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_cycles.h>
18 #include <rte_ethdev.h>
20 #include <rte_string_fns.h>
21 #include <rte_malloc.h>
22 #include <rte_vhost.h>
25 #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 /* Maximum long option length for option parsing. */
59 #define MAX_LONG_OPT_SZ 64
61 /* mask of enabled ports */
62 static uint32_t enabled_port_mask = 0;
64 /* Promiscuous mode */
65 static uint32_t promiscuous;
67 /* number of devices/queues to support*/
68 static uint32_t num_queues = 0;
69 static uint32_t num_devices;
71 static struct rte_mempool *mbuf_pool;
74 /* Enable VM2VM communications. If this is disabled then the MAC address compare is skipped. */
81 static vm2vm_type vm2vm_mode = VM2VM_SOFTWARE;
84 static uint32_t enable_stats = 0;
85 /* Enable retries on RX. */
86 static uint32_t enable_retry = 1;
88 /* Disable TX checksum offload */
89 static uint32_t enable_tx_csum;
91 /* Disable TSO offload */
92 static uint32_t enable_tso;
94 static int client_mode;
96 static int builtin_net_driver;
98 static int async_vhost_driver;
100 static char dma_type[MAX_LONG_OPT_SZ];
102 /* Specify timeout (in useconds) between retries on RX. */
103 static uint32_t burst_rx_delay_time = BURST_RX_WAIT_US;
104 /* Specify the number of retries on RX. */
105 static uint32_t burst_rx_retry_num = BURST_RX_RETRIES;
107 /* Socket file paths. Can be set by user */
108 static char *socket_files;
109 static int nb_sockets;
111 /* empty vmdq configuration structure. Filled in programatically */
112 static struct rte_eth_conf vmdq_conf_default = {
114 .mq_mode = ETH_MQ_RX_VMDQ_ONLY,
117 * VLAN strip is necessary for 1G NIC such as I350,
118 * this fixes bug of ipv4 forwarding in guest can't
119 * forward pakets from one virtio dev to another virtio dev.
121 .offloads = DEV_RX_OFFLOAD_VLAN_STRIP,
125 .mq_mode = ETH_MQ_TX_NONE,
126 .offloads = (DEV_TX_OFFLOAD_IPV4_CKSUM |
127 DEV_TX_OFFLOAD_TCP_CKSUM |
128 DEV_TX_OFFLOAD_VLAN_INSERT |
129 DEV_TX_OFFLOAD_MULTI_SEGS |
130 DEV_TX_OFFLOAD_TCP_TSO),
134 * should be overridden separately in code with
138 .nb_queue_pools = ETH_8_POOLS,
139 .enable_default_pool = 0,
142 .pool_map = {{0, 0},},
148 static unsigned lcore_ids[RTE_MAX_LCORE];
149 static uint16_t ports[RTE_MAX_ETHPORTS];
150 static unsigned num_ports = 0; /**< The number of ports specified in command line */
151 static uint16_t num_pf_queues, num_vmdq_queues;
152 static uint16_t vmdq_pool_base, vmdq_queue_base;
153 static uint16_t queues_per_pool;
155 const uint16_t vlan_tags[] = {
156 1000, 1001, 1002, 1003, 1004, 1005, 1006, 1007,
157 1008, 1009, 1010, 1011, 1012, 1013, 1014, 1015,
158 1016, 1017, 1018, 1019, 1020, 1021, 1022, 1023,
159 1024, 1025, 1026, 1027, 1028, 1029, 1030, 1031,
160 1032, 1033, 1034, 1035, 1036, 1037, 1038, 1039,
161 1040, 1041, 1042, 1043, 1044, 1045, 1046, 1047,
162 1048, 1049, 1050, 1051, 1052, 1053, 1054, 1055,
163 1056, 1057, 1058, 1059, 1060, 1061, 1062, 1063,
166 /* ethernet addresses of ports */
167 static struct rte_ether_addr vmdq_ports_eth_addr[RTE_MAX_ETHPORTS];
169 static struct vhost_dev_tailq_list vhost_dev_list =
170 TAILQ_HEAD_INITIALIZER(vhost_dev_list);
172 static struct lcore_info lcore_info[RTE_MAX_LCORE];
174 /* Used for queueing bursts of TX packets. */
178 struct rte_mbuf *m_table[MAX_PKT_BURST];
181 struct vhost_bufftable {
184 struct rte_mbuf *m_table[MAX_PKT_BURST];
187 /* TX queue for each data core. */
188 struct mbuf_table lcore_tx_queue[RTE_MAX_LCORE];
191 * Vhost TX buffer for each data core.
192 * Every data core maintains a TX buffer for every vhost device,
193 * which is used for batch pkts enqueue for higher performance.
195 struct vhost_bufftable *vhost_txbuff[RTE_MAX_LCORE * MAX_VHOST_DEVICE];
197 #define MBUF_TABLE_DRAIN_TSC ((rte_get_tsc_hz() + US_PER_S - 1) \
198 / US_PER_S * BURST_TX_DRAIN_US)
202 open_dma(const char *value)
204 if (strncmp(dma_type, "ioat", 4) == 0)
205 return open_ioat(value);
211 * Builds up the correct configuration for VMDQ VLAN pool map
212 * according to the pool & queue limits.
