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>
23 #include <rte_vhost.h>
26 #include <rte_pause.h>
32 #define MAX_QUEUES 128
35 /* the maximum number of external ports supported */
36 #define MAX_SUP_PORTS 1
38 #define MBUF_CACHE_SIZE 128
39 #define MBUF_DATA_SIZE RTE_MBUF_DEFAULT_BUF_SIZE
41 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
43 #define BURST_RX_WAIT_US 15 /* Defines how long we wait between retries on RX */
44 #define BURST_RX_RETRIES 4 /* Number of retries on RX. */
46 #define JUMBO_FRAME_MAX_SIZE 0x2600
47 #define MAX_MTU (JUMBO_FRAME_MAX_SIZE - (RTE_ETHER_HDR_LEN + RTE_ETHER_CRC_LEN))
49 /* State of virtio device. */
50 #define DEVICE_MAC_LEARNING 0
52 #define DEVICE_SAFE_REMOVE 2
54 /* Configurable number of RX/TX ring descriptors */
55 #define RTE_TEST_RX_DESC_DEFAULT 1024
56 #define RTE_TEST_TX_DESC_DEFAULT 512
58 #define INVALID_PORT_ID 0xFF
60 /* mask of enabled ports */
61 static uint32_t enabled_port_mask = 0;
63 /* Promiscuous mode */
64 static uint32_t promiscuous;
66 /* number of devices/queues to support*/
67 static uint32_t num_queues = 0;
68 static uint32_t num_devices;
70 static struct rte_mempool *mbuf_pool;
73 /* Enable VM2VM communications. If this is disabled then the MAC address compare is skipped. */
80 static vm2vm_type vm2vm_mode = VM2VM_SOFTWARE;
83 static uint32_t enable_stats = 0;
84 /* Enable retries on RX. */
85 static uint32_t enable_retry = 1;
87 /* Disable TX checksum offload */
88 static uint32_t enable_tx_csum;
90 /* Disable TSO offload */
91 static uint32_t enable_tso;
93 static int client_mode;
95 static int builtin_net_driver;
97 static int async_vhost_driver;
99 static char *dma_type;
101 /* Specify timeout (in useconds) between retries on RX. */
102 static uint32_t burst_rx_delay_time = BURST_RX_WAIT_US;
103 /* Specify the number of retries on RX. */
104 static uint32_t burst_rx_retry_num = BURST_RX_RETRIES;
106 /* Socket file paths. Can be set by user */
107 static char *socket_files;
108 static int nb_sockets;
110 /* empty vmdq configuration structure. Filled in programatically */
111 static struct rte_eth_conf vmdq_conf_default = {
113 .mq_mode = RTE_ETH_MQ_RX_VMDQ_ONLY,
116 * VLAN strip is necessary for 1G NIC such as I350,
117 * this fixes bug of ipv4 forwarding in guest can't
118 * forward pakets from one virtio dev to another virtio dev.
120 .offloads = RTE_ETH_RX_OFFLOAD_VLAN_STRIP,
124 .mq_mode = RTE_ETH_MQ_TX_NONE,
125 .offloads = (RTE_ETH_TX_OFFLOAD_IPV4_CKSUM |
126 RTE_ETH_TX_OFFLOAD_TCP_CKSUM |
127 RTE_ETH_TX_OFFLOAD_VLAN_INSERT |
128 RTE_ETH_TX_OFFLOAD_MULTI_SEGS |
129 RTE_ETH_TX_OFFLOAD_TCP_TSO),
133 * should be overridden separately in code with
137 .nb_queue_pools = RTE_ETH_8_POOLS,
138 .enable_default_pool = 0,
141 .pool_map = {{0, 0},},
147 static unsigned lcore_ids[RTE_MAX_LCORE];
148 static uint16_t ports[RTE_MAX_ETHPORTS];
149 static unsigned num_ports = 0; /**< The number of ports specified in command line */
150 static uint16_t num_pf_queues, num_vmdq_queues;
151 static uint16_t vmdq_pool_base, vmdq_queue_base;
152 static uint16_t queues_per_pool;
154 const uint16_t vlan_tags[] = {
155 1000, 1001, 1002, 1003, 1004, 1005, 1006, 1007,
156 1008, 1009, 1010, 1011, 1012, 1013, 1014, 1015,
157 1016, 1017, 1018, 1019, 1020, 1021, 1022, 1023,
158 1024, 1025, 1026, 1027, 1028, 1029, 1030, 1031,
159 1032, 1033, 1034, 1035, 1036, 1037, 1038, 1039,
160 1040, 1041, 1042, 1043, 1044, 1045, 1046, 1047,
161 1048, 1049, 1050, 1051, 1052, 1053, 1054, 1055,
162 1056, 1057, 1058, 1059, 1060, 1061, 1062, 1063,
165 /* ethernet addresses of ports */
166 static struct rte_ether_addr vmdq_ports_eth_addr[RTE_MAX_ETHPORTS];
168 static struct vhost_dev_tailq_list vhost_dev_list =
169 TAILQ_HEAD_INITIALIZER(vhost_dev_list);
171 static struct lcore_info lcore_info[RTE_MAX_LCORE];
173 /* Used for queueing bursts of TX packets. */
177 struct rte_mbuf *m_table[MAX_PKT_BURST];
180 struct vhost_bufftable {
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];
190 * Vhost TX buffer for each data core.
191 * Every data core maintains a TX buffer for every vhost device,
192 * which is used for batch pkts enqueue for higher performance.
194 struct vhost_bufftable *vhost_txbuff[RTE_MAX_LCORE * MAX_VHOST_DEVICE];
196 #define MBUF_TABLE_DRAIN_TSC ((rte_get_tsc_hz() + US_PER_S - 1) \
197 / US_PER_S * BURST_TX_DRAIN_US)
201 open_dma(const char *value)
203 if (dma_type != NULL && strncmp(dma_type, "ioat", 4) == 0)
204 return open_ioat(value);
210 * Builds up the correct configuration for VMDQ VLAN pool map
211 * according to the pool & queue limits.
