examples/vhost/main.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2010-2017 Intel Corporation
   3  */
   4
   5 #include <arpa/inet.h>
   6 #include <getopt.h>
   7 #include <linux/if_ether.h>
   8 #include <linux/if_vlan.h>
   9 #include <linux/virtio_net.h>
  10 #include <linux/virtio_ring.h>
  11 #include <signal.h>
  12 #include <stdint.h>
  13 #include <sys/eventfd.h>
  14 #include <sys/param.h>
  15 #include <unistd.h>
  16
  17 #include <rte_cycles.h>
  18 #include <rte_ethdev.h>
  19 #include <rte_log.h>
  20 #include <rte_string_fns.h>
  21 #include <rte_malloc.h>
  22 #include <rte_net.h>
  23 #include <rte_vhost.h>
  24 #include <rte_ip.h>
  25 #include <rte_tcp.h>
  26 #include <rte_pause.h>
  27 #include <rte_dmadev.h>
  28 #include <rte_vhost_async.h>
  29
  30 #include "main.h"
  31
  32 #ifndef MAX_QUEUES
  33 #define MAX_QUEUES 128
  34 #endif
  35
  36 #define NUM_MBUFS_DEFAULT 0x24000
  37
  38 /* the maximum number of external ports supported */
  39 #define MAX_SUP_PORTS 1
  40
  41 #define MBUF_CACHE_SIZE 128
  42 #define MBUF_DATA_SIZE  RTE_MBUF_DEFAULT_BUF_SIZE
  43
  44 #define BURST_TX_DRAIN_US 100   /* TX drain every ~100us */
  45
  46 #define BURST_RX_WAIT_US 15     /* Defines how long we wait between retries on RX */
  47 #define BURST_RX_RETRIES 4              /* Number of retries on RX. */
  48
  49 #define JUMBO_FRAME_MAX_SIZE    0x2600
  50 #define MAX_MTU (JUMBO_FRAME_MAX_SIZE - (RTE_ETHER_HDR_LEN + RTE_ETHER_CRC_LEN))
  51
  52 /* State of virtio device. */
  53 #define DEVICE_MAC_LEARNING 0
  54 #define DEVICE_RX                       1
  55 #define DEVICE_SAFE_REMOVE      2
  56
  57 /* Configurable number of RX/TX ring descriptors */
  58 #define RTE_TEST_RX_DESC_DEFAULT 1024
  59 #define RTE_TEST_TX_DESC_DEFAULT 512
  60
  61 #define INVALID_PORT_ID 0xFF
  62 #define INVALID_DMA_ID -1
  63
  64 #define DMA_RING_SIZE 4096
  65
  66 /* number of mbufs in all pools - if specified on command-line. */
  67 static int total_num_mbufs = NUM_MBUFS_DEFAULT;
  68
  69 struct dma_for_vhost dma_bind[RTE_MAX_VHOST_DEVICE];
  70 int16_t dmas_id[RTE_DMADEV_DEFAULT_MAX];
  71 static int dma_count;
  72
  73 /* mask of enabled ports */
  74 static uint32_t enabled_port_mask = 0;
  75
  76 /* Promiscuous mode */
  77 static uint32_t promiscuous;
  78
  79 /* number of devices/queues to support*/
  80 static uint32_t num_queues = 0;
  81 static uint32_t num_devices;
  82
  83 static struct rte_mempool *mbuf_pool;
  84 static int mergeable;
  85
  86 /* Enable VM2VM communications. If this is disabled then the MAC address compare is skipped. */
  87 typedef enum {
  88         VM2VM_DISABLED = 0,
  89         VM2VM_SOFTWARE = 1,
  90         VM2VM_HARDWARE = 2,
  91         VM2VM_LAST
  92 } vm2vm_type;
  93 static vm2vm_type vm2vm_mode = VM2VM_SOFTWARE;
  94
  95 /* Enable stats. */
  96 static uint32_t enable_stats = 0;
  97 /* Enable retries on RX. */
  98 static uint32_t enable_retry = 1;
  99
 100 /* Disable TX checksum offload */
 101 static uint32_t enable_tx_csum;
 102
 103 /* Disable TSO offload */
 104 static uint32_t enable_tso;
 105
 106 static int client_mode;
 107
 108 static int builtin_net_driver;
 109
 110 /* Specify timeout (in useconds) between retries on RX. */
 111 static uint32_t burst_rx_delay_time = BURST_RX_WAIT_US;
 112 /* Specify the number of retries on RX. */
 113 static uint32_t burst_rx_retry_num = BURST_RX_RETRIES;
 114
 115 /* Socket file paths. Can be set by user */
 116 static char *socket_files;
 117 static int nb_sockets;
 118
 119 /* empty VMDq configuration structure. Filled in programmatically */
 120 static struct rte_eth_conf vmdq_conf_default = {
 121         .rxmode = {
 122                 .mq_mode        = RTE_ETH_MQ_RX_VMDQ_ONLY,
 123                 .split_hdr_size = 0,
 124                 /*
 125                  * VLAN strip is necessary for 1G NIC such as I350,
 126                  * this fixes bug of ipv4 forwarding in guest can't
 127                  * forward packets from one virtio dev to another virtio dev.
 128                  */
 129                 .offloads = RTE_ETH_RX_OFFLOAD_VLAN_STRIP,
 130         },
 131
 132         .txmode = {
 133                 .mq_mode = RTE_ETH_MQ_TX_NONE,
 134                 .offloads = (RTE_ETH_TX_OFFLOAD_IPV4_CKSUM |
 135                              RTE_ETH_TX_OFFLOAD_TCP_CKSUM |
 136                              RTE_ETH_TX_OFFLOAD_VLAN_INSERT |
 137                              RTE_ETH_TX_OFFLOAD_MULTI_SEGS |
 138                              RTE_ETH_TX_OFFLOAD_TCP_TSO),
 139         },
 140         .rx_adv_conf = {
 141                 /*
 142                  * should be overridden separately in code with
 143                  * appropriate values
 144                  */
 145                 .vmdq_rx_conf = {
 146                         .nb_queue_pools = RTE_ETH_8_POOLS,
 147                         .enable_default_pool = 0,
 148                         .default_pool = 0,
 149                         .nb_pool_maps = 0,
 150                         .pool_map = {{0, 0},},
 151                 },
 152         },
 153 };
 154
 155
 156 static unsigned lcore_ids[RTE_MAX_LCORE];
 157 static uint16_t ports[RTE_MAX_ETHPORTS];
 158 static unsigned num_ports = 0; /**< The number of ports specified in command line */
 159 static uint16_t num_pf_queues, num_vmdq_queues;
 160 static uint16_t vmdq_pool_base, vmdq_queue_base;
 161 static uint16_t queues_per_pool;
 162
 163 const uint16_t vlan_tags[] = {
 164         1000, 1001, 1002, 1003, 1004, 1005, 1006, 1007,
 165         1008, 1009, 1010, 1011, 1012, 1013, 1014, 1015,
 166         1016, 1017, 1018, 1019, 1020, 1021, 1022, 1023,
 167         1024, 1025, 1026, 1027, 1028, 1029, 1030, 1031,
 168         1032, 1033, 1034, 1035, 1036, 1037, 1038, 1039,
 169         1040, 1041, 1042, 1043, 1044, 1045, 1046, 1047,
 170         1048, 1049, 1050, 1051, 1052, 1053, 1054, 1055,
 171         1056, 1057, 1058, 1059, 1060, 1061, 1062, 1063,
 172 };
 173
 174 /* ethernet addresses of ports */
 175 static struct rte_ether_addr vmdq_ports_eth_addr[RTE_MAX_ETHPORTS];
 176
 177 static struct vhost_dev_tailq_list vhost_dev_list =
 178         TAILQ_HEAD_INITIALIZER(vhost_dev_list);
 179
 180 static struct lcore_info lcore_info[RTE_MAX_LCORE];
 181
 182 /* Used for queueing bursts of TX packets. */
 183 struct mbuf_table {
 184         unsigned len;
 185         unsigned txq_id;
 186         struct rte_mbuf *m_table[MAX_PKT_BURST];
 187 };
 188
 189 struct vhost_bufftable {
 190         uint32_t len;
 191         uint64_t pre_tsc;
 192         struct rte_mbuf *m_table[MAX_PKT_BURST];
 193 };
 194
 195 /* TX queue for each data core. */
 196 struct mbuf_table lcore_tx_queue[RTE_MAX_LCORE];
 197
 198 /*
 199  * Vhost TX buffer for each data core.
 200  * Every data core maintains a TX buffer for every vhost device,
 201  * which is used for batch pkts enqueue for higher performance.
