1 /*- 2 * BSD LICENSE 3 * 4 * Copyright(c) 2010-2014 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 52 #include "main.h" 53 #include "virtio-net.h" 54 #include "vhost-net-cdev.h" 55 56 #define MAX_QUEUES 128 57 58 /* the maximum number of external ports supported */ 59 #define MAX_SUP_PORTS 1 60 61 /* 62 * Calculate the number of buffers needed per port 63 */ 64 #define NUM_MBUFS_PER_PORT ((MAX_QUEUES*RTE_TEST_RX_DESC_DEFAULT) + \ 65 (num_switching_cores*MAX_PKT_BURST) + \ 66 (num_switching_cores*RTE_TEST_TX_DESC_DEFAULT) +\ 67 (num_switching_cores*MBUF_CACHE_SIZE)) 68 69 #define MBUF_CACHE_SIZE 128 70 #define MBUF_SIZE (2048 + sizeof(struct rte_mbuf) + RTE_PKTMBUF_HEADROOM) 71 72 /* 73 * RX and TX Prefetch, Host, and Write-back threshold values should be 74 * carefully set for optimal performance. Consult the network 75 * controller's datasheet and supporting DPDK documentation for guidance 76 * on how these parameters should be set. 77 */ 78 #define RX_PTHRESH 8 /* Default values of RX prefetch threshold reg. */ 79 #define RX_HTHRESH 8 /* Default values of RX host threshold reg. */ 80 #define RX_WTHRESH 4 /* Default values of RX write-back threshold reg. */ 81 82 /* 83 * These default values are optimized for use with the Intel(R) 82599 10 GbE 84 * Controller and the DPDK ixgbe PMD. Consider using other values for other 85 * network controllers and/or network drivers. 86 */ 87 #define TX_PTHRESH 36 /* Default values of TX prefetch threshold reg. */ 88 #define TX_HTHRESH 0 /* Default values of TX host threshold reg. */ 89 #define TX_WTHRESH 0 /* Default values of TX write-back threshold reg. */ 90 91 #define MAX_PKT_BURST 32 /* Max burst size for RX/TX */ 92 #define MAX_MRG_PKT_BURST 16 /* Max burst for merge buffers. Set to 1 due to performance issue. */ 93 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */ 94 95 #define BURST_RX_WAIT_US 15 /* Defines how long we wait between retries on RX */ 96 #define BURST_RX_RETRIES 4 /* Number of retries on RX. */ 97 98 /* State of virtio device. */ 99 #define DEVICE_MAC_LEARNING 0 100 #define DEVICE_RX 1 101 #define DEVICE_SAFE_REMOVE 2 102 103 /* Config_core_flag status definitions. */ 104 #define REQUEST_DEV_REMOVAL 1 105 #define ACK_DEV_REMOVAL 0 106 107 /* Configurable number of RX/TX ring descriptors */ 108 #define RTE_TEST_RX_DESC_DEFAULT 1024 109 #define RTE_TEST_TX_DESC_DEFAULT 512 110 111 #define INVALID_PORT_ID 0xFF 112 113 /* Max number of devices. Limited by vmdq. */ 114 #define MAX_DEVICES 64 115 116 /* Size of buffers used for rte_snprintfs. */ 117 #define MAX_PRINT_BUFF 6072 118 119 /* Maximum character device basename size. */ 120 #define MAX_BASENAME_SZ 10 121 122 /* Maximum long option length for option parsing. */ 123 #define MAX_LONG_OPT_SZ 64 124 125 /* Used to compare MAC addresses. */ 126 #define MAC_ADDR_CMP 0xFFFFFFFFFFFFULL 127 128 /* Number of descriptors per cacheline. */ 129 #define DESC_PER_CACHELINE (CACHE_LINE_SIZE / sizeof(struct vring_desc)) 130 131 /* mask of enabled ports */ 132 static uint32_t enabled_port_mask = 0; 133 134 /*Number of switching cores enabled*/ 135 static uint32_t num_switching_cores = 0; 136 137 /* number of devices/queues to support*/ 138 static uint32_t num_queues = 0; 139 uint32_t num_devices = 0; 140 141 /* Enable VM2VM communications. If this is disabled then the MAC address compare is skipped. */ 142 static uint32_t enable_vm2vm = 1; 143 /* Enable stats. */ 144 static uint32_t enable_stats = 0; 145 /* Enable retries on RX. */ 146 static uint32_t enable_retry = 1; 147 /* Specify timeout (in useconds) between retries on RX. */ 148 static uint32_t burst_rx_delay_time = BURST_RX_WAIT_US; 149 /* Specify the number of retries on RX. */ 150 static uint32_t burst_rx_retry_num = BURST_RX_RETRIES; 151 152 /* Character device basename. Can be set by user. */ 153 static char dev_basename[MAX_BASENAME_SZ] = "vhost-net"; 154 155 /* Charater device index. Can be set by user. */ 156 static uint32_t dev_index = 0; 157 158 /* This can be set by the user so it is made available here. */ 159 extern uint64_t VHOST_FEATURES; 160 161 /* Default configuration for rx and tx thresholds etc. */ 162 static const struct rte_eth_rxconf rx_conf_default = { 163 .rx_thresh = { 164 .pthresh = RX_PTHRESH, 165 .hthresh = RX_HTHRESH, 166 .wthresh = RX_WTHRESH, 167 }, 168 .rx_drop_en = 1, 169 }; 170 171 /* 172 * These default values are optimized for use with the Intel(R) 82599 10 GbE 173 * Controller and the DPDK ixgbe/igb PMD. Consider using other values for other 174 * network controllers and/or network drivers. 175 */ 176 static const struct rte_eth_txconf tx_conf_default = { 177 .tx_thresh = { 178 .pthresh = TX_PTHRESH, 179 .hthresh = TX_HTHRESH, 180 .wthresh = TX_WTHRESH, 181 }, 182 .tx_free_thresh = 0, /* Use PMD default values */ 183 .tx_rs_thresh = 0, /* Use PMD default values */ 184 }; 185 186 /* empty vmdq configuration structure. Filled in programatically */ 187 static const struct rte_eth_conf vmdq_conf_default = { 188 .rxmode = { 189 .mq_mode = ETH_MQ_RX_VMDQ_ONLY, 190 .split_hdr_size = 0, 191 .header_split = 0, /**< Header Split disabled */ 192 .hw_ip_checksum = 0, /**< IP checksum offload disabled */ 193 .hw_vlan_filter = 0, /**< VLAN filtering disabled */ 194 /* 195 * It is necessary for 1G NIC such as I350, 196 * this fixes bug of ipv4 forwarding in guest can't 197 * forward pakets from one virtio dev to another virtio dev. 198 */ 199 .hw_vlan_strip = 1, /**< VLAN strip enabled. */ 200 .jumbo_frame = 0, /**< Jumbo Frame Support disabled */ 201 .hw_strip_crc = 0, /**< CRC stripped by hardware */ 202 }, 203 204 .txmode = { 205 .mq_mode = ETH_MQ_TX_NONE, 206 }, 207 .rx_adv_conf = { 208 /* 209 * should be overridden separately in code with 210 * appropriate values 211 */ 212 .vmdq_rx_conf = { 213 .nb_queue_pools = ETH_8_POOLS, 214 .enable_default_pool = 0, 215 .default_pool = 0, 216 .nb_pool_maps = 0, 217 .pool_map = {{0, 0},}, 218 }, 219 }, 220 }; 221 222 static unsigned lcore_ids[RTE_MAX_LCORE]; 223 static uint8_t ports[RTE_MAX_ETHPORTS]; 224 static unsigned num_ports = 0; /**< The number of ports specified in command line */ 225 226 const uint16_t vlan_tags[] = { 227 1000, 1001, 1002, 1003, 1004, 1005, 1006, 1007, 228 1008, 1009, 1010, 1011, 1012, 1013, 1014, 1015, 229 1016, 1017, 1018, 1019, 1020, 1021, 1022, 1023, 230 1024, 1025, 1026, 1027, 1028, 1029, 1030, 1031, 231 1032, 1033, 1034, 1035, 1036, 1037, 1038, 1039, 232 1040, 1041, 1042, 1043, 1044, 1045, 1046, 1047, 233 1048, 1049, 1050, 1051, 1052, 1053, 1054, 1055, 234 1056, 1057, 1058, 1059, 1060, 1061, 1062, 1063, 235 }; 236 237 /* ethernet addresses of ports */ 238 static struct ether_addr vmdq_ports_eth_addr[RTE_MAX_ETHPORTS]; 239 240 /* heads for the main used and free linked lists for the data path. */ 241 static struct virtio_net_data_ll *ll_root_used = NULL; 242 static struct virtio_net_data_ll *ll_root_free = NULL; 243 244 /* Array of data core structures containing information on individual core linked lists. */ 245 static struct lcore_info lcore_info[RTE_MAX_LCORE]; 246 247 /* Used for queueing bursts of TX packets. */ 248 struct mbuf_table { 249 unsigned len; 250 unsigned txq_id; 251 struct rte_mbuf *m_table[MAX_PKT_BURST]; 252 }; 253 254 /* TX queue for each data core. */ 255 struct mbuf_table lcore_tx_queue[RTE_MAX_LCORE]; 256 257 /* Vlan header struct used to insert vlan tags on TX. */ 258 struct vlan_ethhdr { 259 unsigned char h_dest[ETH_ALEN]; 260 unsigned char h_source[ETH_ALEN]; 261 __be16 h_vlan_proto; 262 __be16 h_vlan_TCI; 263 __be16 h_vlan_encapsulated_proto; 264 }; 265 266 /* Header lengths. */ 267 #define VLAN_HLEN 4 268 #define VLAN_ETH_HLEN 18 269 270 /* Per-device statistics struct */ 271 struct device_statistics { 272 uint64_t tx_total; 273 rte_atomic64_t rx_total; 274 uint64_t tx; 275 rte_atomic64_t rx; 276 } __rte_cache_aligned; 277 struct device_statistics dev_statistics[MAX_DEVICES]; 278 279 /* 280 * Builds up the correct configuration for VMDQ VLAN pool map 281 * according to the pool & queue limits. 282 */ 283 static inline int 284 get_eth_conf(struct rte_eth_conf *eth_conf, uint32_t num_devices) 285 { 286 struct rte_eth_vmdq_rx_conf conf; 287 unsigned i; 288 289 memset(&conf, 0, sizeof(conf)); 290 conf.nb_queue_pools = (enum rte_eth_nb_pools)num_devices; 291 conf.nb_pool_maps = num_devices; 292 293 for (i = 0; i < conf.nb_pool_maps; i++) { 294 conf.pool_map[i].vlan_id = vlan_tags[ i ]; 295 conf.pool_map[i].pools = (1UL << i); 296 } 297 298 (void)(rte_memcpy(eth_conf, &vmdq_conf_default, sizeof(*eth_conf))); 299 (void)(rte_memcpy(ð_conf->rx_adv_conf.vmdq_rx_conf, &conf, 300 sizeof(eth_conf->rx_adv_conf.vmdq_rx_conf))); 301 return 0; 302 } 303 304 /* 305 * Validate the device number according to the max pool number gotten form dev_info 306 * If the device number is invalid, give the error message and return -1. 307 * Each device must have its own pool. 308 */ 309 static inline int 310 validate_num_devices(uint32_t max_nb_devices) 311 { 312 if (num_devices > max_nb_devices) { 313 RTE_LOG(ERR, PORT, "invalid number of devices\n"); 314 return -1; 315 } 316 return 0; 317 } 318 319 /* 320 * Initialises a given port using global settings and with the rx buffers 321 * coming from the mbuf_pool passed as parameter 322 */ 323 static inline int 324 port_init(uint8_t port, struct rte_mempool *mbuf_pool) 325 { 326 struct rte_eth_dev_info dev_info; 327 struct rte_eth_conf port_conf; 328 uint16_t rx_rings, tx_rings = (uint16_t)rte_lcore_count(); 329 const uint16_t rx_ring_size = RTE_TEST_RX_DESC_DEFAULT, tx_ring_size = RTE_TEST_TX_DESC_DEFAULT; 330 int retval; 331 uint16_t q; 332 333 /* The max pool number from dev_info will be used to validate the pool number specified in cmd line */ 334 rte_eth_dev_info_get (port, &dev_info); 335 336 /*configure the number of supported virtio devices based on VMDQ limits */ 337 num_devices = dev_info.max_vmdq_pools; 338 num_queues = dev_info.max_rx_queues; 339 340 retval = validate_num_devices(MAX_DEVICES); 341 if (retval < 0) 342 return retval; 343 344 /* Get port configuration. */ 345 retval = get_eth_conf(&port_conf, num_devices); 346 if (retval < 0) 347 return retval; 348 349 if (port >= rte_eth_dev_count()) return -1; 350 351 rx_rings = (uint16_t)num_queues, 352 /* Configure ethernet device. */ 353 retval = rte_eth_dev_configure(port, rx_rings, tx_rings, &port_conf); 354 if (retval != 0) 355 return retval; 356 357 /* Setup the queues. */ 358 for (q = 0; q < rx_rings; q ++) { 359 retval = rte_eth_rx_queue_setup(port, q, rx_ring_size, 360 rte_eth_dev_socket_id(port), &rx_conf_default, 361 mbuf_pool); 362 if (retval < 0) 363 return retval; 364 } 365 for (q = 0; q < tx_rings; q ++) { 366 retval = rte_eth_tx_queue_setup(port, q, tx_ring_size, 367 rte_eth_dev_socket_id(port), &tx_conf_default); 368 if (retval < 0) 369 return retval; 370 } 371 372 /* Start the device. */ 373 retval = rte_eth_dev_start(port); 374 if (retval < 0) 375 return retval; 376 377 rte_eth_macaddr_get(port, &vmdq_ports_eth_addr[port]); 378 RTE_LOG(INFO, PORT, "Max virtio devices supported: %u\n", num_devices); 379 RTE_LOG(INFO, PORT, "Port %u MAC: %02"PRIx8" %02"PRIx8" %02"PRIx8 380 " %02"PRIx8" %02"PRIx8" %02"PRIx8"\n", 381 (unsigned)port, 382 vmdq_ports_eth_addr[port].addr_bytes[0], 383 vmdq_ports_eth_addr[port].addr_bytes[1], 384 vmdq_ports_eth_addr[port].addr_bytes[2], 385 vmdq_ports_eth_addr[port].addr_bytes[3], 386 vmdq_ports_eth_addr[port].addr_bytes[4], 387 vmdq_ports_eth_addr[port].addr_bytes[5]); 388 389 return 0; 390 } 391 392 /* 393 * Set character device basename. 394 */ 395 static int 396 us_vhost_parse_basename(const char *q_arg) 397 { 398 /* parse number string */ 399 400 if (strnlen(q_arg, MAX_BASENAME_SZ) > MAX_BASENAME_SZ) 401 return -1; 402 else 403 rte_snprintf((char*)&dev_basename, MAX_BASENAME_SZ, "%s", q_arg); 404 405 return 0; 406 } 407 408 /* 409 * Parse the portmask provided at run time. 410 */ 411 static int 412 parse_portmask(const char *portmask) 413 { 414 char *end = NULL; 415 unsigned long pm; 416 417 errno = 0; 418 419 /* parse hexadecimal string */ 420 pm = strtoul(portmask, &end, 16); 421 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0') || (errno != 0)) 422 return -1; 423 424 if (pm == 0) 425 return -1; 426 427 return pm; 428 429 } 430 431 /* 432 * Parse num options at run time. 433 */ 434 static int 435 parse_num_opt(const char *q_arg, uint32_t max_valid_value) 436 { 437 char *end = NULL; 438 unsigned long num; 439 440 errno = 0; 441 442 /* parse unsigned int string */ 443 num = strtoul(q_arg, &end, 10); 444 if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0') || (errno != 0)) 445 return -1; 446 447 if (num > max_valid_value) 448 return -1; 449 450 return num; 451 452 } 453 454 /* 455 * Display usage 456 */ 457 static void 458 us_vhost_usage(const char *prgname) 459 { 460 RTE_LOG(INFO, CONFIG, "%s [EAL options] -- -p PORTMASK --vm2vm [0|1] --rx_retry [0|1] --mergeable [0|1] --stats [0-N] --dev-basename <name> --dev-index [0-N] --nb-devices ND\n" 461 " -p PORTMASK: Set mask for ports to be used by application\n" 462 " --vm2vm [0|1]: disable/enable(default) vm2vm comms\n" 463 " --rx-retry [0|1]: disable/enable(default) retries on rx. Enable retry if destintation queue is full\n" 464 " --rx-retry-delay [0-N]: timeout(in usecond) between retries on RX. This makes effect only if retries on rx enabled\n" 465 " --rx-retry-num [0-N]: the number of retries on rx. This makes effect only if retries on rx enabled\n" 466 " --mergeable [0|1]: disable(default)/enable RX mergeable buffers\n" 467 " --stats [0-N]: 0: Disable stats, N: Time in seconds to print stats\n" 468 " --dev-basename: The basename to be used for the character device.\n" 469 " --dev-index [0-N]: Defaults to zero if not used. Index is appended to basename.