1 /*- 2 * BSD LICENSE 3 * 4 * Copyright(c) 2010-2016 Intel Corporation. All rights reserved. 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 11 * * Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * * Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in 15 * the documentation and/or other materials provided with the 16 * distribution. 17 * * Neither the name of Intel Corporation nor the names of its 18 * contributors may be used to endorse or promote products derived 19 * from this software without specific prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 22 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 24 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 25 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 26 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 27 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 28 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 29 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 30 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE 31 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 32 */ 33 34 #include <arpa/inet.h> 35 #include <getopt.h> 36 #include <linux/if_ether.h> 37 #include <linux/if_vlan.h> 38 #include <linux/virtio_net.h> 39 #include <linux/virtio_ring.h> 40 #include <signal.h> 41 #include <stdint.h> 42 #include <sys/eventfd.h> 43 #include <sys/param.h> 44 #include <unistd.h> 45 46 #include <rte_atomic.h> 47 #include <rte_cycles.h> 48 #include <rte_ethdev.h> 49 #include <rte_log.h> 50 #include <rte_string_fns.h> 51 #include <rte_malloc.h> 52 #include <rte_virtio_net.h> 53 #include <rte_ip.h> 54 #include <rte_tcp.h> 55 56 #include "main.h" 57 58 #ifndef MAX_QUEUES 59 #define MAX_QUEUES 128 60 #endif 61 62 /* the maximum number of external ports supported */ 63 #define MAX_SUP_PORTS 1 64 65 #define MBUF_CACHE_SIZE 128 66 #define MBUF_DATA_SIZE RTE_MBUF_DEFAULT_BUF_SIZE 67 68 #define MAX_PKT_BURST 32 /* Max burst size for RX/TX */ 69 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */ 70 71 #define BURST_RX_WAIT_US 15 /* Defines how long we wait between retries on RX */ 72 #define BURST_RX_RETRIES 4 /* Number of retries on RX. */ 73 74 #define JUMBO_FRAME_MAX_SIZE 0x2600 75 76 /* State of virtio device. */ 77 #define DEVICE_MAC_LEARNING 0 78 #define DEVICE_RX 1 79 #define DEVICE_SAFE_REMOVE 2 80 81 /* Configurable number of RX/TX ring descriptors */ 82 #define RTE_TEST_RX_DESC_DEFAULT 1024 83 #define RTE_TEST_TX_DESC_DEFAULT 512 84 85 #define INVALID_PORT_ID 0xFF 86 87 /* Max number of devices. Limited by vmdq. */ 88 #define MAX_DEVICES 64 89 90 /* Size of buffers used for snprintfs. */ 91 #define MAX_PRINT_BUFF 6072 92 93 /* Maximum character device basename size. */ 94 #define MAX_BASENAME_SZ 10 95 96 /* Maximum long option length for option parsing. */ 97 #define MAX_LONG_OPT_SZ 64 98 99 /* mask of enabled ports */ 100 static uint32_t enabled_port_mask = 0; 101 102 /* Promiscuous mode */ 103 static uint32_t promiscuous; 104 105 /* number of devices/queues to support*/ 106 static uint32_t num_queues = 0; 107 static uint32_t num_devices; 108 109 static struct rte_mempool *mbuf_pool; 110 static int mergeable; 111 112 /* Do vlan strip on host, enabled on default */ 113 static uint32_t vlan_strip = 1; 114 115 /* Enable VM2VM communications. If this is disabled then the MAC address compare is skipped. */ 116 typedef enum { 117 VM2VM_DISABLED = 0, 118 VM2VM_SOFTWARE = 1, 119 VM2VM_HARDWARE = 2, 120 VM2VM_LAST 121 } vm2vm_type; 122 static vm2vm_type vm2vm_mode = VM2VM_SOFTWARE; 123 124 /* Enable stats. */ 125 static uint32_t enable_stats = 0; 126 /* Enable retries on RX. */ 127 static uint32_t enable_retry = 1; 128 129 /* Disable TX checksum offload */ 130 static uint32_t enable_tx_csum; 131 132 /* Disable TSO offload */ 133 static uint32_t enable_tso; 134 135 /* Specify timeout (in useconds) between retries on RX. */ 136 static uint32_t burst_rx_delay_time = BURST_RX_WAIT_US; 137 /* Specify the number of retries on RX. */ 138 static uint32_t burst_rx_retry_num = BURST_RX_RETRIES; 139 140 /* Character device basename. Can be set by user. */ 141 static char dev_basename[MAX_BASENAME_SZ] = "vhost-net"; 142 143 /* empty vmdq configuration structure. Filled in programatically */ 144 static struct rte_eth_conf vmdq_conf_default = { 145 .rxmode = { 146 .mq_mode = ETH_MQ_RX_VMDQ_ONLY, 147 .split_hdr_size = 0, 148 .header_split = 0, /**< Header Split disabled */ 149 .hw_ip_checksum = 0, /**< IP checksum offload disabled */ 150 .hw_vlan_filter = 0, /**< VLAN filtering disabled */ 151 /* 152 * It is necessary for 1G NIC such as I350, 153 * this fixes bug of ipv4 forwarding in guest can't 154 * forward pakets from one virtio dev to another virtio dev. 155 */ 156 .hw_vlan_strip = 1, /**< VLAN strip enabled. */ 157 .jumbo_frame = 0, /**< Jumbo Frame Support disabled */ 158 .hw_strip_crc = 0, /**< CRC stripped by hardware */ 159 }, 160 161 .txmode = { 162 .mq_mode = ETH_MQ_TX_NONE, 163 }, 164 .rx_adv_conf = { 165 /* 166 * should be overridden separately in code with 167 * appropriate values 168 */ 169 .vmdq_rx_conf = { 170 .nb_queue_pools = ETH_8_POOLS, 171 .enable_default_pool = 0, 172 .default_pool = 0, 173 .nb_pool_maps = 0, 174 .pool_map = {{0, 0},}, 175 }, 176 }, 177 }; 178 179 static unsigned lcore_ids[RTE_MAX_LCORE]; 180 static uint8_t ports[RTE_MAX_ETHPORTS]; 181 static unsigned num_ports = 0; /**< The number of ports specified in command line */ 182 static uint16_t num_pf_queues, num_vmdq_queues; 183 static uint16_t vmdq_pool_base, vmdq_queue_base; 184 static uint16_t queues_per_pool; 185 186 const uint16_t vlan_tags[] = { 187 1000, 1001, 1002, 1003, 