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
66 * Calculate the number of buffers needed per port
68 #define NUM_MBUFS_PER_PORT ((MAX_QUEUES*RTE_TEST_RX_DESC_DEFAULT) + \
69 (num_switching_cores*MAX_PKT_BURST) + \
70 (num_switching_cores*RTE_TEST_TX_DESC_DEFAULT) +\
71 ((num_switching_cores+1)*MBUF_CACHE_SIZE))
73 #define MBUF_CACHE_SIZE 128
74 #define MBUF_DATA_SIZE RTE_MBUF_DEFAULT_BUF_SIZE
76 #define MAX_PKT_BURST 32 /* Max burst size for RX/TX */
77 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
79 #define BURST_RX_WAIT_US 15 /* Defines how long we wait between retries on RX */
80 #define BURST_RX_RETRIES 4 /* Number of retries on RX. */
82 #define JUMBO_FRAME_MAX_SIZE 0x2600
84 /* State of virtio device. */
85 #define DEVICE_MAC_LEARNING 0
87 #define DEVICE_SAFE_REMOVE 2
89 /* Configurable number of RX/TX ring descriptors */
90 #define RTE_TEST_RX_DESC_DEFAULT 1024
91 #define RTE_TEST_TX_DESC_DEFAULT 512
93 #define INVALID_PORT_ID 0xFF
95 /* Max number of devices. Limited by vmdq. */
96 #define MAX_DEVICES 64
98 /* Size of buffers used for snprintfs. */
99 #define MAX_PRINT_BUFF 6072
101 /* Maximum character device basename size. */
102 #define MAX_BASENAME_SZ 10
104 /* Maximum long option length for option parsing. */
105 #define MAX_LONG_OPT_SZ 64
107 /* mask of enabled ports */
108 static uint32_t enabled_port_mask = 0;
110 /* Promiscuous mode */
111 static uint32_t promiscuous;
113 /*Number of switching cores enabled*/
114 static uint32_t num_switching_cores = 0;
116 /* number of devices/queues to support*/
117 static uint32_t num_queues = 0;
118 static uint32_t num_devices;
120 static struct rte_mempool *mbuf_pool;
121 static int mergeable;
123 /* Do vlan strip on host, enabled on default */
124 static uint32_t vlan_strip = 1;
126 /* Enable VM2VM communications. If this is disabled then the MAC address compare is skipped. */
133 static vm2vm_type vm2vm_mode = VM2VM_SOFTWARE;
136 static uint32_t enable_stats = 0;
137 /* Enable retries on RX. */
138 static uint32_t enable_retry = 1;
140 /* Disable TX checksum offload */
141 static uint32_t enable_tx_csum;
143 /* Disable TSO offload */
144 static uint32_t enable_tso;
146 /* Specify timeout (in useconds) between retries on RX. */
147 static uint32_t burst_rx_delay_time = BURST_RX_WAIT_US;
148 /* Specify the number of retries on RX. */
149 static uint32_t burst_rx_retry_num = BURST_RX_RETRIES;
151 /* Character device basename. Can be set by user. */
152 static char dev_basename[MAX_BASENAME_SZ] = "vhost-net";
154 /* empty vmdq configuration structure. Filled in programatically */
155 static struct rte_eth_conf vmdq_conf_default = {
157 .mq_mode = ETH_MQ_RX_VMDQ_ONLY,
159 .header_split = 0, /**< Header Split disabled */
160 .hw_ip_checksum = 0, /**< IP checksum offload disabled */
161 .hw_vlan_filter = 0, /**< VLAN filtering disabled */
163 * It is necessary for 1G NIC such as I350,
164 * this fixes bug of ipv4 forwarding in guest can't
165 * forward pakets from one virtio dev to another virtio dev.
167 .hw_vlan_strip = 1, /**< VLAN strip enabled. */
168 .jumbo_frame = 0, /**< Jumbo Frame Support disabled */
169 .hw_strip_crc = 0, /**< CRC stripped by hardware */
173 .mq_mode = ETH_MQ_TX_NONE,
177 * should be overridden separately in code with
181 .nb_queue_pools = ETH_8_POOLS,
182 .enable_default_pool = 0,
185 .pool_map = {{0, 0},},
190 static unsigned lcore_ids[RTE_MAX_LCORE];
191 static uint8_t ports[RTE_MAX_ETHPORTS];
192 static unsigned num_ports = 0; /**< The number of ports specified in command line */
193 static uint16_t num_pf_queues, num_vmdq_queues;
194 static uint16_t vmdq_pool_base, vmdq_queue_base;
195 static uint16_t queues_per_pool;
197 const uint16_t vlan_tags[] = {
198 1000, 1001, 1002, 1003, 1004, 1005, 1006, 1007,
199 1008, 1009, 1010, 1011, 1012, 1013, 1014, 1015,
200 1016, 1017, 1018, 1019, 1020, 1021, 1022, 1023,
201 1024, 1025, 1026, 1027, 1028, 1029, 1030, 1031,
202 1032, 1033, 1034, 1035, 1036, 1037, 1038, 1039,
203 1040, 1041, 1042, 1043, 1044, 1045, 1046, 1047,
204 1048, 1049, 1050, 1051, 1052, 1053, 1054, 1055,
205 1056, 1057, 1058, 1059, 1060, 1061, 1062, 1063,
208 /* ethernet addresses of ports */
209 static struct ether_addr vmdq_ports_eth_addr[RTE_MAX_ETHPORTS];
211 static struct vhost_dev_tailq_list vhost_dev_list =
212 TAILQ_HEAD_INITIALIZER(vhost_dev_list);
214 static struct lcore_info lcore_info[RTE_MAX_LCORE];
216 /* Used for queueing bursts of TX packets. */
220 struct rte_mbuf *m_table[MAX_PKT_BURST];
223 /* TX queue for each data core. */
224 struct mbuf_table lcore_tx_queue[RTE_MAX_LCORE];
228 /* Per-device statistics struct */
229 struct device_statistics {
231 rte_atomic64_t rx_total_atomic;
233 rte_atomic64_t rx_atomic;
234 } __rte_cache_aligned;
235 struct device_statistics dev_statistics[MAX_DEVICES];
238 * Builds up the correct configuration for VMDQ VLAN pool map
239 * according to the pool & queue limits.