214 get_eth_conf(struct rte_eth_conf *eth_conf, uint32_t num_devices)
216 struct rte_eth_vmdq_rx_conf conf;
217 struct rte_eth_vmdq_rx_conf *def_conf =
218 &vmdq_conf_default.rx_adv_conf.vmdq_rx_conf;
221 memset(&conf, 0, sizeof(conf));
222 conf.nb_queue_pools = (enum rte_eth_nb_pools)num_devices;
223 conf.nb_pool_maps = num_devices;
224 conf.enable_loop_back = def_conf->enable_loop_back;
225 conf.rx_mode = def_conf->rx_mode;
227 for (i = 0; i < conf.nb_pool_maps; i++) {
228 conf.pool_map[i].vlan_id = vlan_tags[ i ];
229 conf.pool_map[i].pools = (1UL << i);
232 (void)(rte_memcpy(eth_conf, &vmdq_conf_default, sizeof(*eth_conf)));
233 (void)(rte_memcpy(ð_conf->rx_adv_conf.vmdq_rx_conf, &conf,
234 sizeof(eth_conf->rx_adv_conf.vmdq_rx_conf)));
239 * Initialises a given port using global settings and with the rx buffers
240 * coming from the mbuf_pool passed as parameter
243 port_init(uint16_t port)
245 struct rte_eth_dev_info dev_info;
246 struct rte_eth_conf port_conf;
247 struct rte_eth_rxconf *rxconf;
248 struct rte_eth_txconf *txconf;
249 int16_t rx_rings, tx_rings;
250 uint16_t rx_ring_size, tx_ring_size;
254 /* The max pool number from dev_info will be used to validate the pool number specified in cmd line */
255 retval = rte_eth_dev_info_get(port, &dev_info);
257 RTE_LOG(ERR, VHOST_PORT,
258 "Error during getting device (port %u) info: %s\n",
259 port, strerror(-retval));
264 rxconf = &dev_info.default_rxconf;
265 txconf = &dev_info.default_txconf;
266 rxconf->rx_drop_en = 1;
268 /*configure the number of supported virtio devices based on VMDQ limits */
269 num_devices = dev_info.max_vmdq_pools;
271 rx_ring_size = RTE_TEST_RX_DESC_DEFAULT;
272 tx_ring_size = RTE_TEST_TX_DESC_DEFAULT;
274 tx_rings = (uint16_t)rte_lcore_count();
276 /* Get port configuration. */
277 retval = get_eth_conf(&port_conf, num_devices);
280 /* NIC queues are divided into pf queues and vmdq queues. */
281 num_pf_queues = dev_info.max_rx_queues - dev_info.vmdq_queue_num;
282 queues_per_pool = dev_info.vmdq_queue_num / dev_info.max_vmdq_pools;
283 num_vmdq_queues = num_devices * queues_per_pool;
284 num_queues = num_pf_queues + num_vmdq_queues;
285 vmdq_queue_base = dev_info.vmdq_queue_base;
286 vmdq_pool_base = dev_info.vmdq_pool_base;
287 printf("pf queue num: %u, configured vmdq pool num: %u, each vmdq pool has %u queues\n",
288 num_pf_queues, num_devices, queues_per_pool);
290 if (!rte_eth_dev_is_valid_port(port))
293 rx_rings = (uint16_t)dev_info.max_rx_queues;
294 if (dev_info.tx_offload_capa & RTE_ETH_TX_OFFLOAD_MBUF_FAST_FREE)
295 port_conf.txmode.offloads |=
296 RTE_ETH_TX_OFFLOAD_MBUF_FAST_FREE;
297 /* Configure ethernet device. */
298 retval = rte_eth_dev_configure(port, rx_rings, tx_rings, &port_conf);
300 RTE_LOG(ERR, VHOST_PORT, "Failed to configure port %u: %s.\n",
301 port, strerror(-retval));
305 retval = rte_eth_dev_adjust_nb_rx_tx_desc(port, &rx_ring_size,
308 RTE_LOG(ERR, VHOST_PORT, "Failed to adjust number of descriptors "
309 "for port %u: %s.\n", port, strerror(-retval));
312 if (rx_ring_size > RTE_TEST_RX_DESC_DEFAULT) {
313 RTE_LOG(ERR, VHOST_PORT, "Mbuf pool has an insufficient size "
314 "for Rx queues on port %u.\n", port);
318 /* Setup the queues. */
319 rxconf->offloads = port_conf.rxmode.offloads;
320 for (q = 0; q < rx_rings; q ++) {
321 retval = rte_eth_rx_queue_setup(port, q, rx_ring_size,
322 rte_eth_dev_socket_id(port),
326 RTE_LOG(ERR, VHOST_PORT,
327 "Failed to setup rx queue %u of port %u: %s.\n",
328 q, port, strerror(-retval));
332 txconf->offloads = port_conf.txmode.offloads;
333 for (q = 0; q < tx_rings; q ++) {
334 retval = rte_eth_tx_queue_setup(port, q, tx_ring_size,
335 rte_eth_dev_socket_id(port),
338 RTE_LOG(ERR, VHOST_PORT,
339 "Failed to setup tx queue %u of port %u: %s.\n",
340 q, port, strerror(-retval));
345 /* Start the device. */
346 retval = rte_eth_dev_start(port);
348 RTE_LOG(ERR, VHOST_PORT, "Failed to start port %u: %s\n",
349 port, strerror(-retval));
354 retval = rte_eth_promiscuous_enable(port);
356 RTE_LOG(ERR, VHOST_PORT,
357 "Failed to enable promiscuous mode on port %u: %s\n",
358 port, rte_strerror(-retval));
363 retval = rte_eth_macaddr_get(port, &vmdq_ports_eth_addr[port]);
365 RTE_LOG(ERR, VHOST_PORT,
366 "Failed to get MAC address on port %u: %s\n",
367 port, rte_strerror(-retval));
371 RTE_LOG(INFO, VHOST_PORT, "Max virtio devices supported: %u\n", num_devices);
372 RTE_LOG(INFO, VHOST_PORT, "Port %u MAC: %02"PRIx8" %02"PRIx8" %02"PRIx8
373 " %02"PRIx8" %02"PRIx8" %02"PRIx8"\n",
374 port, RTE_ETHER_ADDR_BYTES(&vmdq_ports_eth_addr[port]));
380 * Set socket file path.
383 us_vhost_parse_socket_path(const char *q_arg)
387 /* parse number string */
388 if (strnlen(q_arg, PATH_MAX) == PATH_MAX)
392 socket_files = realloc(socket_files, PATH_MAX * (nb_sockets + 1));
393 if (socket_files == NULL) {
398 strlcpy(socket_files + nb_sockets * PATH_MAX, q_arg, PATH_MAX);
405 * Parse the portmask provided at run time.
408 parse_portmask(const char *portmask)
415 /* parse hexadecimal string */
416 pm = strtoul(portmask, &end, 16);
417 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0') || (errno != 0))
425 * Parse num options at run time.
428 parse_num_opt(const char *q_arg, uint32_t max_valid_value)
435 /* parse unsigned int string */
436 num = strtoul(q_arg, &end, 10);
437 if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0') || (errno != 0))
440 if (num > max_valid_value)
451 us_vhost_usage(const char *prgname)
453 RTE_LOG(INFO, VHOST_CONFIG, "%s [EAL options] -- -p PORTMASK\n"
455 " --rx_retry [0|1] --mergeable [0|1] --stats [0-N]\n"
456 " --socket-file <path>\n"
458 " -p PORTMASK: Set mask for ports to be used by application\n"
459 " --vm2vm [0|1|2]: disable/software(default)/hardware vm2vm comms\n"
460 " --rx-retry [0|1]: disable/enable(default) retries on rx. Enable retry if destintation queue is full\n"
461 " --rx-retry-delay [0-N]: timeout(in usecond) between retries on RX. This makes effect only if retries on rx enabled\n"
462 " --rx-retry-num [0-N]: the number of retries on rx. This makes effect only if retries on rx enabled\n"
463 " --mergeable [0|1]: disable(default)/enable RX mergeable buffers\n"
464 " --stats [0-N]: 0: Disable stats, N: Time in seconds to print stats\n"
465 " --socket-file: The path of the socket file.\n"
466 " --tx-csum [0|1] disable/enable TX checksum offload.\n"
467 " --tso [0|1] disable/enable TCP segment offload.\n"
468 " --client register a vhost-user socket as client mode.\n"
469 " --dma-type register dma type for your vhost async driver. For example \"ioat\" for now.\n"
470 " --dmas register dma channel for specific vhost device.\n",
475 #define OPT_VM2VM "vm2vm"
477 #define OPT_RX_RETRY "rx-retry"
479 #define OPT_RX_RETRY_DELAY "rx-retry-delay"
480 OPT_RX_RETRY_DELAY_NUM,
481 #define OPT_RX_RETRY_NUMB "rx-retry-num"
482 OPT_RX_RETRY_NUMB_NUM,
483 #define OPT_MERGEABLE "mergeable"
485 #define OPT_STATS "stats"
487 #define OPT_SOCKET_FILE "socket-file"
489 #define OPT_TX_CSUM "tx-csum"
491 #define OPT_TSO "tso"
493 #define OPT_CLIENT "client"
495 #define OPT_BUILTIN_NET_DRIVER "builtin-net-driver"
496 OPT_BUILTIN_NET_DRIVER_NUM,
497 #define OPT_DMA_TYPE "dma-type"
499 #define OPT_DMAS "dmas"
504 * Parse the arguments given in the command line of the application.