 202  */
 203 struct vhost_bufftable *vhost_txbuff[RTE_MAX_LCORE * RTE_MAX_VHOST_DEVICE];
 204
 205 #define MBUF_TABLE_DRAIN_TSC    ((rte_get_tsc_hz() + US_PER_S - 1) \
 206                                  / US_PER_S * BURST_TX_DRAIN_US)
 207
 208 static inline bool
 209 is_dma_configured(int16_t dev_id)
 210 {
 211         int i;
 212
 213         for (i = 0; i < dma_count; i++)
 214                 if (dmas_id[i] == dev_id)
 215                         return true;
 216         return false;
 217 }
 218
 219 static inline int
 220 open_dma(const char *value)
 221 {
 222         struct dma_for_vhost *dma_info = dma_bind;
 223         char *input = strndup(value, strlen(value) + 1);
 224         char *addrs = input;
 225         char *ptrs[2];
 226         char *start, *end, *substr;
 227         int64_t vid;
 228
 229         struct rte_dma_info info;
 230         struct rte_dma_conf dev_config = { .nb_vchans = 1 };
 231         struct rte_dma_vchan_conf qconf = {
 232                 .direction = RTE_DMA_DIR_MEM_TO_MEM,
 233                 .nb_desc = DMA_RING_SIZE
 234         };
 235
 236         int dev_id;
 237         int ret = 0;
 238         uint16_t i = 0;
 239         char *dma_arg[RTE_MAX_VHOST_DEVICE];
 240         int args_nr;
 241
 242         while (isblank(*addrs))
 243                 addrs++;
 244         if (*addrs == '\0') {
 245                 ret = -1;
 246                 goto out;
 247         }
 248
 249         /* process DMA devices within bracket. */
 250         addrs++;
 251         substr = strtok(addrs, ";]");
 252         if (!substr) {
 253                 ret = -1;
 254                 goto out;
 255         }
 256
 257         args_nr = rte_strsplit(substr, strlen(substr), dma_arg, RTE_MAX_VHOST_DEVICE, ',');
 258         if (args_nr <= 0) {
 259                 ret = -1;
 260                 goto out;
 261         }
 262
 263         while (i < args_nr) {
 264                 char *arg_temp = dma_arg[i];
 265                 uint8_t sub_nr;
 266
 267                 sub_nr = rte_strsplit(arg_temp, strlen(arg_temp), ptrs, 2, '@');
 268                 if (sub_nr != 2) {
 269                         ret = -1;
 270                         goto out;
 271                 }
 272
 273                 start = strstr(ptrs[0], "txd");
 274                 if (start == NULL) {
 275                         ret = -1;
 276                         goto out;
 277                 }
 278
 279                 start += 3;
 280                 vid = strtol(start, &end, 0);
 281                 if (end == start) {
 282                         ret = -1;
 283                         goto out;
 284                 }
 285
 286                 dev_id = rte_dma_get_dev_id_by_name(ptrs[1]);
 287                 if (dev_id < 0) {
 288                         RTE_LOG(ERR, VHOST_CONFIG, "Fail to find DMA %s.\n", ptrs[1]);
 289                         ret = -1;
 290                         goto out;
 291                 }
 292
 293                 /* DMA device is already configured, so skip */
 294                 if (is_dma_configured(dev_id))
 295                         goto done;
 296
 297                 if (rte_dma_info_get(dev_id, &info) != 0) {
 298                         RTE_LOG(ERR, VHOST_CONFIG, "Error with rte_dma_info_get()\n");
 299                         ret = -1;
 300                         goto out;
 301                 }
 302
 303                 if (info.max_vchans < 1) {
 304                         RTE_LOG(ERR, VHOST_CONFIG, "No channels available on device %d\n", dev_id);
 305                         ret = -1;
 306                         goto out;
 307                 }
 308
 309                 if (rte_dma_configure(dev_id, &dev_config) != 0) {
 310                         RTE_LOG(ERR, VHOST_CONFIG, "Fail to configure DMA %d.\n", dev_id);
 311                         ret = -1;
 312                         goto out;
 313                 }
 314
 315                 /* Check the max desc supported by DMA device */
 316                 rte_dma_info_get(dev_id, &info);
 317                 if (info.nb_vchans != 1) {
 318                         RTE_LOG(ERR, VHOST_CONFIG, "No configured queues reported by DMA %d.\n",
 319                                         dev_id);
 320                         ret = -1;
 321                         goto out;
 322                 }
 323
 324                 qconf.nb_desc = RTE_MIN(DMA_RING_SIZE, info.max_desc);
 325
 326                 if (rte_dma_vchan_setup(dev_id, 0, &qconf) != 0) {
 327                         RTE_LOG(ERR, VHOST_CONFIG, "Fail to set up DMA %d.\n", dev_id);
 328                         ret = -1;
 329                         goto out;
 330                 }
 331
 332                 if (rte_dma_start(dev_id) != 0) {
 333                         RTE_LOG(ERR, VHOST_CONFIG, "Fail to start DMA %u.\n", dev_id);
 334                         ret = -1;
 335                         goto out;
 336                 }
 337
 338                 dmas_id[dma_count++] = dev_id;
 339
 340 done:
 341                 (dma_info + vid)->dmas[VIRTIO_RXQ].dev_id = dev_id;
 342                 i++;
 343         }
 344 out:
 345         free(input);
 346         return ret;
 347 }
 348
 349 /*
 350  * Builds up the correct configuration for VMDQ VLAN pool map
 351  * according to the pool & queue limits.
 352  */
 353 static inline int
 354 get_eth_conf(struct rte_eth_conf *eth_conf, uint32_t num_devices)
 355 {
 356         struct rte_eth_vmdq_rx_conf conf;
 357         struct rte_eth_vmdq_rx_conf *def_conf =
 358                 &vmdq_conf_default.rx_adv_conf.vmdq_rx_conf;
 359         unsigned i;
 360
 361         memset(&conf, 0, sizeof(conf));
 362         conf.nb_queue_pools = (enum rte_eth_nb_pools)num_devices;
 363         conf.nb_pool_maps = num_devices;
 364         conf.enable_loop_back = def_conf->enable_loop_back;
 365         conf.rx_mode = def_conf->rx_mode;
 366
 367         for (i = 0; i < conf.nb_pool_maps; i++) {
 368                 conf.pool_map[i].vlan_id = vlan_tags[ i ];
 369                 conf.pool_map[i].pools = (1UL << i);
 370         }
 371
 372         (void)(rte_memcpy(eth_conf, &vmdq_conf_default, sizeof(*eth_conf)));
 373         (void)(rte_memcpy(&eth_conf->rx_adv_conf.vmdq_rx_conf, &conf,
 374                    sizeof(eth_conf->rx_adv_conf.vmdq_rx_conf)));
 375         return 0;
 376 }
 377
 378 /*
 379  * Initialises a given port using global settings and with the rx buffers
 380  * coming from the mbuf_pool passed as parameter
 381  */
 382 static inline int
 383 port_init(uint16_t port)
 384 {
 385         struct rte_eth_dev_info dev_info;
 386         struct rte_eth_conf port_conf;
 387         struct rte_eth_rxconf *rxconf;
 388         struct rte_eth_txconf *txconf;
 389         int16_t rx_rings, tx_rings;
 390         uint16_t rx_ring_size, tx_ring_size;
 391         int retval;
 392         uint16_t q;
 393
 394         /* The max pool number from dev_info will be used to validate the pool number specified in cmd line */
 395         retval = rte_eth_dev_info_get(port, &dev_info);
 396         if (retval != 0) {
 397                 RTE_LOG(ERR, VHOST_PORT,
 398                         "Error during getting device (port %u) info: %s\n",
 399                         port, strerror(-retval));
 400
 401                 return retval;
 402         }
 403
 404         rxconf = &dev_info.default_rxconf;
 405         txconf = &dev_info.default_txconf;
 406         rxconf->rx_drop_en = 1;
 407
 408         /*configure the number of supported virtio devices based on VMDQ limits */
 409         num_devices = dev_info.max_vmdq_pools;
 410
 411         rx_ring_size = RTE_TEST_RX_DESC_DEFAULT;
 412         tx_ring_size = RTE_TEST_TX_DESC_DEFAULT;
 413
 414         tx_rings = (uint16_t)rte_lcore_count();
 415
 416         if (mergeable) {
 417                 if (dev_info.max_mtu != UINT16_MAX && dev_info.max_rx_pktlen > dev_info.max_mtu)
 418                         vmdq_conf_default.rxmode.mtu = dev_info.max_mtu;
 419                 else
 420                         vmdq_conf_default.rxmode.mtu = MAX_MTU;
 421         }
 422
 423         /* Get port configuration. */
 424         retval = get_eth_conf(&port_conf, num_devices);
 425         if (retval < 0)
 426                 return retval;
 427         /* NIC queues are divided into pf queues and vmdq queues.  */
 428         num_pf_queues = dev_info.max_rx_queues - dev_info.vmdq_queue_num;
 429         queues_per_pool = dev_info.vmdq_queue_num / dev_info.max_vmdq_pools;
 430         num_vmdq_queues = num_devices * queues_per_pool;
 431         num_queues = num_pf_queues + num_vmdq_queues;
 432         vmdq_queue_base = dev_info.vmdq_queue_base;
 433         vmdq_pool_base  = dev_info.vmdq_pool_base;
 434         printf("pf queue num: %u, configured vmdq pool num: %u, each vmdq pool has %u queues\n",
 435                 num_pf_queues, num_devices, queues_per_pool);
 436
 437         if (!rte_eth_dev_is_valid_port(port))
 438                 return -1;
 439
 440         rx_rings = (uint16_t)dev_info.max_rx_queues;
 441         if (dev_info.tx_offload_capa & RTE_ETH_TX_OFFLOAD_MBUF_FAST_FREE)
 442                 port_conf.txmode.offloads |=
 443                         RTE_ETH_TX_OFFLOAD_MBUF_FAST_FREE;
 444         /* Configure ethernet device. */
 445         retval = rte_eth_dev_configure(port, rx_rings, tx_rings, &port_conf);
 446         if (retval != 0) {
 447                 RTE_LOG(ERR, VHOST_PORT, "Failed to configure port %u: %s.\n",
 448                         port, strerror(-retval));
 449                 return retval;
 450         }
 451
 452         retval = rte_eth_dev_adjust_nb_rx_tx_desc(port, &rx_ring_size,
 453                 &tx_ring_size);
 454         if (retval != 0) {
 455                 RTE_LOG(ERR, VHOST_PORT, "Failed to adjust number of descriptors "
 456                         "for port %u: %s.\n", port, strerror(-retval));
 457                 return retval;
 458         }
 459         if (rx_ring_size > RTE_TEST_RX_DESC_DEFAULT) {
 460                 RTE_LOG(ERR, VHOST_PORT, "Mbuf pool has an insufficient size "
 461                         "for Rx queues on port %u.\n", port);
 462                 return -1;
 463         }
 464
 465         /* Setup the queues. */
 466         rxconf->offloads = port_conf.rxmode.offloads;
 467         for (q = 0; q < rx_rings; q ++) {
 468                 retval = rte_eth_rx_queue_setup(port, q, rx_ring_size,
 469                                                 rte_eth_dev_socket_id(port),
 470                                                 rxconf,
 471                                                 mbuf_pool);
 472                 if (retval < 0) {
 473                         RTE_LOG(ERR, VHOST_PORT,
 474                                 "Failed to setup rx queue %u of port %u: %s.\n",
 475                                 q, port, strerror(-retval));
 476                         return retval;
 477                 }
 478         }
 479         txconf->offloads = port_conf.txmode.offloads;
 480         for (q = 0; q < tx_rings; q ++) {
 481                 retval = rte_eth_tx_queue_setup(port, q, tx_ring_size,
 482                                                 rte_eth_dev_socket_id(port),
 483                                                 txconf);
 484                 if (retval < 0) {
 485                         RTE_LOG(ERR, VHOST_PORT,
 486                                 "Failed to setup tx queue %u of port %u: %s.\n",
 487                                 q, port, strerror(-retval));
 488                         return retval;
 489                 }
 490         }
 491
 492         /* Start the device. */
 493         retval  = rte_eth_dev_start(port);
 494         if (retval < 0) {
 495                 RTE_LOG(ERR, VHOST_PORT, "Failed to start port %u: %s\n",
 496                         port, strerror(-retval));
 497                 return retval;
 498         }
 499
 500         if (promiscuous) {
 501                 retval = rte_eth_promiscuous_enable(port);
 502                 if (retval != 0) {
 503                         RTE_LOG(ERR, VHOST_PORT,
 504                                 "Failed to enable promiscuous mode on port %u: %s\n",
 505                                 port, rte_strerror(-retval));
 506                         return retval;
 507                 }
 508         }
 509
 510         retval = rte_eth_macaddr_get(port, &vmdq_ports_eth_addr[port]);
 511         if (retval < 0) {
 512                 RTE_LOG(ERR, VHOST_PORT,
 513                         "Failed to get MAC address on port %u: %s\n",
 514                         port, rte_strerror(-retval));
 515                 return retval;
 516         }
 517
 518         RTE_LOG(INFO, VHOST_PORT, "Max virtio devices supported: %u\n", num_devices);
 519         RTE_LOG(INFO, VHOST_PORT, "Port %u MAC: %02"PRIx8" %02"PRIx8" %02"PRIx8
 520                 " %02"PRIx8" %02"PRIx8" %02"PRIx8"\n",
 521                 port, RTE_ETHER_ADDR_BYTES(&vmdq_ports_eth_addr[port]));
 522
 523         return 0;
 524 }
 525
 526 /*
 527  * Set socket file path.