1004, 1005, 1006, 1007, 188 1008, 1009, 1010, 1011, 1012, 1013, 1014, 1015, 189 1016, 1017, 1018, 1019, 1020, 1021, 1022, 1023, 190 1024, 1025, 1026, 1027, 1028, 1029, 1030, 1031, 191 1032, 1033, 1034, 1035, 1036, 1037, 1038, 1039, 192 1040, 1041, 1042, 1043, 1044, 1045, 1046, 1047, 193 1048, 1049, 1050, 1051, 1052, 1053, 1054, 1055, 194 1056, 1057, 1058, 1059, 1060, 1061, 1062, 1063, 195 }; 196 197 /* ethernet addresses of ports */ 198 static struct ether_addr vmdq_ports_eth_addr[RTE_MAX_ETHPORTS]; 199 200 static struct vhost_dev_tailq_list vhost_dev_list = 201 TAILQ_HEAD_INITIALIZER(vhost_dev_list); 202 203 static struct lcore_info lcore_info[RTE_MAX_LCORE]; 204 205 /* Used for queueing bursts of TX packets. */ 206 struct mbuf_table { 207 unsigned len; 208 unsigned txq_id; 209 struct rte_mbuf *m_table[MAX_PKT_BURST]; 210 }; 211 212 /* TX queue for each data core. */ 213 struct mbuf_table lcore_tx_queue[RTE_MAX_LCORE]; 214 215 #define MBUF_TABLE_DRAIN_TSC ((rte_get_tsc_hz() + US_PER_S - 1) \ 216 / US_PER_S * BURST_TX_DRAIN_US) 217 #define VLAN_HLEN 4 218 219 /* 220 * Builds up the correct configuration for VMDQ VLAN pool map 221 * according to the pool & queue limits. 222 */ 223 static inline int 224 get_eth_conf(struct rte_eth_conf *eth_conf, uint32_t num_devices) 225 { 226 struct rte_eth_vmdq_rx_conf conf; 227 struct rte_eth_vmdq_rx_conf *def_conf = 228 &vmdq_conf_default.rx_adv_conf.vmdq_rx_conf; 229 unsigned i; 230 231 memset(&conf, 0, sizeof(conf)); 232 conf.nb_queue_pools = (enum rte_eth_nb_pools)num_devices; 233 conf.nb_pool_maps = num_devices; 234 conf.enable_loop_back = def_conf->enable_loop_back; 235 conf.rx_mode = def_conf->rx_mode; 236 237 for (i = 0; i < conf.nb_pool_maps; i++) { 238 conf.pool_map[i].vlan_id = vlan_tags[ i ]; 239 conf.pool_map[i].pools = (1UL << i); 240 } 241 242 (void)(rte_memcpy(eth_conf, &vmdq_conf_default, sizeof(*eth_conf))); 243 (void)(rte_memcpy(ð_conf->rx_adv_conf.vmdq_rx_conf, &conf, 244 sizeof(eth_conf->rx_adv_conf.vmdq_rx_conf))); 245 return 0; 246 } 247 248 /* 249 * Validate the device number according to the max pool number gotten form 250 * dev_info. If the device number is invalid, give the error message and 251 * return -1. Each device must have its own pool. 252 */ 253 static inline int 254 validate_num_devices(uint32_t max_nb_devices) 255 { 256 if (num_devices > max_nb_devices) { 257 RTE_LOG(ERR, VHOST_PORT, "invalid number of devices\n"); 258 return -1; 259 } 260 return 0; 261 } 262 263 /* 264 * Initialises a given port using global settings and with the rx buffers 265 * coming from the mbuf_pool passed as parameter 266 */ 267 static inline int 268 port_init(uint8_t port) 269 { 270 struct rte_eth_dev_info dev_info; 271 struct rte_eth_conf port_conf; 272 struct rte_eth_rxconf *rxconf; 273 struct rte_eth_txconf *txconf; 274 int16_t rx_rings, tx_rings; 275 uint16_t rx_ring_size, tx_ring_size; 276 int retval; 277 uint16_t q; 278 279 /* The max pool number from dev_info will be used to validate the pool number specified in cmd line */ 280 rte_eth_dev_info_get (port, &dev_info); 281 282 if (dev_info.max_rx_queues > MAX_QUEUES) { 283 rte_exit(EXIT_FAILURE, 284 "please define MAX_QUEUES no less than %u in %s\n", 285 dev_info.max_rx_queues, __FILE__); 286 } 287 288 rxconf = &dev_info.default_rxconf; 289 txconf = &dev_info.default_txconf; 290 rxconf->rx_drop_en = 1; 291 292 /* Enable vlan offload */ 293 txconf->txq_flags &= ~ETH_TXQ_FLAGS_NOVLANOFFL; 294 295 /*configure the number of supported virtio devices based on VMDQ limits */ 296 num_devices = dev_info.max_vmdq_pools; 297 298 rx_ring_size = RTE_TEST_RX_DESC_DEFAULT; 299 tx_ring_size = RTE_TEST_TX_DESC_DEFAULT; 300 tx_rings = (uint16_t)rte_lcore_count(); 301 302 retval = validate_num_devices(MAX_DEVICES); 303 if (retval < 0) 304 return retval; 305 306 /* Get port configuration. */ 307 retval = get_eth_conf(&port_conf, num_devices); 308 if (retval < 0) 309 return retval; 310 /* NIC queues are divided into pf queues and vmdq queues. */ 311 num_pf_queues = dev_info.max_rx_queues - dev_info.vmdq_queue_num; 312 queues_per_pool = dev_info.vmdq_queue_num / dev_info.max_vmdq_pools; 313 num_vmdq_queues = num_devices * queues_per_pool; 314 num_queues = num_pf_queues + num_vmdq_queues; 315 vmdq_queue_base = dev_info.vmdq_queue_base; 316 vmdq_pool_base = dev_info.vmdq_pool_base; 317 printf("pf queue num: %u, configured vmdq pool num: %u, each vmdq pool has %u queues\n", 318 num_pf_queues, num_devices, queues_per_pool); 319 320 if (port >= rte_eth_dev_count()) return -1; 321 322 if (enable_tx_csum == 0) 323 rte_vhost_feature_disable(1ULL << VIRTIO_NET_F_CSUM); 324 325 if (enable_tso == 0) { 326 rte_vhost_feature_disable(1ULL << VIRTIO_NET_F_HOST_TSO4); 327 rte_vhost_feature_disable(1ULL << VIRTIO_NET_F_HOST_TSO6); 328 } 329 330 rx_rings = (uint16_t)dev_info.max_rx_queues; 331 /* Configure ethernet device. */ 332 retval = rte_eth_dev_configure(port, rx_rings, tx_rings, &port_conf); 333 if (retval != 0) 334 return retval; 335 336 /* Setup the queues. */ 337 for (q = 0; q < rx_rings; q ++) { 338 retval = rte_eth_rx_queue_setup(port, q, rx_ring_size, 339 rte_eth_dev_socket_id(port), 340 rxconf, 341 mbuf_pool); 342 if (retval < 0) 343 return retval; 344 } 345 for (q = 0; q < tx_rings; q ++) { 346 retval = rte_eth_tx_queue_setup(port, q, tx_ring_size, 347 rte_eth_dev_socket_id(port), 348 txconf); 349 if (retval < 0) 350 return retval; 351 } 352 353 /* Start the device. */ 354 retval = rte_eth_dev_start(port); 355 if (retval < 0) { 356 RTE_LOG(ERR, VHOST_DATA, "Failed to start the device.