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 28 #include "ioat.h" 29 #include "main.h" 30 31 #ifndef MAX_QUEUES 32 #define MAX_QUEUES 128 33 #endif 34 35 /* the maximum number of external ports supported */ 36 #define MAX_SUP_PORTS 1 37 38 #define MBUF_CACHE_SIZE 128 39 #define MBUF_DATA_SIZE RTE_MBUF_DEFAULT_BUF_SIZE 40 41 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */ 42 43 #define BURST_RX_WAIT_US 15 /* Defines how long we wait between retries on RX */ 44 #define BURST_RX_RETRIES 4 /* Number of retries on RX. */ 45 46 #define JUMBO_FRAME_MAX_SIZE 0x2600 47 #define MAX_MTU (JUMBO_FRAME_MAX_SIZE - (RTE_ETHER_HDR_LEN + RTE_ETHER_CRC_LEN)) 48 49 /* State of virtio device. */ 50 #define DEVICE_MAC_LEARNING 0 51 #define DEVICE_RX 1 52 #define DEVICE_SAFE_REMOVE 2 53 54 /* Configurable number of RX/TX ring descriptors */ 55 #define RTE_TEST_RX_DESC_DEFAULT 1024 56 #define RTE_TEST_TX_DESC_DEFAULT 512 57 58 #define INVALID_PORT_ID 0xFF 59 60 /* mask of enabled ports */ 61 static uint32_t enabled_port_mask = 0; 62 63 /* Promiscuous mode */ 64 static uint32_t promiscuous; 65 66 /* number of devices/queues to support*/ 67 static uint32_t num_queues = 0; 68 static uint32_t num_devices; 69 70 static struct rte_mempool *mbuf_pool; 71 static int mergeable; 72 73 /* Enable VM2VM communications. If this is disabled then the MAC address compare is skipped. */ 74 typedef enum { 75 VM2VM_DISABLED = 0, 76 VM2VM_SOFTWARE = 1, 77 VM2VM_HARDWARE = 2, 78 VM2VM_LAST 79 } vm2vm_type; 80 static vm2vm_type vm2vm_mode = VM2VM_SOFTWARE; 81 82 /* Enable stats. */ 83 static uint32_t enable_stats = 0; 84 /* Enable retries on RX. */ 85 static uint32_t enable_retry = 1; 86 87 /* Disable TX checksum offload */ 88 static uint32_t enable_tx_csum; 89 90 /* Disable TSO offload */ 91 static uint32_t enable_tso; 92 93 static int client_mode; 94 95 static int builtin_net_driver; 96 97 static int async_vhost_driver; 98 99 static char *dma_type; 100 101 /* Specify timeout (in useconds) between retries on RX. */ 102 static uint32_t burst_rx_delay_time = BURST_RX_WAIT_US; 103 /* Specify the number of retries on RX. */ 104 static uint32_t burst_rx_retry_num = BURST_RX_RETRIES; 105 106 /* Socket file paths. Can be set by user */ 107 static char *socket_files; 108 static int nb_sockets; 109 110 /* empty vmdq configuration structure. Filled in programatically */ 111 static struct rte_eth_conf vmdq_conf_default = { 112 .rxmode = { 113 .mq_mode = RTE_ETH_MQ_RX_VMDQ_ONLY, 114 .split_hdr_size = 0, 115 /* 116 * VLAN strip is necessary for 1G NIC such as I350, 117 * this fixes bug of ipv4 forwarding in guest can't 118 * forward pakets from one virtio dev to another virtio dev. 119 */ 120 .offloads = RTE_ETH_RX_OFFLOAD_VLAN_STRIP, 121 }, 122 123 .txmode = { 124 .mq_mode = RTE_ETH_MQ_TX_NONE, 125 .offloads = (RTE_ETH_TX_OFFLOAD_IPV4_CKSUM | 126 RTE_ETH_TX_OFFLOAD_TCP_CKSUM | 127 RTE_ETH_TX_OFFLOAD_VLAN_INSERT | 128 RTE_ETH_TX_OFFLOAD_MULTI_SEGS | 129 RTE_ETH_TX_OFFLOAD_TCP_TSO), 130 }, 131 .rx_adv_conf = { 132 /* 133 * should be overridden separately in code with 134 * appropriate values 135 */ 136 .vmdq_rx_conf = { 137 .nb_queue_pools = RTE_ETH_8_POOLS, 138 .enable_default_pool = 0, 139 .default_pool = 0, 140 .nb_pool_maps = 0, 141 .pool_map = {{0, 0},}, 142 }, 143 }, 144 }; 145 146 147 static unsigned lcore_ids[RTE_MAX_LCORE]; 148 static uint16_t ports[RTE_MAX_ETHPORTS]; 149 static unsigned num_ports = 0; /**< The number of ports specified in command line */ 150 static uint16_t num_pf_queues, num_vmdq_queues; 151 static uint16_t vmdq_pool_base, vmdq_queue_base; 152 static uint16_t queues_per_pool; 153 154 const uint16_t vlan_tags[] = { 155 1000, 1001, 1002, 1003, 1004, 1005, 1006, 1007, 156 1008, 1009, 1010, 1011, 1012, 1013, 1014, 1015, 157 1016, 1017, 1018, 1019, 1020, 1021, 1022, 1023, 158 1024, 1025, 1026, 1027, 1028, 1029, 1030, 1031, 159 1032, 1033, 1034, 1035, 1036, 1037, 1038, 1039, 160 1040, 1041, 1042, 1043, 1044, 1045, 1046, 1047, 161 1048, 1049, 1050, 1051, 1052, 1053, 1054, 1055, 162 1056, 1057, 1058, 1059, 1060, 1061, 1062, 1063, 163 }; 164 165 /* ethernet addresses of ports */ 166 static struct rte_ether_addr vmdq_ports_eth_addr[RTE_MAX_ETHPORTS]; 167 168 static struct vhost_dev_tailq_list vhost_dev_list = 169 TAILQ_HEAD_INITIALIZER(vhost_dev_list); 170 171 static struct lcore_info lcore_info[RTE_MAX_LCORE]; 172 173 /* Used for queueing bursts of TX packets. */ 174 struct mbuf_table { 175 unsigned len; 176 unsigned txq_id; 177 struct rte_mbuf *m_table[MAX_PKT_BURST]; 178 }; 179 180 struct vhost_bufftable { 181 uint32_t len; 182 uint64_t pre_tsc; 183 struct rte_mbuf *m_table[MAX_PKT_BURST]; 184 }; 185 186 /* TX queue for each data core. */ 187 struct mbuf_table lcore_tx_queue[RTE_MAX_LCORE]; 188 189 /* 190 * Vhost TX buffer for each data core. 191 * Every data core maintains a TX buffer for every vhost device, 192 * which is used for batch pkts enqueue for higher performance. 193 */ 194 struct vhost_bufftable *vhost_txbuff[RTE_MAX_LCORE * MAX_VHOST_DEVICE]; 195 196 #define MBUF_TABLE_DRAIN_TSC ((rte_get_tsc_hz() + US_PER_S - 1) \ 197 / US_PER_S * BURST_TX_DRAIN_US) 198 #define VLAN_HLEN 4 199 200 static inline int 201 open_dma(const char *value) 202 { 203 if (dma_type != NULL && strncmp(dma_type, "ioat", 4) == 0) 204 return open_ioat(value); 205 206 return -1; 207 } 208 209 /* 210 * Builds up the correct configuration for VMDQ VLAN pool map 211 * according to the pool & queue limits. 212 */ 213 static inline int 214 get_eth_conf(struct rte_eth_conf *eth_conf, uint32_t num_devices) 215 { 216 struct rte_eth_vmdq_rx_conf conf; 217 struct rte_eth_vmdq_rx_conf *def_conf = 218 &vmdq_conf_default.rx_adv_conf.vmdq_rx_conf; 219 unsigned i; 220 221 memset(&conf, 0, sizeof(conf)); 222 conf.nb_queue_pools = (enum rte_eth_nb_pools)num_devices; 223 conf.nb_pool_maps = num_devices; 224 conf.enable_loop_back = def_conf->enable_loop_back; 225 conf.rx_mode = def_conf->rx_mode; 226 227 for (i = 0; i < conf.nb_pool_maps; i++) { 228 conf.pool_map[i].vlan_id = vlan_tags[ i ]; 229 conf.pool_map[i].pools = (1UL << i); 230 } 231 232 (void)(rte_memcpy(eth_conf, &vmdq_conf_default, sizeof(*eth_conf))); 233 (void)(rte_memcpy(ð_conf->rx_adv_conf.vmdq_rx_conf, &conf, 234 sizeof(eth_conf->rx_adv_conf.vmdq_rx_conf))); 235 return 0; 236 } 237 238 /* 239 * Initialises a given port using global settings and with the rx buffers 240 * coming from the mbuf_pool passed as