\n", 470 prgname); 471 } 472 473 /* 474 * Parse the arguments given in the command line of the application. 475 */ 476 static int 477 us_vhost_parse_args(int argc, char **argv) 478 { 479 int opt, ret; 480 int option_index; 481 unsigned i; 482 const char *prgname = argv[0]; 483 static struct option long_option[] = { 484 {"vm2vm", required_argument, NULL, 0}, 485 {"rx-retry", required_argument, NULL, 0}, 486 {"rx-retry-delay", required_argument, NULL, 0}, 487 {"rx-retry-num", required_argument, NULL, 0}, 488 {"mergeable", required_argument, NULL, 0}, 489 {"stats", required_argument, NULL, 0}, 490 {"dev-basename", required_argument, NULL, 0}, 491 {"dev-index", required_argument, NULL, 0}, 492 {NULL, 0, 0, 0} 493 }; 494 495 /* Parse command line */ 496 while ((opt = getopt_long(argc, argv, "p:",long_option, &option_index)) != EOF) { 497 switch (opt) { 498 /* Portmask */ 499 case 'p': 500 enabled_port_mask = parse_portmask(optarg); 501 if (enabled_port_mask == 0) { 502 RTE_LOG(INFO, CONFIG, "Invalid portmask\n"); 503 us_vhost_usage(prgname); 504 return -1; 505 } 506 break; 507 508 case 0: 509 /* Enable/disable vm2vm comms. */ 510 if (!strncmp(long_option[option_index].name, "vm2vm", MAX_LONG_OPT_SZ)) { 511 ret = parse_num_opt(optarg, 1); 512 if (ret == -1) { 513 RTE_LOG(INFO, CONFIG, "Invalid argument for vm2vm [0|1]\n"); 514 us_vhost_usage(prgname); 515 return -1; 516 } else { 517 enable_vm2vm = ret; 518 } 519 } 520 521 /* Enable/disable retries on RX. */ 522 if (!strncmp(long_option[option_index].name, "rx-retry", MAX_LONG_OPT_SZ)) { 523 ret = parse_num_opt(optarg, 1); 524 if (ret == -1) { 525 RTE_LOG(INFO, CONFIG, "Invalid argument for rx-retry [0|1]\n"); 526 us_vhost_usage(prgname); 527 return -1; 528 } else { 529 enable_retry = ret; 530 } 531 } 532 533 /* Specify the retries delay time (in useconds) on RX. */ 534 if (!strncmp(long_option[option_index].name, "rx-retry-delay", MAX_LONG_OPT_SZ)) { 535 ret = parse_num_opt(optarg, INT32_MAX); 536 if (ret == -1) { 537 RTE_LOG(INFO, CONFIG, "Invalid argument for rx-retry-delay [0-N]\n"); 538 us_vhost_usage(prgname); 539 return -1; 540 } else { 541 burst_rx_delay_time = ret; 542 } 543 } 544 545 /* Specify the retries number on RX. */ 546 if (!strncmp(long_option[option_index].name, "rx-retry-num", MAX_LONG_OPT_SZ)) { 547 ret = parse_num_opt(optarg, INT32_MAX); 548 if (ret == -1) { 549 RTE_LOG(INFO, CONFIG, "Invalid argument for rx-retry-num [0-N]\n"); 550 us_vhost_usage(prgname); 551 return -1; 552 } else { 553 burst_rx_retry_num = ret; 554 } 555 } 556 557 /* Enable/disable RX mergeable buffers. */ 558 if (!strncmp(long_option[option_index].name, "mergeable", MAX_LONG_OPT_SZ)) { 559 ret = parse_num_opt(optarg, 1); 560 if (ret == -1) { 561 RTE_LOG(INFO, CONFIG, "Invalid argument for mergeable [0|1]\n"); 562 us_vhost_usage(prgname); 563 return -1; 564 } else { 565 if (ret) 566 VHOST_FEATURES = (1ULL << VIRTIO_NET_F_MRG_RXBUF); 567 } 568 } 569 570 /* Enable/disable stats. */ 571 if (!strncmp(long_option[option_index].name, "stats", MAX_LONG_OPT_SZ)) { 572 ret = parse_num_opt(optarg, INT32_MAX); 573 if (ret == -1) { 574 RTE_LOG(INFO, CONFIG, "Invalid argument for stats [0..N]\n"); 575 us_vhost_usage(prgname); 576 return -1; 577 } else { 578 enable_stats = ret; 579 } 580 } 581 582 /* Set character device basename. */ 583 if (!strncmp(long_option[option_index].name, "dev-basename", MAX_LONG_OPT_SZ)) { 584 if (us_vhost_parse_basename(optarg) == -1) { 585 RTE_LOG(INFO, CONFIG, "Invalid argument for character device basename (Max %d characters)\n", MAX_BASENAME_SZ); 586 us_vhost_usage(prgname); 587 return -1; 588 } 589 } 590 591 /* Set character device index. */ 592 if (!strncmp(long_option[option_index].name, "dev-index", MAX_LONG_OPT_SZ)) { 593 ret = parse_num_opt(optarg, INT32_MAX); 594 if (ret == -1) { 595 RTE_LOG(INFO, CONFIG, "Invalid argument for character device index [0..N]\n"); 596 us_vhost_usage(prgname); 597 return -1; 598 } else { 599 dev_index = ret; 600 } 601 } 602 603 break; 604 605 /* Invalid option - print options. */ 606 default: 607 us_vhost_usage(prgname); 608 return -1; 609 } 610 } 611 612 for (i = 0; i < RTE_MAX_ETHPORTS; i++) { 613 if (enabled_port_mask & (1 << i)) 614 ports[num_ports++] = (uint8_t)i; 615 } 616 617 if ((num_ports == 0) || (num_ports > MAX_SUP_PORTS)) { 618 RTE_LOG(INFO, PORT, "Current enabled port number is %u," 619 "but only %u port can be enabled\n",num_ports, MAX_SUP_PORTS); 620 return -1; 621 } 622 623 return 0; 624 } 625 626 /* 627 * Update the global var NUM_PORTS and array PORTS according to system ports number 628 * and return valid ports number 629 */ 630 static unsigned check_ports_num(unsigned nb_ports) 631 { 632 unsigned valid_num_ports = num_ports; 633 unsigned portid; 634 635 if (num_ports > nb_ports) { 636 RTE_LOG(INFO, PORT, "\nSpecified port number(%u) exceeds total system port number(%u)\n", 637 num_ports, nb_ports); 638 num_ports = nb_ports; 639 } 640 641 for (portid = 0; portid < num_ports; portid ++) { 642 if (ports[portid] >= nb_ports) { 643 RTE_LOG(INFO, PORT, "\nSpecified port ID(%u) exceeds max system port ID(%u)\n", 644 ports[portid], (nb_ports - 1)); 645 ports[portid] = INVALID_PORT_ID; 646 valid_num_ports--; 647 } 648 } 649 return valid_num_ports; 650 } 651 652 /* 653 * Macro to print out packet contents. Wrapped in debug define so that the 654 * data path is not effected when debug is disabled. 655 */ 656 #ifdef DEBUG 657 #define PRINT_PACKET(device, addr, size, header) do { \ 658 char *pkt_addr = (char*)(addr); \ 659 unsigned int index; \ 660 char packet[MAX_PRINT_BUFF]; \ 661 \ 662 if ((header)) \ 663 rte_snprintf(packet, MAX_PRINT_BUFF, "(%"PRIu64") Header size %d: ", (device->device_fh), (size)); \ 664 else \ 665 rte_snprintf(packet, MAX_PRINT_BUFF, "(%"PRIu64") Packet size %d: ", (device->device_fh), (size)); \ 666 for (index = 0; index < (size); index++) { \ 667 rte_snprintf(packet + strnlen(packet, MAX_PRINT_BUFF), MAX_PRINT_BUFF - strnlen(packet, MAX_PRINT_BUFF), \ 668 "%02hhx ", pkt_addr[index]); \ 669 } \ 670 rte_snprintf(packet + strnlen(packet, MAX_PRINT_BUFF), MAX_PRINT_BUFF - strnlen(packet, MAX_PRINT_BUFF), "\n"); \ 671 \ 672 LOG_DEBUG(DATA, "%s", packet); \ 673 } while(0) 674 #else 675 #define PRINT_PACKET(device, addr, size, header) do{} while(0) 676 #endif 677 678 /* 679 * Function to convert guest physical addresses to vhost virtual addresses. This 680 * is used to convert virtio buffer addresses. 