507 us_vhost_parse_args(int argc, char **argv)
512 const char *prgname = argv[0];
513 static struct option long_option[] = {
514 {OPT_VM2VM, required_argument,
515 NULL, OPT_VM2VM_NUM},
516 {OPT_RX_RETRY, required_argument,
517 NULL, OPT_RX_RETRY_NUM},
518 {OPT_RX_RETRY_DELAY, required_argument,
519 NULL, OPT_RX_RETRY_DELAY_NUM},
520 {OPT_RX_RETRY_NUMB, required_argument,
521 NULL, OPT_RX_RETRY_NUMB_NUM},
522 {OPT_MERGEABLE, required_argument,
523 NULL, OPT_MERGEABLE_NUM},
524 {OPT_STATS, required_argument,
525 NULL, OPT_STATS_NUM},
526 {OPT_SOCKET_FILE, required_argument,
527 NULL, OPT_SOCKET_FILE_NUM},
528 {OPT_TX_CSUM, required_argument,
529 NULL, OPT_TX_CSUM_NUM},
530 {OPT_TSO, required_argument,
532 {OPT_CLIENT, no_argument,
533 NULL, OPT_CLIENT_NUM},
534 {OPT_BUILTIN_NET_DRIVER, no_argument,
535 NULL, OPT_BUILTIN_NET_DRIVER_NUM},
536 {OPT_DMA_TYPE, required_argument,
537 NULL, OPT_DMA_TYPE_NUM},
538 {OPT_DMAS, required_argument,
543 /* Parse command line */
544 while ((opt = getopt_long(argc, argv, "p:P",
545 long_option, &option_index)) != EOF) {
549 enabled_port_mask = parse_portmask(optarg);
550 if (enabled_port_mask == 0) {
551 RTE_LOG(INFO, VHOST_CONFIG, "Invalid portmask\n");
552 us_vhost_usage(prgname);
559 vmdq_conf_default.rx_adv_conf.vmdq_rx_conf.rx_mode =
560 RTE_ETH_VMDQ_ACCEPT_BROADCAST |
561 RTE_ETH_VMDQ_ACCEPT_MULTICAST;
565 ret = parse_num_opt(optarg, (VM2VM_LAST - 1));
567 RTE_LOG(INFO, VHOST_CONFIG,
568 "Invalid argument for "
570 us_vhost_usage(prgname);
573 vm2vm_mode = (vm2vm_type)ret;
576 case OPT_RX_RETRY_NUM:
577 ret = parse_num_opt(optarg, 1);
579 RTE_LOG(INFO, VHOST_CONFIG, "Invalid argument for rx-retry [0|1]\n");
580 us_vhost_usage(prgname);
586 case OPT_TX_CSUM_NUM:
587 ret = parse_num_opt(optarg, 1);
589 RTE_LOG(INFO, VHOST_CONFIG, "Invalid argument for tx-csum [0|1]\n");
590 us_vhost_usage(prgname);
593 enable_tx_csum = ret;
597 ret = parse_num_opt(optarg, 1);
599 RTE_LOG(INFO, VHOST_CONFIG, "Invalid argument for tso [0|1]\n");
600 us_vhost_usage(prgname);
606 case OPT_RX_RETRY_DELAY_NUM:
607 ret = parse_num_opt(optarg, INT32_MAX);
609 RTE_LOG(INFO, VHOST_CONFIG, "Invalid argument for rx-retry-delay [0-N]\n");
610 us_vhost_usage(prgname);
613 burst_rx_delay_time = ret;
616 case OPT_RX_RETRY_NUMB_NUM:
617 ret = parse_num_opt(optarg, INT32_MAX);
619 RTE_LOG(INFO, VHOST_CONFIG, "Invalid argument for rx-retry-num [0-N]\n");
620 us_vhost_usage(prgname);
623 burst_rx_retry_num = ret;
626 case OPT_MERGEABLE_NUM:
627 ret = parse_num_opt(optarg, 1);
629 RTE_LOG(INFO, VHOST_CONFIG, "Invalid argument for mergeable [0|1]\n");
630 us_vhost_usage(prgname);
635 vmdq_conf_default.rxmode.mtu = MAX_MTU;
639 ret = parse_num_opt(optarg, INT32_MAX);
641 RTE_LOG(INFO, VHOST_CONFIG,
642 "Invalid argument for stats [0..N]\n");
643 us_vhost_usage(prgname);
649 /* Set socket file path. */
650 case OPT_SOCKET_FILE_NUM:
651 if (us_vhost_parse_socket_path(optarg) == -1) {
652 RTE_LOG(INFO, VHOST_CONFIG,
653 "Invalid argument for socket name (Max %d characters)\n",
655 us_vhost_usage(prgname);
660 case OPT_DMA_TYPE_NUM:
665 if (open_dma(optarg) == -1) {
666 RTE_LOG(INFO, VHOST_CONFIG,
668 us_vhost_usage(prgname);
671 async_vhost_driver = 1;
678 case OPT_BUILTIN_NET_DRIVER_NUM:
679 builtin_net_driver = 1;
682 /* Invalid option - print options. */
684 us_vhost_usage(prgname);
689 for (i = 0; i < RTE_MAX_ETHPORTS; i++) {
690 if (enabled_port_mask & (1 << i))
691 ports[num_ports++] = i;
694 if ((num_ports == 0) || (num_ports > MAX_SUP_PORTS)) {
695 RTE_LOG(INFO, VHOST_PORT, "Current enabled port number is %u,"
696 "but only %u port can be enabled\n",num_ports, MAX_SUP_PORTS);
704 * Update the global var NUM_PORTS and array PORTS according to system ports number
705 * and return valid ports number
707 static unsigned check_ports_num(unsigned nb_ports)
709 unsigned valid_num_ports = num_ports;
712 if (num_ports > nb_ports) {
713 RTE_LOG(INFO, VHOST_PORT, "\nSpecified port number(%u) exceeds total system port number(%u)\n",
714 num_ports, nb_ports);
715 num_ports = nb_ports;
718 for (portid = 0; portid < num_ports; portid ++) {
719 if (!rte_eth_dev_is_valid_port(ports[portid])) {
720 RTE_LOG(INFO, VHOST_PORT,
721 "\nSpecified port ID(%u) is not valid\n",
723 ports[portid] = INVALID_PORT_ID;
727 return valid_num_ports;
730 static __rte_always_inline struct vhost_dev *
731 find_vhost_dev(struct rte_ether_addr *mac)
733 struct vhost_dev *vdev;
735 TAILQ_FOREACH(vdev, &vhost_dev_list, global_vdev_entry) {
736 if (vdev->ready == DEVICE_RX &&
737 rte_is_same_ether_addr(mac, &vdev->mac_address))
745 * This function learns the MAC address of the device and registers this along with a
746 * vlan tag to a VMDQ.