parameter 241 */ 242 static inline int 243 port_init(uint16_t port) 244 { 245 struct rte_eth_dev_info dev_info; 246 struct rte_eth_conf port_conf; 247 struct rte_eth_rxconf *rxconf; 248 struct rte_eth_txconf *txconf; 249 int16_t rx_rings, tx_rings; 250 uint16_t rx_ring_size, tx_ring_size; 251 int retval; 252 uint16_t q; 253 254 /* The max pool number from dev_info will be used to validate the pool number specified in cmd line */ 255 retval = rte_eth_dev_info_get(port, &dev_info); 256 if (retval != 0) { 257 RTE_LOG(ERR, VHOST_PORT, 258 "Error during getting device (port %u) info: %s\n", 259 port, strerror(-retval)); 260 261 return retval; 262 } 263 264 rxconf = &dev_info.default_rxconf; 265 txconf = &dev_info.default_txconf; 266 rxconf->rx_drop_en = 1; 267 268 /*configure the number of supported virtio devices based on VMDQ limits */ 269 num_devices = dev_info.max_vmdq_pools; 270 271 rx_ring_size = RTE_TEST_RX_DESC_DEFAULT; 272 tx_ring_size = RTE_TEST_TX_DESC_DEFAULT; 273 274 tx_rings = (uint16_t)rte_lcore_count(); 275 276 /* Get port configuration. */ 277 retval = get_eth_conf(&port_conf, num_devices); 278 if (retval < 0) 279 return retval; 280 /* NIC queues are divided into pf queues and vmdq queues. */ 281 num_pf_queues = dev_info.max_rx_queues - dev_info.vmdq_queue_num; 282 queues_per_pool = dev_info.vmdq_queue_num / dev_info.max_vmdq_pools; 283 num_vmdq_queues = num_devices * queues_per_pool; 284 num_queues = num_pf_queues + num_vmdq_queues; 285 vmdq_queue_base = dev_info.vmdq_queue_base; 286 vmdq_pool_base = dev_info.vmdq_pool_base; 287 printf("pf queue num: %u, configured vmdq pool num: %u, each vmdq pool has %u queues\n", 288 num_pf_queues, num_devices, queues_per_pool); 289 290 if (!rte_eth_dev_is_valid_port(port)) 291 return -1; 292 293 rx_rings = (uint16_t)dev_info.max_rx_queues; 294 if (dev_info.tx_offload_capa & RTE_ETH_TX_OFFLOAD_MBUF_FAST_FREE) 295 port_conf.txmode.offloads |= 296 RTE_ETH_TX_OFFLOAD_MBUF_FAST_FREE; 297 /* Configure ethernet device. */ 298 retval = rte_eth_dev_configure(port, rx_rings, tx_rings, &port_conf); 299 if (retval != 0) { 300 RTE_LOG(ERR, VHOST_PORT, "Failed to configure port %u: %s.\n", 301 port, strerror(-retval)); 302 return retval; 303 } 304 305 retval = rte_eth_dev_adjust_nb_rx_tx_desc(port, &rx_ring_size, 306 &tx_ring_size); 307 if (retval != 0) { 308 RTE_LOG(ERR, VHOST_PORT, "Failed to adjust number of descriptors " 309 "for port %u: %s.\n", port, strerror(-retval)); 310 return retval; 311 } 312 if (rx_ring_size > RTE_TEST_RX_DESC_DEFAULT) { 313 RTE_LOG(ERR, VHOST_PORT, "Mbuf pool has an insufficient size " 314 "for Rx queues on port %u.\n", port); 315 return -1; 316 } 317 318 /* Setup the queues. */ 319 rxconf->offloads = port_conf.rxmode.offloads; 320 for (q = 0; q < rx_rings; q ++) { 321 retval = rte_eth_rx_queue_setup(port, q, rx_ring_size, 322 rte_eth_dev_socket_id(port), 323 rxconf, 324 mbuf_pool); 325 if (retval < 0) { 326 RTE_LOG(ERR, VHOST_PORT, 327 "Failed to setup rx queue %u of port %u: %s.\n", 328 q, port, strerror(-retval)); 329 return retval; 330 } 331 } 332 txconf->offloads = port_conf.txmode.offloads; 333 for (q = 0; q < tx_rings; q ++) { 334 retval = rte_eth_tx_queue_setup(port, q, tx_ring_size, 335 rte_eth_dev_socket_id(port), 336 txconf); 337 if (retval < 0) { 338 RTE_LOG(ERR, VHOST_PORT, 339 "Failed to setup tx queue %u of port %u: %s.\n", 340 q, port, strerror(-retval)); 341 return retval; 342 } 343 } 344 345 /* Start the device. */ 346 retval = rte_eth_dev_start(port); 347 if (retval < 0) { 348 RTE_LOG(ERR, VHOST_PORT, "Failed to start port %u: %s\n", 349 port, strerror(-retval)); 350 return retval; 351 } 352 353 if (promiscuous) { 354 retval = rte_eth_promiscuous_enable(port); 355 if (retval != 0) { 356 RTE_LOG(ERR, VHOST_PORT, 357 "Failed to enable promiscuous mode on port %u: %s\n", 358 port, rte_strerror(-retval)); 359 return retval; 360 } 361 } 362 363 retval = rte_eth_macaddr_get(port, &vmdq_ports_eth_addr[port]); 364 if (retval < 0) { 365 RTE_LOG(ERR, VHOST_PORT, 366 "Failed to get MAC address on port %u: %s\n", 367 port, rte_strerror(-retval)); 368 return retval; 369 } 370 371 RTE_LOG(INFO, VHOST_PORT, "Max virtio devices supported: %u\n", num_devices); 372 RTE_LOG(INFO, VHOST_PORT, "Port %u MAC: %02"PRIx8" %02"PRIx8" %02"PRIx8 373 " %02"PRIx8" %02"PRIx8" %02"PRIx8"\n", 374 port, RTE_ETHER_ADDR_BYTES(&vmdq_ports_eth_addr[port])); 375 376 return 0; 377 } 378 379 /* 380 * Set socket file path. 381 */ 382 static int 383 us_vhost_parse_socket_path(const char *q_arg) 384 { 385 char *old; 386 387 /* parse number string */ 388 if (strnlen(q_arg, PATH_MAX) == PATH_MAX) 389 return -1; 390 391 old = socket_files; 392 socket_files = realloc(socket_files, PATH_MAX * (nb_sockets + 1)); 393 if (socket_files == NULL) { 394 free(old); 395 return -1; 396 } 397 398 strlcpy(socket_files + nb_sockets * PATH_MAX, q_arg, PATH_MAX); 399 nb_sockets++; 400 401 return 0; 402 } 403 404 /* 405 * Parse the portmask provided at run time. 406 */ 407 static int 408 parse_portmask(const char *portmask) 409 { 410 char *end = NULL; 411 unsigned long pm; 412 413 errno = 0; 414 415 /* parse hexadecimal string */ 416 pm = strtoul(portmask, &end, 16); 417 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0') || (errno != 0)) 418 return 0; 419 420 return pm; 421 422 } 423 424 /* 425 * Parse num options at run time. 426 */ 427 static int 428 parse_num_opt(const char *q_arg, uint32_t max_valid_value) 429 { 430 char *end = NULL; 431 unsigned long num; 432 433 errno = 0; 434 435 /* parse unsigned int string */ 436 num = strtoul(q_arg, &end, 10); 437 if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0') || (errno != 0)) 438 return -1; 439 440 if (num > max_valid_value) 441 return -1; 442 443 return num; 444 445 } 446 447 /* 448 * Display usage 449 */ 450 static void 451 us_vhost_usage(const char *prgname) 452 { 453 RTE_LOG(INFO, VHOST_CONFIG, "%s [EAL options] -- -p PORTMASK\n" 454 " --vm2vm [0|1|2]\n" 455 " --rx_retry [0|1] --mergeable [0|1] --stats [0-N]\n" 456 " --socket-file <path>\n" 457 " --nb-devices ND\n" 458 " -p PORTMASK: Set mask for ports to be used by application\n" 459 " --vm2vm [0|1|2]: disable/software(default)/hardware vm2vm comms\n" 460 " --rx-retry [0|1]: disable/enable(default) retries on rx. Enable retry if destintation queue is full\n" 461 " --rx-retry-delay [0-N]: timeout(in usecond) between retries on RX. This makes effect only if retries on rx enabled\n" 462 " --rx-retry-num [0-N]: the number of retries on rx. This makes effect only if retries on rx enabled\n" 463 " --mergeable [0|1]: disable(default)/enable RX mergeable buffers\n" 464 " --stats [0-N]: 0: Disable stats, N: Time in seconds to print stats\n" 465 " --socket-file: The path of the socket file.