 528  */
 529 static int
 530 us_vhost_parse_socket_path(const char *q_arg)
 531 {
 532         char *old;
 533
 534         /* parse number string */
 535         if (strnlen(q_arg, PATH_MAX) == PATH_MAX)
 536                 return -1;
 537
 538         old = socket_files;
 539         socket_files = realloc(socket_files, PATH_MAX * (nb_sockets + 1));
 540         if (socket_files == NULL) {
 541                 free(old);
 542                 return -1;
 543         }
 544
 545         strlcpy(socket_files + nb_sockets * PATH_MAX, q_arg, PATH_MAX);
 546         nb_sockets++;
 547
 548         return 0;
 549 }
 550
 551 /*
 552  * Parse the portmask provided at run time.
 553  */
 554 static int
 555 parse_portmask(const char *portmask)
 556 {
 557         char *end = NULL;
 558         unsigned long pm;
 559
 560         errno = 0;
 561
 562         /* parse hexadecimal string */
 563         pm = strtoul(portmask, &end, 16);
 564         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0') || (errno != 0))
 565                 return 0;
 566
 567         return pm;
 568
 569 }
 570
 571 /*
 572  * Parse num options at run time.
 573  */
 574 static int
 575 parse_num_opt(const char *q_arg, uint32_t max_valid_value)
 576 {
 577         char *end = NULL;
 578         unsigned long num;
 579
 580         errno = 0;
 581
 582         /* parse unsigned int string */
 583         num = strtoul(q_arg, &end, 10);
 584         if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0') || (errno != 0))
 585                 return -1;
 586
 587         if (num > max_valid_value)
 588                 return -1;
 589
 590         return num;
 591
 592 }
 593
 594 /*
 595  * Display usage
 596  */
 597 static void
 598 us_vhost_usage(const char *prgname)
 599 {
 600         RTE_LOG(INFO, VHOST_CONFIG, "%s [EAL options] -- -p PORTMASK\n"
 601         "               --vm2vm [0|1|2]\n"
 602         "               --rx_retry [0|1] --mergeable [0|1] --stats [0-N]\n"
 603         "               --socket-file <path>\n"
 604         "               --nb-devices ND\n"
 605         "               -p PORTMASK: Set mask for ports to be used by application\n"
 606         "               --vm2vm [0|1|2]: disable/software(default)/hardware vm2vm comms\n"
 607         "               --rx-retry [0|1]: disable/enable(default) retries on Rx. Enable retry if destination queue is full\n"
 608         "               --rx-retry-delay [0-N]: timeout(in usecond) between retries on RX. This makes effect only if retries on rx enabled\n"
 609         "               --rx-retry-num [0-N]: the number of retries on rx. This makes effect only if retries on rx enabled\n"
 610         "               --mergeable [0|1]: disable(default)/enable RX mergeable buffers\n"
 611         "               --stats [0-N]: 0: Disable stats, N: Time in seconds to print stats\n"
 612         "               --socket-file: The path of the socket file.\n"
 613         "               --tx-csum [0|1] disable/enable TX checksum offload.\n"
 614         "               --tso [0|1] disable/enable TCP segment offload.\n"
 615         "               --client register a vhost-user socket as client mode.\n"
 616         "               --dmas register dma channel for specific vhost device.\n"
 617         "               --total-num-mbufs [0-N] set the number of mbufs to be allocated in mbuf pools, the default value is 147456.\n",
 618                prgname);
 619 }
 620
 621 enum {
 622 #define OPT_VM2VM               "vm2vm"
 623         OPT_VM2VM_NUM = 256,
 624 #define OPT_RX_RETRY            "rx-retry"
 625         OPT_RX_RETRY_NUM,
 626 #define OPT_RX_RETRY_DELAY      "rx-retry-delay"
 627         OPT_RX_RETRY_DELAY_NUM,
 628 #define OPT_RX_RETRY_NUMB       "rx-retry-num"
 629         OPT_RX_RETRY_NUMB_NUM,
 630 #define OPT_MERGEABLE           "mergeable"
 631         OPT_MERGEABLE_NUM,
 632 #define OPT_STATS               "stats"
 633         OPT_STATS_NUM,
 634 #define OPT_SOCKET_FILE         "socket-file"
 635         OPT_SOCKET_FILE_NUM,
 636 #define OPT_TX_CSUM             "tx-csum"
 637         OPT_TX_CSUM_NUM,
 638 #define OPT_TSO                 "tso"
 639         OPT_TSO_NUM,
 640 #define OPT_CLIENT              "client"
 641         OPT_CLIENT_NUM,
 642 #define OPT_BUILTIN_NET_DRIVER  "builtin-net-driver"
 643         OPT_BUILTIN_NET_DRIVER_NUM,
 644 #define OPT_DMAS                "dmas"
 645         OPT_DMAS_NUM,
 646 #define OPT_NUM_MBUFS           "total-num-mbufs"
 647         OPT_NUM_MBUFS_NUM,
 648 };
 649
 650 /*
 651  * Parse the arguments given in the command line of the application.
 652  */
 653 static int
 654 us_vhost_parse_args(int argc, char **argv)
 655 {
 656         int opt, ret;
 657         int option_index;
 658         unsigned i;
 659         const char *prgname = argv[0];
 660         static struct option long_option[] = {
 661                 {OPT_VM2VM, required_argument,
 662                                 NULL, OPT_VM2VM_NUM},
 663                 {OPT_RX_RETRY, required_argument,
 664                                 NULL, OPT_RX_RETRY_NUM},
 665                 {OPT_RX_RETRY_DELAY, required_argument,
 666                                 NULL, OPT_RX_RETRY_DELAY_NUM},
 667                 {OPT_RX_RETRY_NUMB, required_argument,
 668                                 NULL, OPT_RX_RETRY_NUMB_NUM},
 669                 {OPT_MERGEABLE, required_argument,
 670                                 NULL, OPT_MERGEABLE_NUM},
 671                 {OPT_STATS, required_argument,
 672                                 NULL, OPT_STATS_NUM},
 673                 {OPT_SOCKET_FILE, required_argument,
 674                                 NULL, OPT_SOCKET_FILE_NUM},
 675                 {OPT_TX_CSUM, required_argument,
 676                                 NULL, OPT_TX_CSUM_NUM},
 677                 {OPT_TSO, required_argument,
 678                                 NULL, OPT_TSO_NUM},
 679                 {OPT_CLIENT, no_argument,
 680                                 NULL, OPT_CLIENT_NUM},
 681                 {OPT_BUILTIN_NET_DRIVER, no_argument,
 682                                 NULL, OPT_BUILTIN_NET_DRIVER_NUM},
 683                 {OPT_DMAS, required_argument,
 684                                 NULL, OPT_DMAS_NUM},
 685                 {OPT_NUM_MBUFS, required_argument,
 686                                 NULL, OPT_NUM_MBUFS_NUM},
 687                 {NULL, 0, 0, 0},
 688         };
 689
 690         /* Parse command line */
 691         while ((opt = getopt_long(argc, argv, "p:P",
 692                         long_option, &option_index)) != EOF) {
 693                 switch (opt) {
 694                 /* Portmask */
 695                 case 'p':
 696                         enabled_port_mask = parse_portmask(optarg);
 697                         if (enabled_port_mask == 0) {
 698                                 RTE_LOG(INFO, VHOST_CONFIG, "Invalid portmask\n");
 699                                 us_vhost_usage(prgname);
 700                                 return -1;
 701                         }
 702                         break;
 703
 704                 case 'P':
 705                         promiscuous = 1;
 706                         vmdq_conf_default.rx_adv_conf.vmdq_rx_conf.rx_mode =
 707                                 RTE_ETH_VMDQ_ACCEPT_BROADCAST |
 708                                 RTE_ETH_VMDQ_ACCEPT_MULTICAST;
 709                         break;
 710
 711                 case OPT_VM2VM_NUM:
 712                         ret = parse_num_opt(optarg, (VM2VM_LAST - 1));
 713                         if (ret == -1) {
 714                                 RTE_LOG(INFO, VHOST_CONFIG,
 715                                         "Invalid argument for "
 716                                         "vm2vm [0|1|2]\n");
 717                                 us_vhost_usage(prgname);
 718                                 return -1;
 719                         }
 720                         vm2vm_mode = (vm2vm_type)ret;
 721                         break;
 722
 723                 case OPT_RX_RETRY_NUM:
 724                         ret = parse_num_opt(optarg, 1);
 725                         if (ret == -1) {
 726                                 RTE_LOG(INFO, VHOST_CONFIG, "Invalid argument for rx-retry [0|1]\n");
 727                                 us_vhost_usage(prgname);
 728                                 return -1;
 729                         }
 730                         enable_retry = ret;
 731                         break;
 732
 733                 case OPT_TX_CSUM_NUM:
 734                         ret = parse_num_opt(optarg, 1);
 735                         if (ret == -1) {
 736                                 RTE_LOG(INFO, VHOST_CONFIG, "Invalid argument for tx-csum [0|1]\n");
 737                                 us_vhost_usage(prgname);