215 get_eth_conf(struct rte_eth_conf *eth_conf, uint32_t num_devices)
217 struct rte_eth_vmdq_rx_conf conf;
218 struct rte_eth_vmdq_rx_conf *def_conf =
219 &vmdq_conf_default.rx_adv_conf.vmdq_rx_conf;
222 memset(&conf, 0, sizeof(conf));
223 conf.nb_queue_pools = (enum rte_eth_nb_pools)num_devices;
224 conf.nb_pool_maps = num_devices;
225 conf.enable_loop_back = def_conf->enable_loop_back;
226 conf.rx_mode = def_conf->rx_mode;
228 for (i = 0; i < conf.nb_pool_maps; i++) {
229 conf.pool_map[i].vlan_id = vlan_tags[ i ];
230 conf.pool_map[i].pools = (1UL << i);
233 (void)(rte_memcpy(eth_conf, &vmdq_conf_default, sizeof(*eth_conf)));
234 (void)(rte_memcpy(ð_conf->rx_adv_conf.vmdq_rx_conf, &conf,
235 sizeof(eth_conf->rx_adv_conf.vmdq_rx_conf)));
240 * Initialises a given port using global settings and with the rx buffers
241 * coming from the mbuf_pool passed as parameter
244 port_init(uint16_t port)
246 struct rte_eth_dev_info dev_info;
247 struct rte_eth_conf port_conf;
248 struct rte_eth_rxconf *rxconf;
249 struct rte_eth_txconf *txconf;
250 int16_t rx_rings, tx_rings;
251 uint16_t rx_ring_size, tx_ring_size;
255 /* The max pool number from dev_info will be used to validate the pool number specified in cmd line */
256 retval = rte_eth_dev_info_get(port, &dev_info);
258 RTE_LOG(ERR, VHOST_PORT,
259 "Error during getting device (port %u) info: %s\n",
260 port, strerror(-retval));
265 rxconf = &dev_info.default_rxconf;
266 txconf = &dev_info.default_txconf;
267 rxconf->rx_drop_en = 1;
269 /*configure the number of supported virtio devices based on VMDQ limits */
270 num_devices = dev_info.max_vmdq_pools;
272 rx_ring_size = RTE_TEST_RX_DESC_DEFAULT;
273 tx_ring_size = RTE_TEST_TX_DESC_DEFAULT;
275 tx_rings = (uint16_t)rte_lcore_count();
277 /* Get port configuration. */
278 retval = get_eth_conf(&port_conf, num_devices);
281 /* NIC queues are divided into pf queues and vmdq queues. */
282 num_pf_queues = dev_info.max_rx_queues - dev_info.vmdq_queue_num;
283 queues_per_pool = dev_info.vmdq_queue_num / dev_info.max_vmdq_pools;
284 num_vmdq_queues = num_devices * queues_per_pool;
285 num_queues = num_pf_queues + num_vmdq_queues;
286 vmdq_queue_base = dev_info.vmdq_queue_base;
287 vmdq_pool_base = dev_info.vmdq_pool_base;
288 printf("pf queue num: %u, configured vmdq pool num: %u, each vmdq pool has %u queues\n",
289 num_pf_queues, num_devices, queues_per_pool);
291 if (!rte_eth_dev_is_valid_port(port))
294 rx_rings = (uint16_t)dev_info.max_rx_queues;
295 if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
296 port_conf.txmode.offloads |=
297 DEV_TX_OFFLOAD_MBUF_FAST_FREE;
298 /* Configure ethernet device. */
299 retval = rte_eth_dev_configure(port, rx_rings, tx_rings, &port_conf);
301 RTE_LOG(ERR, VHOST_PORT, "Failed to configure port %u: %s.\n",
302 port, strerror(-retval));
306 retval = rte_eth_dev_adjust_nb_rx_tx_desc(port, &rx_ring_size,
309 RTE_LOG(ERR, VHOST_PORT, "Failed to adjust number of descriptors "
310 "for port %u: %s.\n", port, strerror(-retval));
313 if (rx_ring_size > RTE_TEST_RX_DESC_DEFAULT) {
314 RTE_LOG(ERR, VHOST_PORT, "Mbuf pool has an insufficient size "
315 "for Rx queues on port %u.\n", port);
319 /* Setup the queues. */
320 rxconf->offloads = port_conf.rxmode.offloads;
321 for (q = 0; q < rx_rings; q ++) {
322 retval = rte_eth_rx_queue_setup(port, q, rx_ring_size,
323 rte_eth_dev_socket_id(port),
327 RTE_LOG(ERR, VHOST_PORT,
328 "Failed to setup rx queue %u of port %u: %s.\n",
329 q, port, strerror(-retval));
333 txconf->offloads = port_conf.txmode.offloads;
334 for (q = 0; q < tx_rings; q ++) {
335 retval = rte_eth_tx_queue_setup(port, q, tx_ring_size,
336 rte_eth_dev_socket_id(port),
339 RTE_LOG(ERR, VHOST_PORT,
340 "Failed to setup tx queue %u of port %u: %s.\n",
341 q, port, strerror(-retval));
346 /* Start the device. */
347 retval = rte_eth_dev_start(port);
349 RTE_LOG(ERR, VHOST_PORT, "Failed to start port %u: %s\n",
350 port, strerror(-retval));
355 retval = rte_eth_promiscuous_enable(port);
357 RTE_LOG(ERR, VHOST_PORT,
358 "Failed to enable promiscuous mode on port %u: %s\n",
359 port, rte_strerror(-retval));
364 retval = rte_eth_macaddr_get(port, &vmdq_ports_eth_addr[port]);
366 RTE_LOG(ERR, VHOST_PORT,
367 "Failed to get MAC address on port %u: %s\n",
368 port, rte_strerror(-retval));
372 RTE_LOG(INFO, VHOST_PORT, "Max virtio devices supported: %u\n", num_devices);
373 RTE_LOG(INFO, VHOST_PORT, "Port %u MAC: %02"PRIx8" %02"PRIx8" %02"PRIx8
374 " %02"PRIx8" %02"PRIx8" %02"PRIx8"\n",
376 vmdq_ports_eth_addr[port].addr_bytes[0],
377 vmdq_ports_eth_addr[port].addr_bytes[1],
378 vmdq_ports_eth_addr[port].addr_bytes[2],
379 vmdq_ports_eth_addr[port].addr_bytes[3],
380 vmdq_ports_eth_addr[port].addr_bytes[4],
381 vmdq_ports_eth_addr[port].addr_bytes[5]);
387 * Set socket file path.
390 us_vhost_parse_socket_path(const char *q_arg)
394 /* parse number string */
395 if (strnlen(q_arg, PATH_MAX) == PATH_MAX)
399 socket_files = realloc(socket_files, PATH_MAX * (nb_sockets + 1));
400 if (socket_files == NULL) {
405 strlcpy(socket_files + nb_sockets * PATH_MAX, q_arg, PATH_MAX);
412 * Parse the portmask provided at run time.
415 parse_portmask(const char *portmask)
422 /* parse hexadecimal string */
423 pm = strtoul(portmask, &end, 16);
424 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0') || (errno != 0))
432 * Parse num options at run time.
435 parse_num_opt(const char *q_arg, uint32_t max_valid_value)
442 /* parse unsigned int string */
443 num = strtoul(q_arg, &end, 10);
444 if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0') || (errno != 0))
447 if (num > max_valid_value)
458 us_vhost_usage(const char *prgname)
460 RTE_LOG(INFO, VHOST_CONFIG, "%s [EAL options] -- -p PORTMASK\n"
462 " --rx_retry [0|1] --mergeable [0|1] --stats [0-N]\n"
463 " --socket-file <path>\n"
465 " -p PORTMASK: Set mask for ports to be used by application\n"
466 " --vm2vm [0|1|2]: disable/software(default)/hardware vm2vm comms\n"
467 " --rx-retry [0|1]: disable/enable(default) retries on rx. Enable retry if destintation queue is full\n"
468 " --rx-retry-delay [0-N]: timeout(in usecond) between retries on RX. This makes effect only if retries on rx enabled\n"
469 " --rx-retry-num [0-N]: the number of retries on rx. This makes effect only if retries on rx enabled\n"
470 " --mergeable [0|1]: disable(default)/enable RX mergeable buffers\n"
471 " --stats [0-N]: 0: Disable stats, N: Time in seconds to print stats\n"
472 " --socket-file: The path of the socket file.\n"
473 " --tx-csum [0|1] disable/enable TX checksum offload.\n"
474 " --tso [0|1] disable/enable TCP segment offload.\n"
475 " --client register a vhost-user socket as client mode.\n"
476 " --dma-type register dma type for your vhost async driver. For example \"ioat\" for now.\n"
477 " --dmas register dma channel for specific vhost device.\n",
482 * Parse the arguments given in the command line of the application.
