xref: /dpdk/examples/vhost/main.c (revision bdb19b771e6f849d828d34aff48c91c6a3030fe0)
1 /*-
2  *   BSD LICENSE
3  *
4  *   Copyright(c) 2010-2016 Intel Corporation. All rights reserved.
5  *   All rights reserved.
6  *
7  *   Redistribution and use in source and binary forms, with or without
8  *   modification, are permitted provided that the following conditions
9  *   are met:
10  *
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
16  *       distribution.
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.
20  *
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.
32  */
33 
34 #include <arpa/inet.h>
35 #include <getopt.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>
40 #include <signal.h>
41 #include <stdint.h>
42 #include <sys/eventfd.h>
43 #include <sys/param.h>
44 #include <unistd.h>
45 
46 #include <rte_atomic.h>
47 #include <rte_cycles.h>
48 #include <rte_ethdev.h>
49 #include <rte_log.h>
50 #include <rte_string_fns.h>
51 #include <rte_malloc.h>
52 #include <rte_virtio_net.h>
53 #include <rte_ip.h>
54 #include <rte_tcp.h>
55 
56 #include "main.h"
57 
58 #ifndef MAX_QUEUES
59 #define MAX_QUEUES 128
60 #endif
61 
62 /* the maximum number of external ports supported */
63 #define MAX_SUP_PORTS 1
64 
65 #define MBUF_CACHE_SIZE	128
66 #define MBUF_DATA_SIZE	RTE_MBUF_DEFAULT_BUF_SIZE
67 
68 #define MAX_PKT_BURST 32		/* Max burst size for RX/TX */
69 #define BURST_TX_DRAIN_US 100	/* TX drain every ~100us */
70 
71 #define BURST_RX_WAIT_US 15	/* Defines how long we wait between retries on RX */
72 #define BURST_RX_RETRIES 4		/* Number of retries on RX. */
73 
74 #define JUMBO_FRAME_MAX_SIZE    0x2600
75 
76 /* State of virtio device. */
77 #define DEVICE_MAC_LEARNING 0
78 #define DEVICE_RX			1
79 #define DEVICE_SAFE_REMOVE	2
80 
81 /* Configurable number of RX/TX ring descriptors */
82 #define RTE_TEST_RX_DESC_DEFAULT 1024
83 #define RTE_TEST_TX_DESC_DEFAULT 512
84 
85 #define INVALID_PORT_ID 0xFF
86 
87 /* Max number of devices. Limited by vmdq. */
88 #define MAX_DEVICES 64
89 
90 /* Size of buffers used for snprintfs. */
91 #define MAX_PRINT_BUFF 6072
92 
93 /* Maximum character device basename size. */
94 #define MAX_BASENAME_SZ 10
95 
96 /* Maximum long option length for option parsing. */
97 #define MAX_LONG_OPT_SZ 64
98 
99 /* mask of enabled ports */
100 static uint32_t enabled_port_mask = 0;
101 
102 /* Promiscuous mode */
103 static uint32_t promiscuous;
104 
105 /* number of devices/queues to support*/
106 static uint32_t num_queues = 0;
107 static uint32_t num_devices;
108 
109 static struct rte_mempool *mbuf_pool;
110 static int mergeable;
111 
112 /* Do vlan strip on host, enabled on default */
113 static uint32_t vlan_strip = 1;
114 
115 /* Enable VM2VM communications. If this is disabled then the MAC address compare is skipped. */
116 typedef enum {
117 	VM2VM_DISABLED = 0,
118 	VM2VM_SOFTWARE = 1,
119 	VM2VM_HARDWARE = 2,
120 	VM2VM_LAST
121 } vm2vm_type;
122 static vm2vm_type vm2vm_mode = VM2VM_SOFTWARE;
123 
124 /* Enable stats. */
125 static uint32_t enable_stats = 0;
126 /* Enable retries on RX. */
127 static uint32_t enable_retry = 1;
128 
129 /* Disable TX checksum offload */
130 static uint32_t enable_tx_csum;
131 
132 /* Disable TSO offload */
133 static uint32_t enable_tso;
134 
135 /* Specify timeout (in useconds) between retries on RX. */
136 static uint32_t burst_rx_delay_time = BURST_RX_WAIT_US;
137 /* Specify the number of retries on RX. */
138 static uint32_t burst_rx_retry_num = BURST_RX_RETRIES;
139 
140 /* Character device basename. Can be set by user. */
141 static char dev_basename[MAX_BASENAME_SZ] = "vhost-net";
142 
143 /* empty vmdq configuration structure. Filled in programatically */
144 static struct rte_eth_conf vmdq_conf_default = {
145 	.rxmode = {
146 		.mq_mode        = ETH_MQ_RX_VMDQ_ONLY,
147 		.split_hdr_size = 0,
148 		.header_split   = 0, /**< Header Split disabled */
149 		.hw_ip_checksum = 0, /**< IP checksum offload disabled */
150 		.hw_vlan_filter = 0, /**< VLAN filtering disabled */
151 		/*
152 		 * It is necessary for 1G NIC such as I350,
153 		 * this fixes bug of ipv4 forwarding in guest can't
154 		 * forward pakets from one virtio dev to another virtio dev.
155 		 */
156 		.hw_vlan_strip  = 1, /**< VLAN strip enabled. */
157 		.jumbo_frame    = 0, /**< Jumbo Frame Support disabled */
158 		.hw_strip_crc   = 0, /**< CRC stripped by hardware */
159 	},
160 
161 	.txmode = {
162 		.mq_mode = ETH_MQ_TX_NONE,
163 	},
164 	.rx_adv_conf = {
165 		/*
166 		 * should be overridden separately in code with
167 		 * appropriate values
168 		 */
169 		.vmdq_rx_conf = {
170 			.nb_queue_pools = ETH_8_POOLS,
171 			.enable_default_pool = 0,
172 			.default_pool = 0,
173 			.nb_pool_maps = 0,
174 			.pool_map = {{0, 0},},
175 		},
176 	},
177 };
178 
179 static unsigned lcore_ids[RTE_MAX_LCORE];
180 static uint8_t ports[RTE_MAX_ETHPORTS];
181 static unsigned num_ports = 0; /**< The number of ports specified in command line */
182 static uint16_t num_pf_queues, num_vmdq_queues;
183 static uint16_t vmdq_pool_base, vmdq_queue_base;
184 static uint16_t queues_per_pool;
185 
186 const uint16_t vlan_tags[] = {
187 	1000, 1001, 1002, 1003, 1004, 1005, 1006, 1007,
188 	1008, 1009, 1010, 1011,	1012, 1013, 1014, 1015,
189 	1016, 1017, 1018, 1019, 1020, 1021, 1022, 1023,
190 	1024, 1025, 1026, 1027, 1028, 1029, 1030, 1031,
191 	1032, 1033, 1034, 1035, 1036, 1037, 1038, 1039,
192 	1040, 1041, 1042, 1043, 1044, 1045, 1046, 1047,
193 	1048, 1049, 1050, 1051, 1052, 1053, 1054, 1055,
194 	1056, 1057, 1058, 1059, 1060, 1061, 1062, 1063,
195 };
196 
197 /* ethernet addresses of ports */
198 static struct ether_addr vmdq_ports_eth_addr[RTE_MAX_ETHPORTS];
199 
200 static struct vhost_dev_tailq_list vhost_dev_list =
201 	TAILQ_HEAD_INITIALIZER(vhost_dev_list);
202 
203 static struct lcore_info lcore_info[RTE_MAX_LCORE];
204 
205 /* Used for queueing bursts of TX packets. */
206 struct mbuf_table {
207 	unsigned len;
208 	unsigned txq_id;
209 	struct rte_mbuf *m_table[MAX_PKT_BURST];
210 };
211 
212 /* TX queue for each data core. */
213 struct mbuf_table lcore_tx_queue[RTE_MAX_LCORE];
214 
215 #define VLAN_HLEN       4
216 
217 /* Per-device statistics struct */
218 struct device_statistics {
219 	uint64_t tx_total;
220 	rte_atomic64_t rx_total_atomic;
221 	uint64_t tx;
222 	rte_atomic64_t rx_atomic;
223 } __rte_cache_aligned;
224 struct device_statistics dev_statistics[MAX_DEVICES];
225 
226 /*
227  * Builds up the correct configuration for VMDQ VLAN pool map
228  * according to the pool & queue limits.