\n" 466 " --tx-csum [0|1] disable/enable TX checksum offload.\n" 467 " --tso [0|1] disable/enable TCP segment offload.\n" 468 " --client register a vhost-user socket as client mode.\n" 469 " --dma-type register dma type for your vhost async driver. For example \"ioat\" for now.\n" 470 " --dmas register dma channel for specific vhost device.\n", 471 prgname); 472 } 473 474 enum { 475 #define OPT_VM2VM "vm2vm" 476 OPT_VM2VM_NUM = 256, 477 #define OPT_RX_RETRY "rx-retry" 478 OPT_RX_RETRY_NUM, 479 #define OPT_RX_RETRY_DELAY "rx-retry-delay" 480 OPT_RX_RETRY_DELAY_NUM, 481 #define OPT_RX_RETRY_NUMB "rx-retry-num" 482 OPT_RX_RETRY_NUMB_NUM, 483 #define OPT_MERGEABLE "mergeable" 484 OPT_MERGEABLE_NUM, 485 #define OPT_STATS "stats" 486 OPT_STATS_NUM, 487 #define OPT_SOCKET_FILE "socket-file" 488 OPT_SOCKET_FILE_NUM, 489 #define OPT_TX_CSUM "tx-csum" 490 OPT_TX_CSUM_NUM, 491 #define OPT_TSO "tso" 492 OPT_TSO_NUM, 493 #define OPT_CLIENT "client" 494 OPT_CLIENT_NUM, 495 #define OPT_BUILTIN_NET_DRIVER "builtin-net-driver" 496 OPT_BUILTIN_NET_DRIVER_NUM, 497 #define OPT_DMA_TYPE "dma-type" 498 OPT_DMA_TYPE_NUM, 499 #define OPT_DMAS "dmas" 500 OPT_DMAS_NUM, 501 }; 502 503 /* 504 * Parse the arguments given in the command line of the application. 505 */ 506 static int 507 us_vhost_parse_args(int argc, char **argv) 508 { 509 int opt, ret; 510 int option_index; 511 unsigned i; 512 const char *prgname = argv[0]; 513 static struct option long_option[] = { 514 {OPT_VM2VM, required_argument, 515 NULL, OPT_VM2VM_NUM}, 516 {OPT_RX_RETRY, required_argument, 517 NULL, OPT_RX_RETRY_NUM}, 518 {OPT_RX_RETRY_DELAY, required_argument, 519 NULL, OPT_RX_RETRY_DELAY_NUM}, 520 {OPT_RX_RETRY_NUMB, required_argument, 521 NULL, OPT_RX_RETRY_NUMB_NUM}, 522 {OPT_MERGEABLE, required_argument, 523 NULL, OPT_MERGEABLE_NUM}, 524 {OPT_STATS, required_argument, 525 NULL, OPT_STATS_NUM}, 526 {OPT_SOCKET_FILE, required_argument, 527 NULL, OPT_SOCKET_FILE_NUM}, 528 {OPT_TX_CSUM, required_argument, 529 NULL, OPT_TX_CSUM_NUM}, 530 {OPT_TSO, required_argument, 531 NULL, OPT_TSO_NUM}, 532 {OPT_CLIENT, no_argument, 533 NULL, OPT_CLIENT_NUM}, 534 {OPT_BUILTIN_NET_DRIVER, no_argument, 535 NULL, OPT_BUILTIN_NET_DRIVER_NUM}, 536 {OPT_DMA_TYPE, required_argument, 537 NULL, OPT_DMA_TYPE_NUM}, 538 {OPT_DMAS, required_argument, 539 NULL, OPT_DMAS_NUM}, 540 {NULL, 0, 0, 0}, 541 }; 542 543 /* Parse command line */ 544 while ((opt = getopt_long(argc, argv, "p:P", 545 long_option, &option_index)) != EOF) { 546 switch (opt) { 547 /* Portmask */ 548 case 'p': 549 enabled_port_mask = parse_portmask(optarg); 550 if (enabled_port_mask == 0) { 551 RTE_LOG(INFO, VHOST_CONFIG, "Invalid portmask\n"); 552 us_vhost_usage(prgname); 553 return -1; 554 } 555 break; 556 557 case 'P': 558 promiscuous = 1; 559 vmdq_conf_default.rx_adv_conf.vmdq_rx_conf.rx_mode = 560 RTE_ETH_VMDQ_ACCEPT_BROADCAST | 561 RTE_ETH_VMDQ_ACCEPT_MULTICAST; 562 break; 563 564 case OPT_VM2VM_NUM: 565 ret = parse_num_opt(optarg, (VM2VM_LAST - 1)); 566 if (ret == -1) { 567 RTE_LOG(INFO, VHOST_CONFIG, 568 "Invalid argument for " 569 "vm2vm [0|1|2]\n"); 570 us_vhost_usage(prgname); 571 return -1; 572 } 573 vm2vm_mode = (vm2vm_type)ret; 574 break; 575 576 case OPT_RX_RETRY_NUM: 577 ret = parse_num_opt(optarg, 1); 578 if (ret == -1) { 579 RTE_LOG(INFO, VHOST_CONFIG, "Invalid argument for rx-retry [0|1]\n"); 580 us_vhost_usage(prgname); 581 return -1; 582 } 583 enable_retry = ret; 584 break; 585 586 case OPT_TX_CSUM_NUM: 587 ret = parse_num_opt(optarg, 1); 588 if (ret == -1) { 589 RTE_LOG(INFO, VHOST_CONFIG, "Invalid argument for tx-csum [0|1]\n"); 590 us_vhost_usage(prgname); 591 return -1; 592 } 593 enable_tx_csum = ret; 594 break; 595 596 case OPT_TSO_NUM: 597 ret = parse_num_opt(optarg, 1); 598 if (ret == -1) { 599 RTE_LOG(INFO, VHOST_CONFIG, "Invalid argument for tso [0|1]\n"); 600 us_vhost_usage(prgname); 601 return -1; 602 } 603 enable_tso = ret; 604 break; 605 606 case OPT_RX_RETRY_DELAY_NUM: 607 ret = parse_num_opt(optarg, INT32_MAX); 608 if (ret == -1) { 609 RTE_LOG(INFO, VHOST_CONFIG, "Invalid argument for rx-retry-delay [0-N]\n"); 610 us_vhost_usage(prgname); 611 return -1; 612 } 613 burst_rx_delay_time = ret; 614 break; 615 616 case OPT_RX_RETRY_NUMB_NUM: 617 ret = parse_num_opt(optarg, INT32_MAX); 618 if (ret == -1) { 619 RTE_LOG(INFO, VHOST_CONFIG, "Invalid argument for rx-retry-num [0-N]\n"); 620 us_vhost_usage(prgname); 621 return -1; 622 } 623 burst_rx_retry_num = ret; 624 break; 625 626 case OPT_MERGEABLE_NUM: 627 ret = parse_num_opt(optarg, 1); 628 if (ret == -1) { 629 RTE_LOG(INFO, VHOST_CONFIG, "Invalid argument for mergeable [0|1]\n"); 630 us_vhost_usage(prgname); 631 return -1; 632 } 633 mergeable = !!ret; 634 if (ret) 635 vmdq_conf_default.rxmode.mtu = MAX_MTU; 636 break; 637 638 case OPT_STATS_NUM: 639 ret = parse_num_opt(optarg, INT32_MAX); 640 if (ret == -1) { 641 RTE_LOG(INFO, VHOST_CONFIG, 642 "Invalid argument for stats [0..N]\n"); 643 us_vhost_usage(prgname); 644 return -1; 645 } 646 enable_stats = ret; 647 break; 648 649 /* Set socket file path. */ 650 case OPT_SOCKET_FILE_NUM: 651 if (us_vhost_parse_socket_path(optarg) == -1) { 652 RTE_LOG(INFO, VHOST_CONFIG, 653 "Invalid argument for socket name (Max %d characters)\n", 654 PATH_MAX); 655 us_vhost_usage(prgname); 656 return -1; 657 } 658 break; 659 660 case OPT_DMA_TYPE_NUM: 661 dma_type = optarg; 662 break; 663 664 case OPT_DMAS_NUM: 665 if (open_dma(optarg) == -1) { 666 RTE_LOG(INFO, VHOST_CONFIG, 667 "Wrong DMA args\n"); 668 us_vhost_usage(prgname); 669 return -1; 670 } 671 async_vhost_driver = 1; 672 break; 673 674 case OPT_CLIENT_NUM: 675 client_mode = 1; 676 break; 677 678 case OPT_BUILTIN_NET_DRIVER_NUM: 679 builtin_net_driver = 1; 680 break; 681 682 /* Invalid option - print options. */ 683 default: 684 us_vhost_usage(prgname); 685 return -1; 686 } 687 } 688 689 for (i = 0; i < RTE_MAX_ETHPORTS; i++) { 690 if (enabled_port_mask & (1 << i)) 691 ports[num_ports++] = i; 692 } 693 694 if ((num_ports == 0) || (num_ports > MAX_SUP_PORTS)) { 695 RTE_LOG(INFO, VHOST_PORT, "Current enabled port number is %u," 696 "but only %u port can be