485 us_vhost_parse_args(int argc, char **argv)
490 const char *prgname = argv[0];
491 static struct option long_option[] = {
492 {"vm2vm", required_argument, NULL, 0},
493 {"rx-retry", required_argument, NULL, 0},
494 {"rx-retry-delay", required_argument, NULL, 0},
495 {"rx-retry-num", required_argument, NULL, 0},
496 {"mergeable", required_argument, NULL, 0},
497 {"stats", required_argument, NULL, 0},
498 {"socket-file", required_argument, NULL, 0},
499 {"tx-csum", required_argument, NULL, 0},
500 {"tso", required_argument, NULL, 0},
501 {"client", no_argument, &client_mode, 1},
502 {"builtin-net-driver", no_argument, &builtin_net_driver, 1},
503 {"dma-type", required_argument, NULL, 0},
504 {"dmas", required_argument, NULL, 0},
508 /* Parse command line */
509 while ((opt = getopt_long(argc, argv, "p:P",
510 long_option, &option_index)) != EOF) {
514 enabled_port_mask = parse_portmask(optarg);
515 if (enabled_port_mask == 0) {
516 RTE_LOG(INFO, VHOST_CONFIG, "Invalid portmask\n");
517 us_vhost_usage(prgname);
524 vmdq_conf_default.rx_adv_conf.vmdq_rx_conf.rx_mode =
525 ETH_VMDQ_ACCEPT_BROADCAST |
526 ETH_VMDQ_ACCEPT_MULTICAST;
531 /* Enable/disable vm2vm comms. */
532 if (!strncmp(long_option[option_index].name, "vm2vm",
534 ret = parse_num_opt(optarg, (VM2VM_LAST - 1));
536 RTE_LOG(INFO, VHOST_CONFIG,
537 "Invalid argument for "
539 us_vhost_usage(prgname);
542 vm2vm_mode = (vm2vm_type)ret;
546 /* Enable/disable retries on RX. */
547 if (!strncmp(long_option[option_index].name, "rx-retry", MAX_LONG_OPT_SZ)) {
548 ret = parse_num_opt(optarg, 1);
550 RTE_LOG(INFO, VHOST_CONFIG, "Invalid argument for rx-retry [0|1]\n");
551 us_vhost_usage(prgname);
558 /* Enable/disable TX checksum offload. */
559 if (!strncmp(long_option[option_index].name, "tx-csum", MAX_LONG_OPT_SZ)) {
560 ret = parse_num_opt(optarg, 1);
562 RTE_LOG(INFO, VHOST_CONFIG, "Invalid argument for tx-csum [0|1]\n");
563 us_vhost_usage(prgname);
566 enable_tx_csum = ret;
569 /* Enable/disable TSO offload. */
570 if (!strncmp(long_option[option_index].name, "tso", MAX_LONG_OPT_SZ)) {
571 ret = parse_num_opt(optarg, 1);
573 RTE_LOG(INFO, VHOST_CONFIG, "Invalid argument for tso [0|1]\n");
574 us_vhost_usage(prgname);
580 /* Specify the retries delay time (in useconds) on RX. */
581 if (!strncmp(long_option[option_index].name, "rx-retry-delay", MAX_LONG_OPT_SZ)) {
582 ret = parse_num_opt(optarg, INT32_MAX);
584 RTE_LOG(INFO, VHOST_CONFIG, "Invalid argument for rx-retry-delay [0-N]\n");
585 us_vhost_usage(prgname);
588 burst_rx_delay_time = ret;
592 /* Specify the retries number on RX. */
593 if (!strncmp(long_option[option_index].name, "rx-retry-num", MAX_LONG_OPT_SZ)) {
594 ret = parse_num_opt(optarg, INT32_MAX);
596 RTE_LOG(INFO, VHOST_CONFIG, "Invalid argument for rx-retry-num [0-N]\n");
597 us_vhost_usage(prgname);
600 burst_rx_retry_num = ret;
604 /* Enable/disable RX mergeable buffers. */
605 if (!strncmp(long_option[option_index].name, "mergeable", MAX_LONG_OPT_SZ)) {
606 ret = parse_num_opt(optarg, 1);
608 RTE_LOG(INFO, VHOST_CONFIG, "Invalid argument for mergeable [0|1]\n");
609 us_vhost_usage(prgname);
614 vmdq_conf_default.rxmode.offloads |=
615 DEV_RX_OFFLOAD_JUMBO_FRAME;
616 vmdq_conf_default.rxmode.max_rx_pkt_len
617 = JUMBO_FRAME_MAX_SIZE;
622 /* Enable/disable stats. */
623 if (!strncmp(long_option[option_index].name, "stats", MAX_LONG_OPT_SZ)) {
624 ret = parse_num_opt(optarg, INT32_MAX);
626 RTE_LOG(INFO, VHOST_CONFIG,
627 "Invalid argument for stats [0..N]\n");
628 us_vhost_usage(prgname);
635 /* Set socket file path. */
636 if (!strncmp(long_option[option_index].name,
637 "socket-file", MAX_LONG_OPT_SZ)) {
638 if (us_vhost_parse_socket_path(optarg) == -1) {
639 RTE_LOG(INFO, VHOST_CONFIG,
640 "Invalid argument for socket name (Max %d characters)\n",
642 us_vhost_usage(prgname);
647 if (!strncmp(long_option[option_index].name,
648 "dma-type", MAX_LONG_OPT_SZ)) {
649 if (strlen(optarg) >= MAX_LONG_OPT_SZ) {
650 RTE_LOG(INFO, VHOST_CONFIG,
652 us_vhost_usage(prgname);
655 strcpy(dma_type, optarg);
658 if (!strncmp(long_option[option_index].name,
659 "dmas", MAX_LONG_OPT_SZ)) {
660 if (open_dma(optarg) == -1) {
661 RTE_LOG(INFO, VHOST_CONFIG,
663 us_vhost_usage(prgname);
666 async_vhost_driver = 1;
671 /* Invalid option - print options. */
673 us_vhost_usage(prgname);
678 for (i = 0; i < RTE_MAX_ETHPORTS; i++) {
679 if (enabled_port_mask & (1 << i))
680 ports[num_ports++] = i;
683 if ((num_ports == 0) || (num_ports > MAX_SUP_PORTS)) {
684 RTE_LOG(INFO, VHOST_PORT, "Current enabled port number is %u,"
685 "but only %u port can be enabled\n",num_ports, MAX_SUP_PORTS);
693 * Update the global var NUM_PORTS and array PORTS according to system ports number
694 * and return valid ports number
696 static unsigned check_ports_num(unsigned nb_ports)
698 unsigned valid_num_ports = num_ports;
701 if (num_ports > nb_ports) {
702 RTE_LOG(INFO, VHOST_PORT, "\nSpecified port number(%u) exceeds total system port number(%u)\n",
703 num_ports, nb_ports);
704 num_ports = nb_ports;
707 for (portid = 0; portid < num_ports; portid ++) {
708 if (!rte_eth_dev_is_valid_port(ports[portid])) {
709 RTE_LOG(INFO, VHOST_PORT,
710 "\nSpecified port ID(%u) is not valid\n",
712 ports[portid] = INVALID_PORT_ID;
716 return valid_num_ports;
719 static __rte_always_inline struct vhost_dev *
720 find_vhost_dev(struct rte_ether_addr *mac)
722 struct vhost_dev *vdev;
724 TAILQ_FOREACH(vdev, &vhost_dev_list, global_vdev_entry) {
725 if (vdev->ready == DEVICE_RX &&
726 rte_is_same_ether_addr(mac, &vdev->mac_address))
734 * This function learns the MAC address of the device and registers this along with a
735 * vlan tag to a VMDQ.
