4 * Copyright(c) 2010-2016 Intel Corporation. All rights reserved.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
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
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.
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.
34 #include <arpa/inet.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>
42 #include <sys/eventfd.h>
43 #include <sys/param.h>
46 #include <rte_atomic.h>
47 #include <rte_cycles.h>
48 #include <rte_ethdev.h>
50 #include <rte_string_fns.h>
51 #include <rte_malloc.h>
52 #include <rte_virtio_net.h>
59 #define MAX_QUEUES 128
62 /* the maximum number of external ports supported */
63 #define MAX_SUP_PORTS 1
65 #define MBUF_CACHE_SIZE 128
66 #define MBUF_DATA_SIZE RTE_MBUF_DEFAULT_BUF_SIZE
68 #define MAX_PKT_BURST 32 /* Max burst size for RX/TX */
69 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
71 #define BURST_RX_WAIT_US 15 /* Defines how long we wait between retries on RX */
72 #define BURST_RX_RETRIES 4 /* Number of retries on RX. */
74 #define JUMBO_FRAME_MAX_SIZE 0x2600
76 /* State of virtio device. */
77 #define DEVICE_MAC_LEARNING 0
79 #define DEVICE_SAFE_REMOVE 2
81 /* Configurable number of RX/TX ring descriptors */
82 #define RTE_TEST_RX_DESC_DEFAULT 1024
83 #define RTE_TEST_TX_DESC_DEFAULT 512
85 #define INVALID_PORT_ID 0xFF
87 /* Max number of devices. Limited by vmdq. */
88 #define MAX_DEVICES 64
90 /* Size of buffers used for snprintfs. */
91 #define MAX_PRINT_BUFF 6072
93 /* Maximum long option length for option parsing. */
94 #define MAX_LONG_OPT_SZ 64
96 /* mask of enabled ports */
97 static uint32_t enabled_port_mask = 0;
99 /* Promiscuous mode */
100 static uint32_t promiscuous;
102 /* number of devices/queues to support*/
103 static uint32_t num_queues = 0;
104 static uint32_t num_devices;
106 static struct rte_mempool *mbuf_pool;
107 static int mergeable;
109 /* Enable VM2VM communications. If this is disabled then the MAC address compare is skipped. */
116 static vm2vm_type vm2vm_mode = VM2VM_SOFTWARE;
119 static uint32_t enable_stats = 0;
120 /* Enable retries on RX. */
121 static uint32_t enable_retry = 1;
123 /* Disable TX checksum offload */
124 static uint32_t enable_tx_csum;
126 /* Disable TSO offload */
127 static uint32_t enable_tso;
129 static int client_mode;
130 static int dequeue_zero_copy;
132 /* Specify timeout (in useconds) between retries on RX. */
133 static uint32_t burst_rx_delay_time = BURST_RX_WAIT_US;
134 /* Specify the number of retries on RX. */
135 static uint32_t burst_rx_retry_num = BURST_RX_RETRIES;
137 /* Socket file paths. Can be set by user */
138 static char *socket_files;
139 static int nb_sockets;
141 /* empty vmdq configuration structure. Filled in programatically */
142 static struct rte_eth_conf vmdq_conf_default = {
144 .mq_mode = ETH_MQ_RX_VMDQ_ONLY,
146 .header_split = 0, /**< Header Split disabled */
147 .hw_ip_checksum = 0, /**< IP checksum offload disabled */
148 .hw_vlan_filter = 0, /**< VLAN filtering disabled */
150 * It is necessary for 1G NIC such as I350,
151 * this fixes bug of ipv4 forwarding in guest can't
152 * forward pakets from one virtio dev to another virtio dev.
154 .hw_vlan_strip = 1, /**< VLAN strip enabled. */
155 .jumbo_frame = 0, /**< Jumbo Frame Support disabled */
156 .hw_strip_crc = 0, /**< CRC stripped by hardware */
160 .mq_mode = ETH_MQ_TX_NONE,
164 * should be overridden separately in code with
168 .nb_queue_pools = ETH_8_POOLS,
169 .enable_default_pool = 0,
172 .pool_map = {{0, 0},},
177 static unsigned lcore_ids[RTE_MAX_LCORE];
178 static uint8_t ports[RTE_MAX_ETHPORTS];
179 static unsigned num_ports = 0; /**< The number of ports specified in command line */
180 static uint16_t num_pf_queues, num_vmdq_queues;
181 static uint16_t vmdq_pool_base, vmdq_queue_base;
182 static uint16_t queues_per_pool;
184 const uint16_t vlan_tags[] = {
185 1000, 1001, 1002, 1003, 1004, 1005, 1006, 1007,
186 1008, 1009, 1010, 1011, 1012, 1013, 1014, 1015,
187 1016, 1017, 1018, 1019, 1020, 1021, 1022, 1023,
188 1024, 1025, 1026, 1027, 1028, 1029, 1030, 1031,
189 1032, 1033, 1034, 1035, 1036, 1037, 1038, 1039,
190 1040, 1041, 1042, 1043, 1044, 1045, 1046, 1047,
191 1048, 1049, 1050, 1051, 1052, 1053, 1054, 1055,
192 1056, 1057, 1058, 1059, 1060, 1061, 1062, 1063,
195 /* ethernet addresses of ports */
196 static struct ether_addr vmdq_ports_eth_addr[RTE_MAX_ETHPORTS];
198 static struct vhost_dev_tailq_list vhost_dev_list =
199 TAILQ_HEAD_INITIALIZER(vhost_dev_list);
201 static struct lcore_info lcore_info[RTE_MAX_LCORE];
203 /* Used for queueing bursts of TX packets. */
207 struct rte_mbuf *m_table[MAX_PKT_BURST];
210 /* TX queue for each data core. */
211 struct mbuf_table lcore_tx_queue[RTE_MAX_LCORE];
213 #define MBUF_TABLE_DRAIN_TSC ((rte_get_tsc_hz() + US_PER_S - 1) \
214 / US_PER_S * BURST_TX_DRAIN_US)
218 * Builds up the correct configuration for VMDQ VLAN pool map
219 * according to the pool & queue limits.