242 get_eth_conf(struct rte_eth_conf *eth_conf, uint32_t num_devices)
244 struct rte_eth_vmdq_rx_conf conf;
245 struct rte_eth_vmdq_rx_conf *def_conf =
246 &vmdq_conf_default.rx_adv_conf.vmdq_rx_conf;
249 memset(&conf, 0, sizeof(conf));
250 conf.nb_queue_pools = (enum rte_eth_nb_pools)num_devices;
251 conf.nb_pool_maps = num_devices;
252 conf.enable_loop_back = def_conf->enable_loop_back;
253 conf.rx_mode = def_conf->rx_mode;
255 for (i = 0; i < conf.nb_pool_maps; i++) {
256 conf.pool_map[i].vlan_id = vlan_tags[ i ];
257 conf.pool_map[i].pools = (1UL << i);
260 (void)(rte_memcpy(eth_conf, &vmdq_conf_default, sizeof(*eth_conf)));
261 (void)(rte_memcpy(ð_conf->rx_adv_conf.vmdq_rx_conf, &conf,
262 sizeof(eth_conf->rx_adv_conf.vmdq_rx_conf)));
267 * Validate the device number according to the max pool number gotten form
268 * dev_info. If the device number is invalid, give the error message and
269 * return -1. Each device must have its own pool.
272 validate_num_devices(uint32_t max_nb_devices)
274 if (num_devices > max_nb_devices) {
275 RTE_LOG(ERR, VHOST_PORT, "invalid number of devices\n");
282 * Initialises a given port using global settings and with the rx buffers
283 * coming from the mbuf_pool passed as parameter
286 port_init(uint8_t port)
288 struct rte_eth_dev_info dev_info;
289 struct rte_eth_conf port_conf;
290 struct rte_eth_rxconf *rxconf;
291 struct rte_eth_txconf *txconf;
292 int16_t rx_rings, tx_rings;
293 uint16_t rx_ring_size, tx_ring_size;
297 /* The max pool number from dev_info will be used to validate the pool number specified in cmd line */
298 rte_eth_dev_info_get (port, &dev_info);
300 if (dev_info.max_rx_queues > MAX_QUEUES) {
301 rte_exit(EXIT_FAILURE,
302 "please define MAX_QUEUES no less than %u in %s\n",
303 dev_info.max_rx_queues, __FILE__);
306 rxconf = &dev_info.default_rxconf;
307 txconf = &dev_info.default_txconf;
308 rxconf->rx_drop_en = 1;
310 /* Enable vlan offload */
311 txconf->txq_flags &= ~ETH_TXQ_FLAGS_NOVLANOFFL;
313 /*configure the number of supported virtio devices based on VMDQ limits */
314 num_devices = dev_info.max_vmdq_pools;
316 rx_ring_size = RTE_TEST_RX_DESC_DEFAULT;
317 tx_ring_size = RTE_TEST_TX_DESC_DEFAULT;
318 tx_rings = (uint16_t)rte_lcore_count();
320 retval = validate_num_devices(MAX_DEVICES);
324 /* Get port configuration. */
325 retval = get_eth_conf(&port_conf, num_devices);
328 /* NIC queues are divided into pf queues and vmdq queues. */
329 num_pf_queues = dev_info.max_rx_queues - dev_info.vmdq_queue_num;
330 queues_per_pool = dev_info.vmdq_queue_num / dev_info.max_vmdq_pools;
331 num_vmdq_queues = num_devices * queues_per_pool;
332 num_queues = num_pf_queues + num_vmdq_queues;
333 vmdq_queue_base = dev_info.vmdq_queue_base;
334 vmdq_pool_base = dev_info.vmdq_pool_base;
335 printf("pf queue num: %u, configured vmdq pool num: %u, each vmdq pool has %u queues\n",
336 num_pf_queues, num_devices, queues_per_pool);
338 if (port >= rte_eth_dev_count()) return -1;
340 if (enable_tx_csum == 0)
341 rte_vhost_feature_disable(1ULL << VIRTIO_NET_F_CSUM);
343 if (enable_tso == 0) {
344 rte_vhost_feature_disable(1ULL << VIRTIO_NET_F_HOST_TSO4);
345 rte_vhost_feature_disable(1ULL << VIRTIO_NET_F_HOST_TSO6);
348 rx_rings = (uint16_t)dev_info.max_rx_queues;
349 /* Configure ethernet device. */
350 retval = rte_eth_dev_configure(port, rx_rings, tx_rings, &port_conf);
354 /* Setup the queues. */
355 for (q = 0; q < rx_rings; q ++) {
356 retval = rte_eth_rx_queue_setup(port, q, rx_ring_size,
357 rte_eth_dev_socket_id(port),
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),
371 /* Start the device. */
372 retval = rte_eth_dev_start(port);
374 RTE_LOG(ERR, VHOST_DATA, "Failed to start the device.\n");
379 rte_eth_promiscuous_enable(port);
381 rte_eth_macaddr_get(port, &vmdq_ports_eth_addr[port]);
382 RTE_LOG(INFO, VHOST_PORT, "Max virtio devices supported: %u\n", num_devices);
383 RTE_LOG(INFO, VHOST_PORT, "Port %u MAC: %02"PRIx8" %02"PRIx8" %02"PRIx8
384 " %02"PRIx8" %02"PRIx8" %02"PRIx8"\n",
386 vmdq_ports_eth_addr[port].addr_bytes[0],
387 vmdq_ports_eth_addr[port].addr_bytes[1],
388 vmdq_ports_eth_addr[port].addr_bytes[2],
389 vmdq_ports_eth_addr[port].addr_bytes[3],
390 vmdq_ports_eth_addr[port].addr_bytes[4],
391 vmdq_ports_eth_addr[port].addr_bytes[5]);
397 * Set character device basename.
