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_vhost.h>
  53 #include <rte_pause.h>
  54
  55 #include "main.h"
  56 #include "vxlan.h"
  57 #include "vxlan_setup.h"
  58
  59 /* the maximum number of external ports supported */
  60 #define MAX_SUP_PORTS 1
  61
  62 /**
  63  * Calculate the number of buffers needed per port
  64  */
  65 #define NUM_MBUFS_PER_PORT ((MAX_QUEUES * RTE_TEST_RX_DESC_DEFAULT) +\
  66                                 (nb_switching_cores * MAX_PKT_BURST) +\
  67                                 (nb_switching_cores * \
  68                                 RTE_TEST_TX_DESC_DEFAULT) +\
  69                                 (nb_switching_cores * MBUF_CACHE_SIZE))
  70
  71 #define MBUF_CACHE_SIZE 128
  72 #define MBUF_DATA_SIZE RTE_MBUF_DEFAULT_BUF_SIZE
  73
  74 #define MAX_PKT_BURST 32        /* Max burst size for RX/TX */
  75 #define BURST_TX_DRAIN_US 100   /* TX drain every ~100us */
  76
  77 /* Defines how long we wait between retries on RX */
  78 #define BURST_RX_WAIT_US 15
  79
  80 #define BURST_RX_RETRIES 4      /* Number of retries on RX. */
  81
  82 #define JUMBO_FRAME_MAX_SIZE    0x2600
  83
  84 /* State of virtio device. */
  85 #define DEVICE_MAC_LEARNING 0
  86 #define DEVICE_RX           1
  87 #define DEVICE_SAFE_REMOVE  2
  88
  89 /* Config_core_flag status definitions. */
  90 #define REQUEST_DEV_REMOVAL 1
  91 #define ACK_DEV_REMOVAL     0
  92
  93 /* Configurable number of RX/TX ring descriptors */
  94 #define RTE_TEST_RX_DESC_DEFAULT 1024
  95 #define RTE_TEST_TX_DESC_DEFAULT 512
  96
  97 /* Get first 4 bytes in mbuf headroom. */
  98 #define MBUF_HEADROOM_UINT32(mbuf) (*(uint32_t *)((uint8_t *)(mbuf) \
  99                 + sizeof(struct rte_mbuf)))
 100
 101 #define INVALID_PORT_ID 0xFFFF
 102
 103 /* Size of buffers used for snprintfs. */
 104 #define MAX_PRINT_BUFF 6072
 105
 106 /* Maximum character device basename size. */
 107 #define MAX_BASENAME_SZ 20
 108
 109 /* Maximum long option length for option parsing. */
 110 #define MAX_LONG_OPT_SZ 64
 111
 112 /* Used to compare MAC addresses. */
 113 #define MAC_ADDR_CMP 0xFFFFFFFFFFFFULL
 114
 115 #define CMD_LINE_OPT_NB_DEVICES "nb-devices"
 116 #define CMD_LINE_OPT_UDP_PORT "udp-port"
 117 #define CMD_LINE_OPT_TX_CHECKSUM "tx-checksum"
 118 #define CMD_LINE_OPT_TSO_SEGSZ "tso-segsz"
 119 #define CMD_LINE_OPT_FILTER_TYPE "filter-type"
 120 #define CMD_LINE_OPT_ENCAP "encap"
 121 #define CMD_LINE_OPT_DECAP "decap"
 122 #define CMD_LINE_OPT_RX_RETRY "rx-retry"
 123 #define CMD_LINE_OPT_RX_RETRY_DELAY "rx-retry-delay"
 124 #define CMD_LINE_OPT_RX_RETRY_NUM "rx-retry-num"
 125 #define CMD_LINE_OPT_STATS "stats"
 126 #define CMD_LINE_OPT_DEV_BASENAME "dev-basename"
 127
 128 /* mask of enabled ports */
 129 static uint32_t enabled_port_mask;
 130
 131 /*Number of switching cores enabled*/
 132 static uint32_t nb_switching_cores;
 133
 134 /* number of devices/queues to support*/
 135 uint16_t nb_devices = 2;
 136
 137 /* max ring descriptor, ixgbe, i40e, e1000 all are 4096. */
 138 #define MAX_RING_DESC 4096
 139
 140 struct vpool {
 141         struct rte_mempool *pool;
 142         struct rte_ring *ring;
 143         uint32_t buf_size;
 144 } vpool_array[MAX_QUEUES+MAX_QUEUES];
 145
 146 /* UDP tunneling port */
 147 uint16_t udp_port = 4789;
 148
 149 /* enable/disable inner TX checksum */
 150 uint8_t tx_checksum = 0;
 151
 152 /* TCP segment size */
 153 uint16_t tso_segsz = 0;
 154
 155 /* enable/disable decapsulation */
 156 uint8_t rx_decap = 1;
 157
 158 /* enable/disable encapsulation */
 159 uint8_t tx_encap = 1;
 160
 161 /* RX filter type for tunneling packet */
 162 uint8_t filter_idx = 1;
 163
 164 /* overlay packet operation */
 165 struct ol_switch_ops overlay_options = {
 166         .port_configure = vxlan_port_init,
 167         .tunnel_setup = vxlan_link,
 168         .tunnel_destroy = vxlan_unlink,
 169         .tx_handle = vxlan_tx_pkts,
 170         .rx_handle = vxlan_rx_pkts,
 171         .param_handle = NULL,
 172 };
 173
 174 /* Enable stats. */
 175 uint32_t enable_stats = 0;
 176 /* Enable retries on RX. */
 177 static uint32_t enable_retry = 1;
 178 /* Specify timeout (in useconds) between retries on RX. */
 179 static uint32_t burst_rx_delay_time = BURST_RX_WAIT_US;
 180 /* Specify the number of retries on RX. */
 181 static uint32_t burst_rx_retry_num = BURST_RX_RETRIES;
 182
 183 /* Character device basename. Can be set by user. */
 184 static char dev_basename[MAX_BASENAME_SZ] = "vhost-net";
 185
 186 static unsigned lcore_ids[RTE_MAX_LCORE];
 187 uint16_t ports[RTE_MAX_ETHPORTS];
 188
 189 static unsigned nb_ports; /**< The number of ports specified in command line */
 190
 191 /* ethernet addresses of ports */
 192 struct ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
 193
 194 /* heads for the main used and free linked lists for the data path. */
 195 static struct virtio_net_data_ll *ll_root_used;
 196 static struct virtio_net_data_ll *ll_root_free;
 197
 198 /**
 199  * Array of data core structures containing information on
 200  * individual core linked lists.