 738                                 return -1;
 739                         }
 740                         enable_tx_csum = ret;
 741                         break;
 742
 743                 case OPT_TSO_NUM:
 744                         ret = parse_num_opt(optarg, 1);
 745                         if (ret == -1) {
 746                                 RTE_LOG(INFO, VHOST_CONFIG, "Invalid argument for tso [0|1]\n");
 747                                 us_vhost_usage(prgname);
 748                                 return -1;
 749                         }
 750                         enable_tso = ret;
 751                         break;
 752
 753                 case OPT_RX_RETRY_DELAY_NUM:
 754                         ret = parse_num_opt(optarg, INT32_MAX);
 755                         if (ret == -1) {
 756                                 RTE_LOG(INFO, VHOST_CONFIG, "Invalid argument for rx-retry-delay [0-N]\n");
 757                                 us_vhost_usage(prgname);
 758                                 return -1;
 759                         }
 760                         burst_rx_delay_time = ret;
 761                         break;
 762
 763                 case OPT_RX_RETRY_NUMB_NUM:
 764                         ret = parse_num_opt(optarg, INT32_MAX);
 765                         if (ret == -1) {
 766                                 RTE_LOG(INFO, VHOST_CONFIG, "Invalid argument for rx-retry-num [0-N]\n");
 767                                 us_vhost_usage(prgname);
 768                                 return -1;
 769                         }
 770                         burst_rx_retry_num = ret;
 771                         break;
 772
 773                 case OPT_MERGEABLE_NUM:
 774                         ret = parse_num_opt(optarg, 1);
 775                         if (ret == -1) {
 776                                 RTE_LOG(INFO, VHOST_CONFIG, "Invalid argument for mergeable [0|1]\n");
 777                                 us_vhost_usage(prgname);
 778                                 return -1;
 779                         }
 780                         mergeable = !!ret;
 781                         break;
 782
 783                 case OPT_STATS_NUM:
 784                         ret = parse_num_opt(optarg, INT32_MAX);
 785                         if (ret == -1) {
 786                                 RTE_LOG(INFO, VHOST_CONFIG,
 787                                         "Invalid argument for stats [0..N]\n");
 788                                 us_vhost_usage(prgname);
 789                                 return -1;
 790                         }
 791                         enable_stats = ret;
 792                         break;
 793
 794                 /* Set socket file path. */
 795                 case OPT_SOCKET_FILE_NUM:
 796                         if (us_vhost_parse_socket_path(optarg) == -1) {
 797                                 RTE_LOG(INFO, VHOST_CONFIG,
 798                                 "Invalid argument for socket name (Max %d characters)\n",
 799                                 PATH_MAX);
 800                                 us_vhost_usage(prgname);
 801                                 return -1;
 802                         }
 803                         break;
 804
 805                 case OPT_DMAS_NUM:
 806                         if (open_dma(optarg) == -1) {
 807                                 RTE_LOG(INFO, VHOST_CONFIG,
 808                                         "Wrong DMA args\n");
 809                                 us_vhost_usage(prgname);
 810                                 return -1;
 811                         }
 812                         break;
 813
 814                 case OPT_NUM_MBUFS_NUM:
 815                         ret = parse_num_opt(optarg, INT32_MAX);
 816                         if (ret == -1) {
 817                                 RTE_LOG(INFO, VHOST_CONFIG,
 818                                         "Invalid argument for total-num-mbufs [0..N]\n");
 819                                 us_vhost_usage(prgname);
 820                                 return -1;
 821                         }
 822
 823                         if (total_num_mbufs < ret)
 824                                 total_num_mbufs = ret;
 825                         break;
 826
 827                 case OPT_CLIENT_NUM:
 828                         client_mode = 1;
 829                         break;
 830
 831                 case OPT_BUILTIN_NET_DRIVER_NUM:
 832                         builtin_net_driver = 1;
 833                         break;
 834
 835                 /* Invalid option - print options. */
 836                 default:
 837                         us_vhost_usage(prgname);
 838                         return -1;
 839                 }
 840         }
 841
 842         for (i = 0; i < RTE_MAX_ETHPORTS; i++) {
 843                 if (enabled_port_mask & (1 << i))
 844                         ports[num_ports++] = i;
 845         }
 846
 847         if ((num_ports ==  0) || (num_ports > MAX_SUP_PORTS)) {
 848                 RTE_LOG(INFO, VHOST_PORT, "Current enabled port number is %u,"
 849                         "but only %u port can be enabled\n",num_ports, MAX_SUP_PORTS);
 850                 return -1;
 851         }
 852
 853         return 0;
 854 }
 855
 856 /*
 857  * Update the global var NUM_PORTS and array PORTS according to system ports number
 858  * and return valid ports number
 859  */
 860 static unsigned check_ports_num(unsigned nb_ports)
 861 {
 862         unsigned valid_num_ports = num_ports;
 863         unsigned portid;
 864
 865         if (num_ports > nb_ports) {
 866                 RTE_LOG(INFO, VHOST_PORT, "\nSpecified port number(%u) exceeds total system port number(%u)\n",
 867                         num_ports, nb_ports);
 868                 num_ports = nb_ports;
 869         }
 870
 871         for (portid = 0; portid < num_ports; portid ++) {
 872                 if (!rte_eth_dev_is_valid_port(ports[portid])) {
 873                         RTE_LOG(INFO, VHOST_PORT,
 874                                 "\nSpecified port ID(%u) is not valid\n",
 875                                 ports[portid]);
 876                         ports[portid] = INVALID_PORT_ID;
 877                         valid_num_ports--;
 878                 }
 879         }
 880         return valid_num_ports;
 881 }
 882
 883 static __rte_always_inline struct vhost_dev *
 884 find_vhost_dev(struct rte_ether_addr *mac)
 885 {
 886         struct vhost_dev *vdev;
 887
 888         TAILQ_FOREACH(vdev, &vhost_dev_list, global_vdev_entry) {
 889                 if (vdev->ready == DEVICE_RX &&
 890                     rte_is_same_ether_addr(mac, &vdev->mac_address))
 891                         return vdev;
 892         }
 893
 894         return NULL;
 895 }
 896
 897 /*
 898  * This function learns the MAC address of the device and registers this along with a
 899  * vlan tag to a VMDQ.
 900  */
 901 static int
 902 link_vmdq(struct vhost_dev *vdev, struct rte_mbuf *m)
 903 {
 904         struct rte_ether_hdr *pkt_hdr;
 905         int i, ret;
 906
 907         /* Learn MAC address of guest device from packet */
 908         pkt_hdr = rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
 909
 910         if (find_vhost_dev(&pkt_hdr->src_addr)) {
 911                 RTE_LOG(ERR, VHOST_DATA,
 912                         "(%d) device is using a registered MAC!\n",
 913                         vdev->vid);
 914                 return -1;
 915         }
 916
 917         for (i = 0; i < RTE_ETHER_ADDR_LEN; i++)
 918                 vdev->mac_address.addr_bytes[i] =
 919                         pkt_hdr->src_addr.addr_bytes[i];
 920
 921         /* vlan_tag currently uses the device_id. */
 922         vdev->vlan_tag = vlan_tags[vdev->vid];
 923
 924         /* Print out VMDQ registration info. */
 925         RTE_LOG(INFO, VHOST_DATA,
 926                 "(%d) mac " RTE_ETHER_ADDR_PRT_FMT " and vlan %d registered\n",
 927                 vdev->vid, RTE_ETHER_ADDR_BYTES(&vdev->mac_address),
 928                 vdev->vlan_tag);
 929
 930         /* Register the MAC address. */
 931         ret = rte_eth_dev_mac_addr_add(ports[0], &vdev->mac_address,
 932                                 (uint32_t)vdev->vid + vmdq_pool_base);
 933         if (ret)
 934                 RTE_LOG(ERR, VHOST_DATA,
 935                         "(%d) failed to add device MAC address to VMDQ\n",
 936                         vdev->vid);
 937
 938         rte_eth_dev_set_vlan_strip_on_queue(ports[0], vdev->vmdq_rx_q, 1);
 939
 940         /* Set device as ready for RX. */
 941         vdev->ready = DEVICE_RX;
 942
 943         return 0;
 944 }
 945
 946 /*
 947  * Removes MAC address and vlan tag from VMDQ. Ensures that nothing is adding buffers to the RX
 948  * queue before disabling RX on the device.
