1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2010-2017 Intel Corporation
7 #include <linux/if_ether.h>
8 #include <linux/if_vlan.h>
9 #include <linux/virtio_net.h>
10 #include <linux/virtio_ring.h>
13 #include <sys/eventfd.h>
14 #include <sys/param.h>
17 #include <rte_atomic.h>
18 #include <rte_cycles.h>
19 #include <rte_ethdev.h>
21 #include <rte_string_fns.h>
22 #include <rte_malloc.h>
23 #include <rte_vhost.h>
26 #include <rte_pause.h>
31 #define MAX_QUEUES 128
34 /* the maximum number of external ports supported */
35 #define MAX_SUP_PORTS 1
37 #define MBUF_CACHE_SIZE 128
38 #define MBUF_DATA_SIZE RTE_MBUF_DEFAULT_BUF_SIZE
40 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
42 #define BURST_RX_WAIT_US 15 /* Defines how long we wait between retries on RX */
43 #define BURST_RX_RETRIES 4 /* Number of retries on RX. */
45 #define JUMBO_FRAME_MAX_SIZE 0x2600
47 /* State of virtio device. */
48 #define DEVICE_MAC_LEARNING 0
50 #define DEVICE_SAFE_REMOVE 2
52 /* Configurable number of RX/TX ring descriptors */
53 #define RTE_TEST_RX_DESC_DEFAULT 1024
54 #define RTE_TEST_TX_DESC_DEFAULT 512
56 #define INVALID_PORT_ID 0xFF
58 /* Max number of devices. Limited by vmdq. */
59 #define MAX_DEVICES 64
61 /* Size of buffers used for snprintfs. */
62 #define MAX_PRINT_BUFF 6072
64 /* Maximum long option length for option parsing. */
65 #define MAX_LONG_OPT_SZ 64
67 /* mask of enabled ports */
68 static uint32_t enabled_port_mask = 0;
70 /* Promiscuous mode */
71 static uint32_t promiscuous;
73 /* number of devices/queues to support*/
74 static uint32_t num_queues = 0;
75 static uint32_t num_devices;
77 static struct rte_mempool *mbuf_pool;
80 /* Enable VM2VM communications. If this is disabled then the MAC address compare is skipped. */
87 static vm2vm_type vm2vm_mode = VM2VM_SOFTWARE;
90 static uint32_t enable_stats = 0;
91 /* Enable retries on RX. */
92 static uint32_t enable_retry = 1;
94 /* Disable TX checksum offload */
95 static uint32_t enable_tx_csum;
97 /* Disable TSO offload */
98 static uint32_t enable_tso;
100 static int client_mode;
101 static int dequeue_zero_copy;
103 static int builtin_net_driver;
105 /* Specify timeout (in useconds) between retries on RX. */
106 static uint32_t burst_rx_delay_time = BURST_RX_WAIT_US;
107 /* Specify the number of retries on RX. */
108 static uint32_t burst_rx_retry_num = BURST_RX_RETRIES;
110 /* Socket file paths. Can be set by user */
111 static char *socket_files;
112 static int nb_sockets;
114 /* empty vmdq configuration structure. Filled in programatically */
115 static struct rte_eth_conf vmdq_conf_default = {
117 .mq_mode = ETH_MQ_RX_VMDQ_ONLY,
119 .ignore_offload_bitfield = 1,
121 * VLAN strip is necessary for 1G NIC such as I350,
122 * this fixes bug of ipv4 forwarding in guest can't
123 * forward pakets from one virtio dev to another virtio dev.
125 .offloads = (DEV_RX_OFFLOAD_CRC_STRIP |
126 DEV_RX_OFFLOAD_VLAN_STRIP),
130 .mq_mode = ETH_MQ_TX_NONE,
131 .offloads = (DEV_TX_OFFLOAD_IPV4_CKSUM |
132 DEV_TX_OFFLOAD_TCP_CKSUM |
133 DEV_TX_OFFLOAD_VLAN_INSERT |
134 DEV_TX_OFFLOAD_MULTI_SEGS |
135 DEV_TX_OFFLOAD_TCP_TSO),
139 * should be overridden separately in code with
143 .nb_queue_pools = ETH_8_POOLS,
144 .enable_default_pool = 0,
147 .pool_map = {{0, 0},},
153 static unsigned lcore_ids[RTE_MAX_LCORE];
154 static uint16_t ports[RTE_MAX_ETHPORTS];
155 static unsigned num_ports = 0; /**< The number of ports specified in command line */
156 static uint16_t num_pf_queues, num_vmdq_queues;
157 static uint16_t vmdq_pool_base, vmdq_queue_base;
158 static uint16_t queues_per_pool;
160 const uint16_t vlan_tags[] = {
161 1000, 1001, 1002, 1003, 1004, 1005, 1006, 1007,
162 1008, 1009, 1010, 1011, 1012, 1013, 1014, 1015,
163 1016, 1017, 1018, 1019, 1020, 1021, 1022, 1023,
164 1024, 1025, 1026, 1027, 1028, 1029, 1030, 1031,
165 1032, 1033, 1034, 1035, 1036, 1037, 1038, 1039,
166 1040, 1041, 1042, 1043, 1044, 1045, 1046, 1047,
167 1048, 1049, 1050, 1051, 1052, 1053, 1054, 1055,
168 1056, 1057, 1058, 1059, 1060, 1061, 1062, 1063,
171 /* ethernet addresses of ports */
172 static struct ether_addr vmdq_ports_eth_addr[RTE_MAX_ETHPORTS];
174 static struct vhost_dev_tailq_list vhost_dev_list =
175 TAILQ_HEAD_INITIALIZER(vhost_dev_list);
177 static struct lcore_info lcore_info[RTE_MAX_LCORE];
179 /* Used for queueing bursts of TX packets. */
183 struct rte_mbuf *m_table[MAX_PKT_BURST];
186 /* TX queue for each data core. */
187 struct mbuf_table lcore_tx_queue[RTE_MAX_LCORE];
189 #define MBUF_TABLE_DRAIN_TSC ((rte_get_tsc_hz() + US_PER_S - 1) \
190 / US_PER_S * BURST_TX_DRAIN_US)
194 * Builds up the correct configuration for VMDQ VLAN pool map
195 * according to the pool & queue limits.