 201  */
 202 static struct lcore_info lcore_info[RTE_MAX_LCORE];
 203
 204 /* Used for queueing bursts of TX packets. */
 205 struct mbuf_table {
 206         unsigned len;
 207         unsigned txq_id;
 208         struct rte_mbuf *m_table[MAX_PKT_BURST];
 209 };
 210
 211 /* TX queue for each data core. */
 212 struct mbuf_table lcore_tx_queue[RTE_MAX_LCORE];
 213
 214 struct device_statistics dev_statistics[MAX_DEVICES];
 215
 216 /**
 217  * Set character device basename.
 218  */
 219 static int
 220 us_vhost_parse_basename(const char *q_arg)
 221 {
 222         /* parse number string */
 223         if (strlen(q_arg) >= MAX_BASENAME_SZ)
 224                 return -1;
 225         else
 226                 snprintf((char *)&dev_basename, MAX_BASENAME_SZ, "%s", q_arg);
 227
 228         return 0;
 229 }
 230
 231 /**
 232  * Parse the portmask provided at run time.
 233  */
 234 static int
 235 parse_portmask(const char *portmask)
 236 {
 237         char *end = NULL;
 238         unsigned long pm;
 239
 240         /* parse hexadecimal string */
 241         pm = strtoul(portmask, &end, 16);
 242         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
 243                 return -1;
 244
 245         if (pm == 0)
 246                 return -1;
 247
 248         return pm;
 249 }
 250
 251 /**
 252  * Parse num options at run time.
 253  */
 254 static int
 255 parse_num_opt(const char *q_arg, uint32_t max_valid_value)
 256 {
 257         char *end = NULL;
 258         unsigned long num;
 259
 260         /* parse unsigned int string */
 261         num = strtoul(q_arg, &end, 10);
 262         if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
 263                 return -1;
 264
 265         if (num > max_valid_value)
 266                 return -1;
 267
 268         return num;
 269 }
 270
 271 /**
 272  * Display usage
 273  */
 274 static void
 275 tep_termination_usage(const char *prgname)
 276 {
 277         RTE_LOG(INFO, VHOST_CONFIG, "%s [EAL options] -- -p PORTMASK\n"
 278         "               --udp-port: UDP destination port for VXLAN packet\n"
 279         "               --nb-devices[1-64]: The number of virtIO device\n"
 280         "               --tx-checksum [0|1]: inner Tx checksum offload\n"
 281         "               --tso-segsz [0-N]: TCP segment size\n"
 282         "               --decap [0|1]: tunneling packet decapsulation\n"
 283         "               --encap [0|1]: tunneling packet encapsulation\n"
 284         "               --filter-type[1-3]: filter type for tunneling packet\n"
 285         "                   1: Inner MAC and tenent ID\n"
 286         "                   2: Inner MAC and VLAN, and tenent ID\n"
 287         "                   3: Outer MAC, Inner MAC and tenent ID\n"
 288         "               -p PORTMASK: Set mask for ports to be used by application\n"
 289         "               --rx-retry [0|1]: disable/enable(default) retries on rx."
 290         "                Enable retry if destintation queue is full\n"
 291         "               --rx-retry-delay [0-N]: timeout(in usecond) between retries on RX."
 292         "                This makes effect only if retries on rx enabled\n"
 293         "               --rx-retry-num [0-N]: the number of retries on rx."
 294         "                This makes effect only if retries on rx enabled\n"
 295         "               --stats [0-N]: 0: Disable stats, N: Time in seconds to print stats\n"
 296         "               --dev-basename: The basename to be used for the character device.\n",
 297                prgname);
 298 }
 299
 300 /**
 301  * Parse the arguments given in the command line of the application.
 302  */
 303 static int
 304 tep_termination_parse_args(int argc, char **argv)
 305 {
 306         int opt, ret;
 307         int option_index;
 308         unsigned i;
 309         const char *prgname = argv[0];
 310         static struct option long_option[] = {
 311                 {CMD_LINE_OPT_NB_DEVICES, required_argument, NULL, 0},
 312                 {CMD_LINE_OPT_UDP_PORT, required_argument, NULL, 0},
 313                 {CMD_LINE_OPT_TX_CHECKSUM, required_argument, NULL, 0},
 314                 {CMD_LINE_OPT_TSO_SEGSZ, required_argument, NULL, 0},
 315                 {CMD_LINE_OPT_DECAP, required_argument, NULL, 0},
 316                 {CMD_LINE_OPT_ENCAP, required_argument, NULL, 0},
 317                 {CMD_LINE_OPT_FILTER_TYPE, required_argument, NULL, 0},
 318                 {CMD_LINE_OPT_RX_RETRY, required_argument, NULL, 0},
 319                 {CMD_LINE_OPT_RX_RETRY_DELAY, required_argument, NULL, 0},
 320                 {CMD_LINE_OPT_RX_RETRY_NUM, required_argument, NULL, 0},
 321                 {CMD_LINE_OPT_STATS, required_argument, NULL, 0},
 322                 {CMD_LINE_OPT_DEV_BASENAME, required_argument, NULL, 0},
 323                 {NULL, 0, 0, 0},
 324         };
 325
 326         /* Parse command line */
 327         while ((opt = getopt_long(argc, argv, "p:",
 328                         long_option, &option_index)) != EOF) {
 329                 switch (opt) {
 330                 /* Portmask */
 331                 case 'p':
 332                         enabled_port_mask = parse_portmask(optarg);
 333                         if (enabled_port_mask == 0) {
 334                                 RTE_LOG(INFO, VHOST_CONFIG,
 335                                         "Invalid portmask\n");
 336                                 tep_termination_usage(prgname);
 337                                 return -1;
 338                         }
 339                         break;
 340                 case 0:
 341                         if (!strncmp(long_option[option_index].name,
 342                                 CMD_LINE_OPT_NB_DEVICES,
 343                                 sizeof(CMD_LINE_OPT_NB_DEVICES))) {
 344                                 ret = parse_num_opt(optarg, MAX_DEVICES);
 345                                 if (ret == -1) {
 346                                         RTE_LOG(INFO, VHOST_CONFIG,
 347                                         "Invalid argument for nb-devices [0-%d]\n",
 348                                         MAX_DEVICES);
 349                                         tep_termination_usage(prgname);
 350                                         return -1;
 351                                 } else
 352                                         nb_devices = ret;
 353                         }
 354
 355                         /* Enable/disable retries on RX. */
 356                         if (!strncmp(long_option[option_index].name,
 357                                 CMD_LINE_OPT_RX_RETRY,
 358                                 sizeof(CMD_LINE_OPT_RX_RETRY))) {
 359                                 ret = parse_num_opt(optarg, 1);
 360                                 if (ret == -1) {
 361                                         RTE_LOG(INFO, VHOST_CONFIG,
 362                                                 "Invalid argument for rx-retry [0|1]\n");
 363                                         tep_termination_usage(prgname);
 364                                         return -1;
 365                                 } else
 366                                         enable_retry = ret;
 367                         }
 368
 369                         if (!strncmp(long_option[option_index].name,
 370                                 CMD_LINE_OPT_TSO_SEGSZ,