400 us_vhost_parse_basename(const char *q_arg)
402 /* parse number string */
404 if (strnlen(q_arg, MAX_BASENAME_SZ) > MAX_BASENAME_SZ)
407 snprintf((char*)&dev_basename, MAX_BASENAME_SZ, "%s", q_arg);
413 * Parse the portmask provided at run time.
416 parse_portmask(const char *portmask)
423 /* parse hexadecimal string */
424 pm = strtoul(portmask, &end, 16);
425 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0') || (errno != 0))
436 * Parse num options at run time.
439 parse_num_opt(const char *q_arg, uint32_t max_valid_value)
446 /* parse unsigned int string */
447 num = strtoul(q_arg, &end, 10);
448 if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0') || (errno != 0))
451 if (num > max_valid_value)
462 us_vhost_usage(const char *prgname)
464 RTE_LOG(INFO, VHOST_CONFIG, "%s [EAL options] -- -p PORTMASK\n"
466 " --rx_retry [0|1] --mergeable [0|1] --stats [0-N]\n"
467 " --dev-basename <name>\n"
469 " -p PORTMASK: Set mask for ports to be used by application\n"
470 " --vm2vm [0|1|2]: disable/software(default)/hardware vm2vm comms\n"
471 " --rx-retry [0|1]: disable/enable(default) retries on rx. Enable retry if destintation queue is full\n"
472 " --rx-retry-delay [0-N]: timeout(in usecond) between retries on RX. This makes effect only if retries on rx enabled\n"
473 " --rx-retry-num [0-N]: the number of retries on rx. This makes effect only if retries on rx enabled\n"
474 " --mergeable [0|1]: disable(default)/enable RX mergeable buffers\n"
475 " --vlan-strip [0|1]: disable/enable(default) RX VLAN strip on host\n"
476 " --stats [0-N]: 0: Disable stats, N: Time in seconds to print stats\n"
477 " --dev-basename: The basename to be used for the character device.\n"
478 " --tx-csum [0|1] disable/enable TX checksum offload.\n"
479 " --tso [0|1] disable/enable TCP segment offload.\n",
484 * Parse the arguments given in the command line of the application.
487 us_vhost_parse_args(int argc, char **argv)
492 const char *prgname = argv[0];
493 static struct option long_option[] = {
494 {"vm2vm", required_argument, NULL, 0},
495 {"rx-retry", required_argument, NULL, 0},
496 {"rx-retry-delay", required_argument, NULL, 0},
497 {"rx-retry-num", required_argument, NULL, 0},
498 {"mergeable", required_argument, NULL, 0},
499 {"vlan-strip", required_argument, NULL, 0},
500 {"stats", required_argument, NULL, 0},
501 {"dev-basename", required_argument, NULL, 0},
502 {"tx-csum", required_argument, NULL, 0},
503 {"tso", required_argument, NULL, 0},
507 /* Parse command line */
508 while ((opt = getopt_long(argc, argv, "p:P",
509 long_option, &option_index)) != EOF) {
513 enabled_port_mask = parse_portmask(optarg);
514 if (enabled_port_mask == 0) {
515 RTE_LOG(INFO, VHOST_CONFIG, "Invalid portmask\n");
516 us_vhost_usage(prgname);
523 vmdq_conf_default.rx_adv_conf.vmdq_rx_conf.rx_mode =
524 ETH_VMDQ_ACCEPT_BROADCAST |
525 ETH_VMDQ_ACCEPT_MULTICAST;
526 rte_vhost_feature_enable(1ULL << VIRTIO_NET_F_CTRL_RX);
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.jumbo_frame = 1;
615 vmdq_conf_default.rxmode.max_rx_pkt_len
616 = JUMBO_FRAME_MAX_SIZE;
621 /* Enable/disable RX VLAN strip on host. */
622 if (!strncmp(long_option[option_index].name,
623 "vlan-strip", MAX_LONG_OPT_SZ)) {
624 ret = parse_num_opt(optarg, 1);
626 RTE_LOG(INFO, VHOST_CONFIG,
627 "Invalid argument for VLAN strip [0|1]\n");
628 us_vhost_usage(prgname);
632 vmdq_conf_default.rxmode.hw_vlan_strip =
637 /* Enable/disable stats. */
638 if (!strncmp(long_option[option_index].name, "stats", MAX_LONG_OPT_SZ)) {
639 ret = parse_num_opt(optarg, INT32_MAX);
641 RTE_LOG(INFO, VHOST_CONFIG, "Invalid argument for stats [0..N]\n");
642 us_vhost_usage(prgname);
649 /* Set character device basename. */
650 if (!strncmp(long_option[option_index].name, "dev-basename", MAX_LONG_OPT_SZ)) {
651 if (us_vhost_parse_basename(optarg) == -1) {
652 RTE_LOG(INFO, VHOST_CONFIG, "Invalid argument for character device basename (Max %d characters)\n", MAX_BASENAME_SZ);
653 us_vhost_usage(prgname);
660 /* Invalid option - print options. */
662 us_vhost_usage(prgname);
667 for (i = 0; i < RTE_MAX_ETHPORTS; i++) {