738 link_vmdq(struct vhost_dev *vdev, struct rte_mbuf *m)
740 struct rte_ether_hdr *pkt_hdr;
743 /* Learn MAC address of guest device from packet */
744 pkt_hdr = rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
746 if (find_vhost_dev(&pkt_hdr->s_addr)) {
747 RTE_LOG(ERR, VHOST_DATA,
748 "(%d) device is using a registered MAC!\n",
753 for (i = 0; i < RTE_ETHER_ADDR_LEN; i++)
754 vdev->mac_address.addr_bytes[i] = pkt_hdr->s_addr.addr_bytes[i];
756 /* vlan_tag currently uses the device_id. */
757 vdev->vlan_tag = vlan_tags[vdev->vid];
759 /* Print out VMDQ registration info. */
760 RTE_LOG(INFO, VHOST_DATA,
761 "(%d) mac %02x:%02x:%02x:%02x:%02x:%02x and vlan %d registered\n",
763 vdev->mac_address.addr_bytes[0], vdev->mac_address.addr_bytes[1],
764 vdev->mac_address.addr_bytes[2], vdev->mac_address.addr_bytes[3],
765 vdev->mac_address.addr_bytes[4], vdev->mac_address.addr_bytes[5],
768 /* Register the MAC address. */
769 ret = rte_eth_dev_mac_addr_add(ports[0], &vdev->mac_address,
770 (uint32_t)vdev->vid + vmdq_pool_base);
772 RTE_LOG(ERR, VHOST_DATA,
773 "(%d) failed to add device MAC address to VMDQ\n",
776 rte_eth_dev_set_vlan_strip_on_queue(ports[0], vdev->vmdq_rx_q, 1);
778 /* Set device as ready for RX. */
779 vdev->ready = DEVICE_RX;
785 * Removes MAC address and vlan tag from VMDQ. Ensures that nothing is adding buffers to the RX
786 * queue before disabling RX on the device.
789 unlink_vmdq(struct vhost_dev *vdev)
793 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
795 if (vdev->ready == DEVICE_RX) {
796 /*clear MAC and VLAN settings*/
797 rte_eth_dev_mac_addr_remove(ports[0], &vdev->mac_address);
798 for (i = 0; i < 6; i++)
799 vdev->mac_address.addr_bytes[i] = 0;
803 /*Clear out the receive buffers*/
804 rx_count = rte_eth_rx_burst(ports[0],
805 (uint16_t)vdev->vmdq_rx_q, pkts_burst, MAX_PKT_BURST);
808 for (i = 0; i < rx_count; i++)
809 rte_pktmbuf_free(pkts_burst[i]);
811 rx_count = rte_eth_rx_burst(ports[0],
812 (uint16_t)vdev->vmdq_rx_q, pkts_burst, MAX_PKT_BURST);
815 vdev->ready = DEVICE_MAC_LEARNING;
820 free_pkts(struct rte_mbuf **pkts, uint16_t n)
823 rte_pktmbuf_free(pkts[n]);
826 static __rte_always_inline void
827 complete_async_pkts(struct vhost_dev *vdev)
829 struct rte_mbuf *p_cpl[MAX_PKT_BURST];
830 uint16_t complete_count;
832 complete_count = rte_vhost_poll_enqueue_completed(vdev->vid,
833 VIRTIO_RXQ, p_cpl, MAX_PKT_BURST);
835 free_pkts(p_cpl, complete_count);
838 static __rte_always_inline void
839 sync_virtio_xmit(struct vhost_dev *dst_vdev, struct vhost_dev *src_vdev,
844 if (builtin_net_driver) {
845 ret = vs_enqueue_pkts(dst_vdev, VIRTIO_RXQ, &m, 1);
847 ret = rte_vhost_enqueue_burst(dst_vdev->vid, VIRTIO_RXQ, &m, 1);
851 __atomic_add_fetch(&dst_vdev->stats.rx_total_atomic, 1,
853 __atomic_add_fetch(&dst_vdev->stats.rx_atomic, ret,
855 src_vdev->stats.tx_total++;
856 src_vdev->stats.tx += ret;
860 static __rte_always_inline void
861 drain_vhost(struct vhost_dev *vdev)
864 uint32_t buff_idx = rte_lcore_id() * MAX_VHOST_DEVICE + vdev->vid;
865 uint16_t nr_xmit = vhost_txbuff[buff_idx]->len;
866 struct rte_mbuf **m = vhost_txbuff[buff_idx]->m_table;
868 if (builtin_net_driver) {
869 ret = vs_enqueue_pkts(vdev, VIRTIO_RXQ, m, nr_xmit);
870 } else if (async_vhost_driver) {
871 uint32_t cpu_cpl_nr = 0;
872 uint16_t enqueue_fail = 0;
873 struct rte_mbuf *m_cpu_cpl[nr_xmit];
875 complete_async_pkts(vdev);
876 ret = rte_vhost_submit_enqueue_burst(vdev->vid, VIRTIO_RXQ,
877 m, nr_xmit, m_cpu_cpl, &cpu_cpl_nr);
880 free_pkts(m_cpu_cpl, cpu_cpl_nr);
882 enqueue_fail = nr_xmit - ret;
884 free_pkts(&m[ret], nr_xmit - ret);
886 ret = rte_vhost_enqueue_burst(vdev->vid, VIRTIO_RXQ,
891 __atomic_add_fetch(&vdev->stats.rx_total_atomic, nr_xmit,
893 __atomic_add_fetch(&vdev->stats.rx_atomic, ret,
897 if (!async_vhost_driver)
898 free_pkts(m, nr_xmit);
901 static __rte_always_inline void
902 drain_vhost_table(void)
904 uint16_t lcore_id = rte_lcore_id();
905 struct vhost_bufftable *vhost_txq;
906 struct vhost_dev *vdev;
909 TAILQ_FOREACH(vdev, &vhost_dev_list, global_vdev_entry) {
910 vhost_txq = vhost_txbuff[lcore_id * MAX_VHOST_DEVICE
913 cur_tsc = rte_rdtsc();
914 if (unlikely(cur_tsc - vhost_txq->pre_tsc
915 > MBUF_TABLE_DRAIN_TSC)) {
916 RTE_LOG_DP(DEBUG, VHOST_DATA,
917 "Vhost TX queue drained after timeout with burst size %u\n",
921 vhost_txq->pre_tsc = cur_tsc;
927 * Check if the packet destination MAC address is for a local device. If so then put
928 * the packet on that devices RX queue. If not then return.