229  */
230 static inline int
231 get_eth_conf(struct rte_eth_conf *eth_conf, uint32_t num_devices)
232 {
233 	struct rte_eth_vmdq_rx_conf conf;
234 	struct rte_eth_vmdq_rx_conf *def_conf =
235 		&vmdq_conf_default.rx_adv_conf.vmdq_rx_conf;
236 	unsigned i;
237 
238 	memset(&conf, 0, sizeof(conf));
239 	conf.nb_queue_pools = (enum rte_eth_nb_pools)num_devices;
240 	conf.nb_pool_maps = num_devices;
241 	conf.enable_loop_back = def_conf->enable_loop_back;
242 	conf.rx_mode = def_conf->rx_mode;
243 
244 	for (i = 0; i < conf.nb_pool_maps; i++) {
245 		conf.pool_map[i].vlan_id = vlan_tags[ i ];
246 		conf.pool_map[i].pools = (1UL << i);
247 	}
248 
249 	(void)(rte_memcpy(eth_conf, &vmdq_conf_default, sizeof(*eth_conf)));
250 	(void)(rte_memcpy(&eth_conf->rx_adv_conf.vmdq_rx_conf, &conf,
251 		   sizeof(eth_conf->rx_adv_conf.vmdq_rx_conf)));
252 	return 0;
253 }
254 
255 /*
256  * Validate the device number according to the max pool number gotten form
257  * dev_info. If the device number is invalid, give the error message and
258  * return -1. Each device must have its own pool.
259  */
260 static inline int
261 validate_num_devices(uint32_t max_nb_devices)
262 {
263 	if (num_devices > max_nb_devices) {
264 		RTE_LOG(ERR, VHOST_PORT, "invalid number of devices\n");
265 		return -1;
266 	}
267 	return 0;
268 }
269 
270 /*
271  * Initialises a given port using global settings and with the rx buffers
272  * coming from the mbuf_pool passed as parameter
273  */
274 static inline int
275 port_init(uint8_t port)
276 {
277 	struct rte_eth_dev_info dev_info;
278 	struct rte_eth_conf port_conf;
279 	struct rte_eth_rxconf *rxconf;
280 	struct rte_eth_txconf *txconf;
281 	int16_t rx_rings, tx_rings;
282 	uint16_t rx_ring_size, tx_ring_size;
283 	int retval;
284 	uint16_t q;
285 
286 	/* The max pool number from dev_info will be used to validate the pool number specified in cmd line */
287 	rte_eth_dev_info_get (port, &dev_info);
288 
289 	if (dev_info.max_rx_queues > MAX_QUEUES) {
290 		rte_exit(EXIT_FAILURE,
291 			"please define MAX_QUEUES no less than %u in %s\n",
292 			dev_info.max_rx_queues, __FILE__);
293 	}
294 
295 	rxconf = &dev_info.default_rxconf;
296 	txconf = &dev_info.default_txconf;
297 	rxconf->rx_drop_en = 1;
298 
299 	/* Enable vlan offload */
300 	txconf->txq_flags &= ~ETH_TXQ_FLAGS_NOVLANOFFL;
301 
302 	/*configure the number of supported virtio devices based on VMDQ limits */
303 	num_devices = dev_info.max_vmdq_pools;
304 
305 	rx_ring_size = RTE_TEST_RX_DESC_DEFAULT;
306 	tx_ring_size = RTE_TEST_TX_DESC_DEFAULT;
307 	tx_rings = (uint16_t)rte_lcore_count();
308 
309 	retval = validate_num_devices(MAX_DEVICES);
310 	if (retval < 0)
311 		return retval;
312 
313 	/* Get port configuration. */
314 	retval = get_eth_conf(&port_conf, num_devices);
315 	if (retval < 0)
316 		return retval;
317 	/* NIC queues are divided into pf queues and vmdq queues.  */
318 	num_pf_queues = dev_info.max_rx_queues - dev_info.vmdq_queue_num;
319 	queues_per_pool = dev_info.vmdq_queue_num / dev_info.max_vmdq_pools;
320 	num_vmdq_queues = num_devices * queues_per_pool;
321 	num_queues = num_pf_queues + num_vmdq_queues;
322 	vmdq_queue_base = dev_info.vmdq_queue_base;
323 	vmdq_pool_base  = dev_info.vmdq_pool_base;
324 	printf("pf queue num: %u, configured vmdq pool num: %u, each vmdq pool has %u queues\n",
325 		num_pf_queues, num_devices, queues_per_pool);
326 
327 	if (port >= rte_eth_dev_count()) return -1;
328 
329 	if (enable_tx_csum == 0)
330 		rte_vhost_feature_disable(1ULL << VIRTIO_NET_F_CSUM);
331 
332 	if (enable_tso == 0) {
333 		rte_vhost_feature_disable(1ULL << VIRTIO_NET_F_HOST_TSO4);
334 		rte_vhost_feature_disable(1ULL << VIRTIO_NET_F_HOST_TSO6);
335 	}
336 
337 	rx_rings = (uint16_t)dev_info.max_rx_queues;
338 	/* Configure ethernet device. */
339 	retval = rte_eth_dev_configure(port, rx_rings, tx_rings, &port_conf);
340 	if (retval != 0)
341 		return retval;
342 
343 	/* Setup the queues. */
344 	for (q = 0; q < rx_rings; q ++) {
345 		retval = rte_eth_rx_queue_setup(port, q, rx_ring_size,
346 						rte_eth_dev_socket_id(port),
347 						rxconf,
348 						mbuf_pool);
349 		if (retval < 0)
350 			return retval;
351 	}
352 	for (q = 0; q < tx_rings; q ++) {
353 		retval = rte_eth_tx_queue_setup(port, q, tx_ring_size,
354 						rte_eth_dev_socket_id(port),
355 						txconf);
356 		if (retval < 0)
357 			return retval;
358 	}
359 
360 	/* Start the device. */
361 	retval  = rte_eth_dev_start(port);
362 	if (retval < 0) {
363 		RTE_LOG(ERR, VHOST_DATA, "Failed to start the device.\n");
364 		return retval;
365 	}
366 
367 	if (promiscuous)
368 		rte_eth_promiscuous_enable(port);
369 
370 	rte_eth_macaddr_get(port, &vmdq_ports_eth_addr[port]);
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 			(unsigned)port,
375 			vmdq_ports_eth_addr[port].addr_bytes[0],
376 			vmdq_ports_eth_addr[port].addr_bytes[1],
377 			vmdq_ports_eth_addr[port].addr_bytes[2],
378 			vmdq_ports_eth_addr[port].addr_bytes[3],
379 			vmdq_ports_eth_addr[port].addr_bytes[4],
380 			vmdq_ports_eth_addr[port].addr_bytes[5]);
381 
382 	return 0;
383 }
384 
385 /*
386  * Set character device basename.