668 if (enabled_port_mask & (1 << i))
669 ports[num_ports++] = (uint8_t)i;
672 if ((num_ports == 0) || (num_ports > MAX_SUP_PORTS)) {
673 RTE_LOG(INFO, VHOST_PORT, "Current enabled port number is %u,"
674 "but only %u port can be enabled\n",num_ports, MAX_SUP_PORTS);
682 * Update the global var NUM_PORTS and array PORTS according to system ports number
683 * and return valid ports number
685 static unsigned check_ports_num(unsigned nb_ports)
687 unsigned valid_num_ports = num_ports;
690 if (num_ports > nb_ports) {
691 RTE_LOG(INFO, VHOST_PORT, "\nSpecified port number(%u) exceeds total system port number(%u)\n",
692 num_ports, nb_ports);
693 num_ports = nb_ports;
696 for (portid = 0; portid < num_ports; portid ++) {
697 if (ports[portid] >= nb_ports) {
698 RTE_LOG(INFO, VHOST_PORT, "\nSpecified port ID(%u) exceeds max system port ID(%u)\n",
699 ports[portid], (nb_ports - 1));
700 ports[portid] = INVALID_PORT_ID;
704 return valid_num_ports;
707 static inline struct vhost_dev *__attribute__((always_inline))
708 find_vhost_dev(struct ether_addr *mac)
710 struct vhost_dev *vdev;
712 TAILQ_FOREACH(vdev, &vhost_dev_list, next) {
713 if (vdev->ready == DEVICE_RX &&
714 is_same_ether_addr(mac, &vdev->mac_address))
722 * This function learns the MAC address of the device and registers this along with a
723 * vlan tag to a VMDQ.
726 link_vmdq(struct vhost_dev *vdev, struct rte_mbuf *m)
728 struct ether_hdr *pkt_hdr;
729 struct virtio_net *dev = vdev->dev;
732 /* Learn MAC address of guest device from packet */
733 pkt_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
735 if (find_vhost_dev(&pkt_hdr->s_addr)) {
736 RTE_LOG(ERR, VHOST_DATA,
737 "Device (%" PRIu64 ") is using a registered MAC!\n",
742 for (i = 0; i < ETHER_ADDR_LEN; i++)
743 vdev->mac_address.addr_bytes[i] = pkt_hdr->s_addr.addr_bytes[i];
745 /* vlan_tag currently uses the device_id. */
746 vdev->vlan_tag = vlan_tags[dev->device_fh];
748 /* Print out VMDQ registration info. */
749 RTE_LOG(INFO, VHOST_DATA, "(%"PRIu64") MAC_ADDRESS %02x:%02x:%02x:%02x:%02x:%02x and VLAN_TAG %d registered\n",
751 vdev->mac_address.addr_bytes[0], vdev->mac_address.addr_bytes[1],
752 vdev->mac_address.addr_bytes[2], vdev->mac_address.addr_bytes[3],
753 vdev->mac_address.addr_bytes[4], vdev->mac_address.addr_bytes[5],
756 /* Register the MAC address. */
757 ret = rte_eth_dev_mac_addr_add(ports[0], &vdev->mac_address,
758 (uint32_t)dev->device_fh + vmdq_pool_base);
760 RTE_LOG(ERR, VHOST_DATA, "(%"PRIu64") Failed to add device MAC address to VMDQ\n",
763 /* Enable stripping of the vlan tag as we handle routing. */
765 rte_eth_dev_set_vlan_strip_on_queue(ports[0],
766 (uint16_t)vdev->vmdq_rx_q, 1);
768 /* Set device as ready for RX. */
769 vdev->ready = DEVICE_RX;
775 * Removes MAC address and vlan tag from VMDQ. Ensures that nothing is adding buffers to the RX
776 * queue before disabling RX on the device.
779 unlink_vmdq(struct vhost_dev *vdev)
783 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
785 if (vdev->ready == DEVICE_RX) {
786 /*clear MAC and VLAN settings*/
787 rte_eth_dev_mac_addr_remove(ports[0], &vdev->mac_address);
788 for (i = 0; i < 6; i++)
789 vdev->mac_address.addr_bytes[i] = 0;
793 /*Clear out the receive buffers*/
794 rx_count = rte_eth_rx_burst(ports[0],
795 (uint16_t)vdev->vmdq_rx_q, pkts_burst, MAX_PKT_BURST);
798 for (i = 0; i < rx_count; i++)
799 rte_pktmbuf_free(pkts_burst[i]);
801 rx_count = rte_eth_rx_burst(ports[0],
802 (uint16_t)vdev->vmdq_rx_q, pkts_burst, MAX_PKT_BURST);
805 vdev->ready = DEVICE_MAC_LEARNING;
809 static inline void __attribute__((always_inline))
810 virtio_xmit(struct virtio_net *dst_dev, struct virtio_net *src_dev,
815 ret = rte_vhost_enqueue_burst(dst_dev, VIRTIO_RXQ, &m, 1);
817 rte_atomic64_inc(&dev_statistics[dst_dev->device_fh].rx_total_atomic);
818 rte_atomic64_add(&dev_statistics[dst_dev->device_fh].rx_atomic, ret);
819 dev_statistics[src_dev->device_fh].tx_total++;
820 dev_statistics[src_dev->device_fh].tx += ret;
825 * Check if the packet destination MAC address is for a local device. If so then put
826 * the packet on that devices RX queue. If not then return.