198 get_eth_conf(struct rte_eth_conf *eth_conf, uint32_t num_devices)
200 struct rte_eth_vmdq_rx_conf conf;
201 struct rte_eth_vmdq_rx_conf *def_conf =
202 &vmdq_conf_default.rx_adv_conf.vmdq_rx_conf;
205 memset(&conf, 0, sizeof(conf));
206 conf.nb_queue_pools = (enum rte_eth_nb_pools)num_devices;
207 conf.nb_pool_maps = num_devices;
208 conf.enable_loop_back = def_conf->enable_loop_back;
209 conf.rx_mode = def_conf->rx_mode;
211 for (i = 0; i < conf.nb_pool_maps; i++) {
212 conf.pool_map[i].vlan_id = vlan_tags[ i ];
213 conf.pool_map[i].pools = (1UL << i);
216 (void)(rte_memcpy(eth_conf, &vmdq_conf_default, sizeof(*eth_conf)));
217 (void)(rte_memcpy(ð_conf->rx_adv_conf.vmdq_rx_conf, &conf,
218 sizeof(eth_conf->rx_adv_conf.vmdq_rx_conf)));
223 * Validate the device number according to the max pool number gotten form
224 * dev_info. If the device number is invalid, give the error message and
225 * return -1. Each device must have its own pool.
228 validate_num_devices(uint32_t max_nb_devices)
230 if (num_devices > max_nb_devices) {
231 RTE_LOG(ERR, VHOST_PORT, "invalid number of devices\n");
238 * Initialises a given port using global settings and with the rx buffers
239 * coming from the mbuf_pool passed as parameter
242 port_init(uint16_t port)
244 struct rte_eth_dev_info dev_info;
245 struct rte_eth_conf port_conf;
246 struct rte_eth_rxconf *rxconf;
247 struct rte_eth_txconf *txconf;
248 int16_t rx_rings, tx_rings;
249 uint16_t rx_ring_size, tx_ring_size;
253 /* The max pool number from dev_info will be used to validate the pool number specified in cmd line */
254 rte_eth_dev_info_get (port, &dev_info);
256 rxconf = &dev_info.default_rxconf;
257 txconf = &dev_info.default_txconf;
258 rxconf->rx_drop_en = 1;
259 txconf->txq_flags = ETH_TXQ_FLAGS_IGNORE;
261 /*configure the number of supported virtio devices based on VMDQ limits */
262 num_devices = dev_info.max_vmdq_pools;
264 rx_ring_size = RTE_TEST_RX_DESC_DEFAULT;
265 tx_ring_size = RTE_TEST_TX_DESC_DEFAULT;
268 * When dequeue zero copy is enabled, guest Tx used vring will be
269 * updated only when corresponding mbuf is freed. Thus, the nb_tx_desc
270 * (tx_ring_size here) must be small enough so that the driver will
271 * hit the free threshold easily and free mbufs timely. Otherwise,
272 * guest Tx vring would be starved.
274 if (dequeue_zero_copy)
277 tx_rings = (uint16_t)rte_lcore_count();
279 retval = validate_num_devices(MAX_DEVICES);
283 /* Get port configuration. */
284 retval = get_eth_conf(&port_conf, num_devices);
287 /* NIC queues are divided into pf queues and vmdq queues. */
288 num_pf_queues = dev_info.max_rx_queues - dev_info.vmdq_queue_num;
289 queues_per_pool = dev_info.vmdq_queue_num / dev_info.max_vmdq_pools;
290 num_vmdq_queues = num_devices * queues_per_pool;
291 num_queues = num_pf_queues + num_vmdq_queues;
292 vmdq_queue_base = dev_info.vmdq_queue_base;
293 vmdq_pool_base = dev_info.vmdq_pool_base;
294 printf("pf queue num: %u, configured vmdq pool num: %u, each vmdq pool has %u queues\n",
295 num_pf_queues, num_devices, queues_per_pool);
297 if (port >= rte_eth_dev_count()) return -1;
299 rx_rings = (uint16_t)dev_info.max_rx_queues;
300 if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
301 port_conf.txmode.offloads |=
302 DEV_TX_OFFLOAD_MBUF_FAST_FREE;
303 /* Configure ethernet device. */
304 retval = rte_eth_dev_configure(port, rx_rings, tx_rings, &port_conf);
306 RTE_LOG(ERR, VHOST_PORT, "Failed to configure port %u: %s.\n",
307 port, strerror(-retval));
311 retval = rte_eth_dev_adjust_nb_rx_tx_desc(port, &rx_ring_size,
314 RTE_LOG(ERR, VHOST_PORT, "Failed to adjust number of descriptors "
315 "for port %u: %s.\n", port, strerror(-retval));
318 if (rx_ring_size > RTE_TEST_RX_DESC_DEFAULT) {
319 RTE_LOG(ERR, VHOST_PORT, "Mbuf pool has an insufficient size "
320 "for Rx queues on port %u.\n", port);
324 /* Setup the queues. */
325 rxconf->offloads = port_conf.rxmode.offloads;
326 for (q = 0; q < rx_rings; q ++) {
327 retval = rte_eth_rx_queue_setup(port, q, rx_ring_size,
328 rte_eth_dev_socket_id(port),
332 RTE_LOG(ERR, VHOST_PORT,
333 "Failed to setup rx queue %u of port %u: %s.\n",
334 q, port, strerror(-retval));
338 txconf->offloads = port_conf.txmode.offloads;
339 for (q = 0; q < tx_rings; q ++) {
340 retval = rte_eth_tx_queue_setup(port, q, tx_ring_size,
341 rte_eth_dev_socket_id(port),
344 RTE_LOG(ERR, VHOST_PORT,
345 "Failed to setup tx queue %u of port %u: %s.\n",
346 q, port, strerror(-retval));
351 /* Start the device. */
352 retval = rte_eth_dev_start(port);
354 RTE_LOG(ERR, VHOST_PORT, "Failed to start port %u: %s\n",
355 port, strerror(-retval));
360 rte_eth_promiscuous_enable(port);
362 rte_eth_macaddr_get(port, &vmdq_ports_eth_addr[port]);
363 RTE_LOG(INFO, VHOST_PORT, "Max virtio devices supported: %u\n", num_devices);
364 RTE_LOG(INFO, VHOST_PORT, "Port %u MAC: %02"PRIx8" %02"PRIx8" %02"PRIx8
365 " %02"PRIx8" %02"PRIx8" %02"PRIx8"\n",
367 vmdq_ports_eth_addr[port].addr_bytes[0],
368 vmdq_ports_eth_addr[port].addr_bytes[1],
369 vmdq_ports_eth_addr[port].addr_bytes[2],
370 vmdq_ports_eth_addr[port].addr_bytes[3],
371 vmdq_ports_eth_addr[port].addr_bytes[4],
372 vmdq_ports_eth_addr[port].addr_bytes[5]);
378 * Set socket file path.