387  */
388 static int
389 us_vhost_parse_basename(const char *q_arg)
390 {
391 	/* parse number string */
392 
393 	if (strnlen(q_arg, MAX_BASENAME_SZ) > MAX_BASENAME_SZ)
394 		return -1;
395 	else
396 		snprintf((char*)&dev_basename, MAX_BASENAME_SZ, "%s", q_arg);
397 
398 	return 0;
399 }
400 
401 /*
402  * Parse the portmask provided at run time.
403  */
404 static int
405 parse_portmask(const char *portmask)
406 {
407 	char *end = NULL;
408 	unsigned long pm;
409 
410 	errno = 0;
411 
412 	/* parse hexadecimal string */
413 	pm = strtoul(portmask, &end, 16);
414 	if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0') || (errno != 0))
415 		return -1;
416 
417 	if (pm == 0)
418 		return -1;
419 
420 	return pm;
421 
422 }
423 
424 /*
425  * Parse num options at run time.
426  */
427 static int
428 parse_num_opt(const char *q_arg, uint32_t max_valid_value)
429 {
430 	char *end = NULL;
431 	unsigned long num;
432 
433 	errno = 0;
434 
435 	/* parse unsigned int string */
436 	num = strtoul(q_arg, &end, 10);
437 	if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0') || (errno != 0))
438 		return -1;
439 
440 	if (num > max_valid_value)
441 		return -1;
442 
443 	return num;
444 
445 }
446 
447 /*
448  * Display usage
449  */
450 static void
451 us_vhost_usage(const char *prgname)
452 {
453 	RTE_LOG(INFO, VHOST_CONFIG, "%s [EAL options] -- -p PORTMASK\n"
454 	"		--vm2vm [0|1|2]\n"
455 	"		--rx_retry [0|1] --mergeable [0|1] --stats [0-N]\n"
456 	"		--dev-basename <name>\n"
457 	"		--nb-devices ND\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 	"		--vlan-strip [0|1]: disable/enable(default) RX VLAN strip on host\n"
465 	"		--stats [0-N]: 0: Disable stats, N: Time in seconds to print stats\n"
466 	"		--dev-basename: The basename to be used for the character device.\n"
467 	"		--tx-csum [0|1] disable/enable TX checksum offload.\n"
468 	"		--tso [0|1] disable/enable TCP segment offload.\n",
469 	       prgname);
470 }
471 
472 /*
473  * Parse the arguments given in the command line of the application.
474  */
475 static int
476 us_vhost_parse_args(int argc, char **argv)
477 {
478 	int opt, ret;
479 	int option_index;
480 	unsigned i;
481 	const char *prgname = argv[0];
482 	static struct option long_option[] = {
483 		{"vm2vm", required_argument, NULL, 0},
484 		{"rx-retry", required_argument, NULL, 0},
485 		{"rx-retry-delay", required_argument, NULL, 0},
486 		{"rx-retry-num", required_argument, NULL, 0},
487 		{"mergeable", required_argument, NULL, 0},
488 		{"vlan-strip", required_argument, NULL, 0},
489 		{"stats", required_argument, NULL, 0},
490 		{"dev-basename", required_argument, NULL, 0},
491 		{"tx-csum", required_argument, NULL, 0},
492 		{"tso", required_argument, NULL, 0},
493 		{NULL, 0, 0, 0},
494 	};
495 
496 	/* Parse command line */
497 	while ((opt = getopt_long(argc, argv, "p:P",
498 			long_option, &option_index)) != EOF) {
499 		switch (opt) {
500 		/* Portmask */
501 		case 'p':
502 			enabled_port_mask = parse_portmask(optarg);
503 			if (enabled_port_mask == 0) {
504 				RTE_LOG(INFO, VHOST_CONFIG, "Invalid portmask\n");
505 				us_vhost_usage(prgname);
506 				return -1;
507 			}
508 			break;
509 
510 		case 'P':
511 			promiscuous = 1;
512 			vmdq_conf_default.rx_adv_conf.vmdq_rx_conf.rx_mode =
513 				ETH_VMDQ_ACCEPT_BROADCAST |
514 				ETH_VMDQ_ACCEPT_MULTICAST;
515 			rte_vhost_feature_enable(1ULL << VIRTIO_NET_F_CTRL_RX);
516 
517 			break;
518 
519 		case 0:
520 			/* Enable/disable vm2vm comms. */
521 			if (!strncmp(long_option[option_index].name, "vm2vm",
522 				MAX_LONG_OPT_SZ)) {
523 				ret = parse_num_opt(optarg, (VM2VM_LAST - 1));
524 				if (ret == -1) {
525 					RTE_LOG(INFO, VHOST_CONFIG,
526 						"Invalid argument for "
527 						"vm2vm [0|1|2]\n");
528 					us_vhost_usage(prgname);
529 					return -1;
530 				} else {
531 					vm2vm_mode = (vm2vm_type)ret;
532 				}
533 			}
534 
535 			/* Enable/disable retries on RX. */
536 			if (!strncmp(long_option[option_index].name, "rx-retry", MAX_LONG_OPT_SZ)) {
537 				ret = parse_num_opt(optarg, 1);
538 				if (ret == -1) {
539 					RTE_LOG(INFO, VHOST_CONFIG, "Invalid argument for rx-retry [0|1]\n");
540 					us_vhost_usage(prgname);
541 					return -1;
542 				} else {
543 					enable_retry = ret;
544 				}
545 			}
546 
547 			/* Enable/disable TX checksum offload. */
548 			if (!strncmp(long_option[option_index].name, "tx-csum", MAX_LONG_OPT_SZ)) {
549 				ret = parse_num_opt(optarg, 1);
550 				if (ret == -1) {
551 					RTE_LOG(INFO, VHOST_CONFIG, "Invalid argument for tx-csum [0|1]\n");
552 					us_vhost_usage(prgname);
553 					return -1;
554 				} else
555 					enable_tx_csum = ret;
556 			}
557 
558 			/* Enable/disable TSO offload. */
559 			if (!strncmp(long_option[option_index].name, "tso", MAX_LONG_OPT_SZ)) {