828 static inline int __attribute__((always_inline))
829 virtio_tx_local(struct vhost_dev *vdev, struct rte_mbuf *m)
831 struct ether_hdr *pkt_hdr;
832 struct vhost_dev *dst_vdev;
835 pkt_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
837 dst_vdev = find_vhost_dev(&pkt_hdr->d_addr);
841 fh = dst_vdev->dev->device_fh;
842 if (fh == vdev->dev->device_fh) {
843 RTE_LOG(DEBUG, VHOST_DATA,
844 "(%" PRIu64 ") TX: src and dst MAC is same. "
845 "Dropping packet.\n", fh);
849 RTE_LOG(DEBUG, VHOST_DATA,
850 "(%" PRIu64 ") TX: MAC address is local\n", fh);
852 if (unlikely(dst_vdev->remove)) {
853 RTE_LOG(DEBUG, VHOST_DATA, "(%" PRIu64 ") "
854 "Device is marked for removal\n", fh);
858 virtio_xmit(dst_vdev->dev, vdev->dev, m);
863 * Check if the destination MAC of a packet is one local VM,
864 * and get its vlan tag, and offset if it is.
866 static inline int __attribute__((always_inline))
867 find_local_dest(struct virtio_net *dev, struct rte_mbuf *m,
868 uint32_t *offset, uint16_t *vlan_tag)
870 struct vhost_dev *dst_vdev;
871 struct ether_hdr *pkt_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
873 dst_vdev = find_vhost_dev(&pkt_hdr->d_addr);
877 if (dst_vdev->dev->device_fh == dev->device_fh) {
878 RTE_LOG(DEBUG, VHOST_DATA,
879 "(%" PRIu64 ") TX: src and dst MAC is same. "
880 " Dropping packet.\n", dst_vdev->dev->device_fh);
885 * HW vlan strip will reduce the packet length
886 * by minus length of vlan tag, so need restore
887 * the packet length by plus it.
890 *vlan_tag = vlan_tags[(uint16_t)dst_vdev->dev->device_fh];
892 RTE_LOG(DEBUG, VHOST_DATA,
893 "(%" PRIu64 ") TX: pkt to local VM device id: (%" PRIu64 ") "
895 dev->device_fh, dst_vdev->dev->device_fh, *vlan_tag);
901 get_psd_sum(void *l3_hdr, uint64_t ol_flags)
903 if (ol_flags & PKT_TX_IPV4)
904 return rte_ipv4_phdr_cksum(l3_hdr, ol_flags);
905 else /* assume ethertype == ETHER_TYPE_IPv6 */
906 return rte_ipv6_phdr_cksum(l3_hdr, ol_flags);
909 static void virtio_tx_offload(struct rte_mbuf *m)
912 struct ipv4_hdr *ipv4_hdr = NULL;
913 struct tcp_hdr *tcp_hdr = NULL;
914 struct ether_hdr *eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
916 l3_hdr = (char *)eth_hdr + m->l2_len;
918 if (m->ol_flags & PKT_TX_IPV4) {
920 ipv4_hdr->hdr_checksum = 0;
921 m->ol_flags |= PKT_TX_IP_CKSUM;
924 tcp_hdr = (struct tcp_hdr *)((char *)l3_hdr + m->l3_len);
925 tcp_hdr->cksum = get_psd_sum(l3_hdr, m->ol_flags);
929 * This function routes the TX packet to the correct interface. This may be a local device
930 * or the physical port.
932 static inline void __attribute__((always_inline))
933 virtio_tx_route(struct vhost_dev *vdev, struct rte_mbuf *m, uint16_t vlan_tag)
935 struct mbuf_table *tx_q;
936 struct rte_mbuf **m_table;
937 unsigned len, ret, offset = 0;
938 const uint16_t lcore_id = rte_lcore_id();
939 struct virtio_net *dev = vdev->dev;
940 struct ether_hdr *nh;
943 nh = rte_pktmbuf_mtod(m, struct ether_hdr *);
944 if (unlikely(is_broadcast_ether_addr(&nh->d_addr))) {
945 struct vhost_dev *vdev2;
947 TAILQ_FOREACH(vdev2, &vhost_dev_list, next) {
948 virtio_xmit(vdev2->dev, vdev->dev, m);
953 /*check if destination is local VM*/
954 if ((vm2vm_mode == VM2VM_SOFTWARE) && (virtio_tx_local(vdev, m) == 0)) {
959 if (unlikely(vm2vm_mode == VM2VM_HARDWARE)) {
960 if (unlikely(find_local_dest(dev, m, &offset, &vlan_tag) != 0)) {
966 RTE_LOG(DEBUG, VHOST_DATA, "(%" PRIu64 ") TX: "
967 "MAC address is external\n", dev->device_fh);
971 /*Add packet to the port tx queue*/
972 tx_q = &lcore_tx_queue[lcore_id];
975 nh = rte_pktmbuf_mtod(m, struct ether_hdr *);
976 if (unlikely(nh->ether_type == rte_cpu_to_be_16(ETHER_TYPE_VLAN))) {
977 /* Guest has inserted the vlan tag. */
978 struct vlan_hdr *vh = (struct vlan_hdr *) (nh + 1);
979 uint16_t vlan_tag_be = rte_cpu_to_be_16(vlan_tag);
980 if ((vm2vm_mode == VM2VM_HARDWARE) &&
981 (vh->vlan_tci != vlan_tag_be))
982 vh->vlan_tci = vlan_tag_be;
984 m->ol_flags |= PKT_TX_VLAN_PKT;
987 * Find the right seg to adjust the data len when offset is
988 * bigger than tail room size.