 949  */
 950 static inline void
 951 unlink_vmdq(struct vhost_dev *vdev)
 952 {
 953         unsigned i = 0;
 954         unsigned rx_count;
 955         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
 956
 957         if (vdev->ready == DEVICE_RX) {
 958                 /*clear MAC and VLAN settings*/
 959                 rte_eth_dev_mac_addr_remove(ports[0], &vdev->mac_address);
 960                 for (i = 0; i < 6; i++)
 961                         vdev->mac_address.addr_bytes[i] = 0;
 962
 963                 vdev->vlan_tag = 0;
 964
 965                 /*Clear out the receive buffers*/
 966                 rx_count = rte_eth_rx_burst(ports[0],
 967                                         (uint16_t)vdev->vmdq_rx_q, pkts_burst, MAX_PKT_BURST);
 968
 969                 while (rx_count) {
 970                         for (i = 0; i < rx_count; i++)
 971                                 rte_pktmbuf_free(pkts_burst[i]);
 972
 973                         rx_count = rte_eth_rx_burst(ports[0],
 974                                         (uint16_t)vdev->vmdq_rx_q, pkts_burst, MAX_PKT_BURST);
 975                 }
 976
 977                 vdev->ready = DEVICE_MAC_LEARNING;
 978         }
 979 }
 980
 981 static inline void
 982 free_pkts(struct rte_mbuf **pkts, uint16_t n)
 983 {
 984         while (n--)
 985                 rte_pktmbuf_free(pkts[n]);
 986 }
 987
 988 static __rte_always_inline void
 989 complete_async_pkts(struct vhost_dev *vdev)
 990 {
 991         struct rte_mbuf *p_cpl[MAX_PKT_BURST];
 992         uint16_t complete_count;
 993         int16_t dma_id = dma_bind[vdev->vid].dmas[VIRTIO_RXQ].dev_id;
 994
 995         complete_count = rte_vhost_poll_enqueue_completed(vdev->vid,
 996                                         VIRTIO_RXQ, p_cpl, MAX_PKT_BURST, dma_id, 0);
 997         if (complete_count)
 998                 free_pkts(p_cpl, complete_count);
 999
1000 }
1001
1002 static __rte_always_inline void
1003 sync_virtio_xmit(struct vhost_dev *dst_vdev, struct vhost_dev *src_vdev,
1004             struct rte_mbuf *m)
1005 {
1006         uint16_t ret;
1007
1008         if (builtin_net_driver) {
1009                 ret = vs_enqueue_pkts(dst_vdev, VIRTIO_RXQ, &m, 1);
1010         } else {
1011                 ret = rte_vhost_enqueue_burst(dst_vdev->vid, VIRTIO_RXQ, &m, 1);
1012         }
1013
1014         if (enable_stats) {
1015                 __atomic_add_fetch(&dst_vdev->stats.rx_total_atomic, 1,
1016                                 __ATOMIC_SEQ_CST);
1017                 __atomic_add_fetch(&dst_vdev->stats.rx_atomic, ret,
1018                                 __ATOMIC_SEQ_CST);
1019                 src_vdev->stats.tx_total++;
1020                 src_vdev->stats.tx += ret;
1021         }
1022 }
1023
1024 static __rte_always_inline void
1025 drain_vhost(struct vhost_dev *vdev)
1026 {
1027         uint16_t ret;
1028         uint32_t buff_idx = rte_lcore_id() * RTE_MAX_VHOST_DEVICE + vdev->vid;
1029         uint16_t nr_xmit = vhost_txbuff[buff_idx]->len;
1030         struct rte_mbuf **m = vhost_txbuff[buff_idx]->m_table;
1031
1032         if (builtin_net_driver) {
1033                 ret = vs_enqueue_pkts(vdev, VIRTIO_RXQ, m, nr_xmit);
1034         } else if (dma_bind[vdev->vid].dmas[VIRTIO_RXQ].async_enabled) {
1035                 uint16_t enqueue_fail = 0;
1036                 int16_t dma_id = dma_bind[vdev->vid].dmas[VIRTIO_RXQ].dev_id;
1037
1038                 complete_async_pkts(vdev);
1039                 ret = rte_vhost_submit_enqueue_burst(vdev->vid, VIRTIO_RXQ, m, nr_xmit, dma_id, 0);
1040
1041                 enqueue_fail = nr_xmit - ret;
1042                 if (enqueue_fail)
1043                         free_pkts(&m[ret], nr_xmit - ret);
1044         } else {
1045                 ret = rte_vhost_enqueue_burst(vdev->vid, VIRTIO_RXQ,
1046                                                 m, nr_xmit);
1047         }
1048
1049         if (enable_stats) {
1050                 __atomic_add_fetch(&vdev->stats.rx_total_atomic, nr_xmit,
1051                                 __ATOMIC_SEQ_CST);
1052                 __atomic_add_fetch(&vdev->stats.rx_atomic, ret,
1053                                 __ATOMIC_SEQ_CST);
1054         }
1055
1056         if (!dma_bind[vdev->vid].dmas[VIRTIO_RXQ].async_enabled)
1057                 free_pkts(m, nr_xmit);
1058 }
1059
1060 static __rte_always_inline void
1061 drain_vhost_table(void)
1062 {
1063         uint16_t lcore_id = rte_lcore_id();
1064         struct vhost_bufftable *vhost_txq;
1065         struct vhost_dev *vdev;
1066         uint64_t cur_tsc;
1067
1068         TAILQ_FOREACH(vdev, &vhost_dev_list, global_vdev_entry) {
1069                 if (unlikely(vdev->remove == 1))
1070                         continue;
1071
1072                 vhost_txq = vhost_txbuff[lcore_id * RTE_MAX_VHOST_DEVICE + vdev->vid];
1073
1074                 cur_tsc = rte_rdtsc();
1075                 if (unlikely(cur_tsc - vhost_txq->pre_tsc
1076                                 > MBUF_TABLE_DRAIN_TSC)) {
1077                         RTE_LOG_DP(DEBUG, VHOST_DATA,
1078                                 "Vhost TX queue drained after timeout with burst size %u\n",
1079                                 vhost_txq->len);
1080                         drain_vhost(vdev);
1081                         vhost_txq->len = 0;
1082                         vhost_txq->pre_tsc = cur_tsc;
1083                 }
1084         }
1085 }
1086
1087 /*
1088  * Check if the packet destination MAC address is for a local device. If so then put
1089  * the packet on that devices RX queue. If not then return.
1090  */
1091 static __rte_always_inline int
1092 virtio_tx_local(struct vhost_dev *vdev, struct rte_mbuf *m)
1093 {
1094         struct rte_ether_hdr *pkt_hdr;
1095         struct vhost_dev *dst_vdev;
1096         struct vhost_bufftable *vhost_txq;
1097         uint16_t lcore_id = rte_lcore_id();
1098         pkt_hdr = rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
1099
1100         dst_vdev = find_vhost_dev(&pkt_hdr->dst_addr);
1101         if (!dst_vdev)
1102                 return -1;
1103
1104         if (vdev->vid == dst_vdev->vid) {
1105                 RTE_LOG_DP(DEBUG, VHOST_DATA,
1106                         "(%d) TX: src and dst MAC is same. Dropping packet.\n",
1107                         vdev->vid);
1108                 return 0;
1109         }
1110
1111         RTE_LOG_DP(DEBUG, VHOST_DATA,
1112                 "(%d) TX: MAC address is local\n", dst_vdev->vid);
1113
1114         if (unlikely(dst_vdev->remove)) {
1115                 RTE_LOG_DP(DEBUG, VHOST_DATA,
1116                         "(%d) device is marked for removal\n", dst_vdev->vid);
1117                 return 0;
1118         }
1119
1120         vhost_txq = vhost_txbuff[lcore_id * RTE_MAX_VHOST_DEVICE + dst_vdev->vid];
1121         vhost_txq->m_table[vhost_txq->len++] = m;
1122
1123         if (enable_stats) {
1124                 vdev->stats.tx_total++;
1125                 vdev->stats.tx++;
1126         }
1127
1128         if (unlikely(vhost_txq->len == MAX_PKT_BURST)) {
1129                 drain_vhost(dst_vdev);
1130                 vhost_txq->len = 0;
1131                 vhost_txq->pre_tsc = rte_rdtsc();
1132         }
1133         return 0;
1134 }
1135
1136 /*
1137  * Check if the destination MAC of a packet is one local VM,
1138  * and get its vlan tag, and offset if it is.
1139  */
1140 static __rte_always_inline int
1141 find_local_dest(struct vhost_dev *vdev, struct rte_mbuf *m,
1142         uint32_t *offset, uint16_t *vlan_tag)
1143 {
1144         struct vhost_dev *dst_vdev;
1145         struct rte_ether_hdr *pkt_hdr =
1146                 rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
1147
1148         dst_vdev = find_vhost_dev(&pkt_hdr->dst_addr);
1149         if (!dst_vdev)
1150                 return 0;
1151
1152         if (vdev->vid == dst_vdev->vid) {
1153                 RTE_LOG_DP(DEBUG, VHOST_DATA,
1154                         "(%d) TX: src and dst MAC is same. Dropping packet.\n",
1155                         vdev->vid);
1156                 return -1;
1157         }
1158
1159         /*
1160          * HW vlan strip will reduce the packet length
1161          * by minus length of vlan tag, so need restore
1162          * the packet length by plus it.
1163          */
1164         *offset  = RTE_VLAN_HLEN;
1165         *vlan_tag = vlan_tags[vdev->vid];
1166
1167         RTE_LOG_DP(DEBUG, VHOST_DATA,
1168                 "(%d) TX: pkt to local VM device id: (%d), vlan tag: %u.\n",
1169                 vdev->vid, dst_vdev->vid, *vlan_tag);
1170
1171         return 0;
1172 }
1173
1174 static void virtio_tx_offload(struct rte_mbuf *m)
1175 {
1176         struct rte_net_hdr_lens hdr_lens;
1177         struct rte_ipv4_hdr *ipv4_hdr;
1178         struct rte_tcp_hdr *tcp_hdr;
1179         uint32_t ptype;
1180         void *l3_hdr;
1181
1182         ptype = rte_net_get_ptype(m, &hdr_lens, RTE_PTYPE_ALL_MASK);
1183         m->l2_len = hdr_lens.l2_len;
1184         m->l3_len = hdr_lens.l3_len;
1185         m->l4_len = hdr_lens.l4_len;
1186
1187         l3_hdr = rte_pktmbuf_mtod_offset(m, void *, m->l2_len);
1188         tcp_hdr = rte_pktmbuf_mtod_offset(m, struct rte_tcp_hdr *,
1189                 m->l2_len + m->l3_len);
1190
1191         m->ol_flags |= RTE_MBUF_F_TX_TCP_SEG;
1192         if ((ptype & RTE_PTYPE_L3_MASK) == RTE_PTYPE_L3_IPV4) {
1193                 m->ol_flags |= RTE_MBUF_F_TX_IPV4;
1194                 m->ol_flags |= RTE_MBUF_F_TX_IP_CKSUM;
1195                 ipv4_hdr = l3_hdr;
1196                 ipv4_hdr->hdr_checksum = 0;
1197                 tcp_hdr->cksum = rte_ipv4_phdr_cksum(l3_hdr, m->ol_flags);
1198         } else { /* assume ethertype == RTE_ETHER_TYPE_IPV6 */
1199                 m->ol_flags |= RTE_MBUF_F_TX_IPV6;
1200                 tcp_hdr->cksum = rte_ipv6_phdr_cksum(l3_hdr, m->ol_flags);
1201         }
1202 }
1203
1204 static __rte_always_inline void
1205 do_drain_mbuf_table(struct mbuf_table *tx_q)
1206 {
1207         uint16_t count;
1208
1209         count = rte_eth_tx_burst(ports[0], tx_q->txq_id,
1210                                  tx_q->m_table, tx_q->len);
1211         if (unlikely(count < tx_q->len))
1212                 free_pkts(&tx_q->m_table[count], tx_q->len - count);
1213
1214         tx_q->len = 0;
1215 }
1216
1217 /*
1218  * This function routes the TX packet to the correct interface. This
1219  * may be a local device or the physical port.