222 get_eth_conf(struct rte_eth_conf *eth_conf, uint32_t num_devices)
224 struct rte_eth_vmdq_rx_conf conf;
225 struct rte_eth_vmdq_rx_conf *def_conf =
226 &vmdq_conf_default.rx_adv_conf.vmdq_rx_conf;
229 memset(&conf, 0, sizeof(conf));
230 conf.nb_queue_pools = (enum rte_eth_nb_pools)num_devices;
231 conf.nb_pool_maps = num_devices;
232 conf.enable_loop_back = def_conf->enable_loop_back;
233 conf.rx_mode = def_conf->rx_mode;
235 for (i = 0; i < conf.nb_pool_maps; i++) {
236 conf.pool_map[i].vlan_id = vlan_tags[ i ];
237 conf.pool_map[i].pools = (1UL << i);
240 (void)(rte_memcpy(eth_conf, &vmdq_conf_default, sizeof(*eth_conf)));
241 (void)(rte_memcpy(ð_conf->rx_adv_conf.vmdq_rx_conf, &conf,
242 sizeof(eth_conf->rx_adv_conf.vmdq_rx_conf)));
247 * Validate the device number according to the max pool number gotten form
248 * dev_info. If the device number is invalid, give the error message and
249 * return -1. Each device must have its own pool.
252 validate_num_devices(uint32_t max_nb_devices)
254 if (num_devices > max_nb_devices) {
255 RTE_LOG(ERR, VHOST_PORT, "invalid number of devices\n");
262 * Initialises a given port using global settings and with the rx buffers
263 * coming from the mbuf_pool passed as parameter
266 port_init(uint8_t port)
268 struct rte_eth_dev_info dev_info;
269 struct rte_eth_conf port_conf;
270 struct rte_eth_rxconf *rxconf;
271 struct rte_eth_txconf *txconf;
272 int16_t rx_rings, tx_rings;
273 uint16_t rx_ring_size, tx_ring_size;
277 /* The max pool number from dev_info will be used to validate the pool number specified in cmd line */
278 rte_eth_dev_info_get (port, &dev_info);
280 if (dev_info.max_rx_queues > MAX_QUEUES) {
281 rte_exit(EXIT_FAILURE,
282 "please define MAX_QUEUES no less than %u in %s\n",
283 dev_info.max_rx_queues, __FILE__);
286 rxconf = &dev_info.default_rxconf;
287 txconf = &dev_info.default_txconf;
288 rxconf->rx_drop_en = 1;
290 /* Enable vlan offload */
291 txconf->txq_flags &= ~ETH_TXQ_FLAGS_NOVLANOFFL;
293 /*configure the number of supported virtio devices based on VMDQ limits */
294 num_devices = dev_info.max_vmdq_pools;
296 rx_ring_size = RTE_TEST_RX_DESC_DEFAULT;
297 tx_ring_size = RTE_TEST_TX_DESC_DEFAULT;
300 * When dequeue zero copy is enabled, guest Tx used vring will be
301 * updated only when corresponding mbuf is freed. Thus, the nb_tx_desc
302 * (tx_ring_size here) must be small enough so that the driver will
303 * hit the free threshold easily and free mbufs timely. Otherwise,
304 * guest Tx vring would be starved.
306 if (dequeue_zero_copy)
309 tx_rings = (uint16_t)rte_lcore_count();
311 retval = validate_num_devices(MAX_DEVICES);
315 /* Get port configuration. */
316 retval = get_eth_conf(&port_conf, num_devices);
319 /* NIC queues are divided into pf queues and vmdq queues. */
320 num_pf_queues = dev_info.max_rx_queues - dev_info.vmdq_queue_num;
321 queues_per_pool = dev_info.vmdq_queue_num / dev_info.max_vmdq_pools;
322 num_vmdq_queues = num_devices * queues_per_pool;
323 num_queues = num_pf_queues + num_vmdq_queues;
324 vmdq_queue_base = dev_info.vmdq_queue_base;
325 vmdq_pool_base = dev_info.vmdq_pool_base;
326 printf("pf queue num: %u, configured vmdq pool num: %u, each vmdq pool has %u queues\n",
327 num_pf_queues, num_devices, queues_per_pool);
329 if (port >= rte_eth_dev_count()) return -1;
331 if (enable_tx_csum == 0)
332 rte_vhost_feature_disable(1ULL << VIRTIO_NET_F_CSUM);
334 if (enable_tso == 0) {
335 rte_vhost_feature_disable(1ULL << VIRTIO_NET_F_HOST_TSO4);
336 rte_vhost_feature_disable(1ULL << VIRTIO_NET_F_HOST_TSO6);
337 rte_vhost_feature_disable(1ULL << VIRTIO_NET_F_GUEST_TSO4);
338 rte_vhost_feature_disable(1ULL << VIRTIO_NET_F_GUEST_TSO6);
341 rx_rings = (uint16_t)dev_info.max_rx_queues;
342 /* Configure ethernet device. */
343 retval = rte_eth_dev_configure(port, rx_rings, tx_rings, &port_conf);
345 RTE_LOG(ERR, VHOST_PORT, "Failed to configure port %u: %s.\n",
346 port, strerror(-retval));
350 /* Setup the queues. */
351 for (q = 0; q < rx_rings; q ++) {
352 retval = rte_eth_rx_queue_setup(port, q, rx_ring_size,
353 rte_eth_dev_socket_id(port),
357 RTE_LOG(ERR, VHOST_PORT,
358 "Failed to setup rx queue %u of port %u: %s.\n",
359 q, port, strerror(-retval));
363 for (q = 0; q < tx_rings; q ++) {
364 retval = rte_eth_tx_queue_setup(port, q, tx_ring_size,
365 rte_eth_dev_socket_id(port),
368 RTE_LOG(ERR, VHOST_PORT,
369 "Failed to setup tx queue %u of port %u: %s.\n",
370 q, port, strerror(-retval));
375 /* Start the device. */
376 retval = rte_eth_dev_start(port);
378 RTE_LOG(ERR, VHOST_PORT, "Failed to start port %u: %s\n",
379 port, strerror(-retval));
384 rte_eth_promiscuous_enable(port);
386 rte_eth_macaddr_get(port, &vmdq_ports_eth_addr[port]);
387 RTE_LOG(INFO, VHOST_PORT, "Max virtio devices supported: %u\n", num_devices);
388 RTE_LOG(INFO, VHOST_PORT, "Port %u MAC: %02"PRIx8" %02"PRIx8" %02"PRIx8
389 " %02"PRIx8" %02"PRIx8" %02"PRIx8"\n",
391 vmdq_ports_eth_addr[port].addr_bytes[0],
392 vmdq_ports_eth_addr[port].addr_bytes[1],
393 vmdq_ports_eth_addr[port].addr_bytes[2],
394 vmdq_ports_eth_addr[port].addr_bytes[3],
395 vmdq_ports_eth_addr[port].addr_bytes[4],
396 vmdq_ports_eth_addr[port].addr_bytes[5]);
402 * Set socket file path.