381 us_vhost_parse_socket_path(const char *q_arg)
383 /* parse number string */
384 if (strnlen(q_arg, PATH_MAX) == PATH_MAX)
387 socket_files = realloc(socket_files, PATH_MAX * (nb_sockets + 1));
388 snprintf(socket_files + nb_sockets * PATH_MAX, PATH_MAX, "%s", q_arg);
395 * Parse the portmask provided at run time.
398 parse_portmask(const char *portmask)
405 /* parse hexadecimal string */
406 pm = strtoul(portmask, &end, 16);
407 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0') || (errno != 0))
418 * Parse num options at run time.
421 parse_num_opt(const char *q_arg, uint32_t max_valid_value)
428 /* parse unsigned int string */
429 num = strtoul(q_arg, &end, 10);
430 if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0') || (errno != 0))
433 if (num > max_valid_value)
444 us_vhost_usage(const char *prgname)
446 RTE_LOG(INFO, VHOST_CONFIG, "%s [EAL options] -- -p PORTMASK\n"
448 " --rx_retry [0|1] --mergeable [0|1] --stats [0-N]\n"
449 " --socket-file <path>\n"
451 " -p PORTMASK: Set mask for ports to be used by application\n"
452 " --vm2vm [0|1|2]: disable/software(default)/hardware vm2vm comms\n"
453 " --rx-retry [0|1]: disable/enable(default) retries on rx. Enable retry if destintation queue is full\n"
454 " --rx-retry-delay [0-N]: timeout(in usecond) between retries on RX. This makes effect only if retries on rx enabled\n"
455 " --rx-retry-num [0-N]: the number of retries on rx. This makes effect only if retries on rx enabled\n"
456 " --mergeable [0|1]: disable(default)/enable RX mergeable buffers\n"
457 " --stats [0-N]: 0: Disable stats, N: Time in seconds to print stats\n"
458 " --socket-file: The path of the socket file.\n"
459 " --tx-csum [0|1] disable/enable TX checksum offload.\n"
460 " --tso [0|1] disable/enable TCP segment offload.\n"
461 " --client register a vhost-user socket as client mode.\n"
462 " --dequeue-zero-copy enables dequeue zero copy\n",
467 * Parse the arguments given in the command line of the application.
470 us_vhost_parse_args(int argc, char **argv)
475 const char *prgname = argv[0];
476 static struct option long_option[] = {
477 {"vm2vm", required_argument, NULL, 0},
478 {"rx-retry", required_argument, NULL, 0},
479 {"rx-retry-delay", required_argument, NULL, 0},
480 {"rx-retry-num", required_argument, NULL, 0},
481 {"mergeable", required_argument, NULL, 0},
482 {"stats", required_argument, NULL, 0},
483 {"socket-file", required_argument, NULL, 0},
484 {"tx-csum", required_argument, NULL, 0},
485 {"tso", required_argument, NULL, 0},
486 {"client", no_argument, &client_mode, 1},
487 {"dequeue-zero-copy", no_argument, &dequeue_zero_copy, 1},
488 {"builtin-net-driver", no_argument, &builtin_net_driver, 1},
492 /* Parse command line */
493 while ((opt = getopt_long(argc, argv, "p:P",
494 long_option, &option_index)) != EOF) {
498 enabled_port_mask = parse_portmask(optarg);
499 if (enabled_port_mask == 0) {
500 RTE_LOG(INFO, VHOST_CONFIG, "Invalid portmask\n");
501 us_vhost_usage(prgname);
508 vmdq_conf_default.rx_adv_conf.vmdq_rx_conf.rx_mode =
509 ETH_VMDQ_ACCEPT_BROADCAST |
510 ETH_VMDQ_ACCEPT_MULTICAST;
515 /* Enable/disable vm2vm comms. */
516 if (!strncmp(long_option[option_index].name, "vm2vm",
518 ret = parse_num_opt(optarg, (VM2VM_LAST - 1));
520 RTE_LOG(INFO, VHOST_CONFIG,
521 "Invalid argument for "
523 us_vhost_usage(prgname);
526 vm2vm_mode = (vm2vm_type)ret;
530 /* Enable/disable retries on RX. */
531 if (!strncmp(long_option[option_index].name, "rx-retry", MAX_LONG_OPT_SZ)) {
532 ret = parse_num_opt(optarg, 1);
534 RTE_LOG(INFO, VHOST_CONFIG, "Invalid argument for rx-retry [0|1]\n");
535 us_vhost_usage(prgname);
542 /* Enable/disable TX checksum offload. */
543 if (!strncmp(long_option[option_index].name, "tx-csum", MAX_LONG_OPT_SZ)) {
544 ret = parse_num_opt(optarg, 1);
546 RTE_LOG(INFO, VHOST_CONFIG, "Invalid argument for tx-csum [0|1]\n");
547 us_vhost_usage(prgname);
550 enable_tx_csum = ret;
553 /* Enable/disable TSO offload. */
554 if (!strncmp(long_option[option_index].name, "tso", MAX_LONG_OPT_SZ)) {
555 ret = parse_num_opt(optarg, 1);
557 RTE_LOG(INFO, VHOST_CONFIG, "Invalid argument for tso [0|1]\n");
558 us_vhost_usage(prgname);
564 /* Specify the retries delay time (in useconds) on RX. */
565 if (!strncmp(long_option[option_index].name, "rx-retry-delay", MAX_LONG_OPT_SZ)) {
566 ret = parse_num_opt(optarg, INT32_MAX);
568 RTE_LOG(INFO, VHOST_CONFIG, "Invalid argument for rx-retry-delay [0-N]\n");
569 us_vhost_usage(prgname);
572 burst_rx_delay_time = ret;
576 /* Specify the retries number on RX. */
577 if (!strncmp(long_option[option_index].name, "rx-retry-num", MAX_LONG_OPT_SZ)) {
578 ret = parse_num_opt(optarg, INT32_MAX);
580 RTE_LOG(INFO, VHOST_CONFIG, "Invalid argument for rx-retry-num [0-N]\n");
581 us_vhost_usage(prgname);
584 burst_rx_retry_num = ret;
588 /* Enable/disable RX mergeable buffers. */
589 if (!strncmp(long_option[option_index].name, "mergeable", MAX_LONG_OPT_SZ)) {