1220  */
1221 static __rte_always_inline void
1222 virtio_tx_route(struct vhost_dev *vdev, struct rte_mbuf *m, uint16_t vlan_tag)
1223 {
1224         struct mbuf_table *tx_q;
1225         unsigned offset = 0;
1226         const uint16_t lcore_id = rte_lcore_id();
1227         struct rte_ether_hdr *nh;
1228
1229
1230         nh = rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
1231         if (unlikely(rte_is_broadcast_ether_addr(&nh->dst_addr))) {
1232                 struct vhost_dev *vdev2;
1233
1234                 TAILQ_FOREACH(vdev2, &vhost_dev_list, global_vdev_entry) {
1235                         if (vdev2 != vdev)
1236                                 sync_virtio_xmit(vdev2, vdev, m);
1237                 }
1238                 goto queue2nic;
1239         }
1240
1241         /*check if destination is local VM*/
1242         if ((vm2vm_mode == VM2VM_SOFTWARE) && (virtio_tx_local(vdev, m) == 0))
1243                 return;
1244
1245         if (unlikely(vm2vm_mode == VM2VM_HARDWARE)) {
1246                 if (unlikely(find_local_dest(vdev, m, &offset,
1247                                              &vlan_tag) != 0)) {
1248                         rte_pktmbuf_free(m);
1249                         return;
1250                 }
1251         }
1252
1253         RTE_LOG_DP(DEBUG, VHOST_DATA,
1254                 "(%d) TX: MAC address is external\n", vdev->vid);
1255
1256 queue2nic:
1257
1258         /*Add packet to the port tx queue*/
1259         tx_q = &lcore_tx_queue[lcore_id];
1260
1261         nh = rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
1262         if (unlikely(nh->ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_VLAN))) {
1263                 /* Guest has inserted the vlan tag. */
1264                 struct rte_vlan_hdr *vh = (struct rte_vlan_hdr *) (nh + 1);
1265                 uint16_t vlan_tag_be = rte_cpu_to_be_16(vlan_tag);
1266                 if ((vm2vm_mode == VM2VM_HARDWARE) &&
1267                         (vh->vlan_tci != vlan_tag_be))
1268                         vh->vlan_tci = vlan_tag_be;
1269         } else {
1270                 m->ol_flags |= RTE_MBUF_F_TX_VLAN;
1271
1272                 /*
1273                  * Find the right seg to adjust the data len when offset is
1274                  * bigger than tail room size.
1275                  */
1276                 if (unlikely(vm2vm_mode == VM2VM_HARDWARE)) {
1277                         if (likely(offset <= rte_pktmbuf_tailroom(m)))
1278                                 m->data_len += offset;
1279                         else {
1280                                 struct rte_mbuf *seg = m;
1281
1282                                 while ((seg->next != NULL) &&
1283                                         (offset > rte_pktmbuf_tailroom(seg)))
1284                                         seg = seg->next;
1285
1286                                 seg->data_len += offset;
1287                         }
1288                         m->pkt_len += offset;
1289                 }
1290
1291                 m->vlan_tci = vlan_tag;
1292         }
1293
1294         if (m->ol_flags & RTE_MBUF_F_RX_LRO)
1295                 virtio_tx_offload(m);
1296
1297         tx_q->m_table[tx_q->len++] = m;
1298         if (enable_stats) {
1299                 vdev->stats.tx_total++;
1300                 vdev->stats.tx++;
1301         }
1302
1303         if (unlikely(tx_q->len == MAX_PKT_BURST))
1304                 do_drain_mbuf_table(tx_q);
1305 }
1306
1307
1308 static __rte_always_inline void
1309 drain_mbuf_table(struct mbuf_table *tx_q)
1310 {
1311         static uint64_t prev_tsc;
1312         uint64_t cur_tsc;
1313
1314         if (tx_q->len == 0)
1315                 return;
1316
1317         cur_tsc = rte_rdtsc();
1318         if (unlikely(cur_tsc - prev_tsc > MBUF_TABLE_DRAIN_TSC)) {
1319                 prev_tsc = cur_tsc;
1320
1321                 RTE_LOG_DP(DEBUG, VHOST_DATA,
1322                         "TX queue drained after timeout with burst size %u\n",
1323                         tx_q->len);
1324                 do_drain_mbuf_table(tx_q);
1325         }
1326 }
1327
1328 static __rte_always_inline void
1329 drain_eth_rx(struct vhost_dev *vdev)
1330 {
1331         uint16_t rx_count, enqueue_count;
1332         struct rte_mbuf *pkts[MAX_PKT_BURST];
1333
1334         rx_count = rte_eth_rx_burst(ports[0], vdev->vmdq_rx_q,
1335                                     pkts, MAX_PKT_BURST);
1336
1337         if (!rx_count)
1338                 return;
1339
1340         /*
1341          * When "enable_retry" is set, here we wait and retry when there
1342          * is no enough free slots in the queue to hold @rx_count packets,
1343          * to diminish packet loss.
1344          */
1345         if (enable_retry &&
1346             unlikely(rx_count > rte_vhost_avail_entries(vdev->vid,
1347                         VIRTIO_RXQ))) {
1348                 uint32_t retry;
1349
1350                 for (retry = 0; retry < burst_rx_retry_num; retry++) {
1351                         rte_delay_us(burst_rx_delay_time);
1352                         if (rx_count <= rte_vhost_avail_entries(vdev->vid,
1353                                         VIRTIO_RXQ))
1354                                 break;
1355                 }
1356         }
1357
1358         if (builtin_net_driver) {
1359                 enqueue_count = vs_enqueue_pkts(vdev, VIRTIO_RXQ,
1360                                                 pkts, rx_count);
1361         } else if (dma_bind[vdev->vid].dmas[VIRTIO_RXQ].async_enabled) {
1362                 uint16_t enqueue_fail = 0;
1363                 int16_t dma_id = dma_bind[vdev->vid].dmas[VIRTIO_RXQ].dev_id;
1364
1365                 complete_async_pkts(vdev);
1366                 enqueue_count = rte_vhost_submit_enqueue_burst(vdev->vid,
1367                                         VIRTIO_RXQ, pkts, rx_count, dma_id, 0);
1368
1369                 enqueue_fail = rx_count - enqueue_count;
1370                 if (enqueue_fail)
1371                         free_pkts(&pkts[enqueue_count], enqueue_fail);
1372
1373         } else {
1374                 enqueue_count = rte_vhost_enqueue_burst(vdev->vid, VIRTIO_RXQ,
1375                                                 pkts, rx_count);
1376         }
1377
1378         if (enable_stats) {
1379                 __atomic_add_fetch(&vdev->stats.rx_total_atomic, rx_count,
1380                                 __ATOMIC_SEQ_CST);
1381                 __atomic_add_fetch(&vdev->stats.rx_atomic, enqueue_count,
1382                                 __ATOMIC_SEQ_CST);
1383         }
1384
1385         if (!dma_bind[vdev->vid].dmas[VIRTIO_RXQ].async_enabled)
1386                 free_pkts(pkts, rx_count);
1387 }
1388
1389 static __rte_always_inline void
1390 drain_virtio_tx(struct vhost_dev *vdev)
1391 {
1392         struct rte_mbuf *pkts[MAX_PKT_BURST];
1393         uint16_t count;
1394         uint16_t i;
1395
1396         if (builtin_net_driver) {
1397                 count = vs_dequeue_pkts(vdev, VIRTIO_TXQ, mbuf_pool,
1398                                         pkts, MAX_PKT_BURST);
1399         } else {
1400                 count = rte_vhost_dequeue_burst(vdev->vid, VIRTIO_TXQ,
1401                                         mbuf_pool, pkts, MAX_PKT_BURST);
1402         }
1403
1404         /* setup VMDq for the first packet */
1405         if (unlikely(vdev->ready == DEVICE_MAC_LEARNING) && count) {
1406                 if (vdev->remove || link_vmdq(vdev, pkts[0]) == -1)
1407                         free_pkts(pkts, count);
1408         }
1409
1410         for (i = 0; i < count; ++i)
1411                 virtio_tx_route(vdev, pkts[i], vlan_tags[vdev->vid]);
1412 }
1413
1414 /*
1415  * Main function of vhost-switch. It basically does:
1416  *
1417  * for each vhost device {
1418  *    - drain_eth_rx()
1419  *
1420  *      Which drains the host eth Rx queue linked to the vhost device,
1421  *      and deliver all of them to guest virito Rx ring associated with
1422  *      this vhost device.
1423  *
1424  *    - drain_virtio_tx()
1425  *
1426  *      Which drains the guest virtio Tx queue and deliver all of them
1427  *      to the target, which could be another vhost device, or the
1428  *      physical eth dev. The route is done in function "virtio_tx_route".
1429  * }
1430  */
1431 static int
1432 switch_worker(void *arg __rte_unused)
1433 {
1434         unsigned i;
1435         unsigned lcore_id = rte_lcore_id();
1436         struct vhost_dev *vdev;
1437         struct mbuf_table *tx_q;
1438
1439         RTE_LOG(INFO, VHOST_DATA, "Processing on Core %u started\n", lcore_id);
1440
1441         tx_q = &lcore_tx_queue[lcore_id];
1442         for (i = 0; i < rte_lcore_count(); i++) {
1443                 if (lcore_ids[i] == lcore_id) {
1444                         tx_q->txq_id = i;
1445                         break;
1446                 }
1447         }
1448
1449         while(1) {
1450                 drain_mbuf_table(tx_q);
1451                 drain_vhost_table();
1452                 /*
1453                  * Inform the configuration core that we have exited the
1454                  * linked list and that no devices are in use if requested.