681 */ 682 static inline uint64_t __attribute__((always_inline)) 683 gpa_to_vva(struct virtio_net *dev, uint64_t guest_pa) 684 { 685 struct virtio_memory_regions *region; 686 uint32_t regionidx; 687 uint64_t vhost_va = 0; 688 689 for (regionidx = 0; regionidx < dev->mem->nregions; regionidx++) { 690 region = &dev->mem->regions[regionidx]; 691 if ((guest_pa >= region->guest_phys_address) && 692 (guest_pa <= region->guest_phys_address_end)) { 693 vhost_va = region->address_offset + guest_pa; 694 break; 695 } 696 } 697 LOG_DEBUG(DATA, "(%"PRIu64") GPA %p| VVA %p\n", 698 dev->device_fh, (void*)(uintptr_t)guest_pa, (void*)(uintptr_t)vhost_va); 699 700 return vhost_va; 701 } 702 703 /* 704 * This function adds buffers to the virtio devices RX virtqueue. Buffers can 705 * be received from the physical port or from another virtio device. A packet 706 * count is returned to indicate the number of packets that were succesfully 707 * added to the RX queue. 708 */ 709 static inline uint32_t __attribute__((always_inline)) 710 virtio_dev_rx(struct virtio_net *dev, struct rte_mbuf **pkts, uint32_t count) 711 { 712 struct vhost_virtqueue *vq; 713 struct vring_desc *desc; 714 struct rte_mbuf *buff; 715 /* The virtio_hdr is initialised to 0. */ 716 struct virtio_net_hdr_mrg_rxbuf virtio_hdr = {{0,0,0,0,0,0},0}; 717 uint64_t buff_addr = 0; 718 uint64_t buff_hdr_addr = 0; 719 uint32_t head[MAX_PKT_BURST], packet_len = 0; 720 uint32_t head_idx, packet_success = 0; 721 uint32_t mergeable, mrg_count = 0; 722 uint32_t retry = 0; 723 uint16_t avail_idx, res_cur_idx; 724 uint16_t res_base_idx, res_end_idx; 725 uint16_t free_entries; 726 uint8_t success = 0; 727 728 LOG_DEBUG(DATA, "(%"PRIu64") virtio_dev_rx()\n", dev->device_fh); 729 vq = dev->virtqueue[VIRTIO_RXQ]; 730 count = (count > MAX_PKT_BURST) ? MAX_PKT_BURST : count; 731 /* As many data cores may want access to available buffers, they need to be reserved. */ 732 do { 733 res_base_idx = vq->last_used_idx_res; 734 avail_idx = *((volatile uint16_t *)&vq->avail->idx); 735 736 free_entries = (avail_idx - res_base_idx); 737 /* If retry is enabled and the queue is full then we wait and retry to avoid packet loss. */ 738 if (enable_retry && unlikely(count > free_entries)) { 739 for (retry = 0; retry < burst_rx_retry_num; retry++) { 740 rte_delay_us(burst_rx_delay_time); 741 avail_idx = 742 *((volatile uint16_t *)&vq->avail->idx); 743 free_entries = (avail_idx - res_base_idx); 744 if (count <= free_entries) 745 break; 746 } 747 } 748 749 /*check that we have enough buffers*/ 750 if (unlikely(count > free_entries)) 751 count = free_entries; 752 753 if (count == 0) 754 return 0; 755 756 res_end_idx = res_base_idx + count; 757 /* vq->last_used_idx_res is atomically updated. */ 758 success = rte_atomic16_cmpset(&vq->last_used_idx_res, res_base_idx, 759 res_end_idx); 760 } while (unlikely(success == 0)); 761 res_cur_idx = res_base_idx; 762 LOG_DEBUG(DATA, "(%"PRIu64") Current Index %d| End Index %d\n", dev->device_fh, res_cur_idx, res_end_idx); 763 764 /* Prefetch available ring to retrieve indexes. */ 765 rte_prefetch0(&vq->avail->ring[res_cur_idx & (vq->size - 1)]); 766 767 /* Check if the VIRTIO_NET_F_MRG_RXBUF feature is enabled. */ 768 mergeable = dev->features & (1 << VIRTIO_NET_F_MRG_RXBUF); 769 770 /* Retrieve all of the head indexes first to avoid caching issues. */ 771 for (head_idx = 0; head_idx < count; head_idx++) 772 head[head_idx] = vq->avail->ring[(res_cur_idx + head_idx) & (vq->size - 1)]; 773 774 /*Prefetch descriptor index. */ 775 rte_prefetch0(&vq->desc[head[packet_success]]); 776 777 while (res_cur_idx != res_end_idx) { 778 /* Get descriptor from available ring */ 779 desc = &vq->desc[head[packet_success]]; 780 781 buff = pkts[packet_success]; 782 783 /* Convert from gpa to vva (guest physical addr -> vhost virtual addr) */ 784 buff_addr = gpa_to_vva(dev, desc->addr); 785 /* Prefetch buffer address. */ 786 rte_prefetch0((void*)(uintptr_t)buff_addr); 787 788 if (mergeable && (mrg_count != 0)) { 789 desc->len = packet_len = rte_pktmbuf_data_len(buff); 790 } else { 791 /* Copy virtio_hdr to packet and increment buffer address */ 792 buff_hdr_addr = buff_addr; 793 packet_len = rte_pktmbuf_data_len(buff) + vq->vhost_hlen; 794 795 /* 796 * If the descriptors are chained the header and data are placed in 797 * separate buffers. 798 */ 799 if (desc->flags & VRING_DESC_F_NEXT) { 800 desc->len = vq->vhost_hlen; 801 desc = &vq->desc[desc->next]; 802 /* Buffer address translation. */ 803 buff_addr = gpa_to_vva(dev, desc->addr); 804 desc->len = rte_pktmbuf_data_len(buff); 805 } else { 806 buff_addr += vq->vhost_hlen; 807 desc->len = packet_len; 808 } 809 } 810 811 PRINT_PACKET(dev, (uintptr_t)buff_addr, rte_pktmbuf_data_len(buff), 0); 812 813 /* Update used ring with desc information */ 814 vq->used->ring[res_cur_idx & (vq->size - 1)].id = head[packet_success]; 815 vq->used->ring[res_cur_idx & (vq->size - 1)].len = packet_len; 816 817 /* Copy mbuf data to buffer */ 818 rte_memcpy((void *)(uintptr_t)buff_addr, (const void*)buff->pkt.data, rte_pktmbuf_data_len(buff)); 819 820 res_cur_idx++; 821 packet_success++; 822 823 /* If mergeable is disabled then a header is required per buffer. */ 824 if (!mergeable) { 825 rte_memcpy((void *)(uintptr_t)buff_hdr_addr, (const void*)&virtio_hdr, vq->vhost_hlen); 826 PRINT_PACKET(dev, (uintptr_t)buff_hdr_addr, vq->vhost_hlen, 1); 827 } else { 828 mrg_count++; 829 /* Merge buffer can only handle so many buffers at a time. Tell the guest if this limit is reached. */ 830 if ((mrg_count == MAX_MRG_PKT_BURST) || (res_cur_idx == res_end_idx)) { 831 virtio_hdr.num_buffers = mrg_count; 832 LOG_DEBUG(DATA, "(%"PRIu64") RX: Num merge buffers %d\n", dev->device_fh, virtio_hdr.num_buffers); 833 rte_memcpy((void *)(uintptr_t)buff_hdr_addr, (const void*)&virtio_hdr, vq->vhost_hlen); 834 PRINT_PACKET(dev, (uintptr_t)buff_hdr_addr, vq->vhost_hlen, 1); 835 mrg_count = 0; 836 } 837 } 838 if (res_cur_idx < res_end_idx) { 839 /* Prefetch descriptor index. */ 840 rte_prefetch0(&vq->desc[head[packet_success]]); 841 } 842 } 843 844 rte_compiler_barrier(); 845 846 /* Wait until it's our turn to add our buffer to the used ring. */ 847 while (unlikely(vq->last_used_idx != res_base_idx)) 848 rte_pause(); 849 850 *(volatile uint16_t *)&vq->used->idx += count; 851 vq->last_used_idx = res_end_idx; 852 853 /* Kick the guest if necessary. */ 854 if (!(vq->avail->flags & VRING_AVAIL_F_NO_INTERRUPT)) 855 eventfd_write((int)vq->kickfd, 1); 856 return count; 857 } 858 859 /* 860 * Compares a packet destination MAC address to a device MAC address. 861 */ 862 static inline int __attribute__((always_inline)) 863 ether_addr_cmp(struct ether_addr *ea, struct ether_addr *eb) 864 { 865 return (((*(uint64_t *)ea ^ *(uint64_t *)eb) & MAC_ADDR_CMP) == 0); 866 } 867 868 /* 869 * This function learns the MAC address of the device and registers this along with a 870 * vlan tag to a VMDQ. 871 */ 872 static int 873 link_vmdq(struct virtio_net *dev, struct rte_mbuf *m) 874 { 875 struct ether_hdr *pkt_hdr; 876 struct virtio_net_data_ll *dev_ll; 877 int i, ret; 878 879 /* Learn MAC address of guest device from packet */ 880 pkt_hdr = (struct ether_hdr *)m->pkt.data; 881 882 dev_ll = ll_root_used; 883 884 while (dev_ll != NULL) { 885 if (ether_addr_cmp(&(pkt_hdr->s_addr), &dev_ll->dev->mac_address)) { 886 RTE_LOG(INFO, DATA, "(%"PRIu64") WARNING: This device is using an existing MAC address and has not been registered.