749 link_vmdq(struct vhost_dev *vdev, struct rte_mbuf *m)
751 struct rte_ether_hdr *pkt_hdr;
754 /* Learn MAC address of guest device from packet */
755 pkt_hdr = rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
757 if (find_vhost_dev(&pkt_hdr->src_addr)) {
758 RTE_LOG(ERR, VHOST_DATA,
759 "(%d) device is using a registered MAC!\n",
764 for (i = 0; i < RTE_ETHER_ADDR_LEN; i++)
765 vdev->mac_address.addr_bytes[i] =
766 pkt_hdr->src_addr.addr_bytes[i];
768 /* vlan_tag currently uses the device_id. */
769 vdev->vlan_tag = vlan_tags[vdev->vid];
771 /* Print out VMDQ registration info. */
772 RTE_LOG(INFO, VHOST_DATA,
773 "(%d) mac " RTE_ETHER_ADDR_PRT_FMT " and vlan %d registered\n",
774 vdev->vid, RTE_ETHER_ADDR_BYTES(&vdev->mac_address),
777 /* Register the MAC address. */
778 ret = rte_eth_dev_mac_addr_add(ports[0], &vdev->mac_address,
779 (uint32_t)vdev->vid + vmdq_pool_base);
781 RTE_LOG(ERR, VHOST_DATA,
782 "(%d) failed to add device MAC address to VMDQ\n",
785 rte_eth_dev_set_vlan_strip_on_queue(ports[0], vdev->vmdq_rx_q, 1);
787 /* Set device as ready for RX. */
788 vdev->ready = DEVICE_RX;
794 * Removes MAC address and vlan tag from VMDQ. Ensures that nothing is adding buffers to the RX
795 * queue before disabling RX on the device.
798 unlink_vmdq(struct vhost_dev *vdev)
802 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
804 if (vdev->ready == DEVICE_RX) {
805 /*clear MAC and VLAN settings*/
806 rte_eth_dev_mac_addr_remove(ports[0], &vdev->mac_address);
807 for (i = 0; i < 6; i++)
808 vdev->mac_address.addr_bytes[i] = 0;
812 /*Clear out the receive buffers*/
813 rx_count = rte_eth_rx_burst(ports[0],
814 (uint16_t)vdev->vmdq_rx_q, pkts_burst, MAX_PKT_BURST);
817 for (i = 0; i < rx_count; i++)
818 rte_pktmbuf_free(pkts_burst[i]);
820 rx_count = rte_eth_rx_burst(ports[0],
821 (uint16_t)vdev->vmdq_rx_q, pkts_burst, MAX_PKT_BURST);
824 vdev->ready = DEVICE_MAC_LEARNING;
829 free_pkts(struct rte_mbuf **pkts, uint16_t n)
832 rte_pktmbuf_free(pkts[n]);
835 static __rte_always_inline void
836 complete_async_pkts(struct vhost_dev *vdev)
838 struct rte_mbuf *p_cpl[MAX_PKT_BURST];
839 uint16_t complete_count;
841 complete_count = rte_vhost_poll_enqueue_completed(vdev->vid,
842 VIRTIO_RXQ, p_cpl, MAX_PKT_BURST);
843 if (complete_count) {
844 free_pkts(p_cpl, complete_count);
845 __atomic_sub_fetch(&vdev->pkts_inflight, complete_count, __ATOMIC_SEQ_CST);
850 static __rte_always_inline void
851 sync_virtio_xmit(struct vhost_dev *dst_vdev, struct vhost_dev *src_vdev,
856 if (builtin_net_driver) {
857 ret = vs_enqueue_pkts(dst_vdev, VIRTIO_RXQ, &m, 1);
859 ret = rte_vhost_enqueue_burst(dst_vdev->vid, VIRTIO_RXQ, &m, 1);
863 __atomic_add_fetch(&dst_vdev->stats.rx_total_atomic, 1,
865 __atomic_add_fetch(&dst_vdev->stats.rx_atomic, ret,
867 src_vdev->stats.tx_total++;
868 src_vdev->stats.tx += ret;
872 static __rte_always_inline void
873 drain_vhost(struct vhost_dev *vdev)
876 uint32_t buff_idx = rte_lcore_id() * MAX_VHOST_DEVICE + vdev->vid;
877 uint16_t nr_xmit = vhost_txbuff[buff_idx]->len;
878 struct rte_mbuf **m = vhost_txbuff[buff_idx]->m_table;
880 if (builtin_net_driver) {
881 ret = vs_enqueue_pkts(vdev, VIRTIO_RXQ, m, nr_xmit);
882 } else if (async_vhost_driver) {
883 uint16_t enqueue_fail = 0;
885 complete_async_pkts(vdev);
886 ret = rte_vhost_submit_enqueue_burst(vdev->vid, VIRTIO_RXQ, m, nr_xmit);
887 __atomic_add_fetch(&vdev->pkts_inflight, ret, __ATOMIC_SEQ_CST);
889 enqueue_fail = nr_xmit - ret;
891 free_pkts(&m[ret], nr_xmit - ret);
893 ret = rte_vhost_enqueue_burst(vdev->vid, VIRTIO_RXQ,
898 __atomic_add_fetch(&vdev->stats.rx_total_atomic, nr_xmit,
900 __atomic_add_fetch(&vdev->stats.rx_atomic, ret,
904 if (!async_vhost_driver)
905 free_pkts(m, nr_xmit);
908 static __rte_always_inline void
909 drain_vhost_table(void)
911 uint16_t lcore_id = rte_lcore_id();
912 struct vhost_bufftable *vhost_txq;
913 struct vhost_dev *vdev;
916 TAILQ_FOREACH(vdev, &vhost_dev_list, global_vdev_entry) {
917 if (unlikely(vdev->remove == 1))
920 vhost_txq = vhost_txbuff[lcore_id * MAX_VHOST_DEVICE
923 cur_tsc = rte_rdtsc();
924 if (unlikely(cur_tsc - vhost_txq->pre_tsc
925 > MBUF_TABLE_DRAIN_TSC)) {
926 RTE_LOG_DP(DEBUG, VHOST_DATA,
927 "Vhost TX queue drained after timeout with burst size %u\n",
931 vhost_txq->pre_tsc = cur_tsc;
937 * Check if the packet destination MAC address is for a local device. If so then put
938 * the packet on that devices RX queue. If not then return.