 371                                 sizeof(CMD_LINE_OPT_TSO_SEGSZ))) {
 372                                 ret = parse_num_opt(optarg, INT16_MAX);
 373                                 if (ret == -1) {
 374                                         RTE_LOG(INFO, VHOST_CONFIG,
 375                                                 "Invalid argument for TCP segment size [0-N]\n");
 376                                         tep_termination_usage(prgname);
 377                                         return -1;
 378                                 } else
 379                                         tso_segsz = ret;
 380                         }
 381
 382                         if (!strncmp(long_option[option_index].name,
 383                                         CMD_LINE_OPT_UDP_PORT,
 384                                         sizeof(CMD_LINE_OPT_UDP_PORT))) {
 385                                 ret = parse_num_opt(optarg, INT16_MAX);
 386                                 if (ret == -1) {
 387                                         RTE_LOG(INFO, VHOST_CONFIG,
 388                                                 "Invalid argument for UDP port [0-N]\n");
 389                                         tep_termination_usage(prgname);
 390                                         return -1;
 391                                 } else
 392                                         udp_port = ret;
 393                         }
 394
 395                         /* Specify the retries delay time (in useconds) on RX.*/
 396                         if (!strncmp(long_option[option_index].name,
 397                                 CMD_LINE_OPT_RX_RETRY_DELAY,
 398                                 sizeof(CMD_LINE_OPT_RX_RETRY_DELAY))) {
 399                                 ret = parse_num_opt(optarg, INT32_MAX);
 400                                 if (ret == -1) {
 401                                         RTE_LOG(INFO, VHOST_CONFIG,
 402                                                 "Invalid argument for rx-retry-delay [0-N]\n");
 403                                         tep_termination_usage(prgname);
 404                                         return -1;
 405                                 } else
 406                                         burst_rx_delay_time = ret;
 407                         }
 408
 409                         /* Specify the retries number on RX. */
 410                         if (!strncmp(long_option[option_index].name,
 411                                 CMD_LINE_OPT_RX_RETRY_NUM,
 412                                 sizeof(CMD_LINE_OPT_RX_RETRY_NUM))) {
 413                                 ret = parse_num_opt(optarg, INT32_MAX);
 414                                 if (ret == -1) {
 415                                         RTE_LOG(INFO, VHOST_CONFIG,
 416                                                 "Invalid argument for rx-retry-num [0-N]\n");
 417                                         tep_termination_usage(prgname);
 418                                         return -1;
 419                                 } else
 420                                         burst_rx_retry_num = ret;
 421                         }
 422
 423                         if (!strncmp(long_option[option_index].name,
 424                                 CMD_LINE_OPT_TX_CHECKSUM,
 425                                 sizeof(CMD_LINE_OPT_TX_CHECKSUM))) {
 426                                 ret = parse_num_opt(optarg, 1);
 427                                 if (ret == -1) {
 428                                         RTE_LOG(INFO, VHOST_CONFIG,
 429                                                 "Invalid argument for tx-checksum [0|1]\n");
 430                                         tep_termination_usage(prgname);
 431                                         return -1;
 432                                 } else
 433                                         tx_checksum = ret;
 434                         }
 435
 436                         if (!strncmp(long_option[option_index].name,
 437                                         CMD_LINE_OPT_FILTER_TYPE,
 438                                         sizeof(CMD_LINE_OPT_FILTER_TYPE))) {
 439                                 ret = parse_num_opt(optarg, 3);
 440                                 if ((ret == -1) || (ret == 0)) {
 441                                         RTE_LOG(INFO, VHOST_CONFIG,
 442                                                 "Invalid argument for filter type [1-3]\n");
 443                                         tep_termination_usage(prgname);
 444                                         return -1;
 445                                 } else
 446                                         filter_idx = ret - 1;
 447                         }
 448
 449                         /* Enable/disable encapsulation on RX. */
 450                         if (!strncmp(long_option[option_index].name,
 451                                 CMD_LINE_OPT_DECAP,
 452                                 sizeof(CMD_LINE_OPT_DECAP))) {
 453                                 ret = parse_num_opt(optarg, 1);
 454                                 if (ret == -1) {
 455                                         RTE_LOG(INFO, VHOST_CONFIG,
 456                                                 "Invalid argument for decap [0|1]\n");
 457                                         tep_termination_usage(prgname);
 458                                         return -1;
 459                                 } else
 460                                         rx_decap = ret;
 461                         }
 462
 463                         /* Enable/disable encapsulation on TX. */
 464                         if (!strncmp(long_option[option_index].name,
 465                                 CMD_LINE_OPT_ENCAP,
 466                                 sizeof(CMD_LINE_OPT_ENCAP))) {
 467                                 ret = parse_num_opt(optarg, 1);
 468                                 if (ret == -1) {
 469                                         RTE_LOG(INFO, VHOST_CONFIG,
 470                                                 "Invalid argument for encap [0|1]\n");
 471                                         tep_termination_usage(prgname);
 472                                         return -1;
 473                                 } else
 474                                         tx_encap = ret;
 475                         }
 476
 477                         /* Enable/disable stats. */
 478                         if (!strncmp(long_option[option_index].name,
 479                                 CMD_LINE_OPT_STATS,
 480                                 sizeof(CMD_LINE_OPT_STATS))) {
 481                                 ret = parse_num_opt(optarg, INT32_MAX);
 482                                 if (ret == -1) {
 483                                         RTE_LOG(INFO, VHOST_CONFIG,
 484                                                         "Invalid argument for stats [0..N]\n");
 485                                         tep_termination_usage(prgname);
 486                                         return -1;
 487                                 } else
 488                                         enable_stats = ret;
 489                         }
 490
 491                         /* Set character device basename. */
 492                         if (!strncmp(long_option[option_index].name,
 493                                 CMD_LINE_OPT_DEV_BASENAME,
 494                                 sizeof(CMD_LINE_OPT_DEV_BASENAME))) {