\n", dev->device_fh); 887 return -1; 888 } 889 dev_ll = dev_ll->next; 890 } 891 892 for (i = 0; i < ETHER_ADDR_LEN; i++) 893 dev->mac_address.addr_bytes[i] = pkt_hdr->s_addr.addr_bytes[i]; 894 895 /* vlan_tag currently uses the device_id. */ 896 dev->vlan_tag = vlan_tags[dev->device_fh]; 897 dev->vmdq_rx_q = dev->device_fh * (num_queues/num_devices); 898 899 /* Print out VMDQ registration info. */ 900 RTE_LOG(INFO, DATA, "(%"PRIu64") MAC_ADDRESS %02x:%02x:%02x:%02x:%02x:%02x and VLAN_TAG %d registered\n", 901 dev->device_fh, 902 dev->mac_address.addr_bytes[0], dev->mac_address.addr_bytes[1], 903 dev->mac_address.addr_bytes[2], dev->mac_address.addr_bytes[3], 904 dev->mac_address.addr_bytes[4], dev->mac_address.addr_bytes[5], 905 dev->vlan_tag); 906 907 /* Register the MAC address. */ 908 ret = rte_eth_dev_mac_addr_add(ports[0], &dev->mac_address, (uint32_t)dev->device_fh); 909 if (ret) 910 RTE_LOG(ERR, DATA, "(%"PRIu64") Failed to add device MAC address to VMDQ\n", 911 dev->device_fh); 912 913 /* Enable stripping of the vlan tag as we handle routing. */ 914 rte_eth_dev_set_vlan_strip_on_queue(ports[0], (uint16_t)dev->vmdq_rx_q, 1); 915 916 /* Set device as ready for RX. */ 917 dev->ready = DEVICE_RX; 918 919 return 0; 920 } 921 922 /* 923 * Removes MAC address and vlan tag from VMDQ. Ensures that nothing is adding buffers to the RX 924 * queue before disabling RX on the device. 925 */ 926 static inline void 927 unlink_vmdq(struct virtio_net *dev) 928 { 929 unsigned i = 0; 930 unsigned rx_count; 931 struct rte_mbuf *pkts_burst[MAX_PKT_BURST]; 932 933 if (dev->ready == DEVICE_RX) { 934 /*clear MAC and VLAN settings*/ 935 rte_eth_dev_mac_addr_remove(ports[0], &dev->mac_address); 936 for (i = 0; i < 6; i++) 937 dev->mac_address.addr_bytes[i] = 0; 938 939 dev->vlan_tag = 0; 940 941 /*Clear out the receive buffers*/ 942 rx_count = rte_eth_rx_burst(ports[0], 943 (uint16_t)dev->vmdq_rx_q, pkts_burst, MAX_PKT_BURST); 944 945 while (rx_count) { 946 for (i = 0; i < rx_count; i++) 947 rte_pktmbuf_free(pkts_burst[i]); 948 949 rx_count = rte_eth_rx_burst(ports[0], 950 (uint16_t)dev->vmdq_rx_q, pkts_burst, MAX_PKT_BURST); 951 } 952 953 dev->ready = DEVICE_MAC_LEARNING; 954 } 955 } 956 957 /* 958 * Check if the packet destination MAC address is for a local device. If so then put 959 * the packet on that devices RX queue. If not then return. 960 */ 961 static inline unsigned __attribute__((always_inline)) 962 virtio_tx_local(struct virtio_net *dev, struct rte_mbuf *m) 963 { 964 struct virtio_net_data_ll *dev_ll; 965 struct ether_hdr *pkt_hdr; 966 uint64_t ret = 0; 967 968 pkt_hdr = (struct ether_hdr *)m->pkt.data; 969 970 /*get the used devices list*/ 971 dev_ll = ll_root_used; 972 973 while (dev_ll != NULL) { 974 if ((dev_ll->dev->ready == DEVICE_RX) && ether_addr_cmp(&(pkt_hdr->d_addr), 975 &dev_ll->dev->mac_address)) { 976 977 /* Drop the packet if the TX packet is destined for the TX device. */ 978 if (dev_ll->dev->device_fh == dev->device_fh) { 979 LOG_DEBUG(DATA, "(%"PRIu64") TX: Source and destination MAC addresses are the same. Dropping packet.\n", 980 dev_ll->dev->device_fh); 981 return 0; 982 } 983 984 985 LOG_DEBUG(DATA, "(%"PRIu64") TX: MAC address is local\n", dev_ll->dev->device_fh); 986 987 if (dev_ll->dev->remove) { 988 /*drop the packet if the device is marked for removal*/ 989 LOG_DEBUG(DATA, "(%"PRIu64") Device is marked for removal\n", dev_ll->dev->device_fh); 990 } else { 991 /*send the packet to the local virtio device*/ 992 ret = virtio_dev_rx(dev_ll->dev, &m, 1); 993 if (enable_stats) { 994 rte_atomic64_add(&dev_statistics[dev_ll->dev->device_fh].rx_total, 1); 995 rte_atomic64_add(&dev_statistics[dev_ll->dev->device_fh].rx, ret); 996 dev_statistics[dev->device_fh].tx_total++; 997 dev_statistics[dev->device_fh].tx += ret; 998 } 999 } 1000 1001 return 0; 1002 } 1003 dev_ll = dev_ll->next; 1004 } 1005 1006 return -1; 1007 } 1008 1009 /* 1010 * This function routes the TX packet to the correct interface. This may be a local device 1011 * or the physical port. 1012 */ 1013 static inline void __attribute__((always_inline)) 1014 virtio_tx_route(struct virtio_net* dev, struct rte_mbuf *m, struct rte_mempool *mbuf_pool, uint16_t vlan_tag) 1015 { 1016 struct mbuf_table *tx_q; 1017 struct vlan_ethhdr *vlan_hdr; 1018 struct rte_mbuf **m_table; 1019 struct rte_mbuf *mbuf; 1020 unsigned len, ret; 1021 const uint16_t lcore_id = rte_lcore_id(); 1022 1023 /*check if destination is local VM*/ 1024 if (enable_vm2vm && (virtio_tx_local(dev, m) == 0)) { 1025 return; 1026 } 1027 1028 LOG_DEBUG(DATA, "(%"PRIu64") TX: MAC address is external\n", dev->device_fh); 1029 1030 /*Add packet to the port tx queue*/ 1031 tx_q = &lcore_tx_queue[lcore_id]; 1032 len = tx_q->len; 1033 1034 /* Allocate an mbuf and populate the structure. */ 1035 mbuf = rte_pktmbuf_alloc(mbuf_pool); 1036 if (unlikely(mbuf == NULL)) { 1037 RTE_LOG(ERR, DATA, "Failed to allocate memory for mbuf.\n"); 1038 return; 1039 } 1040 1041 mbuf->pkt.data_len = m->pkt.data_len + VLAN_HLEN; 1042 mbuf->pkt.pkt_len = mbuf->pkt.data_len; 1043 1044 /* Copy ethernet header to mbuf. */ 1045 rte_memcpy((void*)mbuf->pkt.data, (const void*)m->pkt.data, ETH_HLEN); 1046 1047 1048 /* Setup vlan header. Bytes need to be re-ordered for network with htons()*/ 1049 vlan_hdr = (struct vlan_ethhdr *) mbuf->pkt.data; 1050 vlan_hdr->h_vlan_encapsulated_proto = vlan_hdr->h_vlan_proto; 1051 vlan_hdr->h_vlan_proto = htons(ETH_P_8021Q); 1052 vlan_hdr->h_vlan_TCI = htons(vlan_tag); 1053 1054 /* Copy the remaining packet contents to the mbuf. */ 1055 rte_memcpy((void*) ((uint8_t*)mbuf->pkt.data + VLAN_ETH_HLEN), 1056 (const void*) ((uint8_t*)m->pkt.data + ETH_HLEN), (m->pkt.data_len - ETH_HLEN)); 1057 tx_q->m_table[len] = mbuf; 1058 len++; 1059 if (enable_stats) { 1060 dev_statistics[dev->device_fh].tx_total++; 1061 dev_statistics[dev->device_fh].tx++; 1062 } 1063 1064 if (unlikely(len == MAX_PKT_BURST)) { 1065 m_table = (struct rte_mbuf **)tx_q->m_table; 1066 ret = rte_eth_tx_burst(ports[0], (uint16_t)tx_q->txq_id, m_table, (uint16_t) len); 1067 /* Free any buffers not handled by TX and update the port stats. */ 1068 if (unlikely(ret < len)) { 1069 do { 1070 rte_pktmbuf_free(m_table[ret]); 1071 } while (++ret < len); 1072 } 1073 1074 len = 0; 1075 } 1076 1077 tx_q->len = len; 1078 return; 1079 } 1080 1081 static inline void __attribute__((always_inline)) 1082 virtio_dev_tx(struct virtio_net* dev, struct rte_mempool *mbuf_pool) 1083 { 1084 struct rte_mbuf m; 1085 struct vhost_virtqueue *vq; 1086 struct vring_desc *desc; 1087 uint64_t buff_addr = 0; 1088 uint32_t head[MAX_PKT_BURST]; 1089 uint32_t used_idx; 1090 uint32_t i; 1091 uint16_t free_entries, packet_success = 