940 static __rte_always_inline int
941 virtio_tx_local(struct vhost_dev *vdev, struct rte_mbuf *m)
943 struct rte_ether_hdr *pkt_hdr;
944 struct vhost_dev *dst_vdev;
945 struct vhost_bufftable *vhost_txq;
946 uint16_t lcore_id = rte_lcore_id();
947 pkt_hdr = rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
949 dst_vdev = find_vhost_dev(&pkt_hdr->dst_addr);
953 if (vdev->vid == dst_vdev->vid) {
954 RTE_LOG_DP(DEBUG, VHOST_DATA,
955 "(%d) TX: src and dst MAC is same. Dropping packet.\n",
960 RTE_LOG_DP(DEBUG, VHOST_DATA,
961 "(%d) TX: MAC address is local\n", dst_vdev->vid);
963 if (unlikely(dst_vdev->remove)) {
964 RTE_LOG_DP(DEBUG, VHOST_DATA,
965 "(%d) device is marked for removal\n", dst_vdev->vid);
969 vhost_txq = vhost_txbuff[lcore_id * MAX_VHOST_DEVICE + dst_vdev->vid];
970 vhost_txq->m_table[vhost_txq->len++] = m;
973 vdev->stats.tx_total++;
977 if (unlikely(vhost_txq->len == MAX_PKT_BURST)) {
978 drain_vhost(dst_vdev);
980 vhost_txq->pre_tsc = rte_rdtsc();
986 * Check if the destination MAC of a packet is one local VM,
987 * and get its vlan tag, and offset if it is.
989 static __rte_always_inline int
990 find_local_dest(struct vhost_dev *vdev, struct rte_mbuf *m,
991 uint32_t *offset, uint16_t *vlan_tag)
993 struct vhost_dev *dst_vdev;
994 struct rte_ether_hdr *pkt_hdr =
995 rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
997 dst_vdev = find_vhost_dev(&pkt_hdr->dst_addr);
1001 if (vdev->vid == dst_vdev->vid) {
1002 RTE_LOG_DP(DEBUG, VHOST_DATA,
1003 "(%d) TX: src and dst MAC is same. Dropping packet.\n",
1009 * HW vlan strip will reduce the packet length
1010 * by minus length of vlan tag, so need restore
1011 * the packet length by plus it.
1013 *offset = VLAN_HLEN;
1014 *vlan_tag = vlan_tags[vdev->vid];
1016 RTE_LOG_DP(DEBUG, VHOST_DATA,
1017 "(%d) TX: pkt to local VM device id: (%d), vlan tag: %u.\n",
1018 vdev->vid, dst_vdev->vid, *vlan_tag);
1023 static void virtio_tx_offload(struct rte_mbuf *m)
1025 struct rte_net_hdr_lens hdr_lens;
1026 struct rte_ipv4_hdr *ipv4_hdr;
1027 struct rte_tcp_hdr *tcp_hdr;
1031 ptype = rte_net_get_ptype(m, &hdr_lens, RTE_PTYPE_ALL_MASK);
1032 m->l2_len = hdr_lens.l2_len;
1033 m->l3_len = hdr_lens.l3_len;
1034 m->l4_len = hdr_lens.l4_len;
1036 l3_hdr = rte_pktmbuf_mtod_offset(m, void *, m->l2_len);
1037 tcp_hdr = rte_pktmbuf_mtod_offset(m, struct rte_tcp_hdr *,
1038 m->l2_len + m->l3_len);
1040 m->ol_flags |= RTE_MBUF_F_TX_TCP_SEG;
1041 if ((ptype & RTE_PTYPE_L3_MASK) == RTE_PTYPE_L3_IPV4) {
1042 m->ol_flags |= RTE_MBUF_F_TX_IPV4;
1043 m->ol_flags |= RTE_MBUF_F_TX_IP_CKSUM;
1045 ipv4_hdr->hdr_checksum = 0;
1046 tcp_hdr->cksum = rte_ipv4_phdr_cksum(l3_hdr, m->ol_flags);
1047 } else { /* assume ethertype == RTE_ETHER_TYPE_IPV6 */
1048 m->ol_flags |= RTE_MBUF_F_TX_IPV6;
1049 tcp_hdr->cksum = rte_ipv6_phdr_cksum(l3_hdr, m->ol_flags);
1053 static __rte_always_inline void
1054 do_drain_mbuf_table(struct mbuf_table *tx_q)
1058 count = rte_eth_tx_burst(ports[0], tx_q->txq_id,
1059 tx_q->m_table, tx_q->len);
1060 if (unlikely(count < tx_q->len))
1061 free_pkts(&tx_q->m_table[count], tx_q->len - count);
1067 * This function routes the TX packet to the correct interface. This
1068 * may be a local device or the physical port.
1070 static __rte_always_inline void
1071 virtio_tx_route(struct vhost_dev *vdev, struct rte_mbuf *m, uint16_t vlan_tag)
1073 struct mbuf_table *tx_q;
1074 unsigned offset = 0;
1075 const uint16_t lcore_id = rte_lcore_id();
1076 struct rte_ether_hdr *nh;
1079 nh = rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
1080 if (unlikely(rte_is_broadcast_ether_addr(&nh->dst_addr))) {
1081 struct vhost_dev *vdev2;
1083 TAILQ_FOREACH(vdev2, &vhost_dev_list, global_vdev_entry) {
1085 sync_virtio_xmit(vdev2, vdev, m);
1090 /*check if destination is local VM*/
1091 if ((vm2vm_mode == VM2VM_SOFTWARE) && (virtio_tx_local(vdev, m) == 0))
1094 if (unlikely(vm2vm_mode == VM2VM_HARDWARE)) {
1095 if (unlikely(find_local_dest(vdev, m, &offset,
1097 rte_pktmbuf_free(m);
1102 RTE_LOG_DP(DEBUG, VHOST_DATA,
1103 "(%d) TX: MAC address is external\n", vdev->vid);
1107 /*Add packet to the port tx queue*/
1108 tx_q = &lcore_tx_queue[lcore_id];
1110 nh = rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
1111 if (unlikely(nh->ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_VLAN))) {
1112 /* Guest has inserted the vlan tag. */
1113 struct rte_vlan_hdr *vh = (struct rte_vlan_hdr *) (nh + 1);
1114 uint16_t vlan_tag_be = rte_cpu_to_be_16(vlan_tag);
1115 if ((vm2vm_mode == VM2VM_HARDWARE) &&
1116 (vh->vlan_tci != vlan_tag_be))
1117 vh->vlan_tci = vlan_tag_be;
1119 m->ol_flags |= RTE_MBUF_F_TX_VLAN;
1122 * Find the right seg to adjust the data len when offset is
1123 * bigger than tail room size.