560 				ret = parse_num_opt(optarg, 1);
561 				if (ret == -1) {
562 					RTE_LOG(INFO, VHOST_CONFIG, "Invalid argument for tso [0|1]\n");
563 					us_vhost_usage(prgname);
564 					return -1;
565 				} else
566 					enable_tso = ret;
567 			}
568 
569 			/* Specify the retries delay time (in useconds) on RX. */
570 			if (!strncmp(long_option[option_index].name, "rx-retry-delay", MAX_LONG_OPT_SZ)) {
571 				ret = parse_num_opt(optarg, INT32_MAX);
572 				if (ret == -1) {
573 					RTE_LOG(INFO, VHOST_CONFIG, "Invalid argument for rx-retry-delay [0-N]\n");
574 					us_vhost_usage(prgname);
575 					return -1;
576 				} else {
577 					burst_rx_delay_time = ret;
578 				}
579 			}
580 
581 			/* Specify the retries number on RX. */
582 			if (!strncmp(long_option[option_index].name, "rx-retry-num", MAX_LONG_OPT_SZ)) {
583 				ret = parse_num_opt(optarg, INT32_MAX);
584 				if (ret == -1) {
585 					RTE_LOG(INFO, VHOST_CONFIG, "Invalid argument for rx-retry-num [0-N]\n");
586 					us_vhost_usage(prgname);
587 					return -1;
588 				} else {
589 					burst_rx_retry_num = ret;
590 				}
591 			}
592 
593 			/* Enable/disable RX mergeable buffers. */
594 			if (!strncmp(long_option[option_index].name, "mergeable", MAX_LONG_OPT_SZ)) {
595 				ret = parse_num_opt(optarg, 1);
596 				if (ret == -1) {
597 					RTE_LOG(INFO, VHOST_CONFIG, "Invalid argument for mergeable [0|1]\n");
598 					us_vhost_usage(prgname);
599 					return -1;
600 				} else {
601 					mergeable = !!ret;
602 					if (ret) {
603 						vmdq_conf_default.rxmode.jumbo_frame = 1;
604 						vmdq_conf_default.rxmode.max_rx_pkt_len
605 							= JUMBO_FRAME_MAX_SIZE;
606 					}
607 				}
608 			}
609 
610 			/* Enable/disable RX VLAN strip on host. */
611 			if (!strncmp(long_option[option_index].name,
612 				"vlan-strip", MAX_LONG_OPT_SZ)) {
613 				ret = parse_num_opt(optarg, 1);
614 				if (ret == -1) {
615 					RTE_LOG(INFO, VHOST_CONFIG,
616 						"Invalid argument for VLAN strip [0|1]\n");
617 					us_vhost_usage(prgname);
618 					return -1;
619 				} else {
620 					vlan_strip = !!ret;
621 					vmdq_conf_default.rxmode.hw_vlan_strip =
622 						vlan_strip;
623 				}
624 			}
625 
626 			/* Enable/disable stats. */
627 			if (!strncmp(long_option[option_index].name, "stats", MAX_LONG_OPT_SZ)) {
628 				ret = parse_num_opt(optarg, INT32_MAX);
629 				if (ret == -1) {
630 					RTE_LOG(INFO, VHOST_CONFIG, "Invalid argument for stats [0..N]\n");
631 					us_vhost_usage(prgname);
632 					return -1;
633 				} else {
634 					enable_stats = ret;
635 				}
636 			}
637 
638 			/* Set character device basename. */
639 			if (!strncmp(long_option[option_index].name, "dev-basename", MAX_LONG_OPT_SZ)) {
640 				if (us_vhost_parse_basename(optarg) == -1) {
641 					RTE_LOG(INFO, VHOST_CONFIG, "Invalid argument for character device basename (Max %d characters)\n", MAX_BASENAME_SZ);
642 					us_vhost_usage(prgname);
643 					return -1;
644 				}
645 			}
646 
647 			break;
648 
649 			/* Invalid option - print options. */
650 		default:
651 			us_vhost_usage(prgname);
652 			return -1;
653 		}
654 	}
655 
656 	for (i = 0; i < RTE_MAX_ETHPORTS; i++) {
657 		if (enabled_port_mask & (1 << i))
658 			ports[num_ports++] = (uint8_t)i;
659 	}
660 
661 	if ((num_ports ==  0) || (num_ports > MAX_SUP_PORTS)) {
662 		RTE_LOG(INFO, VHOST_PORT, "Current enabled port number is %u,"
663 			"but only %u port can be enabled\n",num_ports, MAX_SUP_PORTS);
664 		return -1;
665 	}
666 
667 	return 0;
668 }
669 
670 /*
671  * Update the global var NUM_PORTS and array PORTS according to system ports number
672  * and return valid ports number
673  */
674 static unsigned check_ports_num(unsigned nb_ports)
675 {
676 	unsigned valid_num_ports = num_ports;
677 	unsigned portid;
678 
679 	if (num_ports > nb_ports) {
680 		RTE_LOG(INFO, VHOST_PORT, "\nSpecified port number(%u) exceeds total system port number(%u)\n",
681 			num_ports, nb_ports);
682 		num_ports = nb_ports;
683 	}
684 
685 	for (portid = 0; portid < num_ports; portid ++) {
686 		if (ports[portid] >= nb_ports) {
687 			RTE_LOG(INFO, VHOST_PORT, "\nSpecified port ID(%u) exceeds max system port ID(%u)\n",
688 				ports[portid], (nb_ports - 1));
689 			ports[portid] = INVALID_PORT_ID;
690 			valid_num_ports--;
691 		}
692 	}
693 	return valid_num_ports;
694 }
695 
696 static inline struct vhost_dev *__attribute__((always_inline))
697 find_vhost_dev(struct ether_addr *mac)
698 {
699 	struct vhost_dev *vdev;
700 
701 	TAILQ_FOREACH(vdev, &vhost_dev_list, next) {
702 		if (vdev->ready == DEVICE_RX &&
703 		    is_same_ether_addr(mac, &vdev->mac_address))
704 			return vdev;
705 	}
706 
707 	return NULL;
708 }
709 
710 /*
711  * This function learns the MAC address of the device and registers this along with a
712  * vlan tag to a VMDQ.