\n"); 357 return retval; 358 } 359 360 if (promiscuous) 361 rte_eth_promiscuous_enable(port); 362 363 rte_eth_macaddr_get(port, &vmdq_ports_eth_addr[port]); 364 RTE_LOG(INFO, VHOST_PORT, "Max virtio devices supported: %u\n", num_devices); 365 RTE_LOG(INFO, VHOST_PORT, "Port %u MAC: %02"PRIx8" %02"PRIx8" %02"PRIx8 366 " %02"PRIx8" %02"PRIx8" %02"PRIx8"\n", 367 (unsigned)port, 368 vmdq_ports_eth_addr[port].addr_bytes[0], 369 vmdq_ports_eth_addr[port].addr_bytes[1], 370 vmdq_ports_eth_addr[port].addr_bytes[2], 371 vmdq_ports_eth_addr[port].addr_bytes[3], 372 vmdq_ports_eth_addr[port].addr_bytes[4], 373 vmdq_ports_eth_addr[port].addr_bytes[5]); 374 375 return 0; 376 } 377 378 /* 379 * Set character device basename. 380 */ 381 static int 382 us_vhost_parse_basename(const char *q_arg) 383 { 384 /* parse number string */ 385 386 if (strnlen(q_arg, MAX_BASENAME_SZ) > MAX_BASENAME_SZ) 387 return -1; 388 else 389 snprintf((char*)&dev_basename, MAX_BASENAME_SZ, "%s", q_arg); 390 391 return 0; 392 } 393 394 /* 395 * Parse the portmask provided at run time. 396 */ 397 static int 398 parse_portmask(const char *portmask) 399 { 400 char *end = NULL; 401 unsigned long pm; 402 403 errno = 0; 404 405 /* parse hexadecimal string */ 406 pm = strtoul(portmask, &end, 16); 407 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0') || (errno != 0)) 408 return -1; 409 410 if (pm == 0) 411 return -1; 412 413 return pm; 414 415 } 416 417 /* 418 * Parse num options at run time. 419 */ 420 static int 421 parse_num_opt(const char *q_arg, uint32_t max_valid_value) 422 { 423 char *end = NULL; 424 unsigned long num; 425 426 errno = 0; 427 428 /* parse unsigned int string */ 429 num = strtoul(q_arg, &end, 10); 430 if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0') || (errno != 0)) 431 return -1; 432 433 if (num > max_valid_value) 434 return -1; 435 436 return num; 437 438 } 439 440 /* 441 * Display usage 442 */ 443 static void 444 us_vhost_usage(const char *prgname) 445 { 446 RTE_LOG(INFO, VHOST_CONFIG, "%s [EAL options] -- -p PORTMASK\n" 447 " --vm2vm [0|1|2]\n" 448 " --rx_retry [0|1] --mergeable [0|1] --stats [0-N]\n" 449 " --dev-basename <name>\n" 450 " --nb-devices ND\n" 451 " -p PORTMASK: Set mask for ports to be used by application\n" 452 " --vm2vm [0|1|2]: disable/software(default)/hardware vm2vm comms\n" 453 " --rx-retry [0|1]: disable/enable(default) retries on rx. Enable retry if destintation queue is full\n" 454 " --rx-retry-delay [0-N]: timeout(in usecond) between retries on RX. This makes effect only if retries on rx enabled\n" 455 " --rx-retry-num [0-N]: the number of retries on rx. This makes effect only if retries on rx enabled\n" 456 " --mergeable [0|1]: disable(default)/enable RX mergeable buffers\n" 457 " --vlan-strip [0|1]: disable/enable(default) RX VLAN strip on host\n" 458 " --stats [0-N]: 0: Disable stats, N: Time in seconds to print stats\n" 459 " --dev-basename: The basename to be used for the character device.\n" 460 " --tx-csum [0|1] disable/enable TX checksum offload.\n" 461 " --tso [0|1] disable/enable TCP segment offload.\n", 462 prgname); 463 } 464 465 /* 466 * Parse the arguments given in the command line of the application. 467 */ 468 static int 469 us_vhost_parse_args(int argc, char **argv) 470 { 471 int opt, ret; 472 int option_index; 473 unsigned i; 474 const char *prgname = argv[0]; 475 static struct option long_option[] = { 476 {"vm2vm", required_argument, NULL, 0}, 477 {"rx-retry", required_argument, NULL, 0}, 478 {"rx-retry-delay", required_argument, NULL, 0}, 479 {"rx-retry-num", required_argument, NULL, 0}, 480 {"mergeable", required_argument, NULL, 0}, 481 {"vlan-strip", required_argument, NULL, 0}, 482 {"stats", required_argument, NULL, 0}, 483 {"dev-basename", required_argument, NULL, 0}, 484 {"tx-csum", required_argument, NULL, 0}, 485 {"tso", required_argument, NULL, 0}, 486 {NULL, 0, 0, 0}, 487 }; 488 489 /* Parse command line */ 490 while ((opt = getopt_long(argc, argv, "p:P", 491 long_option, &option_index)) != EOF) { 492 switch (opt) { 493 /* Portmask */ 494 case 'p': 495 enabled_port_mask = parse_portmask(optarg); 496 if (enabled_port_mask == 0) { 497 RTE_LOG(INFO, VHOST_CONFIG, "Invalid portmask\n"); 498 us_vhost_usage(prgname); 499 return -1; 500 } 501 break; 502 503 case 'P': 504 promiscuous = 1; 505 vmdq_conf_default.rx_adv_conf.vmdq_rx_conf.rx_mode = 506 ETH_VMDQ_ACCEPT_BROADCAST | 507 ETH_VMDQ_ACCEPT_MULTICAST; 508 rte_vhost_feature_enable(1ULL << VIRTIO_NET_F_CTRL_RX); 509 510 break; 511 512 case 0: 513 /* Enable/disable vm2vm comms. */ 514 if (!strncmp(long_option[option_index].name, "vm2vm", 515 MAX_LONG_OPT_SZ)) { 516 ret = parse_num_opt(optarg, (VM2VM_LAST - 1)); 517 if (ret == -1) { 518 RTE_LOG(INFO, VHOST_CONFIG, 519 "Invalid argument for " 520 "vm2vm [0|1|2]\n"); 521 us_vhost_usage(prgname); 522 return -1; 523 } else { 524 vm2vm_mode = (vm2vm_type)ret; 525 } 526 } 527 528 /* Enable/disable retries on RX. */ 529 if (!strncmp(long_option[option_index].name, "rx-retry", MAX_LONG_OPT_SZ)) { 530 ret = parse_num_opt(optarg, 1); 531 if (ret == -1) { 532 RTE_LOG(INFO, VHOST_CONFIG, "Invalid argument for rx-retry [0|1]\n"); 533 us_vhost_usage(prgname); 534 return -1; 535 } else { 536 enable_retry = ret; 537 } 538 } 539 540 /* Enable/disable TX checksum offload. */ 541 if (!strncmp(long_option[option_index].name, "tx-csum", MAX_LONG_OPT_SZ)) { 542 ret = parse_num_opt(optarg, 1); 543 if (ret == -1) { 544 RTE_LOG(INFO, VHOST_CONFIG, "Invalid argument for tx-csum [0|1]\n"); 545 us_vhost_usage(prgname); 546 return -1; 547 } else 548 enable_tx_csum = ret; 549 } 550 551 /* Enable/disable TSO offload. */ 552 if (!strncmp(long_option[option_index].name, "tso", MAX_LONG_OPT_SZ)) { 553 ret = parse_num_opt(optarg, 