930 static __rte_always_inline int
931 virtio_tx_local(struct vhost_dev *vdev, struct rte_mbuf *m)
933 struct rte_ether_hdr *pkt_hdr;
934 struct vhost_dev *dst_vdev;
935 struct vhost_bufftable *vhost_txq;
936 uint16_t lcore_id = rte_lcore_id();
937 pkt_hdr = rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
939 dst_vdev = find_vhost_dev(&pkt_hdr->d_addr);
943 if (vdev->vid == dst_vdev->vid) {
944 RTE_LOG_DP(DEBUG, VHOST_DATA,
945 "(%d) TX: src and dst MAC is same. Dropping packet.\n",
950 RTE_LOG_DP(DEBUG, VHOST_DATA,
951 "(%d) TX: MAC address is local\n", dst_vdev->vid);
953 if (unlikely(dst_vdev->remove)) {
954 RTE_LOG_DP(DEBUG, VHOST_DATA,
955 "(%d) device is marked for removal\n", dst_vdev->vid);
959 vhost_txq = vhost_txbuff[lcore_id * MAX_VHOST_DEVICE + dst_vdev->vid];
960 vhost_txq->m_table[vhost_txq->len++] = m;
963 vdev->stats.tx_total++;
967 if (unlikely(vhost_txq->len == MAX_PKT_BURST)) {
968 drain_vhost(dst_vdev);
970 vhost_txq->pre_tsc = rte_rdtsc();
976 * Check if the destination MAC of a packet is one local VM,
977 * and get its vlan tag, and offset if it is.
979 static __rte_always_inline int
980 find_local_dest(struct vhost_dev *vdev, struct rte_mbuf *m,
981 uint32_t *offset, uint16_t *vlan_tag)
983 struct vhost_dev *dst_vdev;
984 struct rte_ether_hdr *pkt_hdr =
985 rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
987 dst_vdev = find_vhost_dev(&pkt_hdr->d_addr);
991 if (vdev->vid == dst_vdev->vid) {
992 RTE_LOG_DP(DEBUG, VHOST_DATA,
993 "(%d) TX: src and dst MAC is same. Dropping packet.\n",
999 * HW vlan strip will reduce the packet length
1000 * by minus length of vlan tag, so need restore
1001 * the packet length by plus it.
1003 *offset = VLAN_HLEN;
1004 *vlan_tag = vlan_tags[vdev->vid];
1006 RTE_LOG_DP(DEBUG, VHOST_DATA,
1007 "(%d) TX: pkt to local VM device id: (%d), vlan tag: %u.\n",
1008 vdev->vid, dst_vdev->vid, *vlan_tag);
1014 get_psd_sum(void *l3_hdr, uint64_t ol_flags)
1016 if (ol_flags & PKT_TX_IPV4)
1017 return rte_ipv4_phdr_cksum(l3_hdr, ol_flags);
1018 else /* assume ethertype == RTE_ETHER_TYPE_IPV6 */
1019 return rte_ipv6_phdr_cksum(l3_hdr, ol_flags);
1022 static void virtio_tx_offload(struct rte_mbuf *m)
1025 struct rte_ipv4_hdr *ipv4_hdr = NULL;
1026 struct rte_tcp_hdr *tcp_hdr = NULL;
1027 struct rte_ether_hdr *eth_hdr =
1028 rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
1030 l3_hdr = (char *)eth_hdr + m->l2_len;
1032 if (m->ol_flags & PKT_TX_IPV4) {
1034 ipv4_hdr->hdr_checksum = 0;
1035 m->ol_flags |= PKT_TX_IP_CKSUM;
1038 tcp_hdr = (struct rte_tcp_hdr *)((char *)l3_hdr + m->l3_len);
1039 tcp_hdr->cksum = get_psd_sum(l3_hdr, m->ol_flags);
1042 static __rte_always_inline void
1043 do_drain_mbuf_table(struct mbuf_table *tx_q)
1047 count = rte_eth_tx_burst(ports[0], tx_q->txq_id,
1048 tx_q->m_table, tx_q->len);
1049 if (unlikely(count < tx_q->len))
1050 free_pkts(&tx_q->m_table[count], tx_q->len - count);
1056 * This function routes the TX packet to the correct interface. This
1057 * may be a local device or the physical port.
1059 static __rte_always_inline void
1060 virtio_tx_route(struct vhost_dev *vdev, struct rte_mbuf *m, uint16_t vlan_tag)
1062 struct mbuf_table *tx_q;
1063 unsigned offset = 0;
1064 const uint16_t lcore_id = rte_lcore_id();
1065 struct rte_ether_hdr *nh;
1068 nh = rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
1069 if (unlikely(rte_is_broadcast_ether_addr(&nh->d_addr))) {
1070 struct vhost_dev *vdev2;
1072 TAILQ_FOREACH(vdev2, &vhost_dev_list, global_vdev_entry) {
1074 sync_virtio_xmit(vdev2, vdev, m);
1079 /*check if destination is local VM*/
1080 if ((vm2vm_mode == VM2VM_SOFTWARE) && (virtio_tx_local(vdev, m) == 0))
1083 if (unlikely(vm2vm_mode == VM2VM_HARDWARE)) {
1084 if (unlikely(find_local_dest(vdev, m, &offset,
1086 rte_pktmbuf_free(m);
1091 RTE_LOG_DP(DEBUG, VHOST_DATA,
1092 "(%d) TX: MAC address is external\n", vdev->vid);
1096 /*Add packet to the port tx queue*/
1097 tx_q = &lcore_tx_queue[lcore_id];
1099 nh = rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
1100 if (unlikely(nh->ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_VLAN))) {
1101 /* Guest has inserted the vlan tag. */
1102 struct rte_vlan_hdr *vh = (struct rte_vlan_hdr *) (nh + 1);
1103 uint16_t vlan_tag_be = rte_cpu_to_be_16(vlan_tag);
1104 if ((vm2vm_mode == VM2VM_HARDWARE) &&
1105 (vh->vlan_tci != vlan_tag_be))
1106 vh->vlan_tci = vlan_tag_be;
1108 m->ol_flags |= PKT_TX_VLAN_PKT;
1111 * Find the right seg to adjust the data len when offset is
1112 * bigger than tail room size.