713  */
714 static int
715 link_vmdq(struct vhost_dev *vdev, struct rte_mbuf *m)
716 {
717 	struct ether_hdr *pkt_hdr;
718 	struct virtio_net *dev = vdev->dev;
719 	int i, ret;
720 
721 	/* Learn MAC address of guest device from packet */
722 	pkt_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
723 
724 	if (find_vhost_dev(&pkt_hdr->s_addr)) {
725 		RTE_LOG(ERR, VHOST_DATA,
726 			"Device (%" PRIu64 ") is using a registered MAC!\n",
727 			dev->device_fh);
728 		return -1;
729 	}
730 
731 	for (i = 0; i < ETHER_ADDR_LEN; i++)
732 		vdev->mac_address.addr_bytes[i] = pkt_hdr->s_addr.addr_bytes[i];
733 
734 	/* vlan_tag currently uses the device_id. */
735 	vdev->vlan_tag = vlan_tags[dev->device_fh];
736 
737 	/* Print out VMDQ registration info. */
738 	RTE_LOG(INFO, VHOST_DATA, "(%"PRIu64") MAC_ADDRESS %02x:%02x:%02x:%02x:%02x:%02x and VLAN_TAG %d registered\n",
739 		dev->device_fh,
740 		vdev->mac_address.addr_bytes[0], vdev->mac_address.addr_bytes[1],
741 		vdev->mac_address.addr_bytes[2], vdev->mac_address.addr_bytes[3],
742 		vdev->mac_address.addr_bytes[4], vdev->mac_address.addr_bytes[5],
743 		vdev->vlan_tag);
744 
745 	/* Register the MAC address. */
746 	ret = rte_eth_dev_mac_addr_add(ports[0], &vdev->mac_address,
747 				(uint32_t)dev->device_fh + vmdq_pool_base);
748 	if (ret)
749 		RTE_LOG(ERR, VHOST_DATA, "(%"PRIu64") Failed to add device MAC address to VMDQ\n",
750 					dev->device_fh);
751 
752 	/* Enable stripping of the vlan tag as we handle routing. */
753 	if (vlan_strip)
754 		rte_eth_dev_set_vlan_strip_on_queue(ports[0],
755 			(uint16_t)vdev->vmdq_rx_q, 1);
756 
757 	/* Set device as ready for RX. */
758 	vdev->ready = DEVICE_RX;
759 
760 	return 0;
761 }
762 
763 /*
764  * Removes MAC address and vlan tag from VMDQ. Ensures that nothing is adding buffers to the RX
765  * queue before disabling RX on the device.
766  */
767 static inline void
768 unlink_vmdq(struct vhost_dev *vdev)
769 {
770 	unsigned i = 0;
771 	unsigned rx_count;
772 	struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
773 
774 	if (vdev->ready == DEVICE_RX) {
775 		/*clear MAC and VLAN settings*/
776 		rte_eth_dev_mac_addr_remove(ports[0], &vdev->mac_address);
777 		for (i = 0; i < 6; i++)
778 			vdev->mac_address.addr_bytes[i] = 0;
779 
780 		vdev->vlan_tag = 0;
781 
782 		/*Clear out the receive buffers*/
783 		rx_count = rte_eth_rx_burst(ports[0],
784 					(uint16_t)vdev->vmdq_rx_q, pkts_burst, MAX_PKT_BURST);
785 
786 		while (rx_count) {
787 			for (i = 0; i < rx_count; i++)
788 				rte_pktmbuf_free(pkts_burst[i]);
789 
790 			rx_count = rte_eth_rx_burst(ports[0],
791 					(uint16_t)vdev->vmdq_rx_q, pkts_burst, MAX_PKT_BURST);
792 		}
793 
794 		vdev->ready = DEVICE_MAC_LEARNING;
795 	}
796 }
797 
798 static inline void __attribute__((always_inline))
799 virtio_xmit(struct virtio_net *dst_dev, struct virtio_net *src_dev,
800 	    struct rte_mbuf *m)
801 {
802 	uint16_t ret;
803 
804 	ret = rte_vhost_enqueue_burst(dst_dev, VIRTIO_RXQ, &m, 1);
805 	if (enable_stats) {
806 		rte_atomic64_inc(&dev_statistics[dst_dev->device_fh].rx_total_atomic);
807 		rte_atomic64_add(&dev_statistics[dst_dev->device_fh].rx_atomic, ret);
808 		dev_statistics[src_dev->device_fh].tx_total++;
809 		dev_statistics[src_dev->device_fh].tx += ret;
810 	}
811 }
812 
813 /*
814  * Check if the packet destination MAC address is for a local device. If so then put
815  * the packet on that devices RX queue. If not then return.
816  */
817 static inline int __attribute__((always_inline))
818 virtio_tx_local(struct vhost_dev *vdev, struct rte_mbuf *m)
819 {
820 	struct ether_hdr *pkt_hdr;
821 	struct vhost_dev *dst_vdev;
822 	uint64_t fh;
823 
824 	pkt_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
825 
826 	dst_vdev = find_vhost_dev(&pkt_hdr->d_addr);
827 	if (!dst_vdev)
828 		return -1;
829 
830 	fh = dst_vdev->dev->device_fh;
831 	if (fh == vdev->dev->device_fh) {
832 		RTE_LOG(DEBUG, VHOST_DATA,
833 			"(%" PRIu64 ") TX: src and dst MAC is same. "
834 			"Dropping packet.\n", fh);
835 		return 0;
836 	}
837 
838 	RTE_LOG(DEBUG, VHOST_DATA,
839 		"(%" PRIu64 ") TX: MAC address is local\n", fh);
840 
841 	if (unlikely(dst_vdev->remove)) {
842 		RTE_LOG(DEBUG, VHOST_DATA, "(%" PRIu64 ") "
843 			"Device is marked for removal\n", fh);
844 		return 0;
845 	}
846 
847 	virtio_xmit(dst_vdev->dev, vdev->dev, m);
848 	return 0;
849 }
850 
851 /*
852  * Check if the destination MAC of a packet is one local VM,
853  * and get its vlan tag, and offset if it is.
854  */
855 static inline int __attribute__((always_inline))
856 find_local_dest(struct virtio_net *dev, struct rte_mbuf *m,
857 	uint32_t *offset, uint16_t *vlan_tag)
858 {
859 	struct vhost_dev *dst_vdev;
860 	struct ether_hdr *pkt_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
861 
862 	dst_vdev = find_vhost_dev(&pkt_hdr->d_addr);
863 	if (!dst_vdev)
864 		return 0;
865 
866 	if (dst_vdev->dev->device_fh == dev->device_fh) {
867 		RTE_LOG(DEBUG, VHOST_DATA,
868 			"(%" PRIu64 ") TX: src and dst MAC is same. "
869 			" Dropping packet.\n", dst_vdev->dev->device_fh);
870 		return -1;
871 	}
872 
873 	/*
874 	 * HW vlan strip will reduce the packet length
875 	 * by minus length of vlan tag, so need restore
876 	 * the packet length by plus it.