0; 1092 uint16_t avail_idx; 1093 1094 vq = dev->virtqueue[VIRTIO_TXQ]; 1095 avail_idx = *((volatile uint16_t *)&vq->avail->idx); 1096 1097 /* If there are no available buffers then return. */ 1098 if (vq->last_used_idx == avail_idx) 1099 return; 1100 1101 LOG_DEBUG(DATA, "(%"PRIu64") virtio_dev_tx()\n", dev->device_fh); 1102 1103 /* Prefetch available ring to retrieve head indexes. */ 1104 rte_prefetch0(&vq->avail->ring[vq->last_used_idx & (vq->size - 1)]); 1105 1106 /*get the number of free entries in the ring*/ 1107 free_entries = (avail_idx - vq->last_used_idx); 1108 1109 /* Limit to MAX_PKT_BURST. */ 1110 if (free_entries > MAX_PKT_BURST) 1111 free_entries = MAX_PKT_BURST; 1112 1113 LOG_DEBUG(DATA, "(%"PRIu64") Buffers available %d\n", dev->device_fh, free_entries); 1114 /* Retrieve all of the head indexes first to avoid caching issues. */ 1115 for (i = 0; i < free_entries; i++) 1116 head[i] = vq->avail->ring[(vq->last_used_idx + i) & (vq->size - 1)]; 1117 1118 /* Prefetch descriptor index. */ 1119 rte_prefetch0(&vq->desc[head[packet_success]]); 1120 rte_prefetch0(&vq->used->ring[vq->last_used_idx & (vq->size - 1)]); 1121 1122 while (packet_success < free_entries) { 1123 desc = &vq->desc[head[packet_success]]; 1124 1125 /* Discard first buffer as it is the virtio header */ 1126 desc = &vq->desc[desc->next]; 1127 1128 /* Buffer address translation. */ 1129 buff_addr = gpa_to_vva(dev, desc->addr); 1130 /* Prefetch buffer address. */ 1131 rte_prefetch0((void*)(uintptr_t)buff_addr); 1132 1133 used_idx = vq->last_used_idx & (vq->size - 1); 1134 1135 if (packet_success < (free_entries - 1)) { 1136 /* Prefetch descriptor index. */ 1137 rte_prefetch0(&vq->desc[head[packet_success+1]]); 1138 rte_prefetch0(&vq->used->ring[(used_idx + 1) & (vq->size - 1)]); 1139 } 1140 1141 /* Update used index buffer information. */ 1142 vq->used->ring[used_idx].id = head[packet_success]; 1143 vq->used->ring[used_idx].len = 0; 1144 1145 /* Setup dummy mbuf. This is copied to a real mbuf if transmitted out the physical port. */ 1146 m.pkt.data_len = desc->len; 1147 m.pkt.data = (void*)(uintptr_t)buff_addr; 1148 1149 PRINT_PACKET(dev, (uintptr_t)buff_addr, desc->len, 0); 1150 1151 /* If this is the first received packet we need to learn the MAC and setup VMDQ */ 1152 if (dev->ready == DEVICE_MAC_LEARNING) { 1153 if (dev->remove || (link_vmdq(dev, &m) == -1)) { 1154 /*discard frame if device is scheduled for removal or a duplicate MAC address is found. */ 1155 packet_success += free_entries; 1156 vq->last_used_idx += packet_success; 1157 break; 1158 } 1159 } 1160 virtio_tx_route(dev, &m, mbuf_pool, (uint16_t)dev->device_fh); 1161 1162 vq->last_used_idx++; 1163 packet_success++; 1164 } 1165 1166 rte_compiler_barrier(); 1167 vq->used->idx += packet_success; 1168 /* Kick guest if required. */ 1169 if (!(vq->avail->flags & VRING_AVAIL_F_NO_INTERRUPT)) 1170 eventfd_write((int)vq->kickfd, 1); 1171 } 1172 1173 /* 1174 * This function is called by each data core. It handles all RX/TX registered with the 1175 * core. For TX the specific lcore linked list is used. For RX, MAC addresses are compared 1176 * with all devices in the main linked list. 1177 */ 1178 static int 1179 switch_worker(__attribute__((unused)) void *arg) 1180 { 1181 struct rte_mempool *mbuf_pool = arg; 1182 struct virtio_net *dev = NULL; 1183 struct rte_mbuf *pkts_burst[MAX_PKT_BURST]; 1184 struct virtio_net_data_ll *dev_ll; 1185 struct mbuf_table *tx_q; 1186 volatile struct lcore_ll_info *lcore_ll; 1187 const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US; 1188 uint64_t prev_tsc, diff_tsc, cur_tsc, ret_count = 0; 1189 unsigned ret, i; 1190 const uint16_t lcore_id = rte_lcore_id(); 1191 const uint16_t num_cores = (uint16_t)rte_lcore_count(); 1192 uint16_t rx_count = 0; 1193 1194 RTE_LOG(INFO, DATA, "Procesing on Core %u started \n", lcore_id); 1195 lcore_ll = lcore_info[lcore_id].lcore_ll; 1196 prev_tsc = 0; 1197 1198 tx_q = &lcore_tx_queue[lcore_id]; 1199 for (i = 0; i < num_cores; i ++) { 1200 if (lcore_ids[i] == lcore_id) { 1201 tx_q->txq_id = i; 1202 break; 1203 } 1204 } 1205 1206 while(1) { 1207 cur_tsc = rte_rdtsc(); 1208 /* 1209 * TX burst queue drain 1210 */ 1211 diff_tsc = cur_tsc - prev_tsc; 1212 if (unlikely(diff_tsc > drain_tsc)) { 1213 1214 if (tx_q->len) { 1215 LOG_DEBUG(DATA, "TX queue drained after timeout with burst size %u \n", tx_q->len); 1216 1217 /*Tx any packets in the queue*/ 1218 ret = rte_eth_tx_burst(ports[0], (uint16_t)tx_q->txq_id, 1219 (struct rte_mbuf **)tx_q->m_table, 1220 (uint16_t)tx_q->len); 1221 if (unlikely(ret < tx_q->len)) { 1222 do { 1223 rte_pktmbuf_free(tx_q->m_table[ret]); 1224 } while (++ret < tx_q->len); 1225 } 1226 1227 tx_q->len = 0; 1228 } 1229 1230 prev_tsc = cur_tsc; 1231 1232 } 1233 1234 rte_prefetch0(lcore_ll->ll_root_used); 1235 /* 1236 * Inform the configuration core that we have exited the linked list and that no devices are 1237 * in use if requested. 1238 */ 1239 if (lcore_ll->dev_removal_flag == REQUEST_DEV_REMOVAL) 1240 lcore_ll->dev_removal_flag = ACK_DEV_REMOVAL; 1241 1242 /* 1243 * Process devices 1244 */ 1245 dev_ll = lcore_ll->ll_root_used; 1246 1247 while (dev_ll != NULL) { 1248 /*get virtio device ID*/ 1249 dev = dev_ll->dev; 1250 1251 if (dev->remove) { 1252 dev_ll = dev_ll->next; 1253 unlink_vmdq(dev); 1254 dev->ready = DEVICE_SAFE_REMOVE; 1255 continue; 1256 } 1257 if (likely(dev->ready == DEVICE_RX)) { 1258 /*Handle guest RX*/ 1259 rx_count = rte_eth_rx_burst(ports[0], 1260 (uint16_t)dev->vmdq_rx_q, pkts_burst, MAX_PKT_BURST); 1261 1262 if (rx_count) { 1263 ret_count = virtio_dev_rx(dev, pkts_burst, rx_count); 1264 if (enable_stats) { 1265 rte_atomic64_add(&dev_statistics[dev_ll->dev->device_fh].rx_total, rx_count); 1266 rte_atomic64_add(&dev_statistics[dev_ll->dev->device_fh].rx, ret_count); 1267 } 1268 while (likely(rx_count)) { 1269 rx_count--; 1270 rte_pktmbuf_free_seg(pkts_burst[rx_count]); 1271 } 1272 1273 } 1274 } 1275 1276 if (!dev->remove) 1277 /*Handle guest TX*/ 1278 virtio_dev_tx(dev, mbuf_pool); 1279 1280 /*move to the next device in the list*/ 1281 dev_ll = dev_ll->next; 1282 } 1283 } 1284 1285 return 0; 1286 } 1287 1288 /* 1289 * Add an entry to a used linked list. A free entry must first be found in the free linked list 1290 * using get_data_ll_free_entry(); 1291 */ 1292 static void 1293 add_data_ll_entry(struct virtio_net_data_ll **ll_root_addr, struct virtio_net_data_ll *ll_dev) 1294 { 1295 struct virtio_net_data_ll *ll = *ll_root_addr; 1296 1297 /* Set next as NULL and use a compiler barrier to avoid reordering. */ 1298 ll_dev->next = NULL; 1299 rte_compiler_barrier(); 1300 1301 /* If ll == NULL then this is the first device. */ 1302 if (ll) { 1303 /* Increment to the tail of the linked list. */ 1304 while ((ll->next != NULL) ) 1305 ll = ll->next; 1306 1307 ll->next = ll_dev; 1308 } else { 1309 *ll_root_addr = ll_dev; 1310 } 1311 } 1312 1313 /* 1314 * Remove an entry from a used linked list. The entry must then be added to the free linked list 1315 * using put_data_ll_free_entry(). 