1125 if (unlikely(vm2vm_mode == VM2VM_HARDWARE)) {
1126 if (likely(offset <= rte_pktmbuf_tailroom(m)))
1127 m->data_len += offset;
1129 struct rte_mbuf *seg = m;
1131 while ((seg->next != NULL) &&
1132 (offset > rte_pktmbuf_tailroom(seg)))
1135 seg->data_len += offset;
1137 m->pkt_len += offset;
1140 m->vlan_tci = vlan_tag;
1143 if (m->ol_flags & RTE_MBUF_F_RX_LRO)
1144 virtio_tx_offload(m);
1146 tx_q->m_table[tx_q->len++] = m;
1148 vdev->stats.tx_total++;
1152 if (unlikely(tx_q->len == MAX_PKT_BURST))
1153 do_drain_mbuf_table(tx_q);
1157 static __rte_always_inline void
1158 drain_mbuf_table(struct mbuf_table *tx_q)
1160 static uint64_t prev_tsc;
1166 cur_tsc = rte_rdtsc();
1167 if (unlikely(cur_tsc - prev_tsc > MBUF_TABLE_DRAIN_TSC)) {
1170 RTE_LOG_DP(DEBUG, VHOST_DATA,
1171 "TX queue drained after timeout with burst size %u\n",
1173 do_drain_mbuf_table(tx_q);
1177 static __rte_always_inline void
1178 drain_eth_rx(struct vhost_dev *vdev)
1180 uint16_t rx_count, enqueue_count;
1181 struct rte_mbuf *pkts[MAX_PKT_BURST];
1183 rx_count = rte_eth_rx_burst(ports[0], vdev->vmdq_rx_q,
1184 pkts, MAX_PKT_BURST);
1190 * When "enable_retry" is set, here we wait and retry when there
1191 * is no enough free slots in the queue to hold @rx_count packets,
1192 * to diminish packet loss.
1195 unlikely(rx_count > rte_vhost_avail_entries(vdev->vid,
1199 for (retry = 0; retry < burst_rx_retry_num; retry++) {
1200 rte_delay_us(burst_rx_delay_time);
1201 if (rx_count <= rte_vhost_avail_entries(vdev->vid,
1207 if (builtin_net_driver) {
1208 enqueue_count = vs_enqueue_pkts(vdev, VIRTIO_RXQ,
1210 } else if (async_vhost_driver) {
1211 uint16_t enqueue_fail = 0;
1213 complete_async_pkts(vdev);
1214 enqueue_count = rte_vhost_submit_enqueue_burst(vdev->vid,
1215 VIRTIO_RXQ, pkts, rx_count);
1216 __atomic_add_fetch(&vdev->pkts_inflight, enqueue_count, __ATOMIC_SEQ_CST);
1218 enqueue_fail = rx_count - enqueue_count;
1220 free_pkts(&pkts[enqueue_count], enqueue_fail);
1223 enqueue_count = rte_vhost_enqueue_burst(vdev->vid, VIRTIO_RXQ,
1228 __atomic_add_fetch(&vdev->stats.rx_total_atomic, rx_count,
1230 __atomic_add_fetch(&vdev->stats.rx_atomic, enqueue_count,
1234 if (!async_vhost_driver)
1235 free_pkts(pkts, rx_count);
1238 static __rte_always_inline void
1239 drain_virtio_tx(struct vhost_dev *vdev)
1241 struct rte_mbuf *pkts[MAX_PKT_BURST];
1245 if (builtin_net_driver) {
1246 count = vs_dequeue_pkts(vdev, VIRTIO_TXQ, mbuf_pool,
1247 pkts, MAX_PKT_BURST);
1249 count = rte_vhost_dequeue_burst(vdev->vid, VIRTIO_TXQ,
1250 mbuf_pool, pkts, MAX_PKT_BURST);
1253 /* setup VMDq for the first packet */
1254 if (unlikely(vdev->ready == DEVICE_MAC_LEARNING) && count) {
1255 if (vdev->remove || link_vmdq(vdev, pkts[0]) == -1)
1256 free_pkts(pkts, count);
1259 for (i = 0; i < count; ++i)
1260 virtio_tx_route(vdev, pkts[i], vlan_tags[vdev->vid]);
1264 * Main function of vhost-switch. It basically does:
1266 * for each vhost device {
1269 * Which drains the host eth Rx queue linked to the vhost device,
1270 * and deliver all of them to guest virito Rx ring associated with
1271 * this vhost device.
1273 * - drain_virtio_tx()
1275 * Which drains the guest virtio Tx queue and deliver all of them
1276 * to the target, which could be another vhost device, or the
1277 * physical eth dev. The route is done in function "virtio_tx_route".
1281 switch_worker(void *arg __rte_unused)
1284 unsigned lcore_id = rte_lcore_id();
1285 struct vhost_dev *vdev;
1286 struct mbuf_table *tx_q;
1288 RTE_LOG(INFO, VHOST_DATA, "Procesing on Core %u started\n", lcore_id);
1290 tx_q = &lcore_tx_queue[lcore_id];
1291 for (i = 0; i < rte_lcore_count(); i++) {
1292 if (lcore_ids[i] == lcore_id) {
1299 drain_mbuf_table(tx_q);
1300 drain_vhost_table();
1302 * Inform the configuration core that we have exited the
1303 * linked list and that no devices are in use if requested.
1305 if (lcore_info[lcore_id].dev_removal_flag == REQUEST_DEV_REMOVAL)
1306 lcore_info[lcore_id].dev_removal_flag = ACK_DEV_REMOVAL;
1309 * Process vhost devices
1311 TAILQ_FOREACH(vdev, &lcore_info[lcore_id].vdev_list,
1313 if (unlikely(vdev->remove)) {
1315 vdev->ready = DEVICE_SAFE_REMOVE;
1319 if (likely(vdev->ready == DEVICE_RX))
1322 if (likely(!vdev->remove))
1323 drain_virtio_tx(vdev);
1331 * Remove a device from the specific data core linked list and from the
1332 * main linked list. Synchonization occurs through the use of the
1333 * lcore dev_removal_flag. Device is made volatile here to avoid re-ordering
1334 * of dev->remove=1 which can cause an infinite loop in the rte_pause loop.
1337 destroy_device(int vid)
1339 struct vhost_dev *vdev = NULL;
1343 TAILQ_FOREACH(vdev, &vhost_dev_list, global_vdev_entry) {
1344 if (vdev->vid == vid)
1349 /*set the remove flag. */
1351 while(vdev->ready != DEVICE_SAFE_REMOVE) {
1355 for (i = 0; i < RTE_MAX_LCORE; i++)
1356 rte_free(vhost_txbuff[i * MAX_VHOST_DEVICE + vid]);
1358 if (builtin_net_driver)
1359 vs_vhost_net_remove(vdev);
1361 TAILQ_REMOVE(&lcore_info[vdev->coreid].vdev_list, vdev,
1363 TAILQ_REMOVE(&vhost_dev_list, vdev, global_vdev_entry);
1366 /* Set the dev_removal_flag on each lcore. */
1367 RTE_LCORE_FOREACH_WORKER(lcore)
1368 lcore_info[lcore].dev_removal_flag = REQUEST_DEV_REMOVAL;
1371 * Once each core has set the dev_removal_flag to ACK_DEV_REMOVAL
1372 * we can be sure that they can no longer access the device removed
1373 * from the linked lists and that the devices are no longer in use.