1); 554 if (ret == -1) { 555 RTE_LOG(INFO, VHOST_CONFIG, "Invalid argument for tso [0|1]\n"); 556 us_vhost_usage(prgname); 557 return -1; 558 } else 559 enable_tso = ret; 560 } 561 562 /* Specify the retries delay time (in useconds) on RX. */ 563 if (!strncmp(long_option[option_index].name, "rx-retry-delay", MAX_LONG_OPT_SZ)) { 564 ret = parse_num_opt(optarg, INT32_MAX); 565 if (ret == -1) { 566 RTE_LOG(INFO, VHOST_CONFIG, "Invalid argument for rx-retry-delay [0-N]\n"); 567 us_vhost_usage(prgname); 568 return -1; 569 } else { 570 burst_rx_delay_time = ret; 571 } 572 } 573 574 /* Specify the retries number on RX. */ 575 if (!strncmp(long_option[option_index].name, "rx-retry-num", MAX_LONG_OPT_SZ)) { 576 ret = parse_num_opt(optarg, INT32_MAX); 577 if (ret == -1) { 578 RTE_LOG(INFO, VHOST_CONFIG, "Invalid argument for rx-retry-num [0-N]\n"); 579 us_vhost_usage(prgname); 580 return -1; 581 } else { 582 burst_rx_retry_num = ret; 583 } 584 } 585 586 /* Enable/disable RX mergeable buffers. */ 587 if (!strncmp(long_option[option_index].name, "mergeable", MAX_LONG_OPT_SZ)) { 588 ret = parse_num_opt(optarg, 1); 589 if (ret == -1) { 590 RTE_LOG(INFO, VHOST_CONFIG, "Invalid argument for mergeable [0|1]\n"); 591 us_vhost_usage(prgname); 592 return -1; 593 } else { 594 mergeable = !!ret; 595 if (ret) { 596 vmdq_conf_default.rxmode.jumbo_frame = 1; 597 vmdq_conf_default.rxmode.max_rx_pkt_len 598 = JUMBO_FRAME_MAX_SIZE; 599 } 600 } 601 } 602 603 /* Enable/disable RX VLAN strip on host. */ 604 if (!strncmp(long_option[option_index].name, 605 "vlan-strip", MAX_LONG_OPT_SZ)) { 606 ret = parse_num_opt(optarg, 1); 607 if (ret == -1) { 608 RTE_LOG(INFO, VHOST_CONFIG, 609 "Invalid argument for VLAN strip [0|1]\n"); 610 us_vhost_usage(prgname); 611 return -1; 612 } else { 613 vlan_strip = !!ret; 614 vmdq_conf_default.rxmode.hw_vlan_strip = 615 vlan_strip; 616 } 617 } 618 619 /* Enable/disable stats. */ 620 if (!strncmp(long_option[option_index].name, "stats", MAX_LONG_OPT_SZ)) { 621 ret = parse_num_opt(optarg, INT32_MAX); 622 if (ret == -1) { 623 RTE_LOG(INFO, VHOST_CONFIG, "Invalid argument for stats [0..N]\n"); 624 us_vhost_usage(prgname); 625 return -1; 626 } else { 627 enable_stats = ret; 628 } 629 } 630 631 /* Set character device basename. */ 632 if (!strncmp(long_option[option_index].name, "dev-basename", MAX_LONG_OPT_SZ)) { 633 if (us_vhost_parse_basename(optarg) == -1) { 634 RTE_LOG(INFO, VHOST_CONFIG, "Invalid argument for character device basename (Max %d characters)\n", MAX_BASENAME_SZ); 635 us_vhost_usage(prgname); 636 return -1; 637 } 638 } 639 640 break; 641 642 /* Invalid option - print options. */ 643 default: 644 us_vhost_usage(prgname); 645 return -1; 646 } 647 } 648 649 for (i = 0; i < RTE_MAX_ETHPORTS; i++) { 650 if (enabled_port_mask & (1 << i)) 651 ports[num_ports++] = (uint8_t)i; 652 } 653 654 if ((num_ports == 0) || (num_ports > MAX_SUP_PORTS)) { 655 RTE_LOG(INFO, VHOST_PORT, "Current enabled port number is %u," 656 "but only %u port can be enabled\n",num_ports, MAX_SUP_PORTS); 657 return -1; 658 } 659 660 return 0; 661 } 662 663 /* 664 * Update the global var NUM_PORTS and array PORTS according to system ports number 665 * and return valid ports number 666 */ 667 static unsigned check_ports_num(unsigned nb_ports) 668 { 669 unsigned valid_num_ports = num_ports; 670 unsigned portid; 671 672 if (num_ports > nb_ports) { 673 RTE_LOG(INFO, VHOST_PORT, "\nSpecified port number(%u) exceeds total system port number(%u)\n", 674 num_ports, nb_ports); 675 num_ports = nb_ports; 676 } 677 678 for (portid = 0; portid < num_ports; portid ++) { 679 if (ports[portid] >= nb_ports) { 680 RTE_LOG(INFO, VHOST_PORT, "\nSpecified port ID(%u) exceeds max system port ID(%u)\n", 681 ports[portid], (nb_ports - 1)); 682 ports[portid] = INVALID_PORT_ID; 683 valid_num_ports--; 684 } 685 } 686 return valid_num_ports; 687 } 688 689 static inline struct vhost_dev *__attribute__((always_inline)) 690 find_vhost_dev(struct ether_addr *mac) 691 { 692 struct vhost_dev *vdev; 693 694 TAILQ_FOREACH(vdev, &vhost_dev_list, next) { 695 if (vdev->ready == DEVICE_RX && 696 is_same_ether_addr(mac, &vdev->mac_address)) 697 return vdev; 698 } 699 700 return NULL; 701 } 702 703 /* 704 * This function learns the MAC address of the device and registers this along with a 705 * vlan tag to a VMDQ. 706 */ 707 static int 708 link_vmdq(struct vhost_dev *vdev, struct rte_mbuf *m) 709 { 710 struct ether_hdr *pkt_hdr; 711 struct virtio_net *dev = vdev->dev; 712 int i, ret; 713 714 /* Learn MAC address of guest device from packet */ 715 pkt_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *); 716 717 if (find_vhost_dev(&pkt_hdr->s_addr)) { 718 RTE_LOG(ERR, VHOST_DATA, 719 "Device (%" PRIu64 ") is using a registered MAC!\n", 720 dev->device_fh); 721 return -1; 722 } 723 724 for (i = 0; i < ETHER_ADDR_LEN; i++) 725 vdev->mac_address.addr_bytes[i] = pkt_hdr->s_addr.addr_bytes[i]; 726 727 /* vlan_tag currently uses the device_id. */ 728 vdev->vlan_tag = vlan_tags[dev->device_fh]; 729 730 /* Print out VMDQ registration info. */ 731 RTE_LOG(INFO, VHOST_DATA, "(%"PRIu64") MAC_ADDRESS %02x:%02x:%02x:%02x:%02x:%02x and VLAN_TAG %d registered\n", 732 dev->device_fh, 733 vdev->mac_address.addr_bytes[0], vdev->mac_address.addr_bytes[1], 734 vdev->mac_address.addr_bytes[2], vdev->mac_address.addr_bytes[3], 735 vdev->mac_address.addr_bytes[4], vdev->mac_address.addr_bytes[5], 736 vdev->vlan_tag); 737 738 /* Register the MAC address. */ 739 ret = rte_eth_dev_mac_addr_add(ports[0], &vdev->mac_address, 740 (uint32_t)dev->device_fh + vmdq_pool_base); 741 if (ret) 742 RTE_LOG(ERR, VHOST_DATA, "(%"PRIu64") Failed to add device MAC address to VMDQ\n", 743 dev->device_fh); 744 745 /* Enable stripping of the vlan tag as we handle routing. */ 746 if (vlan_strip) 747 rte_eth_dev_set_vlan_strip_on_queue(ports[0], 748 (uint16_t)vdev->vmdq_rx_q, 1); 749 750 /* Set device as ready for RX. */ 751 vdev->ready = DEVICE_RX; 752 753 return 0; 754 } 755 756 /* 757 * Removes MAC address and vlan tag from VMDQ. Ensures that nothing is adding buffers to the RX 758 * queue before disabling RX on the device. 