990 if (unlikely(vm2vm_mode == VM2VM_HARDWARE)) {
991 if (likely(offset <= rte_pktmbuf_tailroom(m)))
992 m->data_len += offset;
994 struct rte_mbuf *seg = m;
996 while ((seg->next != NULL) &&
997 (offset > rte_pktmbuf_tailroom(seg)))
1000 seg->data_len += offset;
1002 m->pkt_len += offset;
1005 m->vlan_tci = vlan_tag;
1008 if (m->ol_flags & PKT_TX_TCP_SEG)
1009 virtio_tx_offload(m);
1011 tx_q->m_table[len] = m;
1014 dev_statistics[dev->device_fh].tx_total++;
1015 dev_statistics[dev->device_fh].tx++;
1018 if (unlikely(len == MAX_PKT_BURST)) {
1019 m_table = (struct rte_mbuf **)tx_q->m_table;
1020 ret = rte_eth_tx_burst(ports[0], (uint16_t)tx_q->txq_id, m_table, (uint16_t) len);
1021 /* Free any buffers not handled by TX and update the port stats. */
1022 if (unlikely(ret < len)) {
1024 rte_pktmbuf_free(m_table[ret]);
1025 } while (++ret < len);
1035 * This function is called by each data core. It handles all RX/TX registered with the
1036 * core. For TX the specific lcore linked list is used. For RX, MAC addresses are compared
1037 * with all devices in the main linked list.
1040 switch_worker(__attribute__((unused)) void *arg)
1042 struct virtio_net *dev = NULL;
1043 struct vhost_dev *vdev = NULL;
1044 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
1045 struct mbuf_table *tx_q;
1046 const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
1047 uint64_t prev_tsc, diff_tsc, cur_tsc, ret_count = 0;
1049 const uint16_t lcore_id = rte_lcore_id();
1050 const uint16_t num_cores = (uint16_t)rte_lcore_count();
1051 uint16_t rx_count = 0;
1055 RTE_LOG(INFO, VHOST_DATA, "Procesing on Core %u started\n", lcore_id);
1058 tx_q = &lcore_tx_queue[lcore_id];
1059 for (i = 0; i < num_cores; i ++) {
1060 if (lcore_ids[i] == lcore_id) {
1067 cur_tsc = rte_rdtsc();
1069 * TX burst queue drain
1071 diff_tsc = cur_tsc - prev_tsc;
1072 if (unlikely(diff_tsc > drain_tsc)) {
1075 RTE_LOG(DEBUG, VHOST_DATA,
1076 "TX queue drained after timeout with burst size %u\n",
1079 /*Tx any packets in the queue*/
1080 ret = rte_eth_tx_burst(ports[0], (uint16_t)tx_q->txq_id,
1081 (struct rte_mbuf **)tx_q->m_table,
1082 (uint16_t)tx_q->len);
1083 if (unlikely(ret < tx_q->len)) {
1085 rte_pktmbuf_free(tx_q->m_table[ret]);
1086 } while (++ret < tx_q->len);
1097 * Inform the configuration core that we have exited the
1098 * linked list and that no devices are in use if requested.
1100 if (lcore_info[lcore_id].dev_removal_flag == REQUEST_DEV_REMOVAL)
1101 lcore_info[lcore_id].dev_removal_flag = ACK_DEV_REMOVAL;
1106 TAILQ_FOREACH(vdev, &lcore_info[lcore_id].vdev_list, next) {
1110 fh = dev->device_fh;
1112 if (unlikely(vdev->remove)) {
1114 vdev->ready = DEVICE_SAFE_REMOVE;
1118 if (likely(vdev->ready == DEVICE_RX)) {
1120 rx_count = rte_eth_rx_burst(ports[0],
1121 vdev->vmdq_rx_q, pkts_burst, MAX_PKT_BURST);
1125 * Retry is enabled and the queue is full then we wait and retry to avoid packet loss
1126 * Here MAX_PKT_BURST must be less than virtio queue size
1128 if (enable_retry && unlikely(rx_count > rte_vring_available_entries(dev, VIRTIO_RXQ))) {
1129 for (retry = 0; retry < burst_rx_retry_num; retry++) {
1130 rte_delay_us(burst_rx_delay_time);
1131 if (rx_count <= rte_vring_available_entries(dev, VIRTIO_RXQ))
1135 ret_count = rte_vhost_enqueue_burst(dev, VIRTIO_RXQ, pkts_burst, rx_count);
1138 &dev_statistics[fh].rx_total_atomic,
1141 &dev_statistics[fh].rx_atomic,
1144 while (likely(rx_count)) {
1146 rte_pktmbuf_free(pkts_burst[rx_count]);
1152 if (likely(!vdev->remove)) {
1153 /* Handle guest TX*/
1154 tx_count = rte_vhost_dequeue_burst(dev, VIRTIO_TXQ, mbuf_pool, pkts_burst, MAX_PKT_BURST);
1155 /* If this is the first received packet we need to learn the MAC and setup VMDQ */
1156 if (unlikely(vdev->ready == DEVICE_MAC_LEARNING) && tx_count) {
1157 if (vdev->remove || (link_vmdq(vdev, pkts_burst[0]) == -1)) {
1159 rte_pktmbuf_free(pkts_burst[--tx_count]);
1162 for (i = 0; i < tx_count; ++i) {
1163 virtio_tx_route(vdev, pkts_burst[i],
1164 vlan_tags[(uint16_t)dev->device_fh]);
1174 * Remove a device from the specific data core linked list and from the
1175 * main linked list. Synchonization occurs through the use of the
1176 * lcore dev_removal_flag. Device is made volatile here to avoid re-ordering
1177 * of dev->remove=1 which can cause an infinite loop in the rte_pause loop.