1316 */ 1317 static void 1318 rm_data_ll_entry(struct virtio_net_data_ll **ll_root_addr, struct virtio_net_data_ll *ll_dev, struct virtio_net_data_ll *ll_dev_last) 1319 { 1320 struct virtio_net_data_ll *ll = *ll_root_addr; 1321 1322 if (unlikely((ll == NULL) || (ll_dev == NULL))) 1323 return; 1324 1325 if (ll_dev == ll) 1326 *ll_root_addr = ll_dev->next; 1327 else 1328 if (likely(ll_dev_last != NULL)) 1329 ll_dev_last->next = ll_dev->next; 1330 else 1331 RTE_LOG(ERR, CONFIG, "Remove entry form ll failed.\n"); 1332 } 1333 1334 /* 1335 * Find and return an entry from the free linked list. 1336 */ 1337 static struct virtio_net_data_ll * 1338 get_data_ll_free_entry(struct virtio_net_data_ll **ll_root_addr) 1339 { 1340 struct virtio_net_data_ll *ll_free = *ll_root_addr; 1341 struct virtio_net_data_ll *ll_dev; 1342 1343 if (ll_free == NULL) 1344 return NULL; 1345 1346 ll_dev = ll_free; 1347 *ll_root_addr = ll_free->next; 1348 1349 return ll_dev; 1350 } 1351 1352 /* 1353 * Place an entry back on to the free linked list. 1354 */ 1355 static void 1356 put_data_ll_free_entry(struct virtio_net_data_ll **ll_root_addr, struct virtio_net_data_ll *ll_dev) 1357 { 1358 struct virtio_net_data_ll *ll_free = *ll_root_addr; 1359 1360 if (ll_dev == NULL) 1361 return; 1362 1363 ll_dev->next = ll_free; 1364 *ll_root_addr = ll_dev; 1365 } 1366 1367 /* 1368 * Creates a linked list of a given size. 1369 */ 1370 static struct virtio_net_data_ll * 1371 alloc_data_ll(uint32_t size) 1372 { 1373 struct virtio_net_data_ll *ll_new; 1374 uint32_t i; 1375 1376 /* Malloc and then chain the linked list. */ 1377 ll_new = malloc(size * sizeof(struct virtio_net_data_ll)); 1378 if (ll_new == NULL) { 1379 RTE_LOG(ERR, CONFIG, "Failed to allocate memory for ll_new.\n"); 1380 return NULL; 1381 } 1382 1383 for (i = 0; i < size - 1; i++) { 1384 ll_new[i].dev = NULL; 1385 ll_new[i].next = &ll_new[i+1]; 1386 } 1387 ll_new[i].next = NULL; 1388 1389 return (ll_new); 1390 } 1391 1392 /* 1393 * Create the main linked list along with each individual cores linked list. A used and a free list 1394 * are created to manage entries. 1395 */ 1396 static int 1397 init_data_ll (void) 1398 { 1399 int lcore; 1400 1401 RTE_LCORE_FOREACH_SLAVE(lcore) { 1402 lcore_info[lcore].lcore_ll = malloc(sizeof(struct lcore_ll_info)); 1403 if (lcore_info[lcore].lcore_ll == NULL) { 1404 RTE_LOG(ERR, CONFIG, "Failed to allocate memory for lcore_ll.\n"); 1405 return -1; 1406 } 1407 1408 lcore_info[lcore].lcore_ll->device_num = 0; 1409 lcore_info[lcore].lcore_ll->dev_removal_flag = ACK_DEV_REMOVAL; 1410 lcore_info[lcore].lcore_ll->ll_root_used = NULL; 1411 if (num_devices % num_switching_cores) 1412 lcore_info[lcore].lcore_ll->ll_root_free = alloc_data_ll((num_devices / num_switching_cores) + 1); 1413 else 1414 lcore_info[lcore].lcore_ll->ll_root_free = alloc_data_ll(num_devices / num_switching_cores); 1415 } 1416 1417 /* Allocate devices up to a maximum of MAX_DEVICES. */ 1418 ll_root_free = alloc_data_ll(MIN((num_devices), MAX_DEVICES)); 1419 1420 return 0; 1421 } 1422 1423 /* 1424 * Set virtqueue flags so that we do not receive interrupts. 1425 */ 1426 static void 1427 set_irq_status (struct virtio_net *dev) 1428 { 1429 dev->virtqueue[VIRTIO_RXQ]->used->flags = VRING_USED_F_NO_NOTIFY; 1430 dev->virtqueue[VIRTIO_TXQ]->used->flags = VRING_USED_F_NO_NOTIFY; 1431 } 1432 1433 /* 1434 * Remove a device from the specific data core linked list and from the main linked list. Synchonization 1435 * occurs through the use of the lcore dev_removal_flag. Device is made volatile here to avoid re-ordering 1436 * of dev->remove=1 which can cause an infinite loop in the rte_pause loop. 1437 */ 1438 static void 1439 destroy_device (volatile struct virtio_net *dev) 1440 { 1441 struct virtio_net_data_ll *ll_lcore_dev_cur; 1442 struct virtio_net_data_ll *ll_main_dev_cur; 1443 struct virtio_net_data_ll *ll_lcore_dev_last = NULL; 1444 struct virtio_net_data_ll *ll_main_dev_last = NULL; 1445 int lcore; 1446 1447 dev->flags &= ~VIRTIO_DEV_RUNNING; 1448 1449 /*set the remove flag. */ 1450 dev->remove = 1; 1451 1452 while(dev->ready != DEVICE_SAFE_REMOVE) { 1453 rte_pause(); 1454 } 1455 1456 /* Search for entry to be removed from lcore ll */ 1457 ll_lcore_dev_cur = lcore_info[dev->coreid].lcore_ll->ll_root_used; 1458 while (ll_lcore_dev_cur != NULL) { 1459 if (ll_lcore_dev_cur->dev == dev) { 1460 break; 1461 } else { 1462 ll_lcore_dev_last = ll_lcore_dev_cur; 1463 ll_lcore_dev_cur = ll_lcore_dev_cur->next; 1464 } 1465 } 1466 1467 if (ll_lcore_dev_cur == NULL) { 1468 RTE_LOG(ERR, CONFIG, "Failed to find the dev to be destroy.\n"); 1469 return; 1470 } 1471 1472 /* Search for entry to be removed from main ll */ 1473 ll_main_dev_cur = ll_root_used; 1474 ll_main_dev_last = NULL; 1475 while (ll_main_dev_cur != NULL) { 1476 if (ll_main_dev_cur->dev == dev) { 1477 break; 1478 } else { 1479 ll_main_dev_last = ll_main_dev_cur; 1480 ll_main_dev_cur = ll_main_dev_cur->next; 1481 } 1482 } 1483 1484 /* Remove entries from the lcore and main ll. */ 1485 rm_data_ll_entry(&lcore_info[ll_lcore_dev_cur->dev->coreid].lcore_ll->ll_root_used, ll_lcore_dev_cur, ll_lcore_dev_last); 1486 rm_data_ll_entry(&ll_root_used, ll_main_dev_cur, ll_main_dev_last); 1487 1488 /* Set the dev_removal_flag on each lcore. */ 1489 RTE_LCORE_FOREACH_SLAVE(lcore) { 1490 lcore_info[lcore].lcore_ll->dev_removal_flag = REQUEST_DEV_REMOVAL; 1491 } 1492 1493 /* 1494 * Once each core has set the dev_removal_flag to ACK_DEV_REMOVAL we can be sure that 1495 * they can no longer access the device removed from the linked lists and that the devices 1496 * are no longer in use. 1497 */ 1498 RTE_LCORE_FOREACH_SLAVE(lcore) { 1499 while (lcore_info[lcore].lcore_ll->dev_removal_flag != ACK_DEV_REMOVAL) { 1500 rte_pause(); 1501 } 1502 } 1503 1504 /* Add the entries back to the lcore and main free ll.