 495                                 if (us_vhost_parse_basename(optarg) == -1) {
 496                                         RTE_LOG(INFO, VHOST_CONFIG,
 497                                                 "Invalid argument for character "
 498                                                 "device basename (Max %d characters)\n",
 499                                                 MAX_BASENAME_SZ);
 500                                         tep_termination_usage(prgname);
 501                                         return -1;
 502                                 }
 503                         }
 504
 505                         break;
 506
 507                         /* Invalid option - print options. */
 508                 default:
 509                         tep_termination_usage(prgname);
 510                         return -1;
 511                 }
 512         }
 513
 514         for (i = 0; i < RTE_MAX_ETHPORTS; i++) {
 515                 if (enabled_port_mask & (1 << i))
 516                         ports[nb_ports++] = (uint8_t)i;
 517         }
 518
 519         if ((nb_ports ==  0) || (nb_ports > MAX_SUP_PORTS)) {
 520                 RTE_LOG(INFO, VHOST_PORT, "Current enabled port number is %u,"
 521                         "but only %u port can be enabled\n", nb_ports,
 522                         MAX_SUP_PORTS);
 523                 return -1;
 524         }
 525
 526         return 0;
 527 }
 528
 529 /**
 530  * Update the global var NB_PORTS and array PORTS
 531  * according to system ports number and return valid ports number
 532  */
 533 static unsigned
 534 check_ports_num(unsigned max_nb_ports)
 535 {
 536         unsigned valid_nb_ports = nb_ports;
 537         unsigned portid;
 538
 539         if (nb_ports > max_nb_ports) {
 540                 RTE_LOG(INFO, VHOST_PORT, "\nSpecified port number(%u) "
 541                         " exceeds total system port number(%u)\n",
 542                         nb_ports, max_nb_ports);
 543                 nb_ports = max_nb_ports;
 544         }
 545
 546         for (portid = 0; portid < nb_ports; portid++) {
 547                 if (ports[portid] >= max_nb_ports) {
 548                         RTE_LOG(INFO, VHOST_PORT,
 549                                 "\nSpecified port ID(%u) exceeds max "
 550                                 " system port ID(%u)\n",
 551                                 ports[portid], (max_nb_ports - 1));
 552                         ports[portid] = INVALID_PORT_ID;
 553                         valid_nb_ports--;
 554                 }
 555         }
 556         return valid_nb_ports;
 557 }
 558
 559 /**
 560  * This function routes the TX packet to the correct interface. This may be a local device
 561  * or the physical port.
 562  */
 563 static __rte_always_inline void
 564 virtio_tx_route(struct vhost_dev *vdev, struct rte_mbuf *m)
 565 {
 566         struct mbuf_table *tx_q;
 567         struct rte_mbuf **m_table;
 568         unsigned len, ret = 0;
 569         const uint16_t lcore_id = rte_lcore_id();
 570
 571         RTE_LOG_DP(DEBUG, VHOST_DATA, "(%d) TX: MAC address is external\n",
 572                 vdev->vid);
 573
 574         /* Add packet to the port tx queue */
 575         tx_q = &lcore_tx_queue[lcore_id];
 576         len = tx_q->len;
 577
 578         tx_q->m_table[len] = m;
 579         len++;
 580         if (enable_stats) {
 581                 dev_statistics[vdev->vid].tx_total++;
 582                 dev_statistics[vdev->vid].tx++;
 583         }
 584
 585         if (unlikely(len == MAX_PKT_BURST)) {
 586                 m_table = (struct rte_mbuf **)tx_q->m_table;
 587                 ret = overlay_options.tx_handle(ports[0],
 588                         (uint16_t)tx_q->txq_id, m_table,
 589                         (uint16_t)tx_q->len);
 590
 591                 /* Free any buffers not handled by TX and update
 592                  * the port stats.
 593                  */
 594                 if (unlikely(ret < len)) {
 595                         do {
 596                                 rte_pktmbuf_free(m_table[ret]);
 597                         } while (++ret < len);
 598                 }
 599
 600                 len = 0;
 601         }
 602
 603         tx_q->len = len;
 604         return;
 605 }
 606
 607 /**
 608  * This function is called by each data core. It handles all
 609  * RX/TX registered with the core. For TX the specific lcore
 610  * linked list is used. For RX, MAC addresses are compared
 611  * with all devices in the main linked list.
 612  */
 613 static int
 614 switch_worker(__rte_unused void *arg)
 615 {
 616         struct rte_mempool *mbuf_pool = arg;
 617         struct vhost_dev *vdev = NULL;
 618         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
 619         struct virtio_net_data_ll *dev_ll;
 620         struct mbuf_table *tx_q;
 621         volatile struct lcore_ll_info *lcore_ll;
 622         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1)
 623                                         / US_PER_S * BURST_TX_DRAIN_US;
 624         uint64_t prev_tsc, diff_tsc, cur_tsc, ret_count = 0;
 625         unsigned i, ret = 0;
 626         const uint16_t lcore_id = rte_lcore_id();
 627         const uint16_t num_cores = (uint16_t)rte_lcore_count();
 628         uint16_t rx_count = 0;
 629         uint16_t tx_count;
 630         uint32_t retry = 0;
 631
 632         RTE_LOG(INFO, VHOST_DATA, "Procesing on Core %u started\n", lcore_id);
 633         lcore_ll = lcore_info[lcore_id].lcore_ll;
 634         prev_tsc = 0;
 635
 636         tx_q = &lcore_tx_queue[lcore_id];
 637         for (i = 0; i < num_cores; i++) {
 638                 if (lcore_ids[i] == lcore_id) {
 639                         tx_q->txq_id = i;
 640                         break;
 641                 }
 642         }
 643
 644         while (1) {
 645                 cur_tsc = rte_rdtsc();
 646                 /*
 647                  * TX burst queue drain
 648                  */
 649                 diff_tsc = cur_tsc - prev_tsc;
 650                 if (unlikely(diff_tsc > drain_tsc)) {
 651
 652                         if (tx_q->len) {
 653                                 RTE_LOG_DP(DEBUG, VHOST_DATA, "TX queue drained after "
 654                                         "timeout with burst size %u\n",
 655                                         tx_q->len);
 656                                 ret = overlay_options.tx_handle(ports[0],
 657                                         (uint16_t)tx_q->txq_id,
 658                                         (struct rte_mbuf **)tx_q->m_table,
 659                                         (uint16_t)tx_q->len);
 660                                 if (unlikely(ret < tx_q->len)) {
 661                                         do {
 662                                                 rte_pktmbuf_free(tx_q->m_table[ret]);
 663                                         } while (++ret < tx_q->len);
 664                                 }
 665
 666                                 tx_q->len = 0;
 667                         }
 668
 669                         prev_tsc = cur_tsc;
 670
 671                 }
 672
 673                 rte_prefetch0(lcore_ll->ll_root_used);
 674
 675                 /**
 676                  * Inform the configuration core that we have exited
 677                  * the linked list and that no devices are
 678                  * in use if requested.