enabled\n",num_ports, MAX_SUP_PORTS); 697 return -1; 698 } 699 700 return 0; 701 } 702 703 /* 704 * Update the global var NUM_PORTS and array PORTS according to system ports number 705 * and return valid ports number 706 */ 707 static unsigned check_ports_num(unsigned nb_ports) 708 { 709 unsigned valid_num_ports = num_ports; 710 unsigned portid; 711 712 if (num_ports > nb_ports) { 713 RTE_LOG(INFO, VHOST_PORT, "\nSpecified port number(%u) exceeds total system port number(%u)\n", 714 num_ports, nb_ports); 715 num_ports = nb_ports; 716 } 717 718 for (portid = 0; portid < num_ports; portid ++) { 719 if (!rte_eth_dev_is_valid_port(ports[portid])) { 720 RTE_LOG(INFO, VHOST_PORT, 721 "\nSpecified port ID(%u) is not valid\n", 722 ports[portid]); 723 ports[portid] = INVALID_PORT_ID; 724 valid_num_ports--; 725 } 726 } 727 return valid_num_ports; 728 } 729 730 static __rte_always_inline struct vhost_dev * 731 find_vhost_dev(struct rte_ether_addr *mac) 732 { 733 struct vhost_dev *vdev; 734 735 TAILQ_FOREACH(vdev, &vhost_dev_list, global_vdev_entry) { 736 if (vdev->ready == DEVICE_RX && 737 rte_is_same_ether_addr(mac, &vdev->mac_address)) 738 return vdev; 739 } 740 741 return NULL; 742 } 743 744 /* 745 * This function learns the MAC address of the device and registers this along with a 746 * vlan tag to a VMDQ. 747 */ 748 static int 749 link_vmdq(struct vhost_dev *vdev, struct rte_mbuf *m) 750 { 751 struct rte_ether_hdr *pkt_hdr; 752 int i, ret; 753 754 /* Learn MAC address of guest device from packet */ 755 pkt_hdr = rte_pktmbuf_mtod(m, struct rte_ether_hdr *); 756 757 if (find_vhost_dev(&pkt_hdr->src_addr)) { 758 RTE_LOG(ERR, VHOST_DATA, 759 "(%d) device is using a registered MAC!\n", 760 vdev->vid); 761 return -1; 762 } 763 764 for (i = 0; i < RTE_ETHER_ADDR_LEN; i++) 765 vdev->mac_address.addr_bytes[i] = 766 pkt_hdr->src_addr.addr_bytes[i]; 767 768 /* vlan_tag currently uses the device_id. */ 769 vdev->vlan_tag = vlan_tags[vdev->vid]; 770 771 /* Print out VMDQ registration info. */ 772 RTE_LOG(INFO, VHOST_DATA, 773 "(%d) mac " RTE_ETHER_ADDR_PRT_FMT " and vlan %d registered\n", 774 vdev->vid, RTE_ETHER_ADDR_BYTES(&vdev->mac_address), 775 vdev->vlan_tag); 776 777 /* Register the MAC address. */ 778 ret = rte_eth_dev_mac_addr_add(ports[0], &vdev->mac_address, 779 (uint32_t)vdev->vid + vmdq_pool_base); 780 if (ret) 781 RTE_LOG(ERR, VHOST_DATA, 782 "(%d) failed to add device MAC address to VMDQ\n", 783 vdev->vid); 784 785 rte_eth_dev_set_vlan_strip_on_queue(ports[0], vdev->vmdq_rx_q, 1); 786 787 /* Set device as ready for RX. */ 788 vdev->ready = DEVICE_RX; 789 790 return 0; 791 } 792 793 /* 794 * Removes MAC address and vlan tag from VMDQ. Ensures that nothing is adding buffers to the RX 795 * queue before disabling RX on the device. 796 */ 797 static inline void 798 unlink_vmdq(struct vhost_dev *vdev) 799 { 800 unsigned i = 0; 801 unsigned rx_count; 802 struct rte_mbuf *pkts_burst[MAX_PKT_BURST]; 803 804 if (vdev->ready == DEVICE_RX) { 805 /*clear MAC and VLAN settings*/ 806 rte_eth_dev_mac_addr_remove(ports[0], &vdev->mac_address); 807 for (i = 0; i < 6; i++) 808 vdev->mac_address.addr_bytes[i] = 0; 809 810 vdev->vlan_tag = 0; 811 812 /*Clear out the receive buffers*/ 813 rx_count = rte_eth_rx_burst(ports[0], 814 (uint16_t)vdev->vmdq_rx_q, pkts_burst, MAX_PKT_BURST); 815 816 while (rx_count) { 817 for (i = 0; i < rx_count; i++) 818 rte_pktmbuf_free(pkts_burst[i]); 819 820 rx_count = rte_eth_rx_burst(ports[0], 821 (uint16_t)vdev->vmdq_rx_q, pkts_burst, MAX_PKT_BURST); 822 } 823 824 vdev->ready = DEVICE_MAC_LEARNING; 825 } 826 } 827 828 static inline void 829 free_pkts(struct rte_mbuf **pkts, uint16_t n) 830 { 831 while (n--) 832 rte_pktmbuf_free(pkts[n]); 833 } 834 835 static __rte_always_inline void 836 complete_async_pkts(struct vhost_dev *vdev) 837 { 838 struct rte_mbuf *p_cpl[MAX_PKT_BURST]; 839 uint16_t complete_count; 840 841 complete_count = rte_vhost_poll_enqueue_completed(vdev->vid, 842 VIRTIO_RXQ, p_cpl, MAX_PKT_BURST); 843 if (complete_count) { 844 free_pkts(p_cpl, complete_count); 845 __atomic_sub_fetch(&vdev->pkts_inflight, complete_count, __ATOMIC_SEQ_CST); 846 } 847 848 } 849 850 static __rte_always_inline void 851 sync_virtio_xmit(struct vhost_dev *dst_vdev, struct vhost_dev *src_vdev, 852 struct rte_mbuf *m) 853 { 854 uint16_t ret; 855 856 if (builtin_net_driver) { 857 ret = vs_enqueue_pkts(dst_vdev, VIRTIO_RXQ, &m, 1); 858 } else { 859 ret = rte_vhost_enqueue_burst(dst_vdev->vid, VIRTIO_RXQ, &m, 1); 860 } 861 862 if (enable_stats) { 863 __atomic_add_fetch(&dst_vdev->stats.rx_total_atomic, 1, 864 __ATOMIC_SEQ_CST); 865 __atomic_add_fetch(&dst_vdev->stats.rx_atomic, ret, 866 __ATOMIC_SEQ_CST); 867 src_vdev->stats.tx_total++; 868 src_vdev->stats.tx += ret; 869 } 870 } 871 872 static __rte_always_inline void 873 drain_vhost(struct vhost_dev *vdev) 874 { 875 uint16_t ret; 876 uint32_t buff_idx = rte_lcore_id() * MAX_VHOST_DEVICE + vdev->vid; 877 uint16_t nr_xmit = vhost_txbuff[buff_idx]->len; 878 struct rte_mbuf **m = vhost_txbuff[buff_idx]->m_table; 879 880 if (builtin_net_driver) { 881 ret = vs_enqueue_pkts(vdev, VIRTIO_RXQ, m, nr_xmit); 882 } else if (async_vhost_driver) { 883 uint16_t enqueue_fail = 0; 884 885 complete_async_pkts(vdev); 886 ret = rte_vhost_submit_enqueue_burst(vdev->vid, VIRTIO_RXQ, m, nr_xmit); 887 __atomic_add_fetch(&vdev->pkts_inflight, ret, __ATOMIC_SEQ_CST); 888 889 enqueue_fail = nr_xmit - ret; 890 if (enqueue_fail) 891 free_pkts(&m[ret], nr_xmit - ret); 892 } else { 893 ret = rte_vhost_enqueue_burst(vdev->vid, VIRTIO_RXQ, 894 m, nr_xmit); 895 } 896 897 if (enable_stats) { 898 __atomic_add_fetch(&vdev->stats.rx_total_atomic, nr_xmit, 899 __ATOMIC_SEQ_CST); 900 __atomic_add_fetch(&vdev->stats.rx_atomic, ret, 901 __ATOMIC_SEQ_CST); 902 } 903 904 if (!async_vhost_driver) 905 free_pkts(m, nr_xmit); 906 } 907 908 static __rte_always_inline void 909 drain_vhost_table(void) 910 { 911 uint16_t lcore_id = rte_lcore_id(); 912 struct vhost_bufftable *vhost_txq; 913 struct vhost_dev *vdev; 914 uint64_t cur_tsc; 915 916 TAILQ_FOREACH(vdev, &vhost_dev_list, global_vdev_entry) { 917 if (unlikely(vdev->remove == 1)) 918 continue; 919 920 vhost_txq = vhost_txbuff[lcore_id * MAX_VHOST_DEVICE 921 + vdev->vid]; 922 923 cur_tsc = rte_rdtsc(); 924 if (unlikely(cur_tsc - vhost_txq->pre_tsc 925 > MBUF_TABLE_DRAIN_TSC)) { 926 RTE_LOG_DP(DEBUG, VHOST_DATA, 927 "Vhost TX queue drained after timeout with burst size %u\n", 928 vhost_txq->len); 929 drain_vhost(vdev); 930 vhost_txq->len = 0; 931 vhost_txq->pre_tsc = cur_tsc; 932 } 933 } 934 } 935 936 /* 937 * Check if the packet destination MAC address is for a local device. If so then put 938 * the packet on that devices RX queue. If not then return. 