877 	 */
878 	*offset  = VLAN_HLEN;
879 	*vlan_tag = vlan_tags[(uint16_t)dst_vdev->dev->device_fh];
880 
881 	RTE_LOG(DEBUG, VHOST_DATA,
882 		"(%" PRIu64 ") TX: pkt to local VM device id: (%" PRIu64 ") "
883 		"vlan tag: %u.\n",
884 		dev->device_fh, dst_vdev->dev->device_fh, *vlan_tag);
885 
886 	return 0;
887 }
888 
889 static uint16_t
890 get_psd_sum(void *l3_hdr, uint64_t ol_flags)
891 {
892 	if (ol_flags & PKT_TX_IPV4)
893 		return rte_ipv4_phdr_cksum(l3_hdr, ol_flags);
894 	else /* assume ethertype == ETHER_TYPE_IPv6 */
895 		return rte_ipv6_phdr_cksum(l3_hdr, ol_flags);
896 }
897 
898 static void virtio_tx_offload(struct rte_mbuf *m)
899 {
900 	void *l3_hdr;
901 	struct ipv4_hdr *ipv4_hdr = NULL;
902 	struct tcp_hdr *tcp_hdr = NULL;
903 	struct ether_hdr *eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
904 
905 	l3_hdr = (char *)eth_hdr + m->l2_len;
906 
907 	if (m->ol_flags & PKT_TX_IPV4) {
908 		ipv4_hdr = l3_hdr;
909 		ipv4_hdr->hdr_checksum = 0;
910 		m->ol_flags |= PKT_TX_IP_CKSUM;
911 	}
912 
913 	tcp_hdr = (struct tcp_hdr *)((char *)l3_hdr + m->l3_len);
914 	tcp_hdr->cksum = get_psd_sum(l3_hdr, m->ol_flags);
915 }
916 
917 /*
918  * This function routes the TX packet to the correct interface. This may be a local device
919  * or the physical port.
920  */
921 static inline void __attribute__((always_inline))
922 virtio_tx_route(struct vhost_dev *vdev, struct rte_mbuf *m, uint16_t vlan_tag)
923 {
924 	struct mbuf_table *tx_q;
925 	struct rte_mbuf **m_table;
926 	unsigned len, ret, offset = 0;
927 	const uint16_t lcore_id = rte_lcore_id();
928 	struct virtio_net *dev = vdev->dev;
929 	struct ether_hdr *nh;
930 
931 
932 	nh = rte_pktmbuf_mtod(m, struct ether_hdr *);
933 	if (unlikely(is_broadcast_ether_addr(&nh->d_addr))) {
934 		struct vhost_dev *vdev2;
935 
936 		TAILQ_FOREACH(vdev2, &vhost_dev_list, next) {
937 			virtio_xmit(vdev2->dev, vdev->dev, m);
938 		}
939 		goto queue2nic;
940 	}
941 
942 	/*check if destination is local VM*/
943 	if ((vm2vm_mode == VM2VM_SOFTWARE) && (virtio_tx_local(vdev, m) == 0)) {
944 		rte_pktmbuf_free(m);
945 		return;
946 	}
947 
948 	if (unlikely(vm2vm_mode == VM2VM_HARDWARE)) {
949 		if (unlikely(find_local_dest(dev, m, &offset, &vlan_tag) != 0)) {
950 			rte_pktmbuf_free(m);
951 			return;
952 		}
953 	}
954 
955 	RTE_LOG(DEBUG, VHOST_DATA, "(%" PRIu64 ") TX: "
956 		"MAC address is external\n", dev->device_fh);
957 
958 queue2nic:
959 
960 	/*Add packet to the port tx queue*/
961 	tx_q = &lcore_tx_queue[lcore_id];
962 	len = tx_q->len;
963 
964 	nh = rte_pktmbuf_mtod(m, struct ether_hdr *);
965 	if (unlikely(nh->ether_type == rte_cpu_to_be_16(ETHER_TYPE_VLAN))) {
966 		/* Guest has inserted the vlan tag. */
967 		struct vlan_hdr *vh = (struct vlan_hdr *) (nh + 1);
968 		uint16_t vlan_tag_be = rte_cpu_to_be_16(vlan_tag);
969 		if ((vm2vm_mode == VM2VM_HARDWARE) &&
970 			(vh->vlan_tci != vlan_tag_be))
971 			vh->vlan_tci = vlan_tag_be;
972 	} else {
973 		m->ol_flags |= PKT_TX_VLAN_PKT;
974 
975 		/*
976 		 * Find the right seg to adjust the data len when offset is
977 		 * bigger than tail room size.
978 		 */
979 		if (unlikely(vm2vm_mode == VM2VM_HARDWARE)) {
980 			if (likely(offset <= rte_pktmbuf_tailroom(m)))
981 				m->data_len += offset;
982 			else {
983 				struct rte_mbuf *seg = m;
984 
985 				while ((seg->next != NULL) &&
986 					(offset > rte_pktmbuf_tailroom(seg)))
987 					seg = seg->next;
988 
989 				seg->data_len += offset;
990 			}
991 			m->pkt_len += offset;
992 		}
993 
994 		m->vlan_tci = vlan_tag;
995 	}
996 
997 	if (m->ol_flags & PKT_TX_TCP_SEG)
998 		virtio_tx_offload(m);
999 
1000 	tx_q->m_table[len] = m;
1001 	len++;
1002 	if (enable_stats) {
1003 		dev_statistics[dev->device_fh].tx_total++;
1004 		dev_statistics[dev->device_fh].tx++;
1005 	}
1006 
1007 	if (unlikely(len == MAX_PKT_BURST)) {
1008 		m_table = (struct rte_mbuf **)tx_q->m_table;
1009 		ret = rte_eth_tx_burst(ports[0], (uint16_t)tx_q->txq_id, m_table, (uint16_t) len);
1010 		/* Free any buffers not handled by TX and update the port stats. */
1011 		if (unlikely(ret < len)) {
1012 			do {
1013 				rte_pktmbuf_free(m_table[ret]);
1014 			} while (++ret < len);
1015 		}
1016 
1017 		len = 0;
1018 	}
1019 
1020 	tx_q->len = len;
1021 	return;
1022 }
1023 /*
1024  * This function is called by each data core. It handles all RX/TX registered with the
1025  * core. For TX the specific lcore linked list is used. For RX, MAC addresses are compared
1026  * with all devices in the main linked list.
1027  */
1028 static int
1029 switch_worker(__attribute__((unused)) void *arg)
1030 {
1031 	struct virtio_net *dev = NULL;
1032 	struct vhost_dev *vdev = NULL;
1033 	struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
1034 	struct mbuf_table *tx_q;
1035 	const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
1036 	uint64_t prev_tsc, diff_tsc, cur_tsc, ret_count = 0;
1037 	unsigned ret, i;
1038 	const uint16_t lcore_id = rte_lcore_id();
1039 	const uint16_t num_cores = (uint16_t)rte_lcore_count();
1040 	uint16_t rx_count = 0;
1041 	uint16_t tx_count;
1042 	uint32_t retry = 0;
1043 
1044 	RTE_LOG(INFO, VHOST_DATA, "Procesing on Core %u started\n", lcore_id);
1045 	prev_tsc = 0;
1046 
1047 	tx_q = &lcore_tx_queue[lcore_id];
1048 	for (i = 0; i < num_cores; i ++) {
1049 		if (lcore_ids[i] == lcore_id) {
1050 			tx_q->txq_id = i;
1051 			break;
1052 		}
1053 	}
1054 
1055 	while(1) {
1056 		cur_tsc = rte_rdtsc();
1057 		/*
1058 		 * TX burst queue drain
1059 		 */
1060 		diff_tsc = cur_tsc - prev_tsc;
1061 		if (unlikely(diff_tsc > drain_tsc)) {
1062 
1063 			if (tx_q->len) {
1064 				RTE_LOG(DEBUG, VHOST_DATA,
1065 					"TX queue drained after timeout with burst size %u\n",
1066 					tx_q->len);
1067 
1068 				/*Tx any packets in the queue*/
1069 				ret = rte_eth_tx_burst(ports[0], (uint16_t)tx_q->txq_id,
1070 									   (struct rte_mbuf **)tx_q->m_table,
1071 									   (uint16_t)tx_q->len);
1072 				if (unlikely(ret < tx_q->len)) {
1073 					do {
1074 						rte_pktmbuf_free(tx_q->m_table[ret]);
1075 					} while (++ret < tx_q->len);
1076 				}
1077 
1078 				tx_q->len = 0;
1079 			}
1080 
1081 			prev_tsc = cur_tsc;
1082 
1083 		}
1084 
1085 		/*
1086 		 * Inform the configuration core that we have exited the
1087 		 * linked list and that no devices are in use if requested.