405 us_vhost_parse_socket_path(const char *q_arg)
407 /* parse number string */
408 if (strnlen(q_arg, PATH_MAX) > PATH_MAX)
411 socket_files = realloc(socket_files, PATH_MAX * (nb_sockets + 1));
412 snprintf(socket_files + nb_sockets * PATH_MAX, PATH_MAX, "%s", q_arg);
419 * Parse the portmask provided at run time.
422 parse_portmask(const char *portmask)
429 /* parse hexadecimal string */
430 pm = strtoul(portmask, &end, 16);
431 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0') || (errno != 0))
442 * Parse num options at run time.
445 parse_num_opt(const char *q_arg, uint32_t max_valid_value)
452 /* parse unsigned int string */
453 num = strtoul(q_arg, &end, 10);
454 if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0') || (errno != 0))
457 if (num > max_valid_value)
468 us_vhost_usage(const char *prgname)
470 RTE_LOG(INFO, VHOST_CONFIG, "%s [EAL options] -- -p PORTMASK\n"
472 " --rx_retry [0|1] --mergeable [0|1] --stats [0-N]\n"
473 " --socket-file <path>\n"
475 " -p PORTMASK: Set mask for ports to be used by application\n"
476 " --vm2vm [0|1|2]: disable/software(default)/hardware vm2vm comms\n"
477 " --rx-retry [0|1]: disable/enable(default) retries on rx. Enable retry if destintation queue is full\n"
478 " --rx-retry-delay [0-N]: timeout(in usecond) between retries on RX. This makes effect only if retries on rx enabled\n"
479 " --rx-retry-num [0-N]: the number of retries on rx. This makes effect only if retries on rx enabled\n"
480 " --mergeable [0|1]: disable(default)/enable RX mergeable buffers\n"
481 " --stats [0-N]: 0: Disable stats, N: Time in seconds to print stats\n"
482 " --socket-file: The path of the socket file.\n"
483 " --tx-csum [0|1] disable/enable TX checksum offload.\n"
484 " --tso [0|1] disable/enable TCP segment offload.\n"
485 " --client register a vhost-user socket as client mode.\n"
486 " --dequeue-zero-copy enables dequeue zero copy\n",
491 * Parse the arguments given in the command line of the application.
494 us_vhost_parse_args(int argc, char **argv)
499 const char *prgname = argv[0];
500 static struct option long_option[] = {
501 {"vm2vm", required_argument, NULL, 0},
502 {"rx-retry", required_argument, NULL, 0},
503 {"rx-retry-delay", required_argument, NULL, 0},
504 {"rx-retry-num", required_argument, NULL, 0},
505 {"mergeable", required_argument, NULL, 0},
506 {"stats", required_argument, NULL, 0},
507 {"socket-file", required_argument, NULL, 0},
508 {"tx-csum", required_argument, NULL, 0},
509 {"tso", required_argument, NULL, 0},
510 {"client", no_argument, &client_mode, 1},
511 {"dequeue-zero-copy", no_argument, &dequeue_zero_copy, 1},
515 /* Parse command line */
516 while ((opt = getopt_long(argc, argv, "p:P",
517 long_option, &option_index)) != EOF) {
521 enabled_port_mask = parse_portmask(optarg);
522 if (enabled_port_mask == 0) {
523 RTE_LOG(INFO, VHOST_CONFIG, "Invalid portmask\n");
524 us_vhost_usage(prgname);
531 vmdq_conf_default.rx_adv_conf.vmdq_rx_conf.rx_mode =
532 ETH_VMDQ_ACCEPT_BROADCAST |
533 ETH_VMDQ_ACCEPT_MULTICAST;
534 rte_vhost_feature_enable(1ULL << VIRTIO_NET_F_CTRL_RX);
539 /* Enable/disable vm2vm comms. */
540 if (!strncmp(long_option[option_index].name, "vm2vm",
542 ret = parse_num_opt(optarg, (VM2VM_LAST - 1));
544 RTE_LOG(INFO, VHOST_CONFIG,
545 "Invalid argument for "
547 us_vhost_usage(prgname);
550 vm2vm_mode = (vm2vm_type)ret;
554 /* Enable/disable retries on RX. */
555 if (!strncmp(long_option[option_index].name, "rx-retry", MAX_LONG_OPT_SZ)) {
556 ret = parse_num_opt(optarg, 1);
558 RTE_LOG(INFO, VHOST_CONFIG, "Invalid argument for rx-retry [0|1]\n");
559 us_vhost_usage(prgname);
566 /* Enable/disable TX checksum offload. */
567 if (!strncmp(long_option[option_index].name, "tx-csum", MAX_LONG_OPT_SZ)) {
568 ret = parse_num_opt(optarg, 1);
570 RTE_LOG(INFO, VHOST_CONFIG, "Invalid argument for tx-csum [0|1]\n");
571 us_vhost_usage(prgname);
574 enable_tx_csum = ret;
577 /* Enable/disable TSO offload. */
578 if (!strncmp(long_option[option_index].name, "tso", MAX_LONG_OPT_SZ)) {
579 ret = parse_num_opt(optarg, 1);
581 RTE_LOG(INFO, VHOST_CONFIG, "Invalid argument for tso [0|1]\n");
582 us_vhost_usage(prgname);
588 /* Specify the retries delay time (in useconds) on RX. */
589 if (!strncmp(long_option[option_index].name, "rx-retry-delay", MAX_LONG_OPT_SZ)) {