939 */ 940 static __rte_always_inline int 941 virtio_tx_local(struct vhost_dev *vdev, struct rte_mbuf *m) 942 { 943 struct rte_ether_hdr *pkt_hdr; 944 struct vhost_dev *dst_vdev; 945 struct vhost_bufftable *vhost_txq; 946 uint16_t lcore_id = rte_lcore_id(); 947 pkt_hdr = rte_pktmbuf_mtod(m, struct rte_ether_hdr *); 948 949 dst_vdev = find_vhost_dev(&pkt_hdr->dst_addr); 950 if (!dst_vdev) 951 return -1; 952 953 if (vdev->vid == dst_vdev->vid) { 954 RTE_LOG_DP(DEBUG, VHOST_DATA, 955 "(%d) TX: src and dst MAC is same. Dropping packet.\n", 956 vdev->vid); 957 return 0; 958 } 959 960 RTE_LOG_DP(DEBUG, VHOST_DATA, 961 "(%d) TX: MAC address is local\n", dst_vdev->vid); 962 963 if (unlikely(dst_vdev->remove)) { 964 RTE_LOG_DP(DEBUG, VHOST_DATA, 965 "(%d) device is marked for removal\n", dst_vdev->vid); 966 return 0; 967 } 968 969 vhost_txq = vhost_txbuff[lcore_id * MAX_VHOST_DEVICE + dst_vdev->vid]; 970 vhost_txq->m_table[vhost_txq->len++] = m; 971 972 if (enable_stats) { 973 vdev->stats.tx_total++; 974 vdev->stats.tx++; 975 } 976 977 if (unlikely(vhost_txq->len == MAX_PKT_BURST)) { 978 drain_vhost(dst_vdev); 979 vhost_txq->len = 0; 980 vhost_txq->pre_tsc = rte_rdtsc(); 981 } 982 return 0; 983 } 984 985 /* 986 * Check if the destination MAC of a packet is one local VM, 987 * and get its vlan tag, and offset if it is. 988 */ 989 static __rte_always_inline int 990 find_local_dest(struct vhost_dev *vdev, struct rte_mbuf *m, 991 uint32_t *offset, uint16_t *vlan_tag) 992 { 993 struct vhost_dev *dst_vdev; 994 struct rte_ether_hdr *pkt_hdr = 995 rte_pktmbuf_mtod(m, struct rte_ether_hdr *); 996 997 dst_vdev = find_vhost_dev(&pkt_hdr->dst_addr); 998 if (!dst_vdev) 999 return 0; 1000 1001 if (vdev->vid == dst_vdev->vid) { 1002 RTE_LOG_DP(DEBUG, VHOST_DATA, 1003 "(%d) TX: src and dst MAC is same. Dropping packet.\n", 1004 vdev->vid); 1005 return -1; 1006 } 1007 1008 /* 1009 * HW vlan strip will reduce the packet length 1010 * by minus length of vlan tag, so need restore 1011 * the packet length by plus it. 1012 */ 1013 *offset = VLAN_HLEN; 1014 *vlan_tag = vlan_tags[vdev->vid]; 1015 1016 RTE_LOG_DP(DEBUG, VHOST_DATA, 1017 "(%d) TX: pkt to local VM device id: (%d), vlan tag: %u.\n", 1018 vdev->vid, dst_vdev->vid, *vlan_tag); 1019 1020 return 0; 1021 } 1022 1023 static void virtio_tx_offload(struct rte_mbuf *m) 1024 { 1025 struct rte_net_hdr_lens hdr_lens; 1026 struct rte_ipv4_hdr *ipv4_hdr; 1027 struct rte_tcp_hdr *tcp_hdr; 1028 uint32_t ptype; 1029 void *l3_hdr; 1030 1031 ptype = rte_net_get_ptype(m, &hdr_lens, RTE_PTYPE_ALL_MASK); 1032 m->l2_len = hdr_lens.l2_len; 1033 m->l3_len = hdr_lens.l3_len; 1034 m->l4_len = hdr_lens.l4_len; 1035 1036 l3_hdr = rte_pktmbuf_mtod_offset(m, void *, m->l2_len); 1037 tcp_hdr = rte_pktmbuf_mtod_offset(m, struct rte_tcp_hdr *, 1038 m->l2_len + m->l3_len); 1039 1040 m->ol_flags |= RTE_MBUF_F_TX_TCP_SEG; 1041 if ((ptype & RTE_PTYPE_L3_MASK) == RTE_PTYPE_L3_IPV4) { 1042 m->ol_flags |= RTE_MBUF_F_TX_IPV4; 1043 m->ol_flags |= RTE_MBUF_F_TX_IP_CKSUM; 1044 ipv4_hdr = l3_hdr; 1045 ipv4_hdr->hdr_checksum = 0; 1046 tcp_hdr->cksum = rte_ipv4_phdr_cksum(l3_hdr, m->ol_flags); 1047 } else { /* assume ethertype == RTE_ETHER_TYPE_IPV6 */ 1048 m->ol_flags |= RTE_MBUF_F_TX_IPV6; 1049 tcp_hdr->cksum = rte_ipv6_phdr_cksum(l3_hdr, m->ol_flags); 1050 } 1051 } 1052 1053 static __rte_always_inline void 1054 do_drain_mbuf_table(struct mbuf_table *tx_q) 1055 { 1056 uint16_t count; 1057 1058 count = rte_eth_tx_burst(ports[0], tx_q->txq_id, 1059 tx_q->m_table, tx_q->len); 1060 if (unlikely(count < tx_q->len)) 1061 free_pkts(&tx_q->m_table[count], tx_q->len - count); 1062 1063 tx_q->len = 0; 1064 } 1065 1066 /* 1067 * This function routes the TX packet to the correct interface. This 1068 * may be a local device or the physical port. 1069 */ 1070 static __rte_always_inline void 1071 virtio_tx_route(struct vhost_dev *vdev, struct rte_mbuf *m, uint16_t vlan_tag) 1072 { 1073 struct mbuf_table *tx_q; 1074 unsigned offset = 0; 1075 const uint16_t lcore_id = rte_lcore_id(); 1076 struct rte_ether_hdr *nh; 1077 1078 1079 nh = rte_pktmbuf_mtod(m, struct rte_ether_hdr *); 1080 if (unlikely(rte_is_broadcast_ether_addr(&nh->dst_addr))) { 1081 struct vhost_dev *vdev2; 1082 1083 TAILQ_FOREACH(vdev2, &vhost_dev_list, global_vdev_entry) { 1084 if (vdev2 != vdev) 1085 sync_virtio_xmit(vdev2, vdev, m); 1086 } 1087 goto queue2nic; 1088 } 1089 1090 /*check if destination is local VM*/ 1091 if ((vm2vm_mode == VM2VM_SOFTWARE) && (virtio_tx_local(vdev, m) == 0)) 1092 return; 1093 1094 if (unlikely(vm2vm_mode == VM2VM_HARDWARE)) { 1095 if (unlikely(find_local_dest(vdev, m, &offset, 1096 &vlan_tag) != 0)) { 1097 rte_pktmbuf_free(m); 1098 return; 1099 } 1100 } 1101 1102 RTE_LOG_DP(DEBUG, VHOST_DATA, 1103 "(%d) TX: MAC address is external\n", vdev->vid); 1104 1105 queue2nic: 1106 1107 /*Add packet to the port tx queue*/ 1108 tx_q = &lcore_tx_queue[lcore_id]; 1109 1110 nh = rte_pktmbuf_mtod(m, struct rte_ether_hdr *); 1111 if (unlikely(nh->ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_VLAN))) { 1112 /* Guest has inserted the vlan tag. */ 1113 struct rte_vlan_hdr *vh = (struct rte_vlan_hdr *) (nh + 1); 1114 uint16_t vlan_tag_be = rte_cpu_to_be_16(vlan_tag); 1115 if ((vm2vm_mode == VM2VM_HARDWARE) && 1116 (vh->vlan_tci != vlan_tag_be)) 1117 vh->vlan_tci = vlan_tag_be; 1118 } else { 1119 m->ol_flags |= RTE_MBUF_F_TX_VLAN; 1120 1121 /* 1122 * Find the right seg to adjust the data len when offset is 1123 * bigger than tail room size. 1124 */ 1125 if (unlikely(vm2vm_mode == VM2VM_HARDWARE)) { 1126 if (likely(offset <= rte_pktmbuf_tailroom(m))) 1127 m->data_len += offset; 1128 else { 1129 struct rte_mbuf *seg = m; 1130 1131 while ((seg->next != NULL) && 1132 (offset > rte_pktmbuf_tailroom(seg))) 1133 seg = seg->next; 1134 1135 seg->data_len += offset; 1136 } 1137 m->pkt_len += offset; 1138 } 1139 1140 m->vlan_tci = vlan_tag; 1141 } 1142 1143 if (m->ol_flags & RTE_MBUF_F_RX_LRO) 1144 virtio_tx_offload(m); 1145 1146 tx_q->m_table[tx_q->len++] = m; 1147 if (enable_stats) { 1148 vdev->stats.tx_total++; 1149 vdev->stats.tx++; 1150 } 1151 1152 if (unlikely(tx_q->len == MAX_PKT_BURST)) 