1455                  */
1456                 if (lcore_info[lcore_id].dev_removal_flag == REQUEST_DEV_REMOVAL)
1457                         lcore_info[lcore_id].dev_removal_flag = ACK_DEV_REMOVAL;
1458
1459                 /*
1460                  * Process vhost devices
1461                  */
1462                 TAILQ_FOREACH(vdev, &lcore_info[lcore_id].vdev_list,
1463                               lcore_vdev_entry) {
1464                         if (unlikely(vdev->remove)) {
1465                                 unlink_vmdq(vdev);
1466                                 vdev->ready = DEVICE_SAFE_REMOVE;
1467                                 continue;
1468                         }
1469
1470                         if (likely(vdev->ready == DEVICE_RX))
1471                                 drain_eth_rx(vdev);
1472
1473                         if (likely(!vdev->remove))
1474                                 drain_virtio_tx(vdev);
1475                 }
1476         }
1477
1478         return 0;
1479 }
1480
1481 /*
1482  * Remove a device from the specific data core linked list and from the
1483  * main linked list. Synchronization  occurs through the use of the
1484  * lcore dev_removal_flag. Device is made volatile here to avoid re-ordering
1485  * of dev->remove=1 which can cause an infinite loop in the rte_pause loop.
1486  */
1487 static void
1488 destroy_device(int vid)
1489 {
1490         struct vhost_dev *vdev = NULL;
1491         int lcore;
1492         uint16_t i;
1493
1494         TAILQ_FOREACH(vdev, &vhost_dev_list, global_vdev_entry) {
1495                 if (vdev->vid == vid)
1496                         break;
1497         }
1498         if (!vdev)
1499                 return;
1500         /*set the remove flag. */
1501         vdev->remove = 1;
1502         while(vdev->ready != DEVICE_SAFE_REMOVE) {
1503                 rte_pause();
1504         }
1505
1506         for (i = 0; i < RTE_MAX_LCORE; i++)
1507                 rte_free(vhost_txbuff[i * RTE_MAX_VHOST_DEVICE + vid]);
1508
1509         if (builtin_net_driver)
1510                 vs_vhost_net_remove(vdev);
1511
1512         TAILQ_REMOVE(&lcore_info[vdev->coreid].vdev_list, vdev,
1513                      lcore_vdev_entry);
1514         TAILQ_REMOVE(&vhost_dev_list, vdev, global_vdev_entry);
1515
1516
1517         /* Set the dev_removal_flag on each lcore. */
1518         RTE_LCORE_FOREACH_WORKER(lcore)
1519                 lcore_info[lcore].dev_removal_flag = REQUEST_DEV_REMOVAL;
1520
1521         /*
1522          * Once each core has set the dev_removal_flag to ACK_DEV_REMOVAL
1523          * we can be sure that they can no longer access the device removed
1524          * from the linked lists and that the devices are no longer in use.
1525          */
1526         RTE_LCORE_FOREACH_WORKER(lcore) {
1527                 while (lcore_info[lcore].dev_removal_flag != ACK_DEV_REMOVAL)
1528                         rte_pause();
1529         }
1530
1531         lcore_info[vdev->coreid].device_num--;
1532
1533         RTE_LOG(INFO, VHOST_DATA,
1534                 "(%d) device has been removed from data core\n",
1535                 vdev->vid);
1536
1537         if (dma_bind[vid].dmas[VIRTIO_RXQ].async_enabled) {
1538                 uint16_t n_pkt = 0;
1539                 int pkts_inflight;
1540                 int16_t dma_id = dma_bind[vid].dmas[VIRTIO_RXQ].dev_id;
1541                 pkts_inflight = rte_vhost_async_get_inflight_thread_unsafe(vid, VIRTIO_RXQ);
1542                 struct rte_mbuf *m_cpl[pkts_inflight];
1543
1544                 while (pkts_inflight) {
1545                         n_pkt = rte_vhost_clear_queue_thread_unsafe(vid, VIRTIO_RXQ,
1546                                                 m_cpl, pkts_inflight, dma_id, 0);
1547                         free_pkts(m_cpl, n_pkt);
1548                         pkts_inflight = rte_vhost_async_get_inflight_thread_unsafe(vid,
1549                                                                                 VIRTIO_RXQ);
1550                 }
1551
1552                 rte_vhost_async_channel_unregister(vid, VIRTIO_RXQ);
1553                 dma_bind[vid].dmas[VIRTIO_RXQ].async_enabled = false;
1554         }
1555
1556         rte_free(vdev);
1557 }
1558
1559 /*
1560  * A new device is added to a data core. First the device is added to the main linked list
1561  * and then allocated to a specific data core.
1562  */
1563 static int
1564 new_device(int vid)
1565 {
1566         int lcore, core_add = 0;
1567         uint16_t i;
1568         uint32_t device_num_min = num_devices;
1569         struct vhost_dev *vdev;
1570         vdev = rte_zmalloc("vhost device", sizeof(*vdev), RTE_CACHE_LINE_SIZE);
1571         if (vdev == NULL) {
1572                 RTE_LOG(INFO, VHOST_DATA,
1573                         "(%d) couldn't allocate memory for vhost dev\n",
1574                         vid);
1575                 return -1;
1576         }
1577         vdev->vid = vid;
1578
1579         for (i = 0; i < RTE_MAX_LCORE; i++) {
1580                 vhost_txbuff[i * RTE_MAX_VHOST_DEVICE + vid]
1581                         = rte_zmalloc("vhost bufftable",
1582                                 sizeof(struct vhost_bufftable),
1583                                 RTE_CACHE_LINE_SIZE);
1584
1585                 if (vhost_txbuff[i * RTE_MAX_VHOST_DEVICE + vid] == NULL) {
1586                         RTE_LOG(INFO, VHOST_DATA,
1587                           "(%d) couldn't allocate memory for vhost TX\n", vid);
1588                         return -1;
1589                 }
1590         }
1591
1592         if (builtin_net_driver)
1593                 vs_vhost_net_setup(vdev);
1594
1595         TAILQ_INSERT_TAIL(&vhost_dev_list, vdev, global_vdev_entry);
1596         vdev->vmdq_rx_q = vid * queues_per_pool + vmdq_queue_base;
1597
1598         /*reset ready flag*/
1599         vdev->ready = DEVICE_MAC_LEARNING;
1600         vdev->remove = 0;
1601
1602         /* Find a suitable lcore to add the device. */
1603         RTE_LCORE_FOREACH_WORKER(lcore) {
1604                 if (lcore_info[lcore].device_num < device_num_min) {
1605                         device_num_min = lcore_info[lcore].device_num;
1606                         core_add = lcore;
1607                 }
1608         }
1609         vdev->coreid = core_add;
1610
1611         TAILQ_INSERT_TAIL(&lcore_info[vdev->coreid].vdev_list, vdev,
1612                           lcore_vdev_entry);
1613         lcore_info[vdev->coreid].device_num++;
1614
1615         /* Disable notifications. */
1616         rte_vhost_enable_guest_notification(vid, VIRTIO_RXQ, 0);
1617         rte_vhost_enable_guest_notification(vid, VIRTIO_TXQ, 0);
1618
1619         RTE_LOG(INFO, VHOST_DATA,
1620                 "(%d) device has been added to data core %d\n",
1621                 vid, vdev->coreid);
1622
1623         if (dma_bind[vid].dmas[VIRTIO_RXQ].dev_id != INVALID_DMA_ID) {
1624                 int ret;
1625
1626                 ret = rte_vhost_async_channel_register(vid, VIRTIO_RXQ);
1627                 if (ret == 0)
1628                         dma_bind[vid].dmas[VIRTIO_RXQ].async_enabled = true;
1629                 return ret;
1630         }
1631
1632         return 0;
1633 }
1634
1635 static int
1636 vring_state_changed(int vid, uint16_t queue_id, int enable)
1637 {
1638         struct vhost_dev *vdev = NULL;
1639
1640         TAILQ_FOREACH(vdev, &vhost_dev_list, global_vdev_entry) {
1641                 if (vdev->vid == vid)
1642                         break;
1643         }
1644         if (!vdev)
1645                 return -1;
1646
1647         if (queue_id != VIRTIO_RXQ)
1648                 return 0;
1649
1650         if (dma_bind[vid].dmas[queue_id].async_enabled) {
1651                 if (!enable) {
1652                         uint16_t n_pkt = 0;
1653                         int pkts_inflight;
1654                         pkts_inflight = rte_vhost_async_get_inflight_thread_unsafe(vid, queue_id);
1655                         int16_t dma_id = dma_bind[vid].dmas[VIRTIO_RXQ].dev_id;
1656                         struct rte_mbuf *m_cpl[pkts_inflight];
1657
1658                         while (pkts_inflight) {
1659                                 n_pkt = rte_vhost_clear_queue_thread_unsafe(vid, queue_id,
1660                                                         m_cpl, pkts_inflight, dma_id, 0);
1661                                 free_pkts(m_cpl, n_pkt);
1662                                 pkts_inflight = rte_vhost_async_get_inflight_thread_unsafe(vid,
1663                                                                                         queue_id);
1664                         }
1665                 }
1666         }
1667
1668         return 0;
1669 }
1670
1671 /*
1672  * These callback allow devices to be added to the data core when configuration
1673  * has been fully complete.
1674  */
1675 static const struct rte_vhost_device_ops virtio_net_device_ops =
1676 {
1677         .new_device =  new_device,
1678         .destroy_device = destroy_device,
1679         .vring_state_changed = vring_state_changed,
1680 };
1681
1682 /*
1683  * This is a thread will wake up after a period to print stats if the user has
1684  * enabled them.