590 ret = parse_num_opt(optarg, INT32_MAX);
592 RTE_LOG(INFO, VHOST_CONFIG, "Invalid argument for rx-retry-delay [0-N]\n");
593 us_vhost_usage(prgname);
596 burst_rx_delay_time = ret;
600 /* Specify the retries number on RX. */
601 if (!strncmp(long_option[option_index].name, "rx-retry-num", MAX_LONG_OPT_SZ)) {
602 ret = parse_num_opt(optarg, INT32_MAX);
604 RTE_LOG(INFO, VHOST_CONFIG, "Invalid argument for rx-retry-num [0-N]\n");
605 us_vhost_usage(prgname);
608 burst_rx_retry_num = ret;
612 /* Enable/disable RX mergeable buffers. */
613 if (!strncmp(long_option[option_index].name, "mergeable", MAX_LONG_OPT_SZ)) {
614 ret = parse_num_opt(optarg, 1);
616 RTE_LOG(INFO, VHOST_CONFIG, "Invalid argument for mergeable [0|1]\n");
617 us_vhost_usage(prgname);
622 vmdq_conf_default.rxmode.jumbo_frame = 1;
623 vmdq_conf_default.rxmode.max_rx_pkt_len
624 = JUMBO_FRAME_MAX_SIZE;
629 /* Enable/disable stats. */
630 if (!strncmp(long_option[option_index].name, "stats", MAX_LONG_OPT_SZ)) {
631 ret = parse_num_opt(optarg, INT32_MAX);
633 RTE_LOG(INFO, VHOST_CONFIG,
634 "Invalid argument for stats [0..N]\n");
635 us_vhost_usage(prgname);
642 /* Set socket file path. */
643 if (!strncmp(long_option[option_index].name,
644 "socket-file", MAX_LONG_OPT_SZ)) {
645 if (us_vhost_parse_socket_path(optarg) == -1) {
646 RTE_LOG(INFO, VHOST_CONFIG,
647 "Invalid argument for socket name (Max %d characters)\n",
649 us_vhost_usage(prgname);
656 /* Invalid option - print options. */
658 us_vhost_usage(prgname);
663 for (i = 0; i < RTE_MAX_ETHPORTS; i++) {
664 if (enabled_port_mask & (1 << i))
665 ports[num_ports++] = (uint8_t)i;
668 if ((num_ports == 0) || (num_ports > MAX_SUP_PORTS)) {
669 RTE_LOG(INFO, VHOST_PORT, "Current enabled port number is %u,"
670 "but only %u port can be enabled\n",num_ports, MAX_SUP_PORTS);
678 * Update the global var NUM_PORTS and array PORTS according to system ports number
679 * and return valid ports number
681 static unsigned check_ports_num(unsigned nb_ports)
683 unsigned valid_num_ports = num_ports;
686 if (num_ports > nb_ports) {
687 RTE_LOG(INFO, VHOST_PORT, "\nSpecified port number(%u) exceeds total system port number(%u)\n",
688 num_ports, nb_ports);
689 num_ports = nb_ports;
692 for (portid = 0; portid < num_ports; portid ++) {
693 if (ports[portid] >= nb_ports) {
694 RTE_LOG(INFO, VHOST_PORT, "\nSpecified port ID(%u) exceeds max system port ID(%u)\n",
695 ports[portid], (nb_ports - 1));
696 ports[portid] = INVALID_PORT_ID;
700 return valid_num_ports;
703 static inline struct vhost_dev *__attribute__((always_inline))
704 find_vhost_dev(struct ether_addr *mac)
706 struct vhost_dev *vdev;
708 TAILQ_FOREACH(vdev, &vhost_dev_list, global_vdev_entry) {
709 if (vdev->ready == DEVICE_RX &&
710 is_same_ether_addr(mac, &vdev->mac_address))
718 * This function learns the MAC address of the device and registers this along with a
719 * vlan tag to a VMDQ.
722 link_vmdq(struct vhost_dev *vdev, struct rte_mbuf *m)
724 struct ether_hdr *pkt_hdr;
727 /* Learn MAC address of guest device from packet */
728 pkt_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
730 if (find_vhost_dev(&pkt_hdr->s_addr)) {
731 RTE_LOG(ERR, VHOST_DATA,
732 "(%d) device is using a registered MAC!\n",
737 for (i = 0; i < ETHER_ADDR_LEN; i++)
738 vdev->mac_address.addr_bytes[i] = pkt_hdr->s_addr.addr_bytes[i];
740 /* vlan_tag currently uses the device_id. */
741 vdev->vlan_tag = vlan_tags[vdev->vid];
743 /* Print out VMDQ registration info. */
744 RTE_LOG(INFO, VHOST_DATA,
745 "(%d) mac %02x:%02x:%02x:%02x:%02x:%02x and vlan %d registered\n",
747 vdev->mac_address.addr_bytes[0], vdev->mac_address.addr_bytes[1],
748 vdev->mac_address.addr_bytes[2], vdev->mac_address.addr_bytes[3],
749 vdev->mac_address.addr_bytes[4], vdev->mac_address.addr_bytes[5],
752 /* Register the MAC address. */
753 ret = rte_eth_dev_mac_addr_add(ports[0], &vdev->mac_address,
754 (uint32_t)vdev->vid + vmdq_pool_base);
756 RTE_LOG(ERR, VHOST_DATA,
757 "(%d) failed to add device MAC address to VMDQ\n",
760 rte_eth_dev_set_vlan_strip_on_queue(ports[0], vdev->vmdq_rx_q, 1);
762 /* Set device as ready for RX. */
763 vdev->ready = DEVICE_RX;
769 * Removes MAC address and vlan tag from VMDQ. Ensures that nothing is adding buffers to the RX
770 * queue before disabling RX on the device.