1375 RTE_LCORE_FOREACH_WORKER(lcore) {
1376 while (lcore_info[lcore].dev_removal_flag != ACK_DEV_REMOVAL)
1380 lcore_info[vdev->coreid].device_num--;
1382 RTE_LOG(INFO, VHOST_DATA,
1383 "(%d) device has been removed from data core\n",
1386 if (async_vhost_driver) {
1388 struct rte_mbuf *m_cpl[vdev->pkts_inflight];
1390 while (vdev->pkts_inflight) {
1391 n_pkt = rte_vhost_clear_queue_thread_unsafe(vid, VIRTIO_RXQ,
1392 m_cpl, vdev->pkts_inflight);
1393 free_pkts(m_cpl, n_pkt);
1394 __atomic_sub_fetch(&vdev->pkts_inflight, n_pkt, __ATOMIC_SEQ_CST);
1397 rte_vhost_async_channel_unregister(vid, VIRTIO_RXQ);
1404 * A new device is added to a data core. First the device is added to the main linked list
1405 * and then allocated to a specific data core.
1410 int lcore, core_add = 0;
1412 uint32_t device_num_min = num_devices;
1413 struct vhost_dev *vdev;
1414 vdev = rte_zmalloc("vhost device", sizeof(*vdev), RTE_CACHE_LINE_SIZE);
1416 RTE_LOG(INFO, VHOST_DATA,
1417 "(%d) couldn't allocate memory for vhost dev\n",
1423 for (i = 0; i < RTE_MAX_LCORE; i++) {
1424 vhost_txbuff[i * MAX_VHOST_DEVICE + vid]
1425 = rte_zmalloc("vhost bufftable",
1426 sizeof(struct vhost_bufftable),
1427 RTE_CACHE_LINE_SIZE);
1429 if (vhost_txbuff[i * MAX_VHOST_DEVICE + vid] == NULL) {
1430 RTE_LOG(INFO, VHOST_DATA,
1431 "(%d) couldn't allocate memory for vhost TX\n", vid);
1436 if (builtin_net_driver)
1437 vs_vhost_net_setup(vdev);
1439 TAILQ_INSERT_TAIL(&vhost_dev_list, vdev, global_vdev_entry);
1440 vdev->vmdq_rx_q = vid * queues_per_pool + vmdq_queue_base;
1442 /*reset ready flag*/
1443 vdev->ready = DEVICE_MAC_LEARNING;
1446 /* Find a suitable lcore to add the device. */
1447 RTE_LCORE_FOREACH_WORKER(lcore) {
1448 if (lcore_info[lcore].device_num < device_num_min) {
1449 device_num_min = lcore_info[lcore].device_num;
1453 vdev->coreid = core_add;
1455 TAILQ_INSERT_TAIL(&lcore_info[vdev->coreid].vdev_list, vdev,
1457 lcore_info[vdev->coreid].device_num++;
1459 /* Disable notifications. */
1460 rte_vhost_enable_guest_notification(vid, VIRTIO_RXQ, 0);
1461 rte_vhost_enable_guest_notification(vid, VIRTIO_TXQ, 0);
1463 RTE_LOG(INFO, VHOST_DATA,
1464 "(%d) device has been added to data core %d\n",
1467 if (async_vhost_driver) {
1468 struct rte_vhost_async_config config = {0};
1469 struct rte_vhost_async_channel_ops channel_ops;
1471 if (dma_type != NULL && strncmp(dma_type, "ioat", 4) == 0) {
1472 channel_ops.transfer_data = ioat_transfer_data_cb;
1473 channel_ops.check_completed_copies =
1474 ioat_check_completed_copies_cb;
1476 config.features = RTE_VHOST_ASYNC_INORDER;
1478 return rte_vhost_async_channel_register(vid, VIRTIO_RXQ,
1479 config, &channel_ops);
1487 vring_state_changed(int vid, uint16_t queue_id, int enable)
1489 struct vhost_dev *vdev = NULL;
1491 TAILQ_FOREACH(vdev, &vhost_dev_list, global_vdev_entry) {
1492 if (vdev->vid == vid)
1498 if (queue_id != VIRTIO_RXQ)
1501 if (async_vhost_driver) {
1504 struct rte_mbuf *m_cpl[vdev->pkts_inflight];
1506 while (vdev->pkts_inflight) {
1507 n_pkt = rte_vhost_clear_queue_thread_unsafe(vid, queue_id,
1508 m_cpl, vdev->pkts_inflight);
1509 free_pkts(m_cpl, n_pkt);
1510 __atomic_sub_fetch(&vdev->pkts_inflight, n_pkt, __ATOMIC_SEQ_CST);
1519 * These callback allow devices to be added to the data core when configuration
1520 * has been fully complete.
1522 static const struct vhost_device_ops virtio_net_device_ops =
1524 .new_device = new_device,
1525 .destroy_device = destroy_device,
1526 .vring_state_changed = vring_state_changed,
1530 * This is a thread will wake up after a period to print stats if the user has
1534 print_stats(__rte_unused void *arg)
1536 struct vhost_dev *vdev;
1537 uint64_t tx_dropped, rx_dropped;
1538 uint64_t tx, tx_total, rx, rx_total;
1539 const char clr[] = { 27, '[', '2', 'J', '\0' };
1540 const char top_left[] = { 27, '[', '1', ';', '1', 'H','\0' };
1543 sleep(enable_stats);
1545 /* Clear screen and move to top left */
1546 printf("%s%s\n", clr, top_left);
1547 printf("Device statistics =================================\n");
1549 TAILQ_FOREACH(vdev, &vhost_dev_list, global_vdev_entry) {
1550 tx_total = vdev->stats.tx_total;
1551 tx = vdev->stats.tx;
1552 tx_dropped = tx_total - tx;
1554 rx_total = __atomic_load_n(&vdev->stats.rx_total_atomic,
1556 rx = __atomic_load_n(&vdev->stats.rx_atomic,
1558 rx_dropped = rx_total - rx;
1560 printf("Statistics for device %d\n"
1561 "-----------------------\n"
1562 "TX total: %" PRIu64 "\n"
1563 "TX dropped: %" PRIu64 "\n"
1564 "TX successful: %" PRIu64 "\n"
1565 "RX total: %" PRIu64 "\n"
1566 "RX dropped: %" PRIu64 "\n"
1567 "RX successful: %" PRIu64 "\n",
1569 tx_total, tx_dropped, tx,
1570 rx_total, rx_dropped, rx);
1573 printf("===================================================\n");
1582 unregister_drivers(int socket_num)
1586 for (i = 0; i < socket_num; i++) {
1587 ret = rte_vhost_driver_unregister(socket_files + i * PATH_MAX);
1589 RTE_LOG(ERR, VHOST_CONFIG,
1590 "Fail to unregister vhost driver for %s.\n",
1591 socket_files + i * PATH_MAX);
1595 /* When we receive a INT signal, unregister vhost driver */
1597 sigint_handler(__rte_unused int signum)
1599 /* Unregister vhost driver. */
1600 unregister_drivers(nb_sockets);
1606 * While creating an mbuf pool, one key thing is to figure out how
1607 * many mbuf entries is enough for our use. FYI, here are some
1610 * - Each rx queue would reserve @nr_rx_desc mbufs at queue setup stage
1612 * - For each switch core (A CPU core does the packet switch), we need
1613 * also make some reservation for receiving the packets from virtio
1614 * Tx queue. How many is enough depends on the usage. It's normally
1615 * a simple calculation like following:
1617 * MAX_PKT_BURST * max packet size / mbuf size
1619 * So, we definitely need allocate more mbufs when TSO is enabled.