1180 destroy_device (volatile struct virtio_net *dev)
1182 struct vhost_dev *vdev;
1185 dev->flags &= ~VIRTIO_DEV_RUNNING;
1187 vdev = (struct vhost_dev *)dev->priv;
1188 /*set the remove flag. */
1190 while(vdev->ready != DEVICE_SAFE_REMOVE) {
1194 TAILQ_REMOVE(&lcore_info[vdev->coreid].vdev_list, vdev, next);
1195 TAILQ_REMOVE(&vhost_dev_list, vdev, next);
1197 /* Set the dev_removal_flag on each lcore. */
1198 RTE_LCORE_FOREACH_SLAVE(lcore)
1199 lcore_info[lcore].dev_removal_flag = REQUEST_DEV_REMOVAL;
1202 * Once each core has set the dev_removal_flag to ACK_DEV_REMOVAL
1203 * we can be sure that they can no longer access the device removed
1204 * from the linked lists and that the devices are no longer in use.
1206 RTE_LCORE_FOREACH_SLAVE(lcore) {
1207 while (lcore_info[lcore].dev_removal_flag != ACK_DEV_REMOVAL)
1211 lcore_info[vdev->coreid].device_num--;
1213 RTE_LOG(INFO, VHOST_DATA,
1214 "(%" PRIu64 ") Device has been removed from data core\n",
1221 * A new device is added to a data core. First the device is added to the main linked list
1222 * and the allocated to a specific data core.
1225 new_device (struct virtio_net *dev)
1227 int lcore, core_add = 0;
1228 uint32_t device_num_min = num_devices;
1229 struct vhost_dev *vdev;
1231 vdev = rte_zmalloc("vhost device", sizeof(*vdev), RTE_CACHE_LINE_SIZE);
1233 RTE_LOG(INFO, VHOST_DATA, "(%"PRIu64") Couldn't allocate memory for vhost dev\n",
1240 TAILQ_INSERT_TAIL(&vhost_dev_list, vdev, next);
1242 = dev->device_fh * queues_per_pool + vmdq_queue_base;
1244 /*reset ready flag*/
1245 vdev->ready = DEVICE_MAC_LEARNING;
1248 /* Find a suitable lcore to add the device. */
1249 RTE_LCORE_FOREACH_SLAVE(lcore) {
1250 if (lcore_info[lcore].device_num < device_num_min) {
1251 device_num_min = lcore_info[lcore].device_num;
1255 vdev->coreid = core_add;
1257 TAILQ_INSERT_TAIL(&lcore_info[vdev->coreid].vdev_list, vdev, next);
1258 lcore_info[vdev->coreid].device_num++;
1260 /* Initialize device stats */
1261 memset(&dev_statistics[dev->device_fh], 0, sizeof(struct device_statistics));
1263 /* Disable notifications. */
1264 rte_vhost_enable_guest_notification(dev, VIRTIO_RXQ, 0);
1265 rte_vhost_enable_guest_notification(dev, VIRTIO_TXQ, 0);
1266 dev->flags |= VIRTIO_DEV_RUNNING;
1268 RTE_LOG(INFO, VHOST_DATA, "(%"PRIu64") Device has been added to data core %d\n", dev->device_fh, vdev->coreid);
1274 * These callback allow devices to be added to the data core when configuration
1275 * has been fully complete.
1277 static const struct virtio_net_device_ops virtio_net_device_ops =
1279 .new_device = new_device,
1280 .destroy_device = destroy_device,
1284 * This is a thread will wake up after a period to print stats if the user has
1290 struct vhost_dev *vdev;
1291 uint64_t tx_dropped, rx_dropped;
1292 uint64_t tx, tx_total, rx, rx_total;
1294 const char clr[] = { 27, '[', '2', 'J', '\0' };
1295 const char top_left[] = { 27, '[', '1', ';', '1', 'H','\0' };
1298 sleep(enable_stats);
1300 /* Clear screen and move to top left */
1301 printf("%s%s", clr, top_left);
1303 printf("\nDevice statistics ====================================");
1305 TAILQ_FOREACH(vdev, &vhost_dev_list, next) {
1306 device_fh = vdev->dev->device_fh;
1307 tx_total = dev_statistics[device_fh].tx_total;
1308 tx = dev_statistics[device_fh].tx;
1309 tx_dropped = tx_total - tx;
1310 rx_total = rte_atomic64_read(
1311 &dev_statistics[device_fh].rx_total_atomic);
1312 rx = rte_atomic64_read(
1313 &dev_statistics[device_fh].rx_atomic);
1314 rx_dropped = rx_total - rx;
1316 printf("\nStatistics for device %"PRIu32" ------------------------------"
1317 "\nTX total: %"PRIu64""
1318 "\nTX dropped: %"PRIu64""
1319 "\nTX successful: %"PRIu64""
1320 "\nRX total: %"PRIu64""
1321 "\nRX dropped: %"PRIu64""
1322 "\nRX successful: %"PRIu64"",
1331 printf("\n======================================================\n");
1335 /* When we receive a INT signal, unregister vhost driver */
1337 sigint_handler(__rte_unused int signum)
1339 /* Unregister vhost driver. */
1340 int ret = rte_vhost_driver_unregister((char *)&dev_basename);
1342 rte_exit(EXIT_FAILURE, "vhost driver unregister failure.\n");
1347 * Main function, does initialisation and calls the per-lcore functions. The CUSE
1348 * device is also registered here to handle the IOCTLs.