773 unlink_vmdq(struct vhost_dev *vdev)
777 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
779 if (vdev->ready == DEVICE_RX) {
780 /*clear MAC and VLAN settings*/
781 rte_eth_dev_mac_addr_remove(ports[0], &vdev->mac_address);
782 for (i = 0; i < 6; i++)
783 vdev->mac_address.addr_bytes[i] = 0;
787 /*Clear out the receive buffers*/
788 rx_count = rte_eth_rx_burst(ports[0],
789 (uint16_t)vdev->vmdq_rx_q, pkts_burst, MAX_PKT_BURST);
792 for (i = 0; i < rx_count; i++)
793 rte_pktmbuf_free(pkts_burst[i]);
795 rx_count = rte_eth_rx_burst(ports[0],
796 (uint16_t)vdev->vmdq_rx_q, pkts_burst, MAX_PKT_BURST);
799 vdev->ready = DEVICE_MAC_LEARNING;
803 static inline void __attribute__((always_inline))
804 virtio_xmit(struct vhost_dev *dst_vdev, struct vhost_dev *src_vdev,
809 ret = rte_vhost_enqueue_burst(dst_vdev->vid, VIRTIO_RXQ, &m, 1);
811 rte_atomic64_inc(&dst_vdev->stats.rx_total_atomic);
812 rte_atomic64_add(&dst_vdev->stats.rx_atomic, ret);
813 src_vdev->stats.tx_total++;
814 src_vdev->stats.tx += ret;
819 * Check if the packet destination MAC address is for a local device. If so then put
820 * the packet on that devices RX queue. If not then return.
822 static inline int __attribute__((always_inline))
823 virtio_tx_local(struct vhost_dev *vdev, struct rte_mbuf *m)
825 struct ether_hdr *pkt_hdr;
826 struct vhost_dev *dst_vdev;
828 pkt_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
830 dst_vdev = find_vhost_dev(&pkt_hdr->d_addr);
834 if (vdev->vid == dst_vdev->vid) {
835 RTE_LOG_DP(DEBUG, VHOST_DATA,
836 "(%d) TX: src and dst MAC is same. Dropping packet.\n",
841 RTE_LOG_DP(DEBUG, VHOST_DATA,
842 "(%d) TX: MAC address is local\n", dst_vdev->vid);
844 if (unlikely(dst_vdev->remove)) {
845 RTE_LOG_DP(DEBUG, VHOST_DATA,
846 "(%d) device is marked for removal\n", dst_vdev->vid);
850 virtio_xmit(dst_vdev, vdev, m);
855 * Check if the destination MAC of a packet is one local VM,
856 * and get its vlan tag, and offset if it is.
858 static inline int __attribute__((always_inline))
859 find_local_dest(struct vhost_dev *vdev, struct rte_mbuf *m,
860 uint32_t *offset, uint16_t *vlan_tag)
862 struct vhost_dev *dst_vdev;
863 struct ether_hdr *pkt_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
865 dst_vdev = find_vhost_dev(&pkt_hdr->d_addr);
869 if (vdev->vid == dst_vdev->vid) {
870 RTE_LOG_DP(DEBUG, VHOST_DATA,
871 "(%d) TX: src and dst MAC is same. Dropping packet.\n",
877 * HW vlan strip will reduce the packet length
878 * by minus length of vlan tag, so need restore
879 * the packet length by plus it.
882 *vlan_tag = vlan_tags[vdev->vid];
884 RTE_LOG_DP(DEBUG, VHOST_DATA,
885 "(%d) TX: pkt to local VM device id: (%d), vlan tag: %u.\n",
886 vdev->vid, dst_vdev->vid, *vlan_tag);
892 get_psd_sum(void *l3_hdr, uint64_t ol_flags)
894 if (ol_flags & PKT_TX_IPV4)
895 return rte_ipv4_phdr_cksum(l3_hdr, ol_flags);
896 else /* assume ethertype == ETHER_TYPE_IPv6 */
897 return rte_ipv6_phdr_cksum(l3_hdr, ol_flags);
900 static void virtio_tx_offload(struct rte_mbuf *m)
903 struct ipv4_hdr *ipv4_hdr = NULL;
904 struct tcp_hdr *tcp_hdr = NULL;
905 struct ether_hdr *eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
907 l3_hdr = (char *)eth_hdr + m->l2_len;
909 if (m->ol_flags & PKT_TX_IPV4) {
911 ipv4_hdr->hdr_checksum = 0;
912 m->ol_flags |= PKT_TX_IP_CKSUM;
915 tcp_hdr = (struct tcp_hdr *)((char *)l3_hdr + m->l3_len);
916 tcp_hdr->cksum = get_psd_sum(l3_hdr, m->ol_flags);
920 free_pkts(struct rte_mbuf **pkts, uint16_t n)
923 rte_pktmbuf_free(pkts[n]);
926 static inline void __attribute__((always_inline))
927 do_drain_mbuf_table(struct mbuf_table *tx_q)
931 count = rte_eth_tx_burst(ports[0], tx_q->txq_id,
932 tx_q->m_table, tx_q->len);
933 if (unlikely(count < tx_q->len))
934 free_pkts(&tx_q->m_table[count], tx_q->len - count);
940 * This function routes the TX packet to the correct interface. This
941 * may be a local device or the physical port.