 679                  */
 680                 if (lcore_ll->dev_removal_flag == REQUEST_DEV_REMOVAL)
 681                         lcore_ll->dev_removal_flag = ACK_DEV_REMOVAL;
 682
 683                 /*
 684                  * Process devices
 685                  */
 686                 dev_ll = lcore_ll->ll_root_used;
 687
 688                 while (dev_ll != NULL) {
 689                         vdev = dev_ll->vdev;
 690
 691                         if (unlikely(vdev->remove)) {
 692                                 dev_ll = dev_ll->next;
 693                                 overlay_options.tunnel_destroy(vdev);
 694                                 vdev->ready = DEVICE_SAFE_REMOVE;
 695                                 continue;
 696                         }
 697                         if (likely(vdev->ready == DEVICE_RX)) {
 698                                 /* Handle guest RX */
 699                                 rx_count = rte_eth_rx_burst(ports[0],
 700                                         vdev->rx_q, pkts_burst, MAX_PKT_BURST);
 701
 702                                 if (rx_count) {
 703                                         /*
 704                                         * Retry is enabled and the queue is
 705                                         * full then we wait and retry to
 706                                         * avoid packet loss. Here MAX_PKT_BURST
 707                                         * must be less than virtio queue size
 708                                         */
 709                                         if (enable_retry && unlikely(rx_count >
 710                                                 rte_vhost_avail_entries(vdev->vid, VIRTIO_RXQ))) {
 711                                                 for (retry = 0; retry < burst_rx_retry_num;
 712                                                         retry++) {
 713                                                         rte_delay_us(burst_rx_delay_time);
 714                                                         if (rx_count <= rte_vhost_avail_entries(vdev->vid, VIRTIO_RXQ))
 715                                                                 break;
 716                                                 }
 717                                         }
 718
 719                                         ret_count = overlay_options.rx_handle(vdev->vid, pkts_burst, rx_count);
 720                                         if (enable_stats) {
 721                                                 rte_atomic64_add(
 722                                                 &dev_statistics[vdev->vid].rx_total_atomic,
 723                                                 rx_count);
 724                                                 rte_atomic64_add(
 725                                                 &dev_statistics[vdev->vid].rx_atomic, ret_count);
 726                                         }
 727                                         while (likely(rx_count)) {
 728                                                 rx_count--;
 729                                                 rte_pktmbuf_free(pkts_burst[rx_count]);
 730                                         }
 731
 732                                 }
 733                         }
 734
 735                         if (likely(!vdev->remove)) {
 736                                 /* Handle guest TX*/
 737                                 tx_count = rte_vhost_dequeue_burst(vdev->vid,
 738                                                 VIRTIO_TXQ, mbuf_pool,
 739                                                 pkts_burst, MAX_PKT_BURST);
 740                                 /* If this is the first received packet we need to learn the MAC */
 741                                 if (unlikely(vdev->ready == DEVICE_MAC_LEARNING) && tx_count) {
 742                                         if (vdev->remove ||
 743                                                 (overlay_options.tunnel_setup(vdev, pkts_burst[0]) == -1)) {
 744                                                 while (tx_count)
 745                                                         rte_pktmbuf_free(pkts_burst[--tx_count]);
 746                                         }
 747                                 }
 748                                 while (tx_count)
 749                                         virtio_tx_route(vdev, pkts_burst[--tx_count]);
 750                         }
 751
 752                         /* move to the next device in the list */
 753                         dev_ll = dev_ll->next;
 754                 }
 755         }
 756
 757         return 0;
 758 }
 759
 760 /**
 761  * Add an entry to a used linked list. A free entry must first be found
 762  * in the free linked list using get_data_ll_free_entry();
 763  */
 764 static void
 765 add_data_ll_entry(struct virtio_net_data_ll **ll_root_addr,
 766         struct virtio_net_data_ll *ll_dev)
 767 {
 768         struct virtio_net_data_ll *ll = *ll_root_addr;
 769
 770         /* Set next as NULL and use a compiler barrier to avoid reordering. */
 771         ll_dev->next = NULL;
 772         rte_compiler_barrier();
 773
 774         /* If ll == NULL then this is the first device. */
 775         if (ll) {
 776                 /* Increment to the tail of the linked list. */
 777                 while (ll->next != NULL)
 778                         ll = ll->next;
 779
 780                 ll->next = ll_dev;
 781         } else {
 782                 *ll_root_addr = ll_dev;
 783         }
 784 }
 785
 786 /**
 787  * Remove an entry from a used linked list. The entry must then be added to
 788  * the free linked list using put_data_ll_free_entry().
 789  */
 790 static void
 791 rm_data_ll_entry(struct virtio_net_data_ll **ll_root_addr,
 792         struct virtio_net_data_ll *ll_dev,
 793         struct virtio_net_data_ll *ll_dev_last)
 794 {
 795         struct virtio_net_data_ll *ll = *ll_root_addr;
 796
 797         if (unlikely((ll == NULL) || (ll_dev == NULL)))
 798                 return;
 799
 800         if (ll_dev == ll)
 801                 *ll_root_addr = ll_dev->next;
 802         else
 803                 if (likely(ll_dev_last != NULL))
 804                         ll_dev_last->next = ll_dev->next;
 805                 else
 806                         RTE_LOG(ERR, VHOST_CONFIG,
 807                                 "Remove entry form ll failed.\n");
 808 }
 809
 810 /**
 811  * Find and return an entry from the free linked list.