590 ret = parse_num_opt(optarg, 1);
592 RTE_LOG(INFO, VHOST_CONFIG, "Invalid argument for mergeable [0|1]\n");
593 us_vhost_usage(prgname);
598 vmdq_conf_default.rxmode.offloads |=
599 DEV_RX_OFFLOAD_JUMBO_FRAME;
600 vmdq_conf_default.rxmode.max_rx_pkt_len
601 = JUMBO_FRAME_MAX_SIZE;
606 /* Enable/disable stats. */
607 if (!strncmp(long_option[option_index].name, "stats", MAX_LONG_OPT_SZ)) {
608 ret = parse_num_opt(optarg, INT32_MAX);
610 RTE_LOG(INFO, VHOST_CONFIG,
611 "Invalid argument for stats [0..N]\n");
612 us_vhost_usage(prgname);
619 /* Set socket file path. */
620 if (!strncmp(long_option[option_index].name,
621 "socket-file", MAX_LONG_OPT_SZ)) {
622 if (us_vhost_parse_socket_path(optarg) == -1) {
623 RTE_LOG(INFO, VHOST_CONFIG,
624 "Invalid argument for socket name (Max %d characters)\n",
626 us_vhost_usage(prgname);
633 /* Invalid option - print options. */
635 us_vhost_usage(prgname);
640 for (i = 0; i < RTE_MAX_ETHPORTS; i++) {
641 if (enabled_port_mask & (1 << i))
642 ports[num_ports++] = i;
645 if ((num_ports == 0) || (num_ports > MAX_SUP_PORTS)) {
646 RTE_LOG(INFO, VHOST_PORT, "Current enabled port number is %u,"
647 "but only %u port can be enabled\n",num_ports, MAX_SUP_PORTS);
655 * Update the global var NUM_PORTS and array PORTS according to system ports number
656 * and return valid ports number
658 static unsigned check_ports_num(unsigned nb_ports)
660 unsigned valid_num_ports = num_ports;
663 if (num_ports > nb_ports) {
664 RTE_LOG(INFO, VHOST_PORT, "\nSpecified port number(%u) exceeds total system port number(%u)\n",
665 num_ports, nb_ports);
666 num_ports = nb_ports;
669 for (portid = 0; portid < num_ports; portid ++) {
670 if (ports[portid] >= nb_ports) {
671 RTE_LOG(INFO, VHOST_PORT, "\nSpecified port ID(%u) exceeds max system port ID(%u)\n",
672 ports[portid], (nb_ports - 1));
673 ports[portid] = INVALID_PORT_ID;
677 return valid_num_ports;
680 static __rte_always_inline struct vhost_dev *
681 find_vhost_dev(struct ether_addr *mac)
683 struct vhost_dev *vdev;
685 TAILQ_FOREACH(vdev, &vhost_dev_list, global_vdev_entry) {
686 if (vdev->ready == DEVICE_RX &&
687 is_same_ether_addr(mac, &vdev->mac_address))
695 * This function learns the MAC address of the device and registers this along with a
696 * vlan tag to a VMDQ.
699 link_vmdq(struct vhost_dev *vdev, struct rte_mbuf *m)
701 struct ether_hdr *pkt_hdr;
704 /* Learn MAC address of guest device from packet */
705 pkt_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
707 if (find_vhost_dev(&pkt_hdr->s_addr)) {
708 RTE_LOG(ERR, VHOST_DATA,
709 "(%d) device is using a registered MAC!\n",
714 for (i = 0; i < ETHER_ADDR_LEN; i++)
715 vdev->mac_address.addr_bytes[i] = pkt_hdr->s_addr.addr_bytes[i];
717 /* vlan_tag currently uses the device_id. */
718 vdev->vlan_tag = vlan_tags[vdev->vid];
720 /* Print out VMDQ registration info. */
721 RTE_LOG(INFO, VHOST_DATA,
722 "(%d) mac %02x:%02x:%02x:%02x:%02x:%02x and vlan %d registered\n",
724 vdev->mac_address.addr_bytes[0], vdev->mac_address.addr_bytes[1],
725 vdev->mac_address.addr_bytes[2], vdev->mac_address.addr_bytes[3],
726 vdev->mac_address.addr_bytes[4], vdev->mac_address.addr_bytes[5],
729 /* Register the MAC address. */
730 ret = rte_eth_dev_mac_addr_add(ports[0], &vdev->mac_address,
731 (uint32_t)vdev->vid + vmdq_pool_base);
733 RTE_LOG(ERR, VHOST_DATA,
734 "(%d) failed to add device MAC address to VMDQ\n",
737 rte_eth_dev_set_vlan_strip_on_queue(ports[0], vdev->vmdq_rx_q, 1);
739 /* Set device as ready for RX. */
740 vdev->ready = DEVICE_RX;
746 * Removes MAC address and vlan tag from VMDQ. Ensures that nothing is adding buffers to the RX
747 * queue before disabling RX on the device.
750 unlink_vmdq(struct vhost_dev *vdev)
754 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
756 if (vdev->ready == DEVICE_RX) {
757 /*clear MAC and VLAN settings*/
758 rte_eth_dev_mac_addr_remove(ports[0], &vdev->mac_address);
759 for (i = 0; i < 6; i++)
760 vdev->mac_address.addr_bytes[i] = 0;
764 /*Clear out the receive buffers*/
765 rx_count = rte_eth_rx_burst(ports[0],
766 (uint16_t)vdev->vmdq_rx_q, pkts_burst, MAX_PKT_BURST);
769 for (i = 0; i < rx_count; i++)
770 rte_pktmbuf_free(pkts_burst[i]);
772 rx_count = rte_eth_rx_burst(ports[0],
773 (uint16_t)vdev->vmdq_rx_q, pkts_burst, MAX_PKT_BURST);
776 vdev->ready = DEVICE_MAC_LEARNING;
780 static __rte_always_inline void
781 virtio_xmit(struct vhost_dev *dst_vdev, struct vhost_dev *src_vdev,
786 if (builtin_net_driver) {
787 ret = vs_enqueue_pkts(dst_vdev, VIRTIO_RXQ, &m, 1);
789 ret = rte_vhost_enqueue_burst(dst_vdev->vid, VIRTIO_RXQ, &m, 1);
793 rte_atomic64_inc(&dst_vdev->stats.rx_total_atomic);
794 rte_atomic64_add(&dst_vdev->stats.rx_atomic, ret);
795 src_vdev->stats.tx_total++;
796 src_vdev->stats.tx += ret;
801 * Check if the packet destination MAC address is for a local device. If so then put
802 * the packet on that devices RX queue. If not then return.