943 static inline void __attribute__((always_inline))
944 virtio_tx_route(struct vhost_dev *vdev, struct rte_mbuf *m, uint16_t vlan_tag)
946 struct mbuf_table *tx_q;
948 const uint16_t lcore_id = rte_lcore_id();
949 struct ether_hdr *nh;
952 nh = rte_pktmbuf_mtod(m, struct ether_hdr *);
953 if (unlikely(is_broadcast_ether_addr(&nh->d_addr))) {
954 struct vhost_dev *vdev2;
956 TAILQ_FOREACH(vdev2, &vhost_dev_list, global_vdev_entry) {
957 virtio_xmit(vdev2, vdev, m);
962 /*check if destination is local VM*/
963 if ((vm2vm_mode == VM2VM_SOFTWARE) && (virtio_tx_local(vdev, m) == 0)) {
968 if (unlikely(vm2vm_mode == VM2VM_HARDWARE)) {
969 if (unlikely(find_local_dest(vdev, m, &offset,
976 RTE_LOG_DP(DEBUG, VHOST_DATA,
977 "(%d) TX: MAC address is external\n", vdev->vid);
981 /*Add packet to the port tx queue*/
982 tx_q = &lcore_tx_queue[lcore_id];
984 nh = rte_pktmbuf_mtod(m, struct ether_hdr *);
985 if (unlikely(nh->ether_type == rte_cpu_to_be_16(ETHER_TYPE_VLAN))) {
986 /* Guest has inserted the vlan tag. */
987 struct vlan_hdr *vh = (struct vlan_hdr *) (nh + 1);
988 uint16_t vlan_tag_be = rte_cpu_to_be_16(vlan_tag);
989 if ((vm2vm_mode == VM2VM_HARDWARE) &&
990 (vh->vlan_tci != vlan_tag_be))
991 vh->vlan_tci = vlan_tag_be;
993 m->ol_flags |= PKT_TX_VLAN_PKT;
996 * Find the right seg to adjust the data len when offset is
997 * bigger than tail room size.
999 if (unlikely(vm2vm_mode == VM2VM_HARDWARE)) {
1000 if (likely(offset <= rte_pktmbuf_tailroom(m)))
1001 m->data_len += offset;
1003 struct rte_mbuf *seg = m;
1005 while ((seg->next != NULL) &&
1006 (offset > rte_pktmbuf_tailroom(seg)))
1009 seg->data_len += offset;
1011 m->pkt_len += offset;
1014 m->vlan_tci = vlan_tag;
1017 if (m->ol_flags & PKT_TX_TCP_SEG)
1018 virtio_tx_offload(m);
1020 tx_q->m_table[tx_q->len++] = m;
1022 vdev->stats.tx_total++;
1026 if (unlikely(tx_q->len == MAX_PKT_BURST))
1027 do_drain_mbuf_table(tx_q);
1031 static inline void __attribute__((always_inline))
1032 drain_mbuf_table(struct mbuf_table *tx_q)
1034 static uint64_t prev_tsc;
1040 cur_tsc = rte_rdtsc();
1041 if (unlikely(cur_tsc - prev_tsc > MBUF_TABLE_DRAIN_TSC)) {
1044 RTE_LOG_DP(DEBUG, VHOST_DATA,
1045 "TX queue drained after timeout with burst size %u\n",
1047 do_drain_mbuf_table(tx_q);
1051 static inline void __attribute__((always_inline))
1052 drain_eth_rx(struct vhost_dev *vdev)
1054 uint16_t rx_count, enqueue_count;
1055 struct rte_mbuf *pkts[MAX_PKT_BURST];
1057 rx_count = rte_eth_rx_burst(ports[0], vdev->vmdq_rx_q,
1058 pkts, MAX_PKT_BURST);
1063 * When "enable_retry" is set, here we wait and retry when there
1064 * is no enough free slots in the queue to hold @rx_count packets,
1065 * to diminish packet loss.
1068 unlikely(rx_count > rte_vhost_avail_entries(vdev->vid,
1072 for (retry = 0; retry < burst_rx_retry_num; retry++) {
1073 rte_delay_us(burst_rx_delay_time);
1074 if (rx_count <= rte_vhost_avail_entries(vdev->vid,
1080 enqueue_count = rte_vhost_enqueue_burst(vdev->vid, VIRTIO_RXQ,
1083 rte_atomic64_add(&vdev->stats.rx_total_atomic, rx_count);
1084 rte_atomic64_add(&vdev->stats.rx_atomic, enqueue_count);
1087 free_pkts(pkts, rx_count);
1090 static inline void __attribute__((always_inline))
1091 drain_virtio_tx(struct vhost_dev *vdev)
1093 struct rte_mbuf *pkts[MAX_PKT_BURST];
1097 count = rte_vhost_dequeue_burst(vdev->vid, VIRTIO_TXQ, mbuf_pool,
1098 pkts, MAX_PKT_BURST);
1100 /* setup VMDq for the first packet */
1101 if (unlikely(vdev->ready == DEVICE_MAC_LEARNING) && count) {
1102 if (vdev->remove || link_vmdq(vdev, pkts[0]) == -1)
1103 free_pkts(pkts, count);
1106 for (i = 0; i < count; ++i)
1107 virtio_tx_route(vdev, pkts[i], vlan_tags[vdev->vid]);
1111 * Main function of vhost-switch. It basically does:
1113 * for each vhost device {
1116 * Which drains the host eth Rx queue linked to the vhost device,
1117 * and deliver all of them to guest virito Rx ring associated with
1118 * this vhost device.
1120 * - drain_virtio_tx()
1122 * Which drains the guest virtio Tx queue and deliver all of them
1123 * to the target, which could be another vhost device, or the
1124 * physical eth dev. The route is done in function "virtio_tx_route".
1128 switch_worker(void *arg __rte_unused)
1131 unsigned lcore_id = rte_lcore_id();
1132 struct vhost_dev *vdev;
1133 struct mbuf_table *tx_q;
1135 RTE_LOG(INFO, VHOST_DATA, "Procesing on Core %u started\n", lcore_id);
1137 tx_q = &lcore_tx_queue[lcore_id];
1138 for (i = 0; i < rte_lcore_count(); i++) {
1139 if (lcore_ids[i] == lcore_id) {
1146 drain_mbuf_table(tx_q);
1149 * Inform the configuration core that we have exited the
1150 * linked list and that no devices are in use if requested.
1152 if (lcore_info[lcore_id].dev_removal_flag == REQUEST_DEV_REMOVAL)
1153 lcore_info[lcore_id].dev_removal_flag = ACK_DEV_REMOVAL;
1156 * Process vhost devices
1158 TAILQ_FOREACH(vdev, &lcore_info[lcore_id].vdev_list,
1160 if (unlikely(vdev->remove)) {
1162 vdev->ready = DEVICE_SAFE_REMOVE;
1166 if (likely(vdev->ready == DEVICE_RX))
1169 if (likely(!vdev->remove))
1170 drain_virtio_tx(vdev);
1178 * Remove a device from the specific data core linked list and from the
1179 * main linked list. Synchonization occurs through the use of the
1180 * lcore dev_removal_flag. Device is made volatile here to avoid re-ordering
1181 * of dev->remove=1 which can cause an infinite loop in the rte_pause loop.