1114 if (unlikely(vm2vm_mode == VM2VM_HARDWARE)) {
1115 if (likely(offset <= rte_pktmbuf_tailroom(m)))
1116 m->data_len += offset;
1118 struct rte_mbuf *seg = m;
1120 while ((seg->next != NULL) &&
1121 (offset > rte_pktmbuf_tailroom(seg)))
1124 seg->data_len += offset;
1126 m->pkt_len += offset;
1129 m->vlan_tci = vlan_tag;
1132 if (m->ol_flags & PKT_TX_TCP_SEG)
1133 virtio_tx_offload(m);
1135 tx_q->m_table[tx_q->len++] = m;
1137 vdev->stats.tx_total++;
1141 if (unlikely(tx_q->len == MAX_PKT_BURST))
1142 do_drain_mbuf_table(tx_q);
1146 static __rte_always_inline void
1147 drain_mbuf_table(struct mbuf_table *tx_q)
1149 static uint64_t prev_tsc;
1155 cur_tsc = rte_rdtsc();
1156 if (unlikely(cur_tsc - prev_tsc > MBUF_TABLE_DRAIN_TSC)) {
1159 RTE_LOG_DP(DEBUG, VHOST_DATA,
1160 "TX queue drained after timeout with burst size %u\n",
1162 do_drain_mbuf_table(tx_q);
1166 static __rte_always_inline void
1167 drain_eth_rx(struct vhost_dev *vdev)
1169 uint16_t rx_count, enqueue_count;
1170 struct rte_mbuf *pkts[MAX_PKT_BURST];
1172 rx_count = rte_eth_rx_burst(ports[0], vdev->vmdq_rx_q,
1173 pkts, MAX_PKT_BURST);
1179 * When "enable_retry" is set, here we wait and retry when there
1180 * is no enough free slots in the queue to hold @rx_count packets,
1181 * to diminish packet loss.
1184 unlikely(rx_count > rte_vhost_avail_entries(vdev->vid,
1188 for (retry = 0; retry < burst_rx_retry_num; retry++) {
1189 rte_delay_us(burst_rx_delay_time);
1190 if (rx_count <= rte_vhost_avail_entries(vdev->vid,
1196 if (builtin_net_driver) {
1197 enqueue_count = vs_enqueue_pkts(vdev, VIRTIO_RXQ,
1199 } else if (async_vhost_driver) {
1200 uint32_t cpu_cpl_nr = 0;
1201 uint16_t enqueue_fail = 0;
1202 struct rte_mbuf *m_cpu_cpl[MAX_PKT_BURST];
1204 complete_async_pkts(vdev);
1205 enqueue_count = rte_vhost_submit_enqueue_burst(vdev->vid,
1206 VIRTIO_RXQ, pkts, rx_count,
1207 m_cpu_cpl, &cpu_cpl_nr);
1209 free_pkts(m_cpu_cpl, cpu_cpl_nr);
1211 enqueue_fail = rx_count - enqueue_count;
1213 free_pkts(&pkts[enqueue_count], enqueue_fail);
1216 enqueue_count = rte_vhost_enqueue_burst(vdev->vid, VIRTIO_RXQ,
1221 __atomic_add_fetch(&vdev->stats.rx_total_atomic, rx_count,
1223 __atomic_add_fetch(&vdev->stats.rx_atomic, enqueue_count,
1227 if (!async_vhost_driver)
1228 free_pkts(pkts, rx_count);
1231 static __rte_always_inline void
1232 drain_virtio_tx(struct vhost_dev *vdev)
1234 struct rte_mbuf *pkts[MAX_PKT_BURST];
1238 if (builtin_net_driver) {
1239 count = vs_dequeue_pkts(vdev, VIRTIO_TXQ, mbuf_pool,
1240 pkts, MAX_PKT_BURST);
1242 count = rte_vhost_dequeue_burst(vdev->vid, VIRTIO_TXQ,
1243 mbuf_pool, pkts, MAX_PKT_BURST);
1246 /* setup VMDq for the first packet */
1247 if (unlikely(vdev->ready == DEVICE_MAC_LEARNING) && count) {
1248 if (vdev->remove || link_vmdq(vdev, pkts[0]) == -1)
1249 free_pkts(pkts, count);
1252 for (i = 0; i < count; ++i)
1253 virtio_tx_route(vdev, pkts[i], vlan_tags[vdev->vid]);
1257 * Main function of vhost-switch. It basically does:
1259 * for each vhost device {
1262 * Which drains the host eth Rx queue linked to the vhost device,
1263 * and deliver all of them to guest virito Rx ring associated with
1264 * this vhost device.
1266 * - drain_virtio_tx()
1268 * Which drains the guest virtio Tx queue and deliver all of them
1269 * to the target, which could be another vhost device, or the
1270 * physical eth dev. The route is done in function "virtio_tx_route".
1274 switch_worker(void *arg __rte_unused)
1277 unsigned lcore_id = rte_lcore_id();
1278 struct vhost_dev *vdev;
1279 struct mbuf_table *tx_q;
1281 RTE_LOG(INFO, VHOST_DATA, "Procesing on Core %u started\n", lcore_id);
1283 tx_q = &lcore_tx_queue[lcore_id];
1284 for (i = 0; i < rte_lcore_count(); i++) {
1285 if (lcore_ids[i] == lcore_id) {
1292 drain_mbuf_table(tx_q);
1293 drain_vhost_table();
1295 * Inform the configuration core that we have exited the
1296 * linked list and that no devices are in use if requested.
1298 if (lcore_info[lcore_id].dev_removal_flag == REQUEST_DEV_REMOVAL)
1299 lcore_info[lcore_id].dev_removal_flag = ACK_DEV_REMOVAL;
1302 * Process vhost devices
1304 TAILQ_FOREACH(vdev, &lcore_info[lcore_id].vdev_list,
1306 if (unlikely(vdev->remove)) {
1308 vdev->ready = DEVICE_SAFE_REMOVE;
1312 if (likely(vdev->ready == DEVICE_RX))
1315 if (likely(!vdev->remove))
1316 drain_virtio_tx(vdev);
1324 * Remove a device from the specific data core linked list and from the
1325 * main linked list. Synchonization occurs through the use of the
1326 * lcore dev_removal_flag. Device is made volatile here to avoid re-ordering
1327 * of dev->remove=1 which can cause an infinite loop in the rte_pause loop.