804 static __rte_always_inline int
805 virtio_tx_local(struct vhost_dev *vdev, struct rte_mbuf *m)
807 struct ether_hdr *pkt_hdr;
808 struct vhost_dev *dst_vdev;
810 pkt_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
812 dst_vdev = find_vhost_dev(&pkt_hdr->d_addr);
816 if (vdev->vid == dst_vdev->vid) {
817 RTE_LOG_DP(DEBUG, VHOST_DATA,
818 "(%d) TX: src and dst MAC is same. Dropping packet.\n",
823 RTE_LOG_DP(DEBUG, VHOST_DATA,
824 "(%d) TX: MAC address is local\n", dst_vdev->vid);
826 if (unlikely(dst_vdev->remove)) {
827 RTE_LOG_DP(DEBUG, VHOST_DATA,
828 "(%d) device is marked for removal\n", dst_vdev->vid);
832 virtio_xmit(dst_vdev, vdev, m);
837 * Check if the destination MAC of a packet is one local VM,
838 * and get its vlan tag, and offset if it is.
840 static __rte_always_inline int
841 find_local_dest(struct vhost_dev *vdev, struct rte_mbuf *m,
842 uint32_t *offset, uint16_t *vlan_tag)
844 struct vhost_dev *dst_vdev;
845 struct ether_hdr *pkt_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
847 dst_vdev = find_vhost_dev(&pkt_hdr->d_addr);
851 if (vdev->vid == dst_vdev->vid) {
852 RTE_LOG_DP(DEBUG, VHOST_DATA,
853 "(%d) TX: src and dst MAC is same. Dropping packet.\n",
859 * HW vlan strip will reduce the packet length
860 * by minus length of vlan tag, so need restore
861 * the packet length by plus it.
864 *vlan_tag = vlan_tags[vdev->vid];
866 RTE_LOG_DP(DEBUG, VHOST_DATA,
867 "(%d) TX: pkt to local VM device id: (%d), vlan tag: %u.\n",
868 vdev->vid, dst_vdev->vid, *vlan_tag);
874 get_psd_sum(void *l3_hdr, uint64_t ol_flags)
876 if (ol_flags & PKT_TX_IPV4)
877 return rte_ipv4_phdr_cksum(l3_hdr, ol_flags);
878 else /* assume ethertype == ETHER_TYPE_IPv6 */
879 return rte_ipv6_phdr_cksum(l3_hdr, ol_flags);
882 static void virtio_tx_offload(struct rte_mbuf *m)
885 struct ipv4_hdr *ipv4_hdr = NULL;
886 struct tcp_hdr *tcp_hdr = NULL;
887 struct ether_hdr *eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
889 l3_hdr = (char *)eth_hdr + m->l2_len;
891 if (m->ol_flags & PKT_TX_IPV4) {
893 ipv4_hdr->hdr_checksum = 0;
894 m->ol_flags |= PKT_TX_IP_CKSUM;
897 tcp_hdr = (struct tcp_hdr *)((char *)l3_hdr + m->l3_len);
898 tcp_hdr->cksum = get_psd_sum(l3_hdr, m->ol_flags);
902 free_pkts(struct rte_mbuf **pkts, uint16_t n)
905 rte_pktmbuf_free(pkts[n]);
908 static __rte_always_inline void
909 do_drain_mbuf_table(struct mbuf_table *tx_q)
913 count = rte_eth_tx_burst(ports[0], tx_q->txq_id,
914 tx_q->m_table, tx_q->len);
915 if (unlikely(count < tx_q->len))
916 free_pkts(&tx_q->m_table[count], tx_q->len - count);
922 * This function routes the TX packet to the correct interface. This
923 * may be a local device or the physical port.
925 static __rte_always_inline void
926 virtio_tx_route(struct vhost_dev *vdev, struct rte_mbuf *m, uint16_t vlan_tag)
928 struct mbuf_table *tx_q;
930 const uint16_t lcore_id = rte_lcore_id();
931 struct ether_hdr *nh;
934 nh = rte_pktmbuf_mtod(m, struct ether_hdr *);
935 if (unlikely(is_broadcast_ether_addr(&nh->d_addr))) {
936 struct vhost_dev *vdev2;
938 TAILQ_FOREACH(vdev2, &vhost_dev_list, global_vdev_entry) {
940 virtio_xmit(vdev2, vdev, m);
945 /*check if destination is local VM*/
946 if ((vm2vm_mode == VM2VM_SOFTWARE) && (virtio_tx_local(vdev, m) == 0)) {
951 if (unlikely(vm2vm_mode == VM2VM_HARDWARE)) {
952 if (unlikely(find_local_dest(vdev, m, &offset,
959 RTE_LOG_DP(DEBUG, VHOST_DATA,
960 "(%d) TX: MAC address is external\n", vdev->vid);
964 /*Add packet to the port tx queue*/
965 tx_q = &lcore_tx_queue[lcore_id];
967 nh = rte_pktmbuf_mtod(m, struct ether_hdr *);
968 if (unlikely(nh->ether_type == rte_cpu_to_be_16(ETHER_TYPE_VLAN))) {
969 /* Guest has inserted the vlan tag. */
970 struct vlan_hdr *vh = (struct vlan_hdr *) (nh + 1);
971 uint16_t vlan_tag_be = rte_cpu_to_be_16(vlan_tag);
972 if ((vm2vm_mode == VM2VM_HARDWARE) &&
973 (vh->vlan_tci != vlan_tag_be))
974 vh->vlan_tci = vlan_tag_be;
976 m->ol_flags |= PKT_TX_VLAN_PKT;
979 * Find the right seg to adjust the data len when offset is
980 * bigger than tail room size.