1184 destroy_device(int vid)
1186 struct vhost_dev *vdev = NULL;
1189 TAILQ_FOREACH(vdev, &vhost_dev_list, global_vdev_entry) {
1190 if (vdev->vid == vid)
1195 /*set the remove flag. */
1197 while(vdev->ready != DEVICE_SAFE_REMOVE) {
1201 TAILQ_REMOVE(&lcore_info[vdev->coreid].vdev_list, vdev,
1203 TAILQ_REMOVE(&vhost_dev_list, vdev, global_vdev_entry);
1206 /* Set the dev_removal_flag on each lcore. */
1207 RTE_LCORE_FOREACH_SLAVE(lcore)
1208 lcore_info[lcore].dev_removal_flag = REQUEST_DEV_REMOVAL;
1211 * Once each core has set the dev_removal_flag to ACK_DEV_REMOVAL
1212 * we can be sure that they can no longer access the device removed
1213 * from the linked lists and that the devices are no longer in use.
1215 RTE_LCORE_FOREACH_SLAVE(lcore) {
1216 while (lcore_info[lcore].dev_removal_flag != ACK_DEV_REMOVAL)
1220 lcore_info[vdev->coreid].device_num--;
1222 RTE_LOG(INFO, VHOST_DATA,
1223 "(%d) device has been removed from data core\n",
1230 * A new device is added to a data core. First the device is added to the main linked list
1231 * and the allocated to a specific data core.
1236 int lcore, core_add = 0;
1237 uint32_t device_num_min = num_devices;
1238 struct vhost_dev *vdev;
1240 vdev = rte_zmalloc("vhost device", sizeof(*vdev), RTE_CACHE_LINE_SIZE);
1242 RTE_LOG(INFO, VHOST_DATA,
1243 "(%d) couldn't allocate memory for vhost dev\n",
1249 TAILQ_INSERT_TAIL(&vhost_dev_list, vdev, global_vdev_entry);
1250 vdev->vmdq_rx_q = vid * queues_per_pool + vmdq_queue_base;
1252 /*reset ready flag*/
1253 vdev->ready = DEVICE_MAC_LEARNING;
1256 /* Find a suitable lcore to add the device. */
1257 RTE_LCORE_FOREACH_SLAVE(lcore) {
1258 if (lcore_info[lcore].device_num < device_num_min) {
1259 device_num_min = lcore_info[lcore].device_num;
1263 vdev->coreid = core_add;
1265 TAILQ_INSERT_TAIL(&lcore_info[vdev->coreid].vdev_list, vdev,
1267 lcore_info[vdev->coreid].device_num++;
1269 /* Disable notifications. */
1270 rte_vhost_enable_guest_notification(vid, VIRTIO_RXQ, 0);
1271 rte_vhost_enable_guest_notification(vid, VIRTIO_TXQ, 0);
1273 RTE_LOG(INFO, VHOST_DATA,
1274 "(%d) device has been added to data core %d\n",
1281 * These callback allow devices to be added to the data core when configuration
1282 * has been fully complete.
1284 static const struct virtio_net_device_ops virtio_net_device_ops =
1286 .new_device = new_device,
1287 .destroy_device = destroy_device,
1291 * This is a thread will wake up after a period to print stats if the user has
1297 struct vhost_dev *vdev;
1298 uint64_t tx_dropped, rx_dropped;
1299 uint64_t tx, tx_total, rx, rx_total;
1300 const char clr[] = { 27, '[', '2', 'J', '\0' };
1301 const char top_left[] = { 27, '[', '1', ';', '1', 'H','\0' };
1304 sleep(enable_stats);
1306 /* Clear screen and move to top left */
1307 printf("%s%s\n", clr, top_left);
1308 printf("Device statistics =================================\n");
1310 TAILQ_FOREACH(vdev, &vhost_dev_list, global_vdev_entry) {
1311 tx_total = vdev->stats.tx_total;
1312 tx = vdev->stats.tx;
1313 tx_dropped = tx_total - tx;
1315 rx_total = rte_atomic64_read(&vdev->stats.rx_total_atomic);
1316 rx = rte_atomic64_read(&vdev->stats.rx_atomic);
1317 rx_dropped = rx_total - rx;
1319 printf("Statistics for device %d\n"
1320 "-----------------------\n"
1321 "TX total: %" PRIu64 "\n"
1322 "TX dropped: %" PRIu64 "\n"
1323 "TX successful: %" PRIu64 "\n"
1324 "RX total: %" PRIu64 "\n"
1325 "RX dropped: %" PRIu64 "\n"
1326 "RX successful: %" PRIu64 "\n",
1328 tx_total, tx_dropped, tx,
1329 rx_total, rx_dropped, rx);
1332 printf("===================================================\n");
1337 unregister_drivers(int socket_num)
1341 for (i = 0; i < socket_num; i++) {
1342 ret = rte_vhost_driver_unregister(socket_files + i * PATH_MAX);
1344 RTE_LOG(ERR, VHOST_CONFIG,
1345 "Fail to unregister vhost driver for %s.\n",
1346 socket_files + i * PATH_MAX);
1350 /* When we receive a INT signal, unregister vhost driver */
1352 sigint_handler(__rte_unused int signum)
1354 /* Unregister vhost driver. */
1355 unregister_drivers(nb_sockets);
1361 * While creating an mbuf pool, one key thing is to figure out how
1362 * many mbuf entries is enough for our use. FYI, here are some
1365 * - Each rx queue would reserve @nr_rx_desc mbufs at queue setup stage
1367 * - For each switch core (A CPU core does the packet switch), we need
1368 * also make some reservation for receiving the packets from virtio
1369 * Tx queue. How many is enough depends on the usage. It's normally
1370 * a simple calculation like following:
1372 * MAX_PKT_BURST * max packet size / mbuf size
1374 * So, we definitely need allocate more mbufs when TSO is enabled.