1153 do_drain_mbuf_table(tx_q); 1154 } 1155 1156 1157 static __rte_always_inline void 1158 drain_mbuf_table(struct mbuf_table *tx_q) 1159 { 1160 static uint64_t prev_tsc; 1161 uint64_t cur_tsc; 1162 1163 if (tx_q->len == 0) 1164 return; 1165 1166 cur_tsc = rte_rdtsc(); 1167 if (unlikely(cur_tsc - prev_tsc > MBUF_TABLE_DRAIN_TSC)) { 1168 prev_tsc = cur_tsc; 1169 1170 RTE_LOG_DP(DEBUG, VHOST_DATA, 1171 "TX queue drained after timeout with burst size %u\n", 1172 tx_q->len); 1173 do_drain_mbuf_table(tx_q); 1174 } 1175 } 1176 1177 static __rte_always_inline void 1178 drain_eth_rx(struct vhost_dev *vdev) 1179 { 1180 uint16_t rx_count, enqueue_count; 1181 struct rte_mbuf *pkts[MAX_PKT_BURST]; 1182 1183 rx_count = rte_eth_rx_burst(ports[0], vdev->vmdq_rx_q, 1184 pkts, MAX_PKT_BURST); 1185 1186 if (!rx_count) 1187 return; 1188 1189 /* 1190 * When "enable_retry" is set, here we wait and retry when there 1191 * is no enough free slots in the queue to hold @rx_count packets, 1192 * to diminish packet loss. 1193 */ 1194 if (enable_retry && 1195 unlikely(rx_count > rte_vhost_avail_entries(vdev->vid, 1196 VIRTIO_RXQ))) { 1197 uint32_t retry; 1198 1199 for (retry = 0; retry < burst_rx_retry_num; retry++) { 1200 rte_delay_us(burst_rx_delay_time); 1201 if (rx_count <= rte_vhost_avail_entries(vdev->vid, 1202 VIRTIO_RXQ)) 1203 break; 1204 } 1205 } 1206 1207 if (builtin_net_driver) { 1208 enqueue_count = vs_enqueue_pkts(vdev, VIRTIO_RXQ, 1209 pkts, rx_count); 1210 } else if (async_vhost_driver) { 1211 uint16_t enqueue_fail = 0; 1212 1213 complete_async_pkts(vdev); 1214 enqueue_count = rte_vhost_submit_enqueue_burst(vdev->vid, 1215 VIRTIO_RXQ, pkts, rx_count); 1216 __atomic_add_fetch(&vdev->pkts_inflight, enqueue_count, __ATOMIC_SEQ_CST); 1217 1218 enqueue_fail = rx_count - enqueue_count; 1219 if (enqueue_fail) 1220 free_pkts(&pkts[enqueue_count], enqueue_fail); 1221 1222 } else { 1223 enqueue_count = rte_vhost_enqueue_burst(vdev->vid, VIRTIO_RXQ, 1224 pkts, rx_count); 1225 } 1226 1227 if (enable_stats) { 1228 __atomic_add_fetch(&vdev->stats.rx_total_atomic, rx_count, 1229 __ATOMIC_SEQ_CST); 1230 __atomic_add_fetch(&vdev->stats.rx_atomic, enqueue_count, 1231 __ATOMIC_SEQ_CST); 1232 } 1233 1234 if (!async_vhost_driver) 1235 free_pkts(pkts, rx_count); 1236 } 1237 1238 static __rte_always_inline void 1239 drain_virtio_tx(struct vhost_dev *vdev) 1240 { 1241 struct rte_mbuf *pkts[MAX_PKT_BURST]; 1242 uint16_t count; 1243 uint16_t i; 1244 1245 if (builtin_net_driver) { 1246 count = vs_dequeue_pkts(vdev, VIRTIO_TXQ, mbuf_pool, 1247 pkts, MAX_PKT_BURST); 1248 } else { 1249 count = rte_vhost_dequeue_burst(vdev->vid, VIRTIO_TXQ, 1250 mbuf_pool, pkts, MAX_PKT_BURST); 1251 } 1252 1253 /* setup VMDq for the first packet */ 1254 if (unlikely(vdev->ready == DEVICE_MAC_LEARNING) && count) { 1255 if (vdev->remove || link_vmdq(vdev, pkts[0]) == -1) 1256 free_pkts(pkts, count); 1257 } 1258 1259 for (i = 0; i < count; ++i) 1260 virtio_tx_route(vdev, pkts[i], vlan_tags[vdev->vid]); 1261 } 1262 1263 /* 1264 * Main function of vhost-switch. It basically does: 1265 * 1266 * for each vhost device { 1267 * - drain_eth_rx() 1268 * 1269 * Which drains the host eth Rx queue linked to the vhost device, 1270 * and deliver all of them to guest virito Rx ring associated with 1271 * this vhost device. 1272 * 1273 * - drain_virtio_tx() 1274 * 1275 * Which drains the guest virtio Tx queue and deliver all of them 1276 * to the target, which could be another vhost device, or the 1277 * physical eth dev. The route is done in function "virtio_tx_route". 1278 * } 1279 */ 1280 static int 1281 switch_worker(void *arg __rte_unused) 1282 { 1283 unsigned i; 1284 unsigned lcore_id = rte_lcore_id(); 1285 struct vhost_dev *vdev; 1286 struct mbuf_table *tx_q; 1287 1288 RTE_LOG(INFO, VHOST_DATA, "Procesing on Core %u started\n", lcore_id); 1289 1290 tx_q = &lcore_tx_queue[lcore_id]; 1291 for (i = 0; i < rte_lcore_count(); i++) { 1292 if (lcore_ids[i] == lcore_id) { 1293 tx_q->txq_id = i; 1294 break; 1295 } 1296 } 1297 1298 while(1) { 1299 drain_mbuf_table(tx_q); 1300 drain_vhost_table(); 1301 /* 1302 * Inform the configuration core that we have exited the 1303 * linked list and that no devices are in use if requested. 1304 */ 1305 if (lcore_info[lcore_id].dev_removal_flag == REQUEST_DEV_REMOVAL) 1306 lcore_info[lcore_id].dev_removal_flag = ACK_DEV_REMOVAL; 1307 1308 /* 1309 * Process vhost devices 1310 */ 1311 TAILQ_FOREACH(vdev, &lcore_info[lcore_id].vdev_list, 1312 lcore_vdev_entry) { 1313 if (unlikely(vdev->remove)) { 1314 unlink_vmdq(vdev); 1315 vdev->ready = DEVICE_SAFE_REMOVE; 1316 continue; 1317 } 1318 1319 if (likely(vdev->ready == DEVICE_RX)) 1320 drain_eth_rx(vdev); 1321 1322 if (likely(!vdev->remove)) 1323 drain_virtio_tx(vdev); 1324 } 1325 } 1326 1327 return 0; 1328 } 1329 1330 /* 1331 * Remove a device from the specific data core linked list and from the 1332 * main linked list. Synchonization occurs through the use of the 1333 * lcore dev_removal_flag. Device is made volatile here to avoid re-ordering 1334 * of dev->remove=1 which can cause an infinite loop in the rte_pause loop. 1335 */ 1336 static void 1337 destroy_device(int vid) 1338 { 1339 struct vhost_dev *vdev = NULL; 1340 int lcore; 1341 uint16_t i; 1342 1343 TAILQ_FOREACH(vdev, &vhost_dev_list, global_vdev_entry) { 1344 if (vdev->vid == vid) 1345 break; 1346 } 1347 if (!vdev) 1348 return; 1349 /*set the remove flag. */ 1350 vdev->remove = 1; 1351 while(vdev->ready != DEVICE_SAFE_REMOVE) { 1352 rte_pause(); 1353 } 1354 1355 for (i = 0; i < RTE_MAX_LCORE; i++) 1356 rte_free(vhost_txbuff[i * MAX_VHOST_DEVICE + vid]); 1357 1358 if (builtin_net_driver) 1359 vs_vhost_net_remove(vdev); 1360 1361 TAILQ_REMOVE(&lcore_info[vdev->coreid].vdev_list, vdev, 1362 lcore_vdev_entry); 1363 TAILQ_REMOVE(&vhost_dev_list, vdev, global_vdev_entry); 1364 1365 1366 /* Set the dev_removal_flag on each lcore. */ 1367 RTE_LCORE_FOREACH_WORKER(lcore) 1368 lcore_info[lcore].dev_removal_flag = REQUEST_DEV_REMOVAL; 1369 1370 /* 1371 * Once each core has set the dev_removal_flag to ACK_DEV_REMOVAL 1372 * we can be sure that they can no longer access the device removed 1373 * from the linked lists and that the devices are no longer in use. 1374 */ 1375 RTE_LCORE_FOREACH_WORKER(lcore) { 1376 while (lcore_info[lcore].dev_removal_flag != ACK_DEV_REMOVAL) 1377 rte_pause(); 1378 } 1379 1380 lcore_info[vdev->coreid].device_num--; 1381 1382 