982 if (unlikely(vm2vm_mode == VM2VM_HARDWARE)) {
983 if (likely(offset <= rte_pktmbuf_tailroom(m)))
984 m->data_len += offset;
986 struct rte_mbuf *seg = m;
988 while ((seg->next != NULL) &&
989 (offset > rte_pktmbuf_tailroom(seg)))
992 seg->data_len += offset;
994 m->pkt_len += offset;
997 m->vlan_tci = vlan_tag;
1000 if (m->ol_flags & PKT_TX_TCP_SEG)
1001 virtio_tx_offload(m);
1003 tx_q->m_table[tx_q->len++] = m;
1005 vdev->stats.tx_total++;
1009 if (unlikely(tx_q->len == MAX_PKT_BURST))
1010 do_drain_mbuf_table(tx_q);
1014 static __rte_always_inline void
1015 drain_mbuf_table(struct mbuf_table *tx_q)
1017 static uint64_t prev_tsc;
1023 cur_tsc = rte_rdtsc();
1024 if (unlikely(cur_tsc - prev_tsc > MBUF_TABLE_DRAIN_TSC)) {
1027 RTE_LOG_DP(DEBUG, VHOST_DATA,
1028 "TX queue drained after timeout with burst size %u\n",
1030 do_drain_mbuf_table(tx_q);
1034 static __rte_always_inline void
1035 drain_eth_rx(struct vhost_dev *vdev)
1037 uint16_t rx_count, enqueue_count;
1038 struct rte_mbuf *pkts[MAX_PKT_BURST];
1040 rx_count = rte_eth_rx_burst(ports[0], vdev->vmdq_rx_q,
1041 pkts, MAX_PKT_BURST);
1046 * When "enable_retry" is set, here we wait and retry when there
1047 * is no enough free slots in the queue to hold @rx_count packets,
1048 * to diminish packet loss.
1051 unlikely(rx_count > rte_vhost_avail_entries(vdev->vid,
1055 for (retry = 0; retry < burst_rx_retry_num; retry++) {
1056 rte_delay_us(burst_rx_delay_time);
1057 if (rx_count <= rte_vhost_avail_entries(vdev->vid,
1063 if (builtin_net_driver) {
1064 enqueue_count = vs_enqueue_pkts(vdev, VIRTIO_RXQ,
1067 enqueue_count = rte_vhost_enqueue_burst(vdev->vid, VIRTIO_RXQ,
1071 rte_atomic64_add(&vdev->stats.rx_total_atomic, rx_count);
1072 rte_atomic64_add(&vdev->stats.rx_atomic, enqueue_count);
1075 free_pkts(pkts, rx_count);
1078 static __rte_always_inline void
1079 drain_virtio_tx(struct vhost_dev *vdev)
1081 struct rte_mbuf *pkts[MAX_PKT_BURST];
1085 if (builtin_net_driver) {
1086 count = vs_dequeue_pkts(vdev, VIRTIO_TXQ, mbuf_pool,
1087 pkts, MAX_PKT_BURST);
1089 count = rte_vhost_dequeue_burst(vdev->vid, VIRTIO_TXQ,
1090 mbuf_pool, pkts, MAX_PKT_BURST);
1093 /* setup VMDq for the first packet */
1094 if (unlikely(vdev->ready == DEVICE_MAC_LEARNING) && count) {
1095 if (vdev->remove || link_vmdq(vdev, pkts[0]) == -1)
1096 free_pkts(pkts, count);
1099 for (i = 0; i < count; ++i)
1100 virtio_tx_route(vdev, pkts[i], vlan_tags[vdev->vid]);
1104 * Main function of vhost-switch. It basically does:
1106 * for each vhost device {
1109 * Which drains the host eth Rx queue linked to the vhost device,
1110 * and deliver all of them to guest virito Rx ring associated with
1111 * this vhost device.
1113 * - drain_virtio_tx()
1115 * Which drains the guest virtio Tx queue and deliver all of them
1116 * to the target, which could be another vhost device, or the
1117 * physical eth dev. The route is done in function "virtio_tx_route".
1121 switch_worker(void *arg __rte_unused)
1124 unsigned lcore_id = rte_lcore_id();
1125 struct vhost_dev *vdev;
1126 struct mbuf_table *tx_q;
1128 RTE_LOG(INFO, VHOST_DATA, "Procesing on Core %u started\n", lcore_id);
1130 tx_q = &lcore_tx_queue[lcore_id];
1131 for (i = 0; i < rte_lcore_count(); i++) {
1132 if (lcore_ids[i] == lcore_id) {
1139 drain_mbuf_table(tx_q);
1142 * Inform the configuration core that we have exited the
1143 * linked list and that no devices are in use if requested.
1145 if (lcore_info[lcore_id].dev_removal_flag == REQUEST_DEV_REMOVAL)
1146 lcore_info[lcore_id].dev_removal_flag = ACK_DEV_REMOVAL;
1149 * Process vhost devices
1151 TAILQ_FOREACH(vdev, &lcore_info[lcore_id].vdev_list,
1153 if (unlikely(vdev->remove)) {
1155 vdev->ready = DEVICE_SAFE_REMOVE;
1159 if (likely(vdev->ready == DEVICE_RX))
1162 if (likely(!vdev->remove))
1163 drain_virtio_tx(vdev);
1171 * Remove a device from the specific data core linked list and from the
1172 * main linked list. Synchonization occurs through the use of the
1173 * lcore dev_removal_flag. Device is made volatile here to avoid re-ordering
1174 * of dev->remove=1 which can cause an infinite loop in the rte_pause loop.
1177 destroy_device(int vid)
1179 struct vhost_dev *vdev = NULL;
1182 TAILQ_FOREACH(vdev, &vhost_dev_list, global_vdev_entry) {
1183 if (vdev->vid == vid)
1188 /*set the remove flag. */
1190 while(vdev->ready != DEVICE_SAFE_REMOVE) {
1194 if (builtin_net_driver)
1195 vs_vhost_net_remove(vdev);
1197 TAILQ_REMOVE(&lcore_info[vdev->coreid].vdev_list, vdev,
1199 TAILQ_REMOVE(&vhost_dev_list, vdev, global_vdev_entry);
1202 /* Set the dev_removal_flag on each lcore. */
1203 RTE_LCORE_FOREACH_SLAVE(lcore)
1204 lcore_info[lcore].dev_removal_flag = REQUEST_DEV_REMOVAL;
1207 * Once each core has set the dev_removal_flag to ACK_DEV_REMOVAL
1208 * we can be sure that they can no longer access the device removed
1209 * from the linked lists and that the devices are no longer in use.