759 */ 760 static inline void 761 unlink_vmdq(struct vhost_dev *vdev) 762 { 763 unsigned i = 0; 764 unsigned rx_count; 765 struct rte_mbuf *pkts_burst[MAX_PKT_BURST]; 766 767 if (vdev->ready == DEVICE_RX) { 768 /*clear MAC and VLAN settings*/ 769 rte_eth_dev_mac_addr_remove(ports[0], &vdev->mac_address); 770 for (i = 0; i < 6; i++) 771 vdev->mac_address.addr_bytes[i] = 0; 772 773 vdev->vlan_tag = 0; 774 775 /*Clear out the receive buffers*/ 776 rx_count = rte_eth_rx_burst(ports[0], 777 (uint16_t)vdev->vmdq_rx_q, pkts_burst, MAX_PKT_BURST); 778 779 while (rx_count) { 780 for (i = 0; i < rx_count; i++) 781 rte_pktmbuf_free(pkts_burst[i]); 782 783 rx_count = rte_eth_rx_burst(ports[0], 784 (uint16_t)vdev->vmdq_rx_q, pkts_burst, MAX_PKT_BURST); 785 } 786 787 vdev->ready = DEVICE_MAC_LEARNING; 788 } 789 } 790 791 static inline void __attribute__((always_inline)) 792 virtio_xmit(struct vhost_dev *dst_vdev, struct vhost_dev *src_vdev, 793 struct rte_mbuf *m) 794 { 795 uint16_t ret; 796 797 ret = rte_vhost_enqueue_burst(dst_vdev->dev, VIRTIO_RXQ, &m, 1); 798 if (enable_stats) { 799 rte_atomic64_inc(&dst_vdev->stats.rx_total_atomic); 800 rte_atomic64_add(&dst_vdev->stats.rx_atomic, ret); 801 src_vdev->stats.tx_total++; 802 src_vdev->stats.tx += ret; 803 } 804 } 805 806 /* 807 * Check if the packet destination MAC address is for a local device. If so then put 808 * the packet on that devices RX queue. If not then return. 809 */ 810 static inline int __attribute__((always_inline)) 811 virtio_tx_local(struct vhost_dev *vdev, struct rte_mbuf *m) 812 { 813 struct ether_hdr *pkt_hdr; 814 struct vhost_dev *dst_vdev; 815 uint64_t fh; 816 817 pkt_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *); 818 819 dst_vdev = find_vhost_dev(&pkt_hdr->d_addr); 820 if (!dst_vdev) 821 return -1; 822 823 fh = dst_vdev->dev->device_fh; 824 if (fh == vdev->dev->device_fh) { 825 RTE_LOG(DEBUG, VHOST_DATA, 826 "(%" PRIu64 ") TX: src and dst MAC is same. " 827 "Dropping packet.\n", fh); 828 return 0; 829 } 830 831 RTE_LOG(DEBUG, VHOST_DATA, 832 "(%" PRIu64 ") TX: MAC address is local\n", fh); 833 834 if (unlikely(dst_vdev->remove)) { 835 RTE_LOG(DEBUG, VHOST_DATA, "(%" PRIu64 ") " 836 "Device is marked for removal\n", fh); 837 return 0; 838 } 839 840 virtio_xmit(dst_vdev, vdev, m); 841 return 0; 842 } 843 844 /* 845 * Check if the destination MAC of a packet is one local VM, 846 * and get its vlan tag, and offset if it is. 847 */ 848 static inline int __attribute__((always_inline)) 849 find_local_dest(struct virtio_net *dev, struct rte_mbuf *m, 850 uint32_t *offset, uint16_t *vlan_tag) 851 { 852 struct vhost_dev *dst_vdev; 853 struct ether_hdr *pkt_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *); 854 855 dst_vdev = find_vhost_dev(&pkt_hdr->d_addr); 856 if (!dst_vdev) 857 return 0; 858 859 if (dst_vdev->dev->device_fh == dev->device_fh) { 860 RTE_LOG(DEBUG, VHOST_DATA, 861 "(%" PRIu64 ") TX: src and dst MAC is same. " 862 " Dropping packet.\n", dst_vdev->dev->device_fh); 863 return -1; 864 } 865 866 /* 867 * HW vlan strip will reduce the packet length 868 * by minus length of vlan tag, so need restore 869 * the packet length by plus it. 870 */ 871 *offset = VLAN_HLEN; 872 *vlan_tag = vlan_tags[(uint16_t)dst_vdev->dev->device_fh]; 873 874 RTE_LOG(DEBUG, VHOST_DATA, 875 "(%" PRIu64 ") TX: pkt to local VM device id: (%" PRIu64 ") " 876 "vlan tag: %u.\n", 877 dev->device_fh, dst_vdev->dev->device_fh, *vlan_tag); 878 879 return 0; 880 } 881 882 static uint16_t 883 get_psd_sum(void *l3_hdr, uint64_t ol_flags) 884 { 885 if (ol_flags & PKT_TX_IPV4) 886 return rte_ipv4_phdr_cksum(l3_hdr, ol_flags); 887 else /* assume ethertype == ETHER_TYPE_IPv6 */ 888 return rte_ipv6_phdr_cksum(l3_hdr, ol_flags); 889 } 890 891 static void virtio_tx_offload(struct rte_mbuf *m) 892 { 893 void *l3_hdr; 894 struct ipv4_hdr *ipv4_hdr = NULL; 895 struct tcp_hdr *tcp_hdr = NULL; 896 struct ether_hdr *eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *); 897 898 l3_hdr = (char *)eth_hdr + m->l2_len; 899 900 if (m->ol_flags & PKT_TX_IPV4) { 901 ipv4_hdr = l3_hdr; 902 ipv4_hdr->hdr_checksum = 0; 903 m->ol_flags |= PKT_TX_IP_CKSUM; 904 } 905 906 tcp_hdr = (struct tcp_hdr *)((char *)l3_hdr + m->l3_len); 907 tcp_hdr->cksum = get_psd_sum(l3_hdr, m->ol_flags); 908 } 909 910 static inline void 911 free_pkts(struct rte_mbuf **pkts, uint16_t n) 912 { 913 while (n--) 914 rte_pktmbuf_free(pkts[n]); 915 } 916 917 static inline void __attribute__((always_inline)) 918 do_drain_mbuf_table(struct mbuf_table *tx_q) 919 { 920 uint16_t count; 921 922 count = rte_eth_tx_burst(ports[0], tx_q->txq_id, 923 tx_q->m_table, tx_q->len); 924 if (unlikely(count < tx_q->len)) 925 free_pkts(&tx_q->m_table[count], tx_q->len - count); 926 927 tx_q->len = 0; 928 } 929 930 /* 931 * This function routes the TX packet to the correct interface. This 932 * may be a local device or the physical port. 933 */ 934 static inline void __attribute__((always_inline)) 935 virtio_tx_route(struct vhost_dev *vdev, struct rte_mbuf *m, uint16_t vlan_tag) 936 { 937 struct mbuf_table *tx_q; 938 unsigned offset = 0; 939 const uint16_t lcore_id = rte_lcore_id(); 940 struct virtio_net *dev = vdev->dev; 941 struct ether_hdr *nh; 942 943 944 nh = rte_pktmbuf_mtod(m, struct ether_hdr *); 945 if (unlikely(is_broadcast_ether_addr(&nh->d_addr))) { 946 struct vhost_dev *vdev2; 947 948 TAILQ_FOREACH(vdev2, &vhost_dev_list, next) { 949 virtio_xmit(vdev2, vdev, m); 950 } 951 goto queue2nic; 952 } 953 954 /*check if destination is local VM*/ 955 if ((vm2vm_mode == VM2VM_SOFTWARE) && (virtio_tx_local(vdev, m) == 0)) { 956 rte_pktmbuf_free(m); 957 return; 958 } 959 960 if (unlikely(vm2vm_mode == VM2VM_HARDWARE)) { 961 if (unlikely(find_local_dest(dev, m, &offset, &vlan_tag) != 0)) { 962 rte_pktmbuf_free(m); 963 return; 964 } 965 } 966 967 RTE_LOG(DEBUG, VHOST_DATA, "(%" PRIu64 ") TX: " 968 "MAC address is external\n", dev->device_fh); 969 970 queue2nic: 971 972 /*Add packet to the port tx queue*/ 973 tx_q = &lcore_tx_queue[lcore_id]; 974 975 nh = rte_pktmbuf_mtod(m, struct ether_hdr *); 976 if (unlikely(nh->ether_type == rte_cpu_to_be_16(ETHER_TYPE_VLAN))) { 977 /* Guest has inserted the vlan tag. */ 978 struct vlan_hdr *vh = (struct vlan_hdr *) (nh + 1); 979 uint16_t vlan_tag_be = rte_cpu_to_be_16(vlan_tag); 980 if ((vm2vm_mode == VM2VM_HARDWARE) && 981 (vh->vlan_tci != vlan_tag_be)) 982 vh->vlan_tci = vlan_tag_be; 983 } else { 984 m->ol_flags |= PKT_TX_VLAN_PKT; 985 986 /* 987 * Find the right seg to adjust the data len when offset is 988 * bigger than tail room size. 989 */ 990 if (unlikely(vm2vm_mode == VM2VM_HARDWARE)) { 991 if (likely(offset <= rte_pktmbuf_tailroom(m))) 992 m->data_len += offset; 993 else { 994 struct rte_mbuf *seg = m; 995 996 while ((seg->next != NULL) && 997 (offset > rte_pktmbuf_tailroom(seg))) 998 seg = seg->next; 999 1000 seg->data_len += offset; 1001 } 1002 m->pkt_len += offset; 1003 } 1004 1005 m->vlan_tci = vlan_tag; 1006 } 1007 1008 if (m->ol_flags & PKT_TX_TCP_SEG) 1009 virtio_tx_offload(m); 1010 1011 tx_q->m_table[tx_q->len++] = m; 1012 if (enable_stats) { 1013 vdev->stats.tx_total++; 1014 vdev->stats.tx++; 1015 } 1016 1017 if (unlikely(tx_q->len == MAX_PKT_BURST)) 1018 do_drain_mbuf_table(tx_q); 1019 } 1020 1021 1022 static inline void __attribute__((always_inline)) 1023 drain_mbuf_table(struct mbuf_table *tx_q) 1024 { 1025 static uint64_t prev_tsc; 1026 uint64_t cur_tsc; 1027 1028 if (tx_q->len == 0) 1029 return; 1030 1031 cur_tsc = rte_rdtsc(); 1032 if (unlikely(cur_tsc - prev_tsc > MBUF_TABLE_DRAIN_TSC)) { 1033 prev_tsc = cur_tsc; 1034 1035 RTE_LOG(DEBUG, VHOST_DATA, 1036 "TX queue drained after timeout with burst size %u\n", 1037 tx_q->len); 1038 do_drain_mbuf_table(tx_q); 1039 } 1040 } 1041 1042 static inline void __attribute__((always_inline)) 1043 drain_eth_rx(struct vhost_dev *vdev) 1044 { 1045 uint16_t rx_count, enqueue_count; 1046 struct virtio_net *dev = vdev->dev; 1047 struct rte_mbuf *pkts[MAX_PKT_BURST]; 1048 1049 rx_count = rte_eth_rx_burst(ports[0], vdev->vmdq_rx_q, 1050 pkts, MAX_PKT_BURST); 1051 if (!rx_count) 1052 return; 1053 1054 /* 1055 * When "enable_retry" is set, here we wait and retry when there 1056 * is no enough free slots in the queue to hold @rx_count packets, 1057 * to diminish packet loss. 1058 */ 1059 if (enable_retry && 1060 unlikely(rx_count > rte_vring_available_entries(dev, 1061 VIRTIO_RXQ))) { 1062 uint32_t retry; 1063 1064 for (retry = 0; retry < burst_rx_retry_num; retry++) { 1065 rte_delay_us(burst_rx_delay_time); 1066 if (rx_count <= rte_vring_available_entries(dev, 1067 VIRTIO_RXQ)) 1068 break; 1069 } 1070 } 1071 1072 enqueue_count = rte_vhost_enqueue_burst(dev, VIRTIO_RXQ, 1073 pkts, rx_count); 1074 if (enable_stats) { 1075 rte_atomic64_add(&vdev->stats.rx_total_atomic, rx_count); 1076 rte_atomic64_add(&vdev->stats.rx_atomic, enqueue_count); 1077 } 1078 1079 free_pkts(pkts, rx_count); 1080 } 1081 1082 static inline void __attribute__((always_inline)) 1083 drain_virtio_tx(struct vhost_dev *vdev) 1084 { 1085 struct rte_mbuf *pkts[MAX_PKT_BURST]; 1086 uint16_t count; 1087 uint16_t i; 1088 1089 count = rte_vhost_dequeue_burst(vdev->dev, VIRTIO_TXQ, mbuf_pool, 1090 pkts, MAX_PKT_BURST); 1091 1092 /* setup VMDq for the first packet */ 1093 if (unlikely(vdev->ready == DEVICE_MAC_LEARNING) && count) { 1094 if (vdev->remove || link_vmdq(vdev, pkts[0]) == -1) 1095 free_pkts(pkts, count); 1096 } 1097 1098 for (i = 0; i < count; ++i) { 1099 virtio_tx_route(vdev, pkts[i], 1100 vlan_tags[(uint16_t)vdev->dev->device_fh]); 1101 } 1102 } 1103 1104 /* 1105 * Main function of vhost-switch. It basically does: 1106 * 1107 * for each vhost device { 1108 * - drain_eth_rx() 1109 * 1110 * Which drains the host eth Rx queue linked to the vhost device, 1111 * and deliver all of them to guest virito Rx ring associated with 1112 * this vhost device. 1113 * 1114 * - drain_virtio_tx() 1115 * 1116 * Which drains the guest virtio Tx queue and deliver all of them 1117 * to the target, which could be another vhost device, or the 1118 * physical eth dev. The route is done in function "virtio_tx_route". 1119 * } 1120 */ 1121 static int 1122 switch_worker(void *arg __rte_unused) 1123 { 1124 unsigned i; 1125 unsigned lcore_id = rte_lcore_id(); 1126 struct vhost_dev *vdev; 1127 struct mbuf_table *tx_q; 1128 1129 RTE_LOG(INFO, VHOST_DATA, "Procesing on Core %u started\n", lcore_id); 1130 1131 tx_q = &lcore_tx_queue[lcore_id]; 1132 for (i = 0; i < rte_lcore_count(); i++) { 1133 if (lcore_ids[i] == lcore_id) { 1134 tx_q->txq_id = i; 1135 break; 1136 } 1137 } 1138 1139 while(1) { 1140 drain_mbuf_table(tx_q); 1141 1142 /* 1143 * Inform the configuration core that we have exited the 1144 * linked list and that no devices are in use if requested. 