*/ 1505 put_data_ll_free_entry(&lcore_info[ll_lcore_dev_cur->dev->coreid].lcore_ll->ll_root_free, ll_lcore_dev_cur); 1506 put_data_ll_free_entry(&ll_root_free, ll_main_dev_cur); 1507 1508 /* Decrement number of device on the lcore. */ 1509 lcore_info[ll_lcore_dev_cur->dev->coreid].lcore_ll->device_num--; 1510 1511 RTE_LOG(INFO, DATA, "(%"PRIu64") Device has been removed from data core\n", dev->device_fh); 1512 } 1513 1514 /* 1515 * A new device is added to a data core. First the device is added to the main linked list 1516 * and the allocated to a specific data core. 1517 */ 1518 static int 1519 new_device (struct virtio_net *dev) 1520 { 1521 struct virtio_net_data_ll *ll_dev; 1522 int lcore, core_add = 0; 1523 uint32_t device_num_min = num_devices; 1524 1525 /* Add device to main ll */ 1526 ll_dev = get_data_ll_free_entry(&ll_root_free); 1527 if (ll_dev == NULL) { 1528 RTE_LOG(INFO, DATA, "(%"PRIu64") No free entry found in linked list. Device limit " 1529 "of %d devices per core has been reached\n", 1530 dev->device_fh, num_devices); 1531 return -1; 1532 } 1533 ll_dev->dev = dev; 1534 add_data_ll_entry(&ll_root_used, ll_dev); 1535 1536 /*reset ready flag*/ 1537 dev->ready = DEVICE_MAC_LEARNING; 1538 dev->remove = 0; 1539 1540 /* Find a suitable lcore to add the device. */ 1541 RTE_LCORE_FOREACH_SLAVE(lcore) { 1542 if (lcore_info[lcore].lcore_ll->device_num < device_num_min) { 1543 device_num_min = lcore_info[lcore].lcore_ll->device_num; 1544 core_add = lcore; 1545 } 1546 } 1547 /* Add device to lcore ll */ 1548 ll_dev->dev->coreid = core_add; 1549 ll_dev = get_data_ll_free_entry(&lcore_info[ll_dev->dev->coreid].lcore_ll->ll_root_free); 1550 if (ll_dev == NULL) { 1551 RTE_LOG(INFO, DATA, "(%"PRIu64") Failed to add device to data core\n", dev->device_fh); 1552 destroy_device(dev); 1553 return -1; 1554 } 1555 ll_dev->dev = dev; 1556 add_data_ll_entry(&lcore_info[ll_dev->dev->coreid].lcore_ll->ll_root_used, ll_dev); 1557 1558 /* Initialize device stats */ 1559 memset(&dev_statistics[dev->device_fh], 0, sizeof(struct device_statistics)); 1560 1561 /* Disable notifications. */ 1562 set_irq_status(dev); 1563 lcore_info[ll_dev->dev->coreid].lcore_ll->device_num++; 1564 dev->flags |= VIRTIO_DEV_RUNNING; 1565 1566 RTE_LOG(INFO, DATA, "(%"PRIu64") Device has been added to data core %d\n", dev->device_fh, dev->coreid); 1567 1568 return 0; 1569 } 1570 1571 /* 1572 * These callback allow devices to be added to the data core when configuration 1573 * has been fully complete. 1574 */ 1575 static const struct virtio_net_device_ops virtio_net_device_ops = 1576 { 1577 .new_device = new_device, 1578 .destroy_device = destroy_device, 1579 }; 1580 1581 /* 1582 * This is a thread will wake up after a period to print stats if the user has 1583 * enabled them. 1584 */ 1585 static void 1586 print_stats(void) 1587 { 1588 struct virtio_net_data_ll *dev_ll; 1589 uint64_t tx_dropped, rx_dropped; 1590 uint64_t tx, tx_total, rx, rx_total; 1591 uint32_t device_fh; 1592 const char clr[] = { 27, '[', '2', 'J', '\0' }; 1593 const char top_left[] = { 27, '[', '1', ';', '1', 'H','\0' }; 1594 1595 while(1) { 1596 sleep(enable_stats); 1597 1598 /* Clear screen and move to top left */ 1599 printf("%s%s", clr, top_left); 1600 1601 printf("\nDevice statistics ===================================="); 1602 1603 dev_ll = ll_root_used; 1604 while (dev_ll != NULL) { 1605 device_fh = (uint32_t)dev_ll->dev->device_fh; 1606 tx_total = dev_statistics[device_fh].tx_total; 1607 tx = dev_statistics[device_fh].tx; 1608 tx_dropped = tx_total - tx; 1609 rx_total = rte_atomic64_read(&dev_statistics[device_fh].rx_total); 1610 rx = rte_atomic64_read(&dev_statistics[device_fh].rx); 1611 rx_dropped = rx_total - rx; 1612 1613 printf("\nStatistics for device %"PRIu32" ------------------------------" 1614 "\nTX total: %"PRIu64"" 1615 "\nTX dropped: %"PRIu64"" 1616 "\nTX successful: %"PRIu64"" 1617 "\nRX total: %"PRIu64"" 1618 "\nRX dropped: %"PRIu64"" 1619 "\nRX successful: %"PRIu64"", 1620 device_fh, 1621 tx_total, 1622 tx_dropped, 1623 tx, 1624 rx_total, 1625 rx_dropped, 1626 rx); 1627 1628 dev_ll = dev_ll->next; 1629 } 1630 printf("\n======================================================\n"); 1631 } 1632 } 1633 1634 /* 1635 * Main function, does initialisation and calls the per-lcore functions. The CUSE 1636 * device is also registered here to handle the IOCTLs. 1637 */ 1638 int 1639 MAIN(int argc, char *argv[]) 1640 { 1641 struct rte_mempool *mbuf_pool; 1642 unsigned lcore_id, core_id = 0; 1643 unsigned nb_ports, valid_num_ports; 1644 int ret; 1645 uint8_t portid; 1646 static pthread_t tid; 1647 1648 /* init EAL */ 1649 ret = rte_eal_init(argc, argv); 1650 if (ret < 0) 1651 rte_exit(EXIT_FAILURE, "Error with EAL initialization\n"); 1652 argc -= ret; 1653 argv += ret; 1654 1655 /* parse app arguments */ 1656 ret = us_vhost_parse_args(argc, argv); 1657 if (ret < 0) 1658 rte_exit(EXIT_FAILURE, "Invalid argument\n"); 1659 1660 if (rte_eal_pci_probe() != 0) 1661 rte_exit(EXIT_FAILURE, "Error with NIC driver initialization\n"); 1662 1663 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id ++) 1664 if (rte_lcore_is_enabled(lcore_id)) 1665 lcore_ids[core_id ++] = lcore_id; 1666 1667 if (rte_lcore_count() > RTE_MAX_LCORE) 1668 rte_exit(EXIT_FAILURE,"Not enough cores\n"); 1669 1670 /*set the number of swithcing cores available*/ 1671 num_switching_cores = rte_lcore_count()-1; 1672 1673 /* Get the number of physical ports. */ 1674 nb_ports = rte_eth_dev_count(); 1675 if (nb_ports > RTE_MAX_ETHPORTS) 1676 nb_ports = RTE_MAX_ETHPORTS; 1677 1678 /* 1679 * Update the global var NUM_PORTS and global array PORTS 1680 * and get value of var VALID_NUM_PORTS according to system ports number 1681 */ 1682 valid_num_ports = check_ports_num(nb_ports); 1683 1684 if ((valid_num_ports == 0) || (valid_num_ports > MAX_SUP_PORTS)) { 1685 RTE_LOG(INFO, PORT, "Current enabled port number is %u," 1686 "but only %u port can be enabled\n",num_ports, MAX_SUP_PORTS); 1687 return -1; 1688 } 1689 1690 /* Create the mbuf pool. */ 1691 mbuf_pool = rte_mempool_create("MBUF_POOL", NUM_MBUFS_PER_PORT * valid_num_ports, 1692 MBUF_SIZE, MBUF_CACHE_SIZE, 1693 sizeof(struct rte_pktmbuf_pool_private), 1694 rte_pktmbuf_pool_init, NULL, 1695 rte_pktmbuf_init, NULL, 1696 rte_socket_id(), 0); 1697 if (mbuf_pool == NULL) 1698 rte_exit(EXIT_FAILURE, "Cannot create mbuf pool\n"); 1699 1700 /* Set log level. */ 1701 rte_set_log_level(LOG_LEVEL); 1702 1703 /* initialize all ports */ 1704 for (portid = 0; portid < nb_ports; portid++) { 1705 /* skip ports that are not enabled */ 1706 if ((enabled_port_mask & (1 << portid)) == 0) { 1707 RTE_LOG(INFO, PORT, "Skipping disabled port %d\n", portid); 1708 continue; 1709 } 1710 if (port_init(portid, mbuf_pool) != 0) 1711 rte_exit(EXIT_FAILURE, "Cannot initialize network ports\n"); 1712 } 1713 1714 /* Initialise all linked lists. */ 1715 if (init_data_ll() == -1) 1716 rte_exit(EXIT_FAILURE, "Failed to initialize linked list\n"); 1717 1718 /* Initialize device stats */ 1719 memset(&dev_statistics, 0, sizeof(dev_statistics)); 1720 1721 /* Enable stats if the user option is set. */ 1722 if (enable_stats) 1723 pthread_create(&tid, NULL, (void*)print_stats, NULL ); 1724 1725 /* Launch all data cores. */ 1726 RTE_LCORE_FOREACH_SLAVE(lcore_id) { 1727 rte_eal_remote_launch(switch_worker, mbuf_pool, lcore_id); 1728 } 1729 1730 /* Register CUSE device to handle IOCTLs. */ 1731 ret = register_cuse_device((char*)&dev_basename, dev_index, get_virtio_net_callbacks()); 1732 if (ret != 0) 1733 rte_exit(EXIT_FAILURE,"CUSE device setup failure.\n"); 1734 1735 init_virtio_net(&virtio_net_device_ops); 1736 1737 /* Start CUSE session. */ 1738 start_cuse_session_loop(); 1739 return 0; 1740 1741 } 1742 1743