1376 * - Similarly, for each switching core, we should serve @nr_rx_desc
1377 * mbufs for receiving the packets from physical NIC device.
1379 * - We also need make sure, for each switch core, we have allocated
1380 * enough mbufs to fill up the mbuf cache.
1383 create_mbuf_pool(uint16_t nr_port, uint32_t nr_switch_core, uint32_t mbuf_size,
1384 uint32_t nr_queues, uint32_t nr_rx_desc, uint32_t nr_mbuf_cache)
1387 uint32_t nr_mbufs_per_core;
1388 uint32_t mtu = 1500;
1395 nr_mbufs_per_core = (mtu + mbuf_size) * MAX_PKT_BURST /
1396 (mbuf_size - RTE_PKTMBUF_HEADROOM);
1397 nr_mbufs_per_core += nr_rx_desc;
1398 nr_mbufs_per_core = RTE_MAX(nr_mbufs_per_core, nr_mbuf_cache);
1400 nr_mbufs = nr_queues * nr_rx_desc;
1401 nr_mbufs += nr_mbufs_per_core * nr_switch_core;
1402 nr_mbufs *= nr_port;
1404 mbuf_pool = rte_pktmbuf_pool_create("MBUF_POOL", nr_mbufs,
1405 nr_mbuf_cache, 0, mbuf_size,
1407 if (mbuf_pool == NULL)
1408 rte_exit(EXIT_FAILURE, "Cannot create mbuf pool\n");
1412 * Main function, does initialisation and calls the per-lcore functions.
1415 main(int argc, char *argv[])
1417 unsigned lcore_id, core_id = 0;
1418 unsigned nb_ports, valid_num_ports;
1421 static pthread_t tid;
1422 char thread_name[RTE_MAX_THREAD_NAME_LEN];
1425 signal(SIGINT, sigint_handler);
1428 ret = rte_eal_init(argc, argv);
1430 rte_exit(EXIT_FAILURE, "Error with EAL initialization\n");
1434 /* parse app arguments */
1435 ret = us_vhost_parse_args(argc, argv);
1437 rte_exit(EXIT_FAILURE, "Invalid argument\n");
1439 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1440 TAILQ_INIT(&lcore_info[lcore_id].vdev_list);
1442 if (rte_lcore_is_enabled(lcore_id))
1443 lcore_ids[core_id++] = lcore_id;
1446 if (rte_lcore_count() > RTE_MAX_LCORE)
1447 rte_exit(EXIT_FAILURE,"Not enough cores\n");
1449 /* Get the number of physical ports. */
1450 nb_ports = rte_eth_dev_count();
1453 * Update the global var NUM_PORTS and global array PORTS
1454 * and get value of var VALID_NUM_PORTS according to system ports number
1456 valid_num_ports = check_ports_num(nb_ports);
1458 if ((valid_num_ports == 0) || (valid_num_ports > MAX_SUP_PORTS)) {
1459 RTE_LOG(INFO, VHOST_PORT, "Current enabled port number is %u,"
1460 "but only %u port can be enabled\n",num_ports, MAX_SUP_PORTS);
1465 * FIXME: here we are trying to allocate mbufs big enough for
1466 * @MAX_QUEUES, but the truth is we're never going to use that
1467 * many queues here. We probably should only do allocation for
1468 * those queues we are going to use.
1470 create_mbuf_pool(valid_num_ports, rte_lcore_count() - 1, MBUF_DATA_SIZE,
1471 MAX_QUEUES, RTE_TEST_RX_DESC_DEFAULT, MBUF_CACHE_SIZE);
1473 if (vm2vm_mode == VM2VM_HARDWARE) {
1474 /* Enable VT loop back to let L2 switch to do it. */
1475 vmdq_conf_default.rx_adv_conf.vmdq_rx_conf.enable_loop_back = 1;
1476 RTE_LOG(DEBUG, VHOST_CONFIG,
1477 "Enable loop back for L2 switch in vmdq.\n");
1480 /* initialize all ports */
1481 for (portid = 0; portid < nb_ports; portid++) {
1482 /* skip ports that are not enabled */
1483 if ((enabled_port_mask & (1 << portid)) == 0) {
1484 RTE_LOG(INFO, VHOST_PORT,
1485 "Skipping disabled port %d\n", portid);
1488 if (port_init(portid) != 0)
1489 rte_exit(EXIT_FAILURE,
1490 "Cannot initialize network ports\n");
1493 /* Enable stats if the user option is set. */
1495 ret = pthread_create(&tid, NULL, (void *)print_stats, NULL);
1497 rte_exit(EXIT_FAILURE,
1498 "Cannot create print-stats thread\n");
1500 /* Set thread_name for aid in debugging. */
1501 snprintf(thread_name, RTE_MAX_THREAD_NAME_LEN, "print-stats");
1502 ret = rte_thread_setname(tid, thread_name);
1504 RTE_LOG(DEBUG, VHOST_CONFIG,
1505 "Cannot set print-stats name\n");
1508 /* Launch all data cores. */
1509 RTE_LCORE_FOREACH_SLAVE(lcore_id)
1510 rte_eal_remote_launch(switch_worker, NULL, lcore_id);
1513 rte_vhost_feature_disable(1ULL << VIRTIO_NET_F_MRG_RXBUF);
1516 flags |= RTE_VHOST_USER_CLIENT;
1518 if (dequeue_zero_copy)
1519 flags |= RTE_VHOST_USER_DEQUEUE_ZERO_COPY;
1521 /* Register vhost user driver to handle vhost messages. */
1522 for (i = 0; i < nb_sockets; i++) {
1523 ret = rte_vhost_driver_register
1524 (socket_files + i * PATH_MAX, flags);
1526 unregister_drivers(i);
1527 rte_exit(EXIT_FAILURE,
1528 "vhost driver register failure.\n");
1532 rte_vhost_driver_callback_register(&virtio_net_device_ops);
1534 rte_vhost_driver_session_start();