 812  */
 813 static struct virtio_net_data_ll *
 814 get_data_ll_free_entry(struct virtio_net_data_ll **ll_root_addr)
 815 {
 816         struct virtio_net_data_ll *ll_free = *ll_root_addr;
 817         struct virtio_net_data_ll *ll_dev;
 818
 819         if (ll_free == NULL)
 820                 return NULL;
 821
 822         ll_dev = ll_free;
 823         *ll_root_addr = ll_free->next;
 824
 825         return ll_dev;
 826 }
 827
 828 /**
 829  * Place an entry back on to the free linked list.
 830  */
 831 static void
 832 put_data_ll_free_entry(struct virtio_net_data_ll **ll_root_addr,
 833         struct virtio_net_data_ll *ll_dev)
 834 {
 835         struct virtio_net_data_ll *ll_free = *ll_root_addr;
 836
 837         if (ll_dev == NULL)
 838                 return;
 839
 840         ll_dev->next = ll_free;
 841         *ll_root_addr = ll_dev;
 842 }
 843
 844 /**
 845  * Creates a linked list of a given size.
 846  */
 847 static struct virtio_net_data_ll *
 848 alloc_data_ll(uint32_t size)
 849 {
 850         struct virtio_net_data_ll *ll_new;
 851         uint32_t i;
 852
 853         /* Malloc and then chain the linked list. */
 854         ll_new = malloc(size * sizeof(struct virtio_net_data_ll));
 855         if (ll_new == NULL) {
 856                 RTE_LOG(ERR, VHOST_CONFIG,
 857                         "Failed to allocate memory for ll_new.\n");
 858                 return NULL;
 859         }
 860
 861         for (i = 0; i < size - 1; i++) {
 862                 ll_new[i].vdev = NULL;
 863                 ll_new[i].next = &ll_new[i+1];
 864         }
 865         ll_new[i].next = NULL;
 866
 867         return ll_new;
 868 }
 869
 870 /**
 871  * Create the main linked list along with each individual cores
 872  * linked list. A used and a free list are created to manage entries.
 873  */
 874 static int
 875 init_data_ll(void)
 876 {
 877         int lcore;
 878
 879         RTE_LCORE_FOREACH_SLAVE(lcore) {
 880                 lcore_info[lcore].lcore_ll =
 881                         malloc(sizeof(struct lcore_ll_info));
 882                 if (lcore_info[lcore].lcore_ll == NULL) {
 883                         RTE_LOG(ERR, VHOST_CONFIG,
 884                                 "Failed to allocate memory for lcore_ll.\n");
 885                         return -1;
 886                 }
 887
 888                 lcore_info[lcore].lcore_ll->device_num = 0;
 889                 lcore_info[lcore].lcore_ll->dev_removal_flag = ACK_DEV_REMOVAL;
 890                 lcore_info[lcore].lcore_ll->ll_root_used = NULL;
 891                 if (nb_devices % nb_switching_cores)
 892                         lcore_info[lcore].lcore_ll->ll_root_free =
 893                                 alloc_data_ll((nb_devices / nb_switching_cores)
 894                                                 + 1);
 895                 else
 896                         lcore_info[lcore].lcore_ll->ll_root_free =
 897                                 alloc_data_ll(nb_devices / nb_switching_cores);
 898         }
 899
 900         /* Allocate devices up to a maximum of MAX_DEVICES. */
 901         ll_root_free = alloc_data_ll(MIN((nb_devices), MAX_DEVICES));
 902
 903         return 0;
 904 }
 905
 906 /**
 907  * Remove a device from the specific data core linked list and
 908  * from the main linked list. Synchonization occurs through the use
 909  * of the lcore dev_removal_flag.
 910  */
 911 static void
 912 destroy_device(int vid)
 913 {
 914         struct virtio_net_data_ll *ll_lcore_dev_cur;
 915         struct virtio_net_data_ll *ll_main_dev_cur;
 916         struct virtio_net_data_ll *ll_lcore_dev_last = NULL;
 917         struct virtio_net_data_ll *ll_main_dev_last = NULL;
 918         struct vhost_dev *vdev = NULL;
 919         int lcore;
 920
 921         ll_main_dev_cur = ll_root_used;
 922         while (ll_main_dev_cur != NULL) {
 923                 if (ll_main_dev_cur->vdev->vid == vid) {
 924                         vdev = ll_main_dev_cur->vdev;
 925                         break;
 926                 }
 927         }
 928         if (!vdev)
 929                 return;
 930
 931         /* set the remove flag. */
 932         vdev->remove = 1;
 933         while (vdev->ready != DEVICE_SAFE_REMOVE)
 934                 rte_pause();
 935
 936         /* Search for entry to be removed from lcore ll */
 937         ll_lcore_dev_cur = lcore_info[vdev->coreid].lcore_ll->ll_root_used;
 938         while (ll_lcore_dev_cur != NULL) {
 939                 if (ll_lcore_dev_cur->vdev == vdev) {
 940                         break;
 941                 } else {
 942                         ll_lcore_dev_last = ll_lcore_dev_cur;
 943                         ll_lcore_dev_cur = ll_lcore_dev_cur->next;
 944                 }
 945         }
 946
 947         if (ll_lcore_dev_cur == NULL) {
 948                 RTE_LOG(ERR, VHOST_CONFIG,
 949                         "(%d) Failed to find the dev to be destroy.\n", vid);
 950                 return;
 951         }
 952
 953         /* Search for entry to be removed from main ll */
 954         ll_main_dev_cur = ll_root_used;
 955         ll_main_dev_last = NULL;
 956         while (ll_main_dev_cur != NULL) {
 957                 if (ll_main_dev_cur->vdev == vdev) {
 958                         break;
 959                 } else {
 960                         ll_main_dev_last = ll_main_dev_cur;
 961                         ll_main_dev_cur = ll_main_dev_cur->next;
 962                 }
 963         }
 964
 965         /* Remove entries from the lcore and main ll. */
 966         rm_data_ll_entry(&lcore_info[vdev->coreid].lcore_ll->ll_root_used,
 967                         ll_lcore_dev_cur, ll_lcore_dev_last);
 968         rm_data_ll_entry(&ll_root_used, ll_main_dev_cur, ll_main_dev_last);
 969
 970         /* Set the dev_removal_flag on each lcore. */
 971         RTE_LCORE_FOREACH_SLAVE(lcore) {
 972                 lcore_info[lcore].lcore_ll->dev_removal_flag =
 973                         REQUEST_DEV_REMOVAL;
 974         }
 975
 976         /*
 977          * Once each core has set the dev_removal_flag to
 978          * ACK_DEV_REMOVAL we can be sure that they can no longer access
 979          * the device removed from the linked lists and that the devices
 980          * are no longer in use.