1685  */
1686 static void *
1687 print_stats(__rte_unused void *arg)
1688 {
1689         struct vhost_dev *vdev;
1690         uint64_t tx_dropped, rx_dropped;
1691         uint64_t tx, tx_total, rx, rx_total;
1692         const char clr[] = { 27, '[', '2', 'J', '\0' };
1693         const char top_left[] = { 27, '[', '1', ';', '1', 'H','\0' };
1694
1695         while(1) {
1696                 sleep(enable_stats);
1697
1698                 /* Clear screen and move to top left */
1699                 printf("%s%s\n", clr, top_left);
1700                 printf("Device statistics =================================\n");
1701
1702                 TAILQ_FOREACH(vdev, &vhost_dev_list, global_vdev_entry) {
1703                         tx_total   = vdev->stats.tx_total;
1704                         tx         = vdev->stats.tx;
1705                         tx_dropped = tx_total - tx;
1706
1707                         rx_total = __atomic_load_n(&vdev->stats.rx_total_atomic,
1708                                 __ATOMIC_SEQ_CST);
1709                         rx         = __atomic_load_n(&vdev->stats.rx_atomic,
1710                                 __ATOMIC_SEQ_CST);
1711                         rx_dropped = rx_total - rx;
1712
1713                         printf("Statistics for device %d\n"
1714                                 "-----------------------\n"
1715                                 "TX total:              %" PRIu64 "\n"
1716                                 "TX dropped:            %" PRIu64 "\n"
1717                                 "TX successful:         %" PRIu64 "\n"
1718                                 "RX total:              %" PRIu64 "\n"
1719                                 "RX dropped:            %" PRIu64 "\n"
1720                                 "RX successful:         %" PRIu64 "\n",
1721                                 vdev->vid,
1722                                 tx_total, tx_dropped, tx,
1723                                 rx_total, rx_dropped, rx);
1724                 }
1725
1726                 printf("===================================================\n");
1727
1728                 fflush(stdout);
1729         }
1730
1731         return NULL;
1732 }
1733
1734 static void
1735 unregister_drivers(int socket_num)
1736 {
1737         int i, ret;
1738
1739         for (i = 0; i < socket_num; i++) {
1740                 ret = rte_vhost_driver_unregister(socket_files + i * PATH_MAX);
1741                 if (ret != 0)
1742                         RTE_LOG(ERR, VHOST_CONFIG,
1743                                 "Fail to unregister vhost driver for %s.\n",
1744                                 socket_files + i * PATH_MAX);
1745         }
1746 }
1747
1748 /* When we receive a INT signal, unregister vhost driver */
1749 static void
1750 sigint_handler(__rte_unused int signum)
1751 {
1752         /* Unregister vhost driver. */
1753         unregister_drivers(nb_sockets);
1754
1755         exit(0);
1756 }
1757
1758 static void
1759 reset_dma(void)
1760 {
1761         int i;
1762
1763         for (i = 0; i < RTE_MAX_VHOST_DEVICE; i++) {
1764                 int j;
1765
1766                 for (j = 0; j < RTE_MAX_QUEUES_PER_PORT * 2; j++) {
1767                         dma_bind[i].dmas[j].dev_id = INVALID_DMA_ID;
1768                         dma_bind[i].dmas[j].async_enabled = false;
1769                 }
1770         }
1771
1772         for (i = 0; i < RTE_DMADEV_DEFAULT_MAX; i++)
1773                 dmas_id[i] = INVALID_DMA_ID;
1774 }
1775
1776 /*
1777  * Main function, does initialisation and calls the per-lcore functions.
1778  */
1779 int
1780 main(int argc, char *argv[])
1781 {
1782         unsigned lcore_id, core_id = 0;
1783         unsigned nb_ports, valid_num_ports;
1784         int ret, i;
1785         uint16_t portid;
1786         static pthread_t tid;
1787         uint64_t flags = RTE_VHOST_USER_NET_COMPLIANT_OL_FLAGS;
1788
1789         signal(SIGINT, sigint_handler);
1790
1791         /* init EAL */
1792         ret = rte_eal_init(argc, argv);
1793         if (ret < 0)
1794                 rte_exit(EXIT_FAILURE, "Error with EAL initialization\n");
1795         argc -= ret;
1796         argv += ret;
1797
1798         /* initialize dma structures */
1799         reset_dma();
1800
1801         /* parse app arguments */
1802         ret = us_vhost_parse_args(argc, argv);
1803         if (ret < 0)
1804                 rte_exit(EXIT_FAILURE, "Invalid argument\n");
1805
1806         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1807                 TAILQ_INIT(&lcore_info[lcore_id].vdev_list);
1808
1809                 if (rte_lcore_is_enabled(lcore_id))
1810                         lcore_ids[core_id++] = lcore_id;
1811         }
1812
1813         if (rte_lcore_count() > RTE_MAX_LCORE)
1814                 rte_exit(EXIT_FAILURE,"Not enough cores\n");
1815
1816         /* Get the number of physical ports. */
1817         nb_ports = rte_eth_dev_count_avail();
1818
1819         /*
1820          * Update the global var NUM_PORTS and global array PORTS
1821          * and get value of var VALID_NUM_PORTS according to system ports number
1822          */
1823         valid_num_ports = check_ports_num(nb_ports);
1824
1825         if ((valid_num_ports ==  0) || (valid_num_ports > MAX_SUP_PORTS)) {
1826                 RTE_LOG(INFO, VHOST_PORT, "Current enabled port number is %u,"
1827                         "but only %u port can be enabled\n",num_ports, MAX_SUP_PORTS);
1828                 return -1;
1829         }
1830
1831         /*
1832          * FIXME: here we are trying to allocate mbufs big enough for
1833          * @MAX_QUEUES, but the truth is we're never going to use that
1834          * many queues here. We probably should only do allocation for
1835          * those queues we are going to use.
1836          */
1837         mbuf_pool = rte_pktmbuf_pool_create("MBUF_POOL", total_num_mbufs,
1838                                             MBUF_CACHE_SIZE, 0, MBUF_DATA_SIZE,
1839                                             rte_socket_id());
1840         if (mbuf_pool == NULL)
1841                 rte_exit(EXIT_FAILURE, "Cannot create mbuf pool\n");
1842
1843         if (vm2vm_mode == VM2VM_HARDWARE) {
1844                 /* Enable VT loop back to let L2 switch to do it. */
1845                 vmdq_conf_default.rx_adv_conf.vmdq_rx_conf.enable_loop_back = 1;
1846                 RTE_LOG(DEBUG, VHOST_CONFIG,
1847                         "Enable loop back for L2 switch in vmdq.\n");
1848         }
1849
1850         /* initialize all ports */
1851         RTE_ETH_FOREACH_DEV(portid) {
1852                 /* skip ports that are not enabled */
1853                 if ((enabled_port_mask & (1 << portid)) == 0) {
1854                         RTE_LOG(INFO, VHOST_PORT,
1855                                 "Skipping disabled port %d\n", portid);
1856                         continue;
1857                 }
1858                 if (port_init(portid) != 0)
1859                         rte_exit(EXIT_FAILURE,
1860                                 "Cannot initialize network ports\n");
1861         }
1862
1863         /* Enable stats if the user option is set. */
1864         if (enable_stats) {
1865                 ret = rte_ctrl_thread_create(&tid, "print-stats", NULL,
1866                                         print_stats, NULL);
1867                 if (ret < 0)
1868                         rte_exit(EXIT_FAILURE,
1869                                 "Cannot create print-stats thread\n");
1870         }
1871
1872         /* Launch all data cores. */
1873         RTE_LCORE_FOREACH_WORKER(lcore_id)
1874                 rte_eal_remote_launch(switch_worker, NULL, lcore_id);
1875
1876         if (client_mode)
1877                 flags |= RTE_VHOST_USER_CLIENT;
1878
1879         for (i = 0; i < dma_count; i++) {
1880                 if (rte_vhost_async_dma_configure(dmas_id[i], 0) < 0) {
1881                         RTE_LOG(ERR, VHOST_PORT, "Failed to configure DMA in vhost.\n");
1882                         rte_exit(EXIT_FAILURE, "Cannot use given DMA device\n");
1883                 }
1884         }
1885
1886         /* Register vhost user driver to handle vhost messages. */
1887         for (i = 0; i < nb_sockets; i++) {
1888                 char *file = socket_files + i * PATH_MAX;
1889
1890                 if (dma_count)
1891                         flags = flags | RTE_VHOST_USER_ASYNC_COPY;
1892
1893                 ret = rte_vhost_driver_register(file, flags);
1894                 if (ret != 0) {
1895                         unregister_drivers(i);
1896                         rte_exit(EXIT_FAILURE,
1897                                 "vhost driver register failure.\n");
1898                 }
1899
1900                 if (builtin_net_driver)
1901                         rte_vhost_driver_set_features(file, VIRTIO_NET_FEATURES);
1902
1903                 if (mergeable == 0) {
1904                         rte_vhost_driver_disable_features(file,
1905                                 1ULL << VIRTIO_NET_F_MRG_RXBUF);
1906                 }
1907
1908                 if (enable_tx_csum == 0) {
1909                         rte_vhost_driver_disable_features(file,
1910                                 1ULL << VIRTIO_NET_F_CSUM);
1911                 }
1912
1913                 if (enable_tso == 0) {
1914                         rte_vhost_driver_disable_features(file,
1915                                 1ULL << VIRTIO_NET_F_HOST_TSO4);
1916                         rte_vhost_driver_disable_features(file,
1917                                 1ULL << VIRTIO_NET_F_HOST_TSO6);
1918                         rte_vhost_driver_disable_features(file,
1919                                 1ULL << VIRTIO_NET_F_GUEST_TSO4);
1920                         rte_vhost_driver_disable_features(file,
1921                                 1ULL << VIRTIO_NET_F_GUEST_TSO6);
1922                 }
1923
1924                 if (promiscuous) {
1925                         rte_vhost_driver_enable_features(file,
1926                                 1ULL << VIRTIO_NET_F_CTRL_RX);
1927                 }
1928
1929                 ret = rte_vhost_driver_callback_register(file,
1930                         &virtio_net_device_ops);
1931                 if (ret != 0) {
1932                         rte_exit(EXIT_FAILURE,
1933                                 "failed to register vhost driver callbacks.\n");
1934                 }
1935
1936                 if (rte_vhost_driver_start(file) < 0) {
1937                         rte_exit(EXIT_FAILURE,
1938                                 "failed to start vhost driver.\n");
1939                 }
1940         }
1941
1942         RTE_LCORE_FOREACH_WORKER(lcore_id)
1943                 rte_eal_wait_lcore(lcore_id);
1944
1945         /* clean up the EAL */
1946         rte_eal_cleanup();
1947
1948         return 0;
1949 }