1211 RTE_LCORE_FOREACH_SLAVE(lcore) {
1212 while (lcore_info[lcore].dev_removal_flag != ACK_DEV_REMOVAL)
1216 lcore_info[vdev->coreid].device_num--;
1218 RTE_LOG(INFO, VHOST_DATA,
1219 "(%d) device has been removed from data core\n",
1226 * A new device is added to a data core. First the device is added to the main linked list
1227 * and the allocated to a specific data core.
1232 int lcore, core_add = 0;
1233 uint32_t device_num_min = num_devices;
1234 struct vhost_dev *vdev;
1236 vdev = rte_zmalloc("vhost device", sizeof(*vdev), RTE_CACHE_LINE_SIZE);
1238 RTE_LOG(INFO, VHOST_DATA,
1239 "(%d) couldn't allocate memory for vhost dev\n",
1245 if (builtin_net_driver)
1246 vs_vhost_net_setup(vdev);
1248 TAILQ_INSERT_TAIL(&vhost_dev_list, vdev, global_vdev_entry);
1249 vdev->vmdq_rx_q = vid * queues_per_pool + vmdq_queue_base;
1251 /*reset ready flag*/
1252 vdev->ready = DEVICE_MAC_LEARNING;
1255 /* Find a suitable lcore to add the device. */
1256 RTE_LCORE_FOREACH_SLAVE(lcore) {
1257 if (lcore_info[lcore].device_num < device_num_min) {
1258 device_num_min = lcore_info[lcore].device_num;
1262 vdev->coreid = core_add;
1264 TAILQ_INSERT_TAIL(&lcore_info[vdev->coreid].vdev_list, vdev,
1266 lcore_info[vdev->coreid].device_num++;
1268 /* Disable notifications. */
1269 rte_vhost_enable_guest_notification(vid, VIRTIO_RXQ, 0);
1270 rte_vhost_enable_guest_notification(vid, VIRTIO_TXQ, 0);
1272 RTE_LOG(INFO, VHOST_DATA,
1273 "(%d) device has been added to data core %d\n",
1280 * These callback allow devices to be added to the data core when configuration
1281 * has been fully complete.
1283 static const struct vhost_device_ops virtio_net_device_ops =
1285 .new_device = new_device,
1286 .destroy_device = destroy_device,
1290 * This is a thread will wake up after a period to print stats if the user has
1296 struct vhost_dev *vdev;
1297 uint64_t tx_dropped, rx_dropped;
1298 uint64_t tx, tx_total, rx, rx_total;
1299 const char clr[] = { 27, '[', '2', 'J', '\0' };
1300 const char top_left[] = { 27, '[', '1', ';', '1', 'H','\0' };
1303 sleep(enable_stats);
1305 /* Clear screen and move to top left */
1306 printf("%s%s\n", clr, top_left);
1307 printf("Device statistics =================================\n");
1309 TAILQ_FOREACH(vdev, &vhost_dev_list, global_vdev_entry) {
1310 tx_total = vdev->stats.tx_total;
1311 tx = vdev->stats.tx;
1312 tx_dropped = tx_total - tx;
1314 rx_total = rte_atomic64_read(&vdev->stats.rx_total_atomic);
1315 rx = rte_atomic64_read(&vdev->stats.rx_atomic);
1316 rx_dropped = rx_total - rx;
1318 printf("Statistics for device %d\n"
1319 "-----------------------\n"
1320 "TX total: %" PRIu64 "\n"
1321 "TX dropped: %" PRIu64 "\n"
1322 "TX successful: %" PRIu64 "\n"
1323 "RX total: %" PRIu64 "\n"
1324 "RX dropped: %" PRIu64 "\n"
1325 "RX successful: %" PRIu64 "\n",
1327 tx_total, tx_dropped, tx,
1328 rx_total, rx_dropped, rx);
1331 printf("===================================================\n");
1336 unregister_drivers(int socket_num)
1340 for (i = 0; i < socket_num; i++) {
1341 ret = rte_vhost_driver_unregister(socket_files + i * PATH_MAX);
1343 RTE_LOG(ERR, VHOST_CONFIG,
1344 "Fail to unregister vhost driver for %s.\n",
1345 socket_files + i * PATH_MAX);
1349 /* When we receive a INT signal, unregister vhost driver */
1351 sigint_handler(__rte_unused int signum)
1353 /* Unregister vhost driver. */
1354 unregister_drivers(nb_sockets);
1360 * While creating an mbuf pool, one key thing is to figure out how
1361 * many mbuf entries is enough for our use. FYI, here are some
1364 * - Each rx queue would reserve @nr_rx_desc mbufs at queue setup stage
1366 * - For each switch core (A CPU core does the packet switch), we need
1367 * also make some reservation for receiving the packets from virtio
1368 * Tx queue. How many is enough depends on the usage. It's normally
1369 * a simple calculation like following:
1371 * MAX_PKT_BURST * max packet size / mbuf size
1373 * So, we definitely need allocate more mbufs when TSO is enabled.
1375 * - Similarly, for each switching core, we should serve @nr_rx_desc
1376 * mbufs for receiving the packets from physical NIC device.
1378 * - We also need make sure, for each switch core, we have allocated
1379 * enough mbufs to fill up the mbuf cache.
1382 create_mbuf_pool(uint16_t nr_port, uint32_t nr_switch_core, uint32_t mbuf_size,
1383 uint32_t nr_queues, uint32_t nr_rx_desc, uint32_t nr_mbuf_cache)
1386 uint32_t nr_mbufs_per_core;
1387 uint32_t mtu = 1500;
1394 nr_mbufs_per_core = (mtu + mbuf_size) * MAX_PKT_BURST /
1395 (mbuf_size - RTE_PKTMBUF_HEADROOM);
1396 nr_mbufs_per_core += nr_rx_desc;
1397 nr_mbufs_per_core = RTE_MAX(nr_mbufs_per_core, nr_mbuf_cache);
1399 nr_mbufs = nr_queues * nr_rx_desc;
1400 nr_mbufs += nr_mbufs_per_core * nr_switch_core;
1401 nr_mbufs *= nr_port;
1403 mbuf_pool = rte_pktmbuf_pool_create("MBUF_POOL", nr_mbufs,
1404 nr_mbuf_cache, 0, mbuf_size,
1406 if (mbuf_pool == NULL)
1407 rte_exit(EXIT_FAILURE, "Cannot create mbuf pool\n");
1411 * Main function, does initialisation and calls the per-lcore functions.