 981          */
 982         RTE_LCORE_FOREACH_SLAVE(lcore) {
 983                 while (lcore_info[lcore].lcore_ll->dev_removal_flag
 984                         != ACK_DEV_REMOVAL)
 985                         rte_pause();
 986         }
 987
 988         /* Add the entries back to the lcore and main free ll.*/
 989         put_data_ll_free_entry(&lcore_info[vdev->coreid].lcore_ll->ll_root_free,
 990                                 ll_lcore_dev_cur);
 991         put_data_ll_free_entry(&ll_root_free, ll_main_dev_cur);
 992
 993         /* Decrement number of device on the lcore. */
 994         lcore_info[vdev->coreid].lcore_ll->device_num--;
 995
 996         RTE_LOG(INFO, VHOST_DATA, "(%d) Device has been removed "
 997                 "from data core\n", vid);
 998
 999         rte_free(vdev);
1000
1001 }
1002
1003 /**
1004  * A new device is added to a data core. First the device is added
1005  * to the main linked list and the allocated to a specific data core.
1006  */
1007 static int
1008 new_device(int vid)
1009 {
1010         struct virtio_net_data_ll *ll_dev;
1011         int lcore, core_add = 0;
1012         uint32_t device_num_min = nb_devices;
1013         struct vhost_dev *vdev;
1014
1015         vdev = rte_zmalloc("vhost device", sizeof(*vdev), RTE_CACHE_LINE_SIZE);
1016         if (vdev == NULL) {
1017                 RTE_LOG(INFO, VHOST_DATA,
1018                         "(%d) Couldn't allocate memory for vhost dev\n", vid);
1019                 return -1;
1020         }
1021         vdev->vid = vid;
1022         /* Add device to main ll */
1023         ll_dev = get_data_ll_free_entry(&ll_root_free);
1024         if (ll_dev == NULL) {
1025                 RTE_LOG(INFO, VHOST_DATA, "(%d) No free entry found in"
1026                         " linked list Device limit of %d devices per core"
1027                         " has been reached\n", vid, nb_devices);
1028                 if (vdev->regions_hpa)
1029                         rte_free(vdev->regions_hpa);
1030                 rte_free(vdev);
1031                 return -1;
1032         }
1033         ll_dev->vdev = vdev;
1034         add_data_ll_entry(&ll_root_used, ll_dev);
1035         vdev->rx_q = vid;
1036
1037         /* reset ready flag */
1038         vdev->ready = DEVICE_MAC_LEARNING;
1039         vdev->remove = 0;
1040
1041         /* Find a suitable lcore to add the device. */
1042         RTE_LCORE_FOREACH_SLAVE(lcore) {
1043                 if (lcore_info[lcore].lcore_ll->device_num < device_num_min) {
1044                         device_num_min = lcore_info[lcore].lcore_ll->device_num;
1045                         core_add = lcore;
1046                 }
1047         }
1048         /* Add device to lcore ll */
1049         ll_dev = get_data_ll_free_entry(&lcore_info[core_add].lcore_ll->ll_root_free);
1050         if (ll_dev == NULL) {
1051                 RTE_LOG(INFO, VHOST_DATA,
1052                         "(%d) Failed to add device to data core\n",
1053                         vid);
1054                 vdev->ready = DEVICE_SAFE_REMOVE;
1055                 destroy_device(vid);
1056                 rte_free(vdev->regions_hpa);
1057                 rte_free(vdev);
1058                 return -1;
1059         }
1060         ll_dev->vdev = vdev;
1061         vdev->coreid = core_add;
1062
1063         add_data_ll_entry(&lcore_info[vdev->coreid].lcore_ll->ll_root_used,
1064                         ll_dev);
1065
1066         /* Initialize device stats */
1067         memset(&dev_statistics[vid], 0,
1068                 sizeof(struct device_statistics));
1069
1070         /* Disable notifications. */
1071         rte_vhost_enable_guest_notification(vid, VIRTIO_RXQ, 0);
1072         rte_vhost_enable_guest_notification(vid, VIRTIO_TXQ, 0);
1073         lcore_info[vdev->coreid].lcore_ll->device_num++;
1074
1075         RTE_LOG(INFO, VHOST_DATA, "(%d) Device has been added to data core %d\n",
1076                 vid, vdev->coreid);
1077
1078         return 0;
1079 }
1080
1081 /**
1082  * These callback allow devices to be added to the data core when configuration
1083  * has been fully complete.
1084  */
1085 static const struct vhost_device_ops virtio_net_device_ops = {
1086         .new_device =  new_device,
1087         .destroy_device = destroy_device,
1088 };
1089
1090 /**
1091  * This is a thread will wake up after a period to print stats if the user has
1092  * enabled them.
1093  */
1094 static void
1095 print_stats(void)
1096 {
1097         struct virtio_net_data_ll *dev_ll;
1098         uint64_t tx_dropped, rx_dropped;
1099         uint64_t tx, tx_total, rx, rx_total, rx_ip_csum, rx_l4_csum;
1100         int vid;
1101         const char clr[] = { 27, '[', '2', 'J', '\0' };
1102         const char top_left[] = { 27, '[', '1', ';', '1', 'H', '\0' };
1103
1104         while (1) {
1105                 sleep(enable_stats);
1106
1107                 /* Clear screen and move to top left */
1108                 printf("%s%s", clr, top_left);
1109
1110                 printf("\nDevice statistics ================================");
1111
1112                 dev_ll = ll_root_used;
1113                 while (dev_ll != NULL) {
1114                         vid = dev_ll->vdev->vid;
1115                         tx_total = dev_statistics[vid].tx_total;
1116                         tx = dev_statistics[vid].tx;
1117                         tx_dropped = tx_total - tx;
1118
1119                         rx_total = rte_atomic64_read(
1120                                 &dev_statistics[vid].rx_total_atomic);
1121                         rx = rte_atomic64_read(
1122                                 &dev_statistics[vid].rx_atomic);
1123                         rx_dropped = rx_total - rx;
1124                         rx_ip_csum = rte_atomic64_read(
1125                                 &dev_statistics[vid].rx_bad_ip_csum);
1126                         rx_l4_csum = rte_atomic64_read(
1127                                 &dev_statistics[vid].rx_bad_l4_csum);
1128
1129                         printf("\nStatistics for device %d ----------"
1130                                         "\nTX total:            %"PRIu64""
1131                                         "\nTX dropped:          %"PRIu64""
1132                                         "\nTX successful:               %"PRIu64""
1133                                         "\nRX total:            %"PRIu64""
1134                                         "\nRX bad IP csum:      %"PRIu64""
1135                                         "\nRX bad L4 csum:      %"PRIu64""
1136                                         "\nRX dropped:          %"PRIu64""
1137                                         "\nRX successful:               %"PRIu64"",
1138                                         vid,
1139                                         tx_total,
1140                                         tx_dropped,
1141                                         tx,
1142                                         rx_total,
1143                                         rx_ip_csum,
1144                                         rx_l4_csum,
1145                                         rx_dropped,
1146                                         rx);
1147
1148                         dev_ll = dev_ll->next;
1149                 }
1150                 printf("\n================================================\n");
1151         }
1152 }
1153
1154 /**
1155  * Main function, does initialisation and calls the per-lcore functions.