1088 		 */
1089 		if (lcore_info[lcore_id].dev_removal_flag == REQUEST_DEV_REMOVAL)
1090 			lcore_info[lcore_id].dev_removal_flag = ACK_DEV_REMOVAL;
1091 
1092 		/*
1093 		 * Process devices
1094 		 */
1095 		TAILQ_FOREACH(vdev, &lcore_info[lcore_id].vdev_list, next) {
1096 			uint64_t fh;
1097 
1098 			dev = vdev->dev;
1099 			fh  = dev->device_fh;
1100 
1101 			if (unlikely(vdev->remove)) {
1102 				unlink_vmdq(vdev);
1103 				vdev->ready = DEVICE_SAFE_REMOVE;
1104 				continue;
1105 			}
1106 
1107 			if (likely(vdev->ready == DEVICE_RX)) {
1108 				/*Handle guest RX*/
1109 				rx_count = rte_eth_rx_burst(ports[0],
1110 					vdev->vmdq_rx_q, pkts_burst, MAX_PKT_BURST);
1111 
1112 				if (rx_count) {
1113 					/*
1114 					* Retry is enabled and the queue is full then we wait and retry to avoid packet loss
1115 					* Here MAX_PKT_BURST must be less than virtio queue size
1116 					*/
1117 					if (enable_retry && unlikely(rx_count > rte_vring_available_entries(dev, VIRTIO_RXQ))) {
1118 						for (retry = 0; retry < burst_rx_retry_num; retry++) {
1119 							rte_delay_us(burst_rx_delay_time);
1120 							if (rx_count <= rte_vring_available_entries(dev, VIRTIO_RXQ))
1121 								break;
1122 						}
1123 					}
1124 					ret_count = rte_vhost_enqueue_burst(dev, VIRTIO_RXQ, pkts_burst, rx_count);
1125 					if (enable_stats) {
1126 						rte_atomic64_add(
1127 							&dev_statistics[fh].rx_total_atomic,
1128 							rx_count);
1129 						rte_atomic64_add(
1130 							&dev_statistics[fh].rx_atomic,
1131 							ret_count);
1132 					}
1133 					while (likely(rx_count)) {
1134 						rx_count--;
1135 						rte_pktmbuf_free(pkts_burst[rx_count]);
1136 					}
1137 
1138 				}
1139 			}
1140 
1141 			if (likely(!vdev->remove)) {
1142 				/* Handle guest TX*/
1143 				tx_count = rte_vhost_dequeue_burst(dev, VIRTIO_TXQ, mbuf_pool, pkts_burst, MAX_PKT_BURST);
1144 				/* If this is the first received packet we need to learn the MAC and setup VMDQ */
1145 				if (unlikely(vdev->ready == DEVICE_MAC_LEARNING) && tx_count) {
1146 					if (vdev->remove || (link_vmdq(vdev, pkts_burst[0]) == -1)) {
1147 						while (tx_count)
1148 							rte_pktmbuf_free(pkts_burst[--tx_count]);
1149 					}
1150 				}
1151 				for (i = 0; i < tx_count; ++i) {
1152 					virtio_tx_route(vdev, pkts_burst[i],
1153 						vlan_tags[(uint16_t)dev->device_fh]);
1154 				}
1155 			}
1156 		}
1157 	}
1158 
1159 	return 0;
1160 }
1161 
1162 /*
1163  * Remove a device from the specific data core linked list and from the
1164  * main linked list. Synchonization  occurs through the use of the
1165  * lcore dev_removal_flag. Device is made volatile here to avoid re-ordering
1166  * of dev->remove=1 which can cause an infinite loop in the rte_pause loop.
1167  */
1168 static void
1169 destroy_device (volatile struct virtio_net *dev)
1170 {
1171 	struct vhost_dev *vdev;
1172 	int lcore;
1173 
1174 	dev->flags &= ~VIRTIO_DEV_RUNNING;
1175 
1176 	vdev = (struct vhost_dev *)dev->priv;
1177 	/*set the remove flag. */
1178 	vdev->remove = 1;
1179 	while(vdev->ready != DEVICE_SAFE_REMOVE) {
1180 		rte_pause();
1181 	}
1182 
1183 	TAILQ_REMOVE(&lcore_info[vdev->coreid].vdev_list, vdev, next);
1184 	TAILQ_REMOVE(&vhost_dev_list, vdev, next);
1185 
1186 	/* Set the dev_removal_flag on each lcore. */
1187 	RTE_LCORE_FOREACH_SLAVE(lcore)
1188 		lcore_info[lcore].dev_removal_flag = REQUEST_DEV_REMOVAL;
1189 
1190 	/*
1191 	 * Once each core has set the dev_removal_flag to ACK_DEV_REMOVAL
1192 	 * we can be sure that they can no longer access the device removed
1193 	 * from the linked lists and that the devices are no longer in use.
1194 	 */
1195 	RTE_LCORE_FOREACH_SLAVE(lcore) {
1196 		while (lcore_info[lcore].dev_removal_flag != ACK_DEV_REMOVAL)
1197 			rte_pause();
1198 	}
1199 
1200 	lcore_info[vdev->coreid].device_num--;
1201 
1202 	RTE_LOG(INFO, VHOST_DATA,
1203 		"(%" PRIu64 ") Device has been removed from data core\n",
1204 		dev->device_fh);
1205 
1206 	rte_free(vdev);
1207 }
1208 
1209 /*
1210  * A new device is added to a data core. First the device is added to the main linked list
1211  * and the allocated to a specific data core.