1145 */ 1146 if (lcore_info[lcore_id].dev_removal_flag == REQUEST_DEV_REMOVAL) 1147 lcore_info[lcore_id].dev_removal_flag = ACK_DEV_REMOVAL; 1148 1149 /* 1150 * Process vhost devices 1151 */ 1152 TAILQ_FOREACH(vdev, &lcore_info[lcore_id].vdev_list, next) { 1153 if (unlikely(vdev->remove)) { 1154 unlink_vmdq(vdev); 1155 vdev->ready = DEVICE_SAFE_REMOVE; 1156 continue; 1157 } 1158 1159 if (likely(vdev->ready == DEVICE_RX)) 1160 drain_eth_rx(vdev); 1161 1162 if (likely(!vdev->remove)) 1163 drain_virtio_tx(vdev); 1164 } 1165 } 1166 1167 return 0; 1168 } 1169 1170 /* 1171 * Remove a device from the specific data core linked list and from the 1172 * main linked list. Synchonization occurs through the use of the 1173 * lcore dev_removal_flag. Device is made volatile here to avoid re-ordering 1174 * of dev->remove=1 which can cause an infinite loop in the rte_pause loop. 1175 */ 1176 static void 1177 destroy_device (volatile struct virtio_net *dev) 1178 { 1179 struct vhost_dev *vdev; 1180 int lcore; 1181 1182 dev->flags &= ~VIRTIO_DEV_RUNNING; 1183 1184 vdev = (struct vhost_dev *)dev->priv; 1185 /*set the remove flag. */ 1186 vdev->remove = 1; 1187 while(vdev->ready != DEVICE_SAFE_REMOVE) { 1188 rte_pause(); 1189 } 1190 1191 TAILQ_REMOVE(&lcore_info[vdev->coreid].vdev_list, vdev, next); 1192 TAILQ_REMOVE(&vhost_dev_list, vdev, next); 1193 1194 /* Set the dev_removal_flag on each lcore. */ 1195 RTE_LCORE_FOREACH_SLAVE(lcore) 1196 lcore_info[lcore].dev_removal_flag = REQUEST_DEV_REMOVAL; 1197 1198 /* 1199 * Once each core has set the dev_removal_flag to ACK_DEV_REMOVAL 1200 * we can be sure that they can no longer access the device removed 1201 * from the linked lists and that the devices are no longer in use. 1202 */ 1203 RTE_LCORE_FOREACH_SLAVE(lcore) { 1204 while (lcore_info[lcore].dev_removal_flag != ACK_DEV_REMOVAL) 1205 rte_pause(); 1206 } 1207 1208 lcore_info[vdev->coreid].device_num--; 1209 1210 RTE_LOG(INFO, VHOST_DATA, 1211 "(%" PRIu64 ") Device has been removed from data core\n", 1212 dev->device_fh); 1213 1214 rte_free(vdev); 1215 } 1216 1217 /* 1218 * A new device is added to a data core. First the device is added to the main linked list 1219 * and the allocated to a specific data core. 1220 */ 1221 static int 1222 new_device (struct virtio_net *dev) 1223 { 1224 int lcore, core_add = 0; 1225 uint32_t device_num_min = num_devices; 1226 struct vhost_dev *vdev; 1227 1228 vdev = rte_zmalloc("vhost device", sizeof(*vdev), RTE_CACHE_LINE_SIZE); 1229 if (vdev == NULL) { 1230 RTE_LOG(INFO, VHOST_DATA, "(%"PRIu64") Couldn't allocate memory for vhost dev\n", 1231 dev->device_fh); 1232 return -1; 1233 } 1234 vdev->dev = dev; 1235 dev->priv = vdev; 1236 1237 TAILQ_INSERT_TAIL(&vhost_dev_list, vdev, next); 1238 vdev->vmdq_rx_q 1239 = dev->device_fh * queues_per_pool + vmdq_queue_base; 1240 1241 /*reset ready flag*/ 1242 vdev->ready = DEVICE_MAC_LEARNING; 1243 vdev->remove = 0; 1244 1245 /* Find a suitable lcore to add the device. */ 1246 RTE_LCORE_FOREACH_SLAVE(lcore) { 1247 if (lcore_info[lcore].device_num < device_num_min) { 1248 device_num_min = lcore_info[lcore].device_num; 1249 core_add = lcore; 1250 } 1251 } 1252 vdev->coreid = core_add; 1253 1254 TAILQ_INSERT_TAIL(&lcore_info[vdev->coreid].vdev_list, vdev, next); 1255 lcore_info[vdev->coreid].device_num++; 1256 1257 /* Disable notifications. */ 1258 rte_vhost_enable_guest_notification(dev, VIRTIO_RXQ, 0); 1259 rte_vhost_enable_guest_notification(dev, VIRTIO_TXQ, 0); 1260 dev->flags |= VIRTIO_DEV_RUNNING; 1261 1262 RTE_LOG(INFO, VHOST_DATA, "(%"PRIu64") Device has been added to data core %d\n", dev->device_fh, vdev->coreid); 1263 1264 return 0; 1265 } 1266 1267 /* 1268 * These callback allow devices to be added to the data core when configuration 1269 * has been fully complete. 1270 */ 1271 static const struct virtio_net_device_ops virtio_net_device_ops = 1272 { 1273 .new_device = new_device, 1274 .destroy_device = destroy_device, 1275 }; 1276 1277 /* 1278 * This is a thread will wake up after a period to print stats if the user has 1279 * enabled them. 1280 */ 1281 static void 1282 print_stats(void) 1283 { 1284 struct vhost_dev *vdev; 1285 uint64_t tx_dropped, rx_dropped; 1286 uint64_t tx, tx_total, rx, rx_total; 1287 const char clr[] = { 27, '[', '2', 'J', '\0' }; 1288 const char top_left[] = { 27, '[', '1', ';', '1', 'H','\0' }; 1289 1290 while(1) { 1291 sleep(enable_stats); 1292 1293 /* Clear screen and move to top left */ 1294 printf("%s%s\n", clr, top_left); 1295 printf("Device statistics =================================\n"); 1296 1297 TAILQ_FOREACH(vdev, &vhost_dev_list, next) { 1298 tx_total = vdev->stats.tx_total; 1299 tx = vdev->stats.tx; 1300 tx_dropped = tx_total - tx; 1301 1302 rx_total = rte_atomic64_read(&vdev->stats.rx_total_atomic); 1303 rx = rte_atomic64_read(&vdev->stats.rx_atomic); 1304 rx_dropped = rx_total - rx; 1305 1306 printf("Statistics for device %" PRIu64 "\n" 1307 "-----------------------\n" 1308 "TX total: %" PRIu64 "\n" 1309 "TX dropped: %" PRIu64 "\n" 1310 "TX successful: %" PRIu64 "\n" 1311 "RX total: %" PRIu64 "\n" 1312 "RX dropped: %" PRIu64 "\n" 1313 "RX successful: %" PRIu64 "\n", 1314 vdev->dev->device_fh, 1315 tx_total, tx_dropped, tx, 1316 rx_total, rx_dropped, rx); 1317 } 1318 1319 printf("===================================================\n"); 1320 } 1321 } 1322 1323 /* When we receive a INT signal, unregister vhost driver */ 1324 static void 1325 sigint_handler(__rte_unused int signum) 1326 { 1327 /* Unregister vhost driver. */ 1328 int ret = rte_vhost_driver_unregister((char *)&dev_basename); 1329 if (ret != 0) 1330 rte_exit(EXIT_FAILURE, "vhost driver unregister failure.