1156  */
1157 int
1158 main(int argc, char *argv[])
1159 {
1160         struct rte_mempool *mbuf_pool = NULL;
1161         unsigned lcore_id, core_id = 0;
1162         unsigned nb_ports, valid_nb_ports;
1163         int ret;
1164         uint16_t portid;
1165         uint16_t queue_id;
1166         static pthread_t tid;
1167         char thread_name[RTE_MAX_THREAD_NAME_LEN];
1168
1169         /* init EAL */
1170         ret = rte_eal_init(argc, argv);
1171         if (ret < 0)
1172                 rte_exit(EXIT_FAILURE, "Error with EAL initialization\n");
1173         argc -= ret;
1174         argv += ret;
1175
1176         /* parse app arguments */
1177         ret = tep_termination_parse_args(argc, argv);
1178         if (ret < 0)
1179                 rte_exit(EXIT_FAILURE, "Invalid argument\n");
1180
1181         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++)
1182                 if (rte_lcore_is_enabled(lcore_id))
1183                         lcore_ids[core_id++] = lcore_id;
1184
1185         /* set the number of swithcing cores available */
1186         nb_switching_cores = rte_lcore_count()-1;
1187
1188         /* Get the number of physical ports. */
1189         nb_ports = rte_eth_dev_count();
1190
1191         /*
1192          * Update the global var NB_PORTS and global array PORTS
1193          * and get value of var VALID_NB_PORTS according to system ports number
1194          */
1195         valid_nb_ports = check_ports_num(nb_ports);
1196
1197         if ((valid_nb_ports == 0) || (valid_nb_ports > MAX_SUP_PORTS)) {
1198                 rte_exit(EXIT_FAILURE, "Current enabled port number is %u,"
1199                         "but only %u port can be enabled\n", nb_ports,
1200                         MAX_SUP_PORTS);
1201         }
1202         /* Create the mbuf pool. */
1203         mbuf_pool = rte_pktmbuf_pool_create(
1204                         "MBUF_POOL",
1205                         NUM_MBUFS_PER_PORT * valid_nb_ports,
1206                         MBUF_CACHE_SIZE,
1207                         0,
1208                         MBUF_DATA_SIZE,
1209                         rte_socket_id());
1210         if (mbuf_pool == NULL)
1211                 rte_exit(EXIT_FAILURE, "Cannot create mbuf pool\n");
1212
1213         for (queue_id = 0; queue_id < MAX_QUEUES + 1; queue_id++)
1214                 vpool_array[queue_id].pool = mbuf_pool;
1215
1216         /* initialize all ports */
1217         for (portid = 0; portid < nb_ports; portid++) {
1218                 /* skip ports that are not enabled */
1219                 if ((enabled_port_mask & (1 << portid)) == 0) {
1220                         RTE_LOG(INFO, VHOST_PORT,
1221                                 "Skipping disabled port %d\n", portid);
1222                         continue;
1223                 }
1224                 if (overlay_options.port_configure(portid, mbuf_pool) != 0)
1225                         rte_exit(EXIT_FAILURE,
1226                                 "Cannot initialize network ports\n");
1227         }
1228
1229         /* Initialise all linked lists. */
1230         if (init_data_ll() == -1)
1231                 rte_exit(EXIT_FAILURE, "Failed to initialize linked list\n");
1232
1233         /* Initialize device stats */
1234         memset(&dev_statistics, 0, sizeof(dev_statistics));
1235
1236         /* Enable stats if the user option is set. */
1237         if (enable_stats) {
1238                 ret = pthread_create(&tid, NULL, (void *)print_stats, NULL);
1239                 if (ret != 0)
1240                         rte_exit(EXIT_FAILURE, "Cannot create print-stats thread\n");
1241                 snprintf(thread_name, RTE_MAX_THREAD_NAME_LEN, "print-stats");
1242                 ret = rte_thread_setname(tid, thread_name);
1243                 if (ret != 0)
1244                         RTE_LOG(DEBUG, VHOST_CONFIG, "Cannot set print-stats name\n");
1245         }
1246
1247         /* Launch all data cores. */
1248         RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1249                 rte_eal_remote_launch(switch_worker,
1250                         mbuf_pool, lcore_id);
1251         }
1252
1253         ret = rte_vhost_driver_register((char *)&dev_basename, 0);
1254         if (ret != 0)
1255                 rte_exit(EXIT_FAILURE, "failed to register vhost driver.\n");
1256
1257         rte_vhost_driver_disable_features(dev_basename,
1258                 1ULL << VIRTIO_NET_F_MRG_RXBUF);
1259
1260         ret = rte_vhost_driver_callback_register(dev_basename,
1261                 &virtio_net_device_ops);
1262         if (ret != 0) {
1263                 rte_exit(EXIT_FAILURE,
1264                         "failed to register vhost driver callbacks.\n");
1265         }
1266
1267         if (rte_vhost_driver_start(dev_basename) < 0) {
1268                 rte_exit(EXIT_FAILURE,
1269                         "failed to start vhost driver.\n");
1270         }
1271
1272         RTE_LCORE_FOREACH_SLAVE(lcore_id)
1273                 rte_eal_wait_lcore(lcore_id);
1274
1275         return 0;
1276 }