RTE_LOG(INFO, VHOST_DATA, 1383 "(%d) device has been removed from data core\n", 1384 vdev->vid); 1385 1386 if (async_vhost_driver) { 1387 uint16_t n_pkt = 0; 1388 struct rte_mbuf *m_cpl[vdev->pkts_inflight]; 1389 1390 while (vdev->pkts_inflight) { 1391 n_pkt = rte_vhost_clear_queue_thread_unsafe(vid, VIRTIO_RXQ, 1392 m_cpl, vdev->pkts_inflight); 1393 free_pkts(m_cpl, n_pkt); 1394 __atomic_sub_fetch(&vdev->pkts_inflight, n_pkt, __ATOMIC_SEQ_CST); 1395 } 1396 1397 rte_vhost_async_channel_unregister(vid, VIRTIO_RXQ); 1398 } 1399 1400 rte_free(vdev); 1401 } 1402 1403 /* 1404 * A new device is added to a data core. First the device is added to the main linked list 1405 * and then allocated to a specific data core. 1406 */ 1407 static int 1408 new_device(int vid) 1409 { 1410 int lcore, core_add = 0; 1411 uint16_t i; 1412 uint32_t device_num_min = num_devices; 1413 struct vhost_dev *vdev; 1414 vdev = rte_zmalloc("vhost device", sizeof(*vdev), RTE_CACHE_LINE_SIZE); 1415 if (vdev == NULL) { 1416 RTE_LOG(INFO, VHOST_DATA, 1417 "(%d) couldn't allocate memory for vhost dev\n", 1418 vid); 1419 return -1; 1420 } 1421 vdev->vid = vid; 1422 1423 for (i = 0; i < RTE_MAX_LCORE; i++) { 1424 vhost_txbuff[i * MAX_VHOST_DEVICE + vid] 1425 = rte_zmalloc("vhost bufftable", 1426 sizeof(struct vhost_bufftable), 1427 RTE_CACHE_LINE_SIZE); 1428 1429 if (vhost_txbuff[i * MAX_VHOST_DEVICE + vid] == NULL) { 1430 RTE_LOG(INFO, VHOST_DATA, 1431 "(%d) couldn't allocate memory for vhost TX\n", vid); 1432 return -1; 1433 } 1434 } 1435 1436 if (builtin_net_driver) 1437 vs_vhost_net_setup(vdev); 1438 1439 TAILQ_INSERT_TAIL(&vhost_dev_list, vdev, global_vdev_entry); 1440 vdev->vmdq_rx_q = vid * queues_per_pool + vmdq_queue_base; 1441 1442 /*reset ready flag*/ 1443 vdev->ready = DEVICE_MAC_LEARNING; 1444 vdev->remove = 0; 1445 1446 /* Find a suitable lcore to add the device. */ 1447 RTE_LCORE_FOREACH_WORKER(lcore) { 1448 if (lcore_info[lcore].device_num < device_num_min) { 1449 device_num_min = lcore_info[lcore].device_num; 1450 core_add = lcore; 1451 } 1452 } 1453 vdev->coreid = core_add; 1454 1455 TAILQ_INSERT_TAIL(&lcore_info[vdev->coreid].vdev_list, vdev, 1456 lcore_vdev_entry); 1457 lcore_info[vdev->coreid].device_num++; 1458 1459 /* Disable notifications. */ 1460 rte_vhost_enable_guest_notification(vid, VIRTIO_RXQ, 0); 1461 rte_vhost_enable_guest_notification(vid, VIRTIO_TXQ, 0); 1462 1463 RTE_LOG(INFO, VHOST_DATA, 1464 "(%d) device has been added to data core %d\n", 1465 vid, vdev->coreid); 1466 1467 if (async_vhost_driver) { 1468 struct rte_vhost_async_config config = {0}; 1469 struct rte_vhost_async_channel_ops channel_ops; 1470 1471 if (dma_type != NULL && strncmp(dma_type, "ioat", 4) == 0) { 1472 channel_ops.transfer_data = ioat_transfer_data_cb; 1473 channel_ops.check_completed_copies = 1474 ioat_check_completed_copies_cb; 1475 1476 config.features = RTE_VHOST_ASYNC_INORDER; 1477 1478 return rte_vhost_async_channel_register(vid, VIRTIO_RXQ, 1479 config, &channel_ops); 1480 } 1481 } 1482 1483 return 0; 1484 } 1485 1486 static int 1487 vring_state_changed(int vid, uint16_t queue_id, int enable) 1488 { 1489 struct vhost_dev *vdev = NULL; 1490 1491 TAILQ_FOREACH(vdev, &vhost_dev_list, global_vdev_entry) { 1492 if (vdev->vid == vid) 1493 break; 1494 } 1495 if (!vdev) 1496 return -1; 1497 1498 if (queue_id != VIRTIO_RXQ) 1499 return 0; 1500 1501 if (async_vhost_driver) { 1502 if (!enable) { 1503 uint16_t n_pkt = 0; 1504 struct rte_mbuf *m_cpl[vdev->pkts_inflight]; 1505 1506 while (vdev->pkts_inflight) { 1507 n_pkt = rte_vhost_clear_queue_thread_unsafe(vid, queue_id, 1508 m_cpl, vdev->pkts_inflight); 1509 free_pkts(m_cpl, n_pkt); 1510 __atomic_sub_fetch(&vdev->pkts_inflight, n_pkt, __ATOMIC_SEQ_CST); 1511 } 1512 } 1513 } 1514 1515 return 0; 1516 } 1517 1518 /* 1519 * These callback allow devices to be added to the data core when configuration 1520 * has been fully complete. 1521 */ 1522 static const struct rte_vhost_device_ops virtio_net_device_ops = 1523 { 1524 .new_device = new_device, 1525 .destroy_device = destroy_device, 1526 .vring_state_changed = vring_state_changed, 1527 }; 1528 1529 /* 1530 * This is a thread will wake up after a period to print stats if the user has 1531 * enabled them. 1532 */ 1533 static void * 1534 print_stats(__rte_unused void *arg) 1535 { 1536 struct vhost_dev *vdev; 1537 uint64_t tx_dropped, rx_dropped; 1538 uint64_t tx, tx_total, rx, rx_total; 1539 const char clr[] = { 27, '[', '2', 'J', '\0' }; 1540 const char top_left[] = { 27, '[', '1', ';', '1', 'H','\0' }; 1541 1542 while(1) { 1543 sleep(enable_stats); 1544 1545 /* Clear screen and move to top left */ 1546 printf("%s%s\n", clr, top_left); 1547 printf("Device statistics =================================\n"); 1548 1549 TAILQ_FOREACH(vdev, &vhost_dev_list, global_vdev_entry) { 1550 tx_total = vdev->stats.tx_total; 1551 tx = vdev->stats.tx; 1552 tx_dropped = tx_total - tx; 1553 1554 rx_total = __atomic_load_n(&vdev->stats.rx_total_atomic, 1555 __ATOMIC_SEQ_CST); 1556 rx = __atomic_load_n(&vdev->stats.rx_atomic, 1557 __ATOMIC_SEQ_CST); 1558 rx_dropped = rx_total - rx; 1559 1560 printf("Statistics for device %d\n" 1561 "-----------------------\n" 1562 "TX total: %" PRIu64 "\n" 1563 "TX dropped: %" PRIu64 "\n" 1564 "TX successful: %" PRIu64 "\n" 1565 "RX total: %" PRIu64 "\n" 1566 "RX dropped: %" PRIu64 "\n" 1567 "RX successful: %" PRIu64 "\n", 1568 vdev->vid, 1569 tx_total, tx_dropped, tx, 1570 rx_total, rx_dropped, rx); 1571 } 1572 1573 printf("===================================================\n"); 1574 1575 fflush(stdout); 1576 } 1577 1578 return NULL; 1579 } 1580 1581 static void 1582 unregister_drivers(int socket_num) 1583 { 1584 int i, ret; 1585 1586 for (i = 0; i < socket_num; i++) { 1587 ret = rte_vhost_driver_unregister(socket_files + i * PATH_MAX); 1588 if (ret != 0) 1589 RTE_LOG(ERR, VHOST_CONFIG, 1590 "Fail to unregister vhost driver for %s.\n", 1591 socket_files + i * PATH_MAX); 1592 } 1593 } 1594 1595 /* When we receive a INT signal, unregister vhost driver */ 1596 static void 1597 sigint_handler(__rte_unused int signum) 1598 { 1599 /* Unregister vhost driver. */ 1600 unregister_drivers(nb_sockets); 1601 1602 exit(0); 1603 } 1604 1605 /* 1606 * While creating an mbuf pool, one key thing is to figure out how 1607 * many mbuf entries is enough for our use. FYI, here are some 1608 * guidelines: 1609 * 1610 * - Each rx queue would reserve @nr_rx_desc mbufs at queue setup stage 1611 * 1612 * - For each switch core (A CPU core does the packet switch), we need 1613 * also make some reservation for receiving the packets from virtio 1614 * Tx queue. How many is enough depends on the usage. It's normally 1615 * a simple calculation like following: 1616 * 1617 * MAX_PKT_BURST * max packet size / mbuf size 1618 * 1619 * So, we definitely need allocate more mbufs when TSO is enabled. 