\n"); 1331 exit(0); 1332 } 1333 1334 /* 1335 * While creating an mbuf pool, one key thing is to figure out how 1336 * many mbuf entries is enough for our use. FYI, here are some 1337 * guidelines: 1338 * 1339 * - Each rx queue would reserve @nr_rx_desc mbufs at queue setup stage 1340 * 1341 * - For each switch core (A CPU core does the packet switch), we need 1342 * also make some reservation for receiving the packets from virtio 1343 * Tx queue. How many is enough depends on the usage. It's normally 1344 * a simple calculation like following: 1345 * 1346 * MAX_PKT_BURST * max packet size / mbuf size 1347 * 1348 * So, we definitely need allocate more mbufs when TSO is enabled. 1349 * 1350 * - Similarly, for each switching core, we should serve @nr_rx_desc 1351 * mbufs for receiving the packets from physical NIC device. 1352 * 1353 * - We also need make sure, for each switch core, we have allocated 1354 * enough mbufs to fill up the mbuf cache. 1355 */ 1356 static void 1357 create_mbuf_pool(uint16_t nr_port, uint32_t nr_switch_core, uint32_t mbuf_size, 1358 uint32_t nr_queues, uint32_t nr_rx_desc, uint32_t nr_mbuf_cache) 1359 { 1360 uint32_t nr_mbufs; 1361 uint32_t nr_mbufs_per_core; 1362 uint32_t mtu = 1500; 1363 1364 if (mergeable) 1365 mtu = 9000; 1366 if (enable_tso) 1367 mtu = 64 * 1024; 1368 1369 nr_mbufs_per_core = (mtu + mbuf_size) * MAX_PKT_BURST / 1370 (mbuf_size - RTE_PKTMBUF_HEADROOM) * MAX_PKT_BURST; 1371 nr_mbufs_per_core += nr_rx_desc; 1372 nr_mbufs_per_core = RTE_MAX(nr_mbufs_per_core, nr_mbuf_cache); 1373 1374 nr_mbufs = nr_queues * nr_rx_desc; 1375 nr_mbufs += nr_mbufs_per_core * nr_switch_core; 1376 nr_mbufs *= nr_port; 1377 1378 mbuf_pool = rte_pktmbuf_pool_create("MBUF_POOL", nr_mbufs, 1379 nr_mbuf_cache, 0, mbuf_size, 1380 rte_socket_id()); 1381 if (mbuf_pool == NULL) 1382 rte_exit(EXIT_FAILURE, "Cannot create mbuf pool\n"); 1383 } 1384 1385 /* 1386 * Main function, does initialisation and calls the per-lcore functions. The CUSE 1387 * device is also registered here to handle the IOCTLs. 1388 */ 1389 int 1390 main(int argc, char *argv[]) 1391 { 1392 unsigned lcore_id, core_id = 0; 1393 unsigned nb_ports, valid_num_ports; 1394 int ret; 1395 uint8_t portid; 1396 static pthread_t tid; 1397 char thread_name[RTE_MAX_THREAD_NAME_LEN]; 1398 1399 signal(SIGINT, sigint_handler); 1400 1401 /* init EAL */ 1402 ret = rte_eal_init(argc, argv); 1403 if (ret < 0) 1404 rte_exit(EXIT_FAILURE, "Error with EAL initialization\n"); 1405 argc -= ret; 1406 argv += ret; 1407 1408 /* parse app arguments */ 1409 ret = us_vhost_parse_args(argc, argv); 1410 if (ret < 0) 1411 rte_exit(EXIT_FAILURE, "Invalid argument\n"); 1412 1413 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id ++) 1414 TAILQ_INIT(&lcore_info[lcore_id].vdev_list); 1415 1416 if (rte_lcore_is_enabled(lcore_id)) 1417 lcore_ids[core_id ++] = lcore_id; 1418 1419 if (rte_lcore_count() > RTE_MAX_LCORE) 1420 rte_exit(EXIT_FAILURE,"Not enough cores\n"); 1421 1422 /* Get the number of physical ports. */ 1423 nb_ports = rte_eth_dev_count(); 1424 1425 /* 1426 * Update the global var NUM_PORTS and global array PORTS 1427 * and get value of var VALID_NUM_PORTS according to system ports number 1428 */ 1429 valid_num_ports = check_ports_num(nb_ports); 1430 1431 if ((valid_num_ports == 0) || (valid_num_ports > MAX_SUP_PORTS)) { 1432 RTE_LOG(INFO, VHOST_PORT, "Current enabled port number is %u," 1433 "but only %u port can be enabled\n",num_ports, MAX_SUP_PORTS); 1434 return -1; 1435 } 1436 1437 /* 1438 * FIXME: here we are trying to allocate mbufs big enough for 1439 * @MAX_QUEUES, but the truth is we're never going to use that 1440 * many queues here. We probably should only do allocation for 1441 * those queues we are going to use. 1442 */ 1443 create_mbuf_pool(valid_num_ports, rte_lcore_count() - 1, MBUF_DATA_SIZE, 1444 MAX_QUEUES, RTE_TEST_RX_DESC_DEFAULT, MBUF_CACHE_SIZE); 1445 1446 if (vm2vm_mode == VM2VM_HARDWARE) { 1447 /* Enable VT loop back to let L2 switch to do it. */ 1448 vmdq_conf_default.rx_adv_conf.vmdq_rx_conf.enable_loop_back = 1; 1449 RTE_LOG(DEBUG, VHOST_CONFIG, 1450 "Enable loop back for L2 switch in vmdq.\n"); 1451 } 1452 1453 /* initialize all ports */ 1454 for (portid = 0; portid < nb_ports; portid++) { 1455 /* skip ports that are not enabled */ 1456 if ((enabled_port_mask & (1 << portid)) == 0) { 1457 RTE_LOG(INFO, VHOST_PORT, 1458 "Skipping disabled port %d\n", portid); 1459 continue; 1460 } 1461 if (port_init(portid) != 0) 1462 rte_exit(EXIT_FAILURE, 1463 "Cannot initialize network ports\n"); 1464 } 1465 1466 /* Enable stats if the user option is set. */ 1467 if (enable_stats) { 1468 ret = pthread_create(&tid, NULL, (void *)print_stats, NULL); 1469 if (ret != 0) 1470 rte_exit(EXIT_FAILURE, 1471 "Cannot create print-stats thread\n"); 1472 1473 /* Set thread_name for aid in debugging. */ 1474 snprintf(thread_name, RTE_MAX_THREAD_NAME_LEN, "print-stats"); 1475 ret = rte_thread_setname(tid, thread_name); 1476 if (ret != 0) 1477 RTE_LOG(ERR, VHOST_CONFIG, 1478 "Cannot set print-stats name\n"); 1479 } 1480 1481 /* Launch all data cores. */ 1482 RTE_LCORE_FOREACH_SLAVE(lcore_id) 1483 rte_eal_remote_launch(switch_worker, NULL, lcore_id); 1484 1485 if (mergeable == 0) 1486 rte_vhost_feature_disable(1ULL << VIRTIO_NET_F_MRG_RXBUF); 1487 1488 /* Register vhost(cuse or user) driver to handle vhost messages. */ 1489 ret = rte_vhost_driver_register((char *)&dev_basename); 1490 if (ret != 0) 1491 rte_exit(EXIT_FAILURE, "vhost driver register failure.\n"); 1492 1493 rte_vhost_driver_callback_register(&virtio_net_device_ops); 1494 1495 /* Start CUSE session. */ 1496 rte_vhost_driver_session_start(); 1497 return 0; 1498 1499 } 1500