1330 destroy_device(int vid)
1332 struct vhost_dev *vdev = NULL;
1336 TAILQ_FOREACH(vdev, &vhost_dev_list, global_vdev_entry) {
1337 if (vdev->vid == vid)
1342 /*set the remove flag. */
1344 while(vdev->ready != DEVICE_SAFE_REMOVE) {
1348 for (i = 0; i < RTE_MAX_LCORE; i++)
1349 rte_free(vhost_txbuff[i * MAX_VHOST_DEVICE + vid]);
1351 if (builtin_net_driver)
1352 vs_vhost_net_remove(vdev);
1354 TAILQ_REMOVE(&lcore_info[vdev->coreid].vdev_list, vdev,
1356 TAILQ_REMOVE(&vhost_dev_list, vdev, global_vdev_entry);
1359 /* Set the dev_removal_flag on each lcore. */
1360 RTE_LCORE_FOREACH_WORKER(lcore)
1361 lcore_info[lcore].dev_removal_flag = REQUEST_DEV_REMOVAL;
1364 * Once each core has set the dev_removal_flag to ACK_DEV_REMOVAL
1365 * we can be sure that they can no longer access the device removed
1366 * from the linked lists and that the devices are no longer in use.
1368 RTE_LCORE_FOREACH_WORKER(lcore) {
1369 while (lcore_info[lcore].dev_removal_flag != ACK_DEV_REMOVAL)
1373 lcore_info[vdev->coreid].device_num--;
1375 RTE_LOG(INFO, VHOST_DATA,
1376 "(%d) device has been removed from data core\n",
1379 if (async_vhost_driver)
1380 rte_vhost_async_channel_unregister(vid, VIRTIO_RXQ);
1386 * A new device is added to a data core. First the device is added to the main linked list
1387 * and then allocated to a specific data core.
1392 int lcore, core_add = 0;
1394 uint32_t device_num_min = num_devices;
1395 struct vhost_dev *vdev;
1396 vdev = rte_zmalloc("vhost device", sizeof(*vdev), RTE_CACHE_LINE_SIZE);
1398 RTE_LOG(INFO, VHOST_DATA,
1399 "(%d) couldn't allocate memory for vhost dev\n",
1405 for (i = 0; i < RTE_MAX_LCORE; i++) {
1406 vhost_txbuff[i * MAX_VHOST_DEVICE + vid]
1407 = rte_zmalloc("vhost bufftable",
1408 sizeof(struct vhost_bufftable),
1409 RTE_CACHE_LINE_SIZE);
1411 if (vhost_txbuff[i * MAX_VHOST_DEVICE + vid] == NULL) {
1412 RTE_LOG(INFO, VHOST_DATA,
1413 "(%d) couldn't allocate memory for vhost TX\n", vid);
1418 if (builtin_net_driver)
1419 vs_vhost_net_setup(vdev);
1421 TAILQ_INSERT_TAIL(&vhost_dev_list, vdev, global_vdev_entry);
1422 vdev->vmdq_rx_q = vid * queues_per_pool + vmdq_queue_base;
1424 /*reset ready flag*/
1425 vdev->ready = DEVICE_MAC_LEARNING;
1428 /* Find a suitable lcore to add the device. */
1429 RTE_LCORE_FOREACH_WORKER(lcore) {
1430 if (lcore_info[lcore].device_num < device_num_min) {
1431 device_num_min = lcore_info[lcore].device_num;
1435 vdev->coreid = core_add;
1437 TAILQ_INSERT_TAIL(&lcore_info[vdev->coreid].vdev_list, vdev,
1439 lcore_info[vdev->coreid].device_num++;
1441 /* Disable notifications. */
1442 rte_vhost_enable_guest_notification(vid, VIRTIO_RXQ, 0);
1443 rte_vhost_enable_guest_notification(vid, VIRTIO_TXQ, 0);
1445 RTE_LOG(INFO, VHOST_DATA,
1446 "(%d) device has been added to data core %d\n",
1449 if (async_vhost_driver) {
1450 struct rte_vhost_async_features f;
1451 struct rte_vhost_async_channel_ops channel_ops;
1453 if (strncmp(dma_type, "ioat", 4) == 0) {
1454 channel_ops.transfer_data = ioat_transfer_data_cb;
1455 channel_ops.check_completed_copies =
1456 ioat_check_completed_copies_cb;
1458 f.async_inorder = 1;
1459 f.async_threshold = 256;
1461 return rte_vhost_async_channel_register(vid, VIRTIO_RXQ,
1462 f.intval, &channel_ops);
1470 * These callback allow devices to be added to the data core when configuration
1471 * has been fully complete.
1473 static const struct vhost_device_ops virtio_net_device_ops =
1475 .new_device = new_device,
1476 .destroy_device = destroy_device,
1480 * This is a thread will wake up after a period to print stats if the user has
1484 print_stats(__rte_unused void *arg)
1486 struct vhost_dev *vdev;
1487 uint64_t tx_dropped, rx_dropped;
1488 uint64_t tx, tx_total, rx, rx_total;
1489 const char clr[] = { 27, '[', '2', 'J', '\0' };
1490 const char top_left[] = { 27, '[', '1', ';', '1', 'H','\0' };
1493 sleep(enable_stats);
1495 /* Clear screen and move to top left */
1496 printf("%s%s\n", clr, top_left);
1497 printf("Device statistics =================================\n");
1499 TAILQ_FOREACH(vdev, &vhost_dev_list, global_vdev_entry) {
1500 tx_total = vdev->stats.tx_total;
1501 tx = vdev->stats.tx;
1502 tx_dropped = tx_total - tx;
1504 rx_total = __atomic_load_n(&vdev->stats.rx_total_atomic,
1506 rx = __atomic_load_n(&vdev->stats.rx_atomic,
1508 rx_dropped = rx_total - rx;
1510 printf("Statistics for device %d\n"
1511 "-----------------------\n"
1512 "TX total: %" PRIu64 "\n"
1513 "TX dropped: %" PRIu64 "\n"
1514 "TX successful: %" PRIu64 "\n"
1515 "RX total: %" PRIu64 "\n"
1516 "RX dropped: %" PRIu64 "\n"
1517 "RX successful: %" PRIu64 "\n",
1519 tx_total, tx_dropped, tx,
1520 rx_total, rx_dropped, rx);
1523 printf("===================================================\n");
1532 unregister_drivers(int socket_num)
1536 for (i = 0; i < socket_num; i++) {
1537 ret = rte_vhost_driver_unregister(socket_files + i * PATH_MAX);
1539 RTE_LOG(ERR, VHOST_CONFIG,
1540 "Fail to unregister vhost driver for %s.\n",
1541 socket_files + i * PATH_MAX);
1545 /* When we receive a INT signal, unregister vhost driver */
1547 sigint_handler(__rte_unused int signum)
1549 /* Unregister vhost driver. */
1550 unregister_drivers(nb_sockets);
1556 * While creating an mbuf pool, one key thing is to figure out how
1557 * many mbuf entries is enough for our use. FYI, here are some
1560 * - Each rx queue would reserve @nr_rx_desc mbufs at queue setup stage
1562 * - For each switch core (A CPU core does the packet switch), we need
1563 * also make some reservation for receiving the packets from virtio
1564 * Tx queue. How many is enough depends on the usage. It's normally
1565 * a simple calculation like following:
1567 * MAX_PKT_BURST * max packet size / mbuf size
1569 * So, we definitely need allocate more mbufs when TSO is enabled.