1620 * 1621 * - Similarly, for each switching core, we should serve @nr_rx_desc 1622 * mbufs for receiving the packets from physical NIC device. 1623 * 1624 * - We also need make sure, for each switch core, we have allocated 1625 * enough mbufs to fill up the mbuf cache. 1626 */ 1627 static void 1628 create_mbuf_pool(uint16_t nr_port, uint32_t nr_switch_core, uint32_t mbuf_size, 1629 uint32_t nr_queues, uint32_t nr_rx_desc, uint32_t nr_mbuf_cache) 1630 { 1631 uint32_t nr_mbufs; 1632 uint32_t nr_mbufs_per_core; 1633 uint32_t mtu = 1500; 1634 1635 if (mergeable) 1636 mtu = 9000; 1637 if (enable_tso) 1638 mtu = 64 * 1024; 1639 1640 nr_mbufs_per_core = (mtu + mbuf_size) * MAX_PKT_BURST / 1641 (mbuf_size - RTE_PKTMBUF_HEADROOM); 1642 nr_mbufs_per_core += nr_rx_desc; 1643 nr_mbufs_per_core = RTE_MAX(nr_mbufs_per_core, nr_mbuf_cache); 1644 1645 nr_mbufs = nr_queues * nr_rx_desc; 1646 nr_mbufs += nr_mbufs_per_core * nr_switch_core; 1647 nr_mbufs *= nr_port; 1648 1649 mbuf_pool = rte_pktmbuf_pool_create("MBUF_POOL", nr_mbufs, 1650 nr_mbuf_cache, 0, mbuf_size, 1651 rte_socket_id()); 1652 if (mbuf_pool == NULL) 1653 rte_exit(EXIT_FAILURE, "Cannot create mbuf pool\n"); 1654 } 1655 1656 /* 1657 * Main function, does initialisation and calls the per-lcore functions. 1658 */ 1659 int 1660 main(int argc, char *argv[]) 1661 { 1662 unsigned lcore_id, core_id = 0; 1663 unsigned nb_ports, valid_num_ports; 1664 int ret, i; 1665 uint16_t portid; 1666 static pthread_t tid; 1667 uint64_t flags = RTE_VHOST_USER_NET_COMPLIANT_OL_FLAGS; 1668 1669 signal(SIGINT, sigint_handler); 1670 1671 /* init EAL */ 1672 ret = rte_eal_init(argc, argv); 1673 if (ret < 0) 1674 rte_exit(EXIT_FAILURE, "Error with EAL initialization\n"); 1675 argc -= ret; 1676 argv += ret; 1677 1678 /* parse app arguments */ 1679 ret = us_vhost_parse_args(argc, argv); 1680 if (ret < 0) 1681 rte_exit(EXIT_FAILURE, "Invalid argument\n"); 1682 1683 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) { 1684 TAILQ_INIT(&lcore_info[lcore_id].vdev_list); 1685 1686 if (rte_lcore_is_enabled(lcore_id)) 1687 lcore_ids[core_id++] = lcore_id; 1688 } 1689 1690 if (rte_lcore_count() > RTE_MAX_LCORE) 1691 rte_exit(EXIT_FAILURE,"Not enough cores\n"); 1692 1693 /* Get the number of physical ports. */ 1694 nb_ports = rte_eth_dev_count_avail(); 1695 1696 /* 1697 * Update the global var NUM_PORTS and global array PORTS 1698 * and get value of var VALID_NUM_PORTS according to system ports number 1699 */ 1700 valid_num_ports = check_ports_num(nb_ports); 1701 1702 if ((valid_num_ports == 0) || (valid_num_ports > MAX_SUP_PORTS)) { 1703 RTE_LOG(INFO, VHOST_PORT, "Current enabled port number is %u," 1704 "but only %u port can be enabled\n",num_ports, MAX_SUP_PORTS); 1705 return -1; 1706 } 1707 1708 /* 1709 * FIXME: here we are trying to allocate mbufs big enough for 1710 * @MAX_QUEUES, but the truth is we're never going to use that 1711 * many queues here. We probably should only do allocation for 1712 * those queues we are going to use. 1713 */ 1714 create_mbuf_pool(valid_num_ports, rte_lcore_count() - 1, MBUF_DATA_SIZE, 1715 MAX_QUEUES, RTE_TEST_RX_DESC_DEFAULT, MBUF_CACHE_SIZE); 1716 1717 if (vm2vm_mode == VM2VM_HARDWARE) { 1718 /* Enable VT loop back to let L2 switch to do it. */ 1719 vmdq_conf_default.rx_adv_conf.vmdq_rx_conf.enable_loop_back = 1; 1720 RTE_LOG(DEBUG, VHOST_CONFIG, 1721 "Enable loop back for L2 switch in vmdq.\n"); 1722 } 1723 1724 /* initialize all ports */ 1725 RTE_ETH_FOREACH_DEV(portid) { 1726 /* skip ports that are not enabled */ 1727 if ((enabled_port_mask & (1 << portid)) == 0) { 1728 RTE_LOG(INFO, VHOST_PORT, 1729 "Skipping disabled port %d\n", portid); 1730 continue; 1731 } 1732 if (port_init(portid) != 0) 1733 rte_exit(EXIT_FAILURE, 1734 "Cannot initialize network ports\n"); 1735 } 1736 1737 /* Enable stats if the user option is set. */ 1738 if (enable_stats) { 1739 ret = rte_ctrl_thread_create(&tid, "print-stats", NULL, 1740 print_stats, NULL); 1741 if (ret < 0) 1742 rte_exit(EXIT_FAILURE, 1743 "Cannot create print-stats thread\n"); 1744 } 1745 1746 /* Launch all data cores. */ 1747 RTE_LCORE_FOREACH_WORKER(lcore_id) 1748 rte_eal_remote_launch(switch_worker, NULL, lcore_id); 1749 1750 if (client_mode) 1751 flags |= RTE_VHOST_USER_CLIENT; 1752 1753 /* Register vhost user driver to handle vhost messages. */ 1754 for (i = 0; i < nb_sockets; i++) { 1755 char *file = socket_files + i * PATH_MAX; 1756 1757 if (async_vhost_driver) 1758 flags = flags | RTE_VHOST_USER_ASYNC_COPY; 1759 1760 ret = rte_vhost_driver_register(file, flags); 1761 if (ret != 0) { 1762 unregister_drivers(i); 1763 rte_exit(EXIT_FAILURE, 1764 "vhost driver register failure.\n"); 1765 } 1766 1767 if (builtin_net_driver) 1768 rte_vhost_driver_set_features(file, VIRTIO_NET_FEATURES); 1769 1770 if (mergeable == 0) { 1771 rte_vhost_driver_disable_features(file, 1772 1ULL << VIRTIO_NET_F_MRG_RXBUF); 1773 } 1774 1775 if (enable_tx_csum == 0) { 1776 rte_vhost_driver_disable_features(file, 1777 1ULL << VIRTIO_NET_F_CSUM); 1778 } 1779 1780 if (enable_tso == 0) { 1781 rte_vhost_driver_disable_features(file, 1782 1ULL << VIRTIO_NET_F_HOST_TSO4); 1783 rte_vhost_driver_disable_features(file, 1784 1ULL << VIRTIO_NET_F_HOST_TSO6); 1785 rte_vhost_driver_disable_features(file, 1786 1ULL << VIRTIO_NET_F_GUEST_TSO4); 1787 rte_vhost_driver_disable_features(file, 1788 1ULL << VIRTIO_NET_F_GUEST_TSO6); 1789 } 1790 1791 if (promiscuous) { 1792 rte_vhost_driver_enable_features(file, 1793 1ULL << VIRTIO_NET_F_CTRL_RX); 1794 } 1795 1796 ret = rte_vhost_driver_callback_register(file, 1797 &virtio_net_device_ops); 1798 if (ret != 0) { 1799 rte_exit(EXIT_FAILURE, 1800 "failed to register vhost driver callbacks.\n"); 1801 } 1802 1803 if (rte_vhost_driver_start(file) < 0) { 1804 rte_exit(EXIT_FAILURE, 1805 "failed to start vhost driver.\n"); 1806 } 1807 } 1808 1809 RTE_LCORE_FOREACH_WORKER(lcore_id) 1810 rte_eal_wait_lcore(lcore_id); 1811 1812 /* clean up the EAL */ 1813 rte_eal_cleanup(); 1814 1815 return 0; 1816 } 1817