1621 * - Similarly, for each switching core, we should serve @nr_rx_desc
1622 * mbufs for receiving the packets from physical NIC device.
1624 * - We also need make sure, for each switch core, we have allocated
1625 * enough mbufs to fill up the mbuf cache.
1628 create_mbuf_pool(uint16_t nr_port, uint32_t nr_switch_core, uint32_t mbuf_size,
1629 uint32_t nr_queues, uint32_t nr_rx_desc, uint32_t nr_mbuf_cache)
1632 uint32_t nr_mbufs_per_core;
1633 uint32_t mtu = 1500;
1640 nr_mbufs_per_core = (mtu + mbuf_size) * MAX_PKT_BURST /
1641 (mbuf_size - RTE_PKTMBUF_HEADROOM);
1642 nr_mbufs_per_core += nr_rx_desc;
1643 nr_mbufs_per_core = RTE_MAX(nr_mbufs_per_core, nr_mbuf_cache);
1645 nr_mbufs = nr_queues * nr_rx_desc;
1646 nr_mbufs += nr_mbufs_per_core * nr_switch_core;
1647 nr_mbufs *= nr_port;
1649 mbuf_pool = rte_pktmbuf_pool_create("MBUF_POOL", nr_mbufs,
1650 nr_mbuf_cache, 0, mbuf_size,
1652 if (mbuf_pool == NULL)
1653 rte_exit(EXIT_FAILURE, "Cannot create mbuf pool\n");
1657 * Main function, does initialisation and calls the per-lcore functions.
1660 main(int argc, char *argv[])
1662 unsigned lcore_id, core_id = 0;
1663 unsigned nb_ports, valid_num_ports;
1666 static pthread_t tid;
1667 uint64_t flags = RTE_VHOST_USER_NET_COMPLIANT_OL_FLAGS;
1669 signal(SIGINT, sigint_handler);
1672 ret = rte_eal_init(argc, argv);
1674 rte_exit(EXIT_FAILURE, "Error with EAL initialization\n");
1678 /* parse app arguments */
1679 ret = us_vhost_parse_args(argc, argv);
1681 rte_exit(EXIT_FAILURE, "Invalid argument\n");
1683 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1684 TAILQ_INIT(&lcore_info[lcore_id].vdev_list);
1686 if (rte_lcore_is_enabled(lcore_id))
1687 lcore_ids[core_id++] = lcore_id;
1690 if (rte_lcore_count() > RTE_MAX_LCORE)
1691 rte_exit(EXIT_FAILURE,"Not enough cores\n");
1693 /* Get the number of physical ports. */
1694 nb_ports = rte_eth_dev_count_avail();
1697 * Update the global var NUM_PORTS and global array PORTS
1698 * and get value of var VALID_NUM_PORTS according to system ports number
1700 valid_num_ports = check_ports_num(nb_ports);
1702 if ((valid_num_ports == 0) || (valid_num_ports > MAX_SUP_PORTS)) {
1703 RTE_LOG(INFO, VHOST_PORT, "Current enabled port number is %u,"
1704 "but only %u port can be enabled\n",num_ports, MAX_SUP_PORTS);
1709 * FIXME: here we are trying to allocate mbufs big enough for
1710 * @MAX_QUEUES, but the truth is we're never going to use that
1711 * many queues here. We probably should only do allocation for
1712 * those queues we are going to use.
1714 create_mbuf_pool(valid_num_ports, rte_lcore_count() - 1, MBUF_DATA_SIZE,
1715 MAX_QUEUES, RTE_TEST_RX_DESC_DEFAULT, MBUF_CACHE_SIZE);
1717 if (vm2vm_mode == VM2VM_HARDWARE) {
1718 /* Enable VT loop back to let L2 switch to do it. */
1719 vmdq_conf_default.rx_adv_conf.vmdq_rx_conf.enable_loop_back = 1;
1720 RTE_LOG(DEBUG, VHOST_CONFIG,
1721 "Enable loop back for L2 switch in vmdq.\n");
1724 /* initialize all ports */
1725 RTE_ETH_FOREACH_DEV(portid) {
1726 /* skip ports that are not enabled */
1727 if ((enabled_port_mask & (1 << portid)) == 0) {
1728 RTE_LOG(INFO, VHOST_PORT,
1729 "Skipping disabled port %d\n", portid);
1732 if (port_init(portid) != 0)
1733 rte_exit(EXIT_FAILURE,
1734 "Cannot initialize network ports\n");
1737 /* Enable stats if the user option is set. */
1739 ret = rte_ctrl_thread_create(&tid, "print-stats", NULL,
1742 rte_exit(EXIT_FAILURE,
1743 "Cannot create print-stats thread\n");
1746 /* Launch all data cores. */
1747 RTE_LCORE_FOREACH_WORKER(lcore_id)
1748 rte_eal_remote_launch(switch_worker, NULL, lcore_id);
1751 flags |= RTE_VHOST_USER_CLIENT;
1753 /* Register vhost user driver to handle vhost messages. */
1754 for (i = 0; i < nb_sockets; i++) {
1755 char *file = socket_files + i * PATH_MAX;
1757 if (async_vhost_driver)
1758 flags = flags | RTE_VHOST_USER_ASYNC_COPY;
1760 ret = rte_vhost_driver_register(file, flags);
1762 unregister_drivers(i);
1763 rte_exit(EXIT_FAILURE,
1764 "vhost driver register failure.\n");
1767 if (builtin_net_driver)
1768 rte_vhost_driver_set_features(file, VIRTIO_NET_FEATURES);
1770 if (mergeable == 0) {
1771 rte_vhost_driver_disable_features(file,
1772 1ULL << VIRTIO_NET_F_MRG_RXBUF);
1775 if (enable_tx_csum == 0) {
1776 rte_vhost_driver_disable_features(file,
1777 1ULL << VIRTIO_NET_F_CSUM);
1780 if (enable_tso == 0) {
1781 rte_vhost_driver_disable_features(file,
1782 1ULL << VIRTIO_NET_F_HOST_TSO4);
1783 rte_vhost_driver_disable_features(file,
1784 1ULL << VIRTIO_NET_F_HOST_TSO6);
1785 rte_vhost_driver_disable_features(file,
1786 1ULL << VIRTIO_NET_F_GUEST_TSO4);
1787 rte_vhost_driver_disable_features(file,
1788 1ULL << VIRTIO_NET_F_GUEST_TSO6);
1792 rte_vhost_driver_enable_features(file,
1793 1ULL << VIRTIO_NET_F_CTRL_RX);
1796 ret = rte_vhost_driver_callback_register(file,
1797 &virtio_net_device_ops);
1799 rte_exit(EXIT_FAILURE,
1800 "failed to register vhost driver callbacks.\n");
1803 if (rte_vhost_driver_start(file) < 0) {
1804 rte_exit(EXIT_FAILURE,
1805 "failed to start vhost driver.\n");
1809 RTE_LCORE_FOREACH_WORKER(lcore_id)
1810 rte_eal_wait_lcore(lcore_id);
1812 /* clean up the EAL */