1571 * - Similarly, for each switching core, we should serve @nr_rx_desc
1572 * mbufs for receiving the packets from physical NIC device.
1574 * - We also need make sure, for each switch core, we have allocated
1575 * enough mbufs to fill up the mbuf cache.
1578 create_mbuf_pool(uint16_t nr_port, uint32_t nr_switch_core, uint32_t mbuf_size,
1579 uint32_t nr_queues, uint32_t nr_rx_desc, uint32_t nr_mbuf_cache)
1582 uint32_t nr_mbufs_per_core;
1583 uint32_t mtu = 1500;
1590 nr_mbufs_per_core = (mtu + mbuf_size) * MAX_PKT_BURST /
1591 (mbuf_size - RTE_PKTMBUF_HEADROOM);
1592 nr_mbufs_per_core += nr_rx_desc;
1593 nr_mbufs_per_core = RTE_MAX(nr_mbufs_per_core, nr_mbuf_cache);
1595 nr_mbufs = nr_queues * nr_rx_desc;
1596 nr_mbufs += nr_mbufs_per_core * nr_switch_core;
1597 nr_mbufs *= nr_port;
1599 mbuf_pool = rte_pktmbuf_pool_create("MBUF_POOL", nr_mbufs,
1600 nr_mbuf_cache, 0, mbuf_size,
1602 if (mbuf_pool == NULL)
1603 rte_exit(EXIT_FAILURE, "Cannot create mbuf pool\n");
1607 * Main function, does initialisation and calls the per-lcore functions.
1610 main(int argc, char *argv[])
1612 unsigned lcore_id, core_id = 0;
1613 unsigned nb_ports, valid_num_ports;
1616 static pthread_t tid;
1619 signal(SIGINT, sigint_handler);
1622 ret = rte_eal_init(argc, argv);
1624 rte_exit(EXIT_FAILURE, "Error with EAL initialization\n");
1628 /* parse app arguments */
1629 ret = us_vhost_parse_args(argc, argv);
1631 rte_exit(EXIT_FAILURE, "Invalid argument\n");
1633 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1634 TAILQ_INIT(&lcore_info[lcore_id].vdev_list);
1636 if (rte_lcore_is_enabled(lcore_id))
1637 lcore_ids[core_id++] = lcore_id;
1640 if (rte_lcore_count() > RTE_MAX_LCORE)
1641 rte_exit(EXIT_FAILURE,"Not enough cores\n");
1643 /* Get the number of physical ports. */
1644 nb_ports = rte_eth_dev_count_avail();
1647 * Update the global var NUM_PORTS and global array PORTS
1648 * and get value of var VALID_NUM_PORTS according to system ports number
1650 valid_num_ports = check_ports_num(nb_ports);
1652 if ((valid_num_ports == 0) || (valid_num_ports > MAX_SUP_PORTS)) {
1653 RTE_LOG(INFO, VHOST_PORT, "Current enabled port number is %u,"
1654 "but only %u port can be enabled\n",num_ports, MAX_SUP_PORTS);
1659 * FIXME: here we are trying to allocate mbufs big enough for
1660 * @MAX_QUEUES, but the truth is we're never going to use that
1661 * many queues here. We probably should only do allocation for
1662 * those queues we are going to use.
1664 create_mbuf_pool(valid_num_ports, rte_lcore_count() - 1, MBUF_DATA_SIZE,
1665 MAX_QUEUES, RTE_TEST_RX_DESC_DEFAULT, MBUF_CACHE_SIZE);
1667 if (vm2vm_mode == VM2VM_HARDWARE) {
1668 /* Enable VT loop back to let L2 switch to do it. */
1669 vmdq_conf_default.rx_adv_conf.vmdq_rx_conf.enable_loop_back = 1;
1670 RTE_LOG(DEBUG, VHOST_CONFIG,
1671 "Enable loop back for L2 switch in vmdq.\n");
1674 /* initialize all ports */
1675 RTE_ETH_FOREACH_DEV(portid) {
1676 /* skip ports that are not enabled */
1677 if ((enabled_port_mask & (1 << portid)) == 0) {
1678 RTE_LOG(INFO, VHOST_PORT,
1679 "Skipping disabled port %d\n", portid);
1682 if (port_init(portid) != 0)
1683 rte_exit(EXIT_FAILURE,
1684 "Cannot initialize network ports\n");
1687 /* Enable stats if the user option is set. */
1689 ret = rte_ctrl_thread_create(&tid, "print-stats", NULL,
1692 rte_exit(EXIT_FAILURE,
1693 "Cannot create print-stats thread\n");
1696 /* Launch all data cores. */
1697 RTE_LCORE_FOREACH_WORKER(lcore_id)
1698 rte_eal_remote_launch(switch_worker, NULL, lcore_id);
1701 flags |= RTE_VHOST_USER_CLIENT;
1703 /* Register vhost user driver to handle vhost messages. */
1704 for (i = 0; i < nb_sockets; i++) {
1705 char *file = socket_files + i * PATH_MAX;
1707 if (async_vhost_driver)
1708 flags = flags | RTE_VHOST_USER_ASYNC_COPY;
1710 ret = rte_vhost_driver_register(file, flags);
1712 unregister_drivers(i);
1713 rte_exit(EXIT_FAILURE,
1714 "vhost driver register failure.\n");
1717 if (builtin_net_driver)
1718 rte_vhost_driver_set_features(file, VIRTIO_NET_FEATURES);
1720 if (mergeable == 0) {
1721 rte_vhost_driver_disable_features(file,
1722 1ULL << VIRTIO_NET_F_MRG_RXBUF);
1725 if (enable_tx_csum == 0) {
1726 rte_vhost_driver_disable_features(file,
1727 1ULL << VIRTIO_NET_F_CSUM);
1730 if (enable_tso == 0) {
1731 rte_vhost_driver_disable_features(file,
1732 1ULL << VIRTIO_NET_F_HOST_TSO4);
1733 rte_vhost_driver_disable_features(file,
1734 1ULL << VIRTIO_NET_F_HOST_TSO6);
1735 rte_vhost_driver_disable_features(file,
1736 1ULL << VIRTIO_NET_F_GUEST_TSO4);
1737 rte_vhost_driver_disable_features(file,
1738 1ULL << VIRTIO_NET_F_GUEST_TSO6);
1742 rte_vhost_driver_enable_features(file,
1743 1ULL << VIRTIO_NET_F_CTRL_RX);
1746 ret = rte_vhost_driver_callback_register(file,
1747 &virtio_net_device_ops);
1749 rte_exit(EXIT_FAILURE,
1750 "failed to register vhost driver callbacks.\n");
1753 if (rte_vhost_driver_start(file) < 0) {
1754 rte_exit(EXIT_FAILURE,
1755 "failed to start vhost driver.\n");
1759 RTE_LCORE_FOREACH_WORKER(lcore_id)
1760 rte_eal_wait_lcore(lcore_id);