1414 main(int argc, char *argv[])
1416 unsigned lcore_id, core_id = 0;
1417 unsigned nb_ports, valid_num_ports;
1420 static pthread_t tid;
1421 char thread_name[RTE_MAX_THREAD_NAME_LEN];
1424 signal(SIGINT, sigint_handler);
1427 ret = rte_eal_init(argc, argv);
1429 rte_exit(EXIT_FAILURE, "Error with EAL initialization\n");
1433 /* parse app arguments */
1434 ret = us_vhost_parse_args(argc, argv);
1436 rte_exit(EXIT_FAILURE, "Invalid argument\n");
1438 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1439 TAILQ_INIT(&lcore_info[lcore_id].vdev_list);
1441 if (rte_lcore_is_enabled(lcore_id))
1442 lcore_ids[core_id++] = lcore_id;
1445 if (rte_lcore_count() > RTE_MAX_LCORE)
1446 rte_exit(EXIT_FAILURE,"Not enough cores\n");
1448 /* Get the number of physical ports. */
1449 nb_ports = rte_eth_dev_count();
1452 * Update the global var NUM_PORTS and global array PORTS
1453 * and get value of var VALID_NUM_PORTS according to system ports number
1455 valid_num_ports = check_ports_num(nb_ports);
1457 if ((valid_num_ports == 0) || (valid_num_ports > MAX_SUP_PORTS)) {
1458 RTE_LOG(INFO, VHOST_PORT, "Current enabled port number is %u,"
1459 "but only %u port can be enabled\n",num_ports, MAX_SUP_PORTS);
1464 * FIXME: here we are trying to allocate mbufs big enough for
1465 * @MAX_QUEUES, but the truth is we're never going to use that
1466 * many queues here. We probably should only do allocation for
1467 * those queues we are going to use.
1469 create_mbuf_pool(valid_num_ports, rte_lcore_count() - 1, MBUF_DATA_SIZE,
1470 MAX_QUEUES, RTE_TEST_RX_DESC_DEFAULT, MBUF_CACHE_SIZE);
1472 if (vm2vm_mode == VM2VM_HARDWARE) {
1473 /* Enable VT loop back to let L2 switch to do it. */
1474 vmdq_conf_default.rx_adv_conf.vmdq_rx_conf.enable_loop_back = 1;
1475 RTE_LOG(DEBUG, VHOST_CONFIG,
1476 "Enable loop back for L2 switch in vmdq.\n");
1479 /* initialize all ports */
1480 for (portid = 0; portid < nb_ports; portid++) {
1481 /* skip ports that are not enabled */
1482 if ((enabled_port_mask & (1 << portid)) == 0) {
1483 RTE_LOG(INFO, VHOST_PORT,
1484 "Skipping disabled port %d\n", portid);
1487 if (port_init(portid) != 0)
1488 rte_exit(EXIT_FAILURE,
1489 "Cannot initialize network ports\n");
1492 /* Enable stats if the user option is set. */
1494 ret = pthread_create(&tid, NULL, (void *)print_stats, NULL);
1496 rte_exit(EXIT_FAILURE,
1497 "Cannot create print-stats thread\n");
1499 /* Set thread_name for aid in debugging. */
1500 snprintf(thread_name, RTE_MAX_THREAD_NAME_LEN, "print-stats");
1501 ret = rte_thread_setname(tid, thread_name);
1503 RTE_LOG(DEBUG, VHOST_CONFIG,
1504 "Cannot set print-stats name\n");
1507 /* Launch all data cores. */
1508 RTE_LCORE_FOREACH_SLAVE(lcore_id)
1509 rte_eal_remote_launch(switch_worker, NULL, lcore_id);
1512 flags |= RTE_VHOST_USER_CLIENT;
1514 if (dequeue_zero_copy)
1515 flags |= RTE_VHOST_USER_DEQUEUE_ZERO_COPY;
1517 /* Register vhost user driver to handle vhost messages. */
1518 for (i = 0; i < nb_sockets; i++) {
1519 char *file = socket_files + i * PATH_MAX;
1520 ret = rte_vhost_driver_register(file, flags);
1522 unregister_drivers(i);
1523 rte_exit(EXIT_FAILURE,
1524 "vhost driver register failure.\n");
1527 if (builtin_net_driver)
1528 rte_vhost_driver_set_features(file, VIRTIO_NET_FEATURES);
1530 if (mergeable == 0) {
1531 rte_vhost_driver_disable_features(file,
1532 1ULL << VIRTIO_NET_F_MRG_RXBUF);
1535 if (enable_tx_csum == 0) {
1536 rte_vhost_driver_disable_features(file,
1537 1ULL << VIRTIO_NET_F_CSUM);
1540 if (enable_tso == 0) {
1541 rte_vhost_driver_disable_features(file,
1542 1ULL << VIRTIO_NET_F_HOST_TSO4);
1543 rte_vhost_driver_disable_features(file,
1544 1ULL << VIRTIO_NET_F_HOST_TSO6);
1545 rte_vhost_driver_disable_features(file,
1546 1ULL << VIRTIO_NET_F_GUEST_TSO4);
1547 rte_vhost_driver_disable_features(file,
1548 1ULL << VIRTIO_NET_F_GUEST_TSO6);
1552 rte_vhost_driver_enable_features(file,
1553 1ULL << VIRTIO_NET_F_CTRL_RX);
1556 ret = rte_vhost_driver_callback_register(file,
1557 &virtio_net_device_ops);
1559 rte_exit(EXIT_FAILURE,
1560 "failed to register vhost driver callbacks.\n");
1563 if (rte_vhost_driver_start(file) < 0) {
1564 rte_exit(EXIT_FAILURE,
1565 "failed to start vhost driver.\n");
1569 RTE_LCORE_FOREACH_SLAVE(lcore_id)
1570 rte_eal_wait_lcore(lcore_id);