1351 main(int argc, char *argv[])
1353 unsigned lcore_id, core_id = 0;
1354 unsigned nb_ports, valid_num_ports;
1357 static pthread_t tid;
1358 char thread_name[RTE_MAX_THREAD_NAME_LEN];
1360 signal(SIGINT, sigint_handler);
1363 ret = rte_eal_init(argc, argv);
1365 rte_exit(EXIT_FAILURE, "Error with EAL initialization\n");
1369 /* parse app arguments */
1370 ret = us_vhost_parse_args(argc, argv);
1372 rte_exit(EXIT_FAILURE, "Invalid argument\n");
1374 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id ++)
1375 TAILQ_INIT(&lcore_info[lcore_id].vdev_list);
1377 if (rte_lcore_is_enabled(lcore_id))
1378 lcore_ids[core_id ++] = lcore_id;
1380 if (rte_lcore_count() > RTE_MAX_LCORE)
1381 rte_exit(EXIT_FAILURE,"Not enough cores\n");
1383 /*set the number of swithcing cores available*/
1384 num_switching_cores = rte_lcore_count()-1;
1386 /* Get the number of physical ports. */
1387 nb_ports = rte_eth_dev_count();
1388 if (nb_ports > RTE_MAX_ETHPORTS)
1389 nb_ports = RTE_MAX_ETHPORTS;
1392 * Update the global var NUM_PORTS and global array PORTS
1393 * and get value of var VALID_NUM_PORTS according to system ports number
1395 valid_num_ports = check_ports_num(nb_ports);
1397 if ((valid_num_ports == 0) || (valid_num_ports > MAX_SUP_PORTS)) {
1398 RTE_LOG(INFO, VHOST_PORT, "Current enabled port number is %u,"
1399 "but only %u port can be enabled\n",num_ports, MAX_SUP_PORTS);
1403 /* Create the mbuf pool. */
1404 mbuf_pool = rte_pktmbuf_pool_create("MBUF_POOL",
1405 NUM_MBUFS_PER_PORT * valid_num_ports, MBUF_CACHE_SIZE,
1406 0, MBUF_DATA_SIZE, rte_socket_id());
1407 if (mbuf_pool == NULL)
1408 rte_exit(EXIT_FAILURE, "Cannot create mbuf pool\n");
1410 if (vm2vm_mode == VM2VM_HARDWARE) {
1411 /* Enable VT loop back to let L2 switch to do it. */
1412 vmdq_conf_default.rx_adv_conf.vmdq_rx_conf.enable_loop_back = 1;
1413 RTE_LOG(DEBUG, VHOST_CONFIG,
1414 "Enable loop back for L2 switch in vmdq.\n");
1417 /* initialize all ports */
1418 for (portid = 0; portid < nb_ports; portid++) {
1419 /* skip ports that are not enabled */
1420 if ((enabled_port_mask & (1 << portid)) == 0) {
1421 RTE_LOG(INFO, VHOST_PORT,
1422 "Skipping disabled port %d\n", portid);
1425 if (port_init(portid) != 0)
1426 rte_exit(EXIT_FAILURE,
1427 "Cannot initialize network ports\n");
1430 /* Initialize device stats */
1431 memset(&dev_statistics, 0, sizeof(dev_statistics));
1433 /* Enable stats if the user option is set. */
1435 ret = pthread_create(&tid, NULL, (void *)print_stats, NULL);
1437 rte_exit(EXIT_FAILURE,
1438 "Cannot create print-stats thread\n");
1440 /* Set thread_name for aid in debugging. */
1441 snprintf(thread_name, RTE_MAX_THREAD_NAME_LEN, "print-stats");
1442 ret = rte_thread_setname(tid, thread_name);
1444 RTE_LOG(ERR, VHOST_CONFIG,
1445 "Cannot set print-stats name\n");
1448 /* Launch all data cores. */
1449 RTE_LCORE_FOREACH_SLAVE(lcore_id)
1450 rte_eal_remote_launch(switch_worker, NULL, lcore_id);
1453 rte_vhost_feature_disable(1ULL << VIRTIO_NET_F_MRG_RXBUF);
1455 /* Register vhost(cuse or user) driver to handle vhost messages. */
1456 ret = rte_vhost_driver_register((char *)&dev_basename);
1458 rte_exit(EXIT_FAILURE, "vhost driver register failure.\n");
1460 rte_vhost_driver_callback_register(&virtio_net_device_ops);
1462 /* Start CUSE session. */
1463 rte_vhost_driver_session_start();