                 }
1193                 if (overlay_options.port_configure(portid, mbuf_pool) != 0)
1194                         rte_exit(EXIT_FAILURE,
1195                                 "Cannot initialize network ports\n");
1196         }
1197
1198         /* Initialise all linked lists. */
1199         if (init_data_ll() == -1)
1200                 rte_exit(EXIT_FAILURE, "Failed to initialize linked list\n");
1201
1202         /* Initialize device stats */
1203         memset(&dev_statistics, 0, sizeof(dev_statistics));
1204
1205         /* Enable stats if the user option is set. */
1206         if (enable_stats)
1207                 pthread_create(&tid, NULL, (void *)print_stats, NULL);
1208
1209         /* Launch all data cores. */
1210         RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1211                 rte_eal_remote_launch(switch_worker,
1212                         mbuf_pool, lcore_id);
1213         }
1214         rte_vhost_feature_disable(1ULL << VIRTIO_NET_F_MRG_RXBUF);
1215
1216         /* Register CUSE device to handle IOCTLs. */
1217         ret = rte_vhost_driver_register((char *)&dev_basename);
1218         if (ret != 0)
1219                 rte_exit(EXIT_FAILURE, "CUSE device setup failure.\n");
1220
1221         rte_vhost_driver_callback_register(&virtio_net_device_ops);
1222
1223         /* Start CUSE session. */
1224         rte_vhost_driver_session_start();
1225
1226         return 0;
1227 }