1212  */
1213 static int
1214 new_device (struct virtio_net *dev)
1215 {
1216 	int lcore, core_add = 0;
1217 	uint32_t device_num_min = num_devices;
1218 	struct vhost_dev *vdev;
1219 
1220 	vdev = rte_zmalloc("vhost device", sizeof(*vdev), RTE_CACHE_LINE_SIZE);
1221 	if (vdev == NULL) {
1222 		RTE_LOG(INFO, VHOST_DATA, "(%"PRIu64") Couldn't allocate memory for vhost dev\n",
1223 			dev->device_fh);
1224 		return -1;
1225 	}
1226 	vdev->dev = dev;
1227 	dev->priv = vdev;
1228 
1229 	TAILQ_INSERT_TAIL(&vhost_dev_list, vdev, next);
1230 	vdev->vmdq_rx_q
1231 		= dev->device_fh * queues_per_pool + vmdq_queue_base;
1232 
1233 	/*reset ready flag*/
1234 	vdev->ready = DEVICE_MAC_LEARNING;
1235 	vdev->remove = 0;
1236 
1237 	/* Find a suitable lcore to add the device. */
1238 	RTE_LCORE_FOREACH_SLAVE(lcore) {
1239 		if (lcore_info[lcore].device_num < device_num_min) {
1240 			device_num_min = lcore_info[lcore].device_num;
1241 			core_add = lcore;
1242 		}
1243 	}
1244 	vdev->coreid = core_add;
1245 
1246 	TAILQ_INSERT_TAIL(&lcore_info[vdev->coreid].vdev_list, vdev, next);
1247 	lcore_info[vdev->coreid].device_num++;
1248 
1249 	/* Initialize device stats */
1250 	memset(&dev_statistics[dev->device_fh], 0, sizeof(struct device_statistics));
1251 
1252 	/* Disable notifications. */
1253 	rte_vhost_enable_guest_notification(dev, VIRTIO_RXQ, 0);
1254 	rte_vhost_enable_guest_notification(dev, VIRTIO_TXQ, 0);
1255 	dev->flags |= VIRTIO_DEV_RUNNING;
1256 
1257 	RTE_LOG(INFO, VHOST_DATA, "(%"PRIu64") Device has been added to data core %d\n", dev->device_fh, vdev->coreid);
1258 
1259 	return 0;
1260 }
1261 
1262 /*
1263  * These callback allow devices to be added to the data core when configuration
1264  * has been fully complete.
1265  */
1266 static const struct virtio_net_device_ops virtio_net_device_ops =
1267 {
1268 	.new_device =  new_device,
1269 	.destroy_device = destroy_device,
1270 };
1271 
1272 /*
1273  * This is a thread will wake up after a period to print stats if the user has
1274  * enabled them.
1275  */
1276 static void
1277 print_stats(void)
1278 {
1279 	struct vhost_dev *vdev;
1280 	uint64_t tx_dropped, rx_dropped;
1281 	uint64_t tx, tx_total, rx, rx_total;
1282 	uint32_t device_fh;
1283 	const char clr[] = { 27, '[', '2', 'J', '\0' };
1284 	const char top_left[] = { 27, '[', '1', ';', '1', 'H','\0' };
1285 
1286 	while(1) {
1287 		sleep(enable_stats);
1288 
1289 		/* Clear screen and move to top left */
1290 		printf("%s%s", clr, top_left);
1291 
1292 		printf("\nDevice statistics ====================================");
1293 
1294 		TAILQ_FOREACH(vdev, &vhost_dev_list, next) {
1295 			device_fh = vdev->dev->device_fh;
1296 			tx_total = dev_statistics[device_fh].tx_total;
1297 			tx = dev_statistics[device_fh].tx;
1298 			tx_dropped = tx_total - tx;
1299 			rx_total = rte_atomic64_read(
1300 				&dev_statistics[device_fh].rx_total_atomic);
1301 			rx = rte_atomic64_read(
1302 				&dev_statistics[device_fh].rx_atomic);
1303 			rx_dropped = rx_total - rx;
1304 
1305 			printf("\nStatistics for device %"PRIu32" ------------------------------"
1306 					"\nTX total: 		%"PRIu64""
1307 					"\nTX dropped: 		%"PRIu64""
1308 					"\nTX successful: 		%"PRIu64""
1309 					"\nRX total: 		%"PRIu64""
1310 					"\nRX dropped: 		%"PRIu64""
1311 					"\nRX successful: 		%"PRIu64"",
1312 					device_fh,
1313 					tx_total,
1314 					tx_dropped,
1315 					tx,
1316 					rx_total,
1317 					rx_dropped,
1318 					rx);
1319 		}
1320 		printf("\n======================================================\n");
1321 	}
1322 }
1323 
1324 /* When we receive a INT signal, unregister vhost driver */
1325 static void
1326 sigint_handler(__rte_unused int signum)
1327 {
1328 	/* Unregister vhost driver. */
1329 	int ret = rte_vhost_driver_unregister((char *)&dev_basename);
1330 	if (ret != 0)
1331 		rte_exit(EXIT_FAILURE, "vhost driver unregister failure.\n");
1332 	exit(0);
1333 }
1334 
1335 /*
1336  * While creating an mbuf pool, one key thing is to figure out how
1337  * many mbuf entries is enough for our use. FYI, here are some
1338  * guidelines:
1339  *
1340  * - Each rx queue would reserve @nr_rx_desc mbufs at queue setup stage
1341  *
1342  * - For each switch core (A CPU core does the packet switch), we need
1343  *   also make some reservation for receiving the packets from virtio
1344  *   Tx queue. How many is enough depends on the usage. It's normally
1345  *   a simple calculation like following:
1346  *
1347  *       MAX_PKT_BURST * max packet size / mbuf size
1348  *
1349  *   So, we definitely need allocate more mbufs when TSO is enabled.
1350  *
1351  * - Similarly, for each switching core, we should serve @nr_rx_desc
1352  *   mbufs for receiving the packets from physical NIC device.
1353  *
1354  * - We also need make sure, for each switch core, we have allocated
1355  *   enough mbufs to fill up the mbuf cache.
1356  */
1357 static void
1358 create_mbuf_pool(uint16_t nr_port, uint32_t nr_switch_core, uint32_t mbuf_size,
1359 	uint32_t nr_queues, uint32_t nr_rx_desc, uint32_t nr_mbuf_cache)
1360 {
1361 	uint32_t nr_mbufs;
1362 	uint32_t nr_mbufs_per_core;
1363 	uint32_t mtu = 1500;
1364 
1365 	if (mergeable)
1366 		mtu = 9000;
1367 	if (enable_tso)
1368 		mtu = 64 * 1024;
1369 
1370 	nr_mbufs_per_core  = (mtu + mbuf_size) * MAX_PKT_BURST /
1371 			(mbuf_size - RTE_PKTMBUF_HEADROOM) * MAX_PKT_BURST;
1372 	nr_mbufs_per_core += nr_rx_desc;
1373 	nr_mbufs_per_core  = RTE_MAX(nr_mbufs_per_core, nr_mbuf_cache);
1374 
1375 	nr_mbufs  = nr_queues * nr_rx_desc;
1376 	nr_mbufs += nr_mbufs_per_core * nr_switch_core;
1377 	nr_mbufs *= nr_port;
1378 
1379 	mbuf_pool = rte_pktmbuf_pool_create("MBUF_POOL", nr_mbufs,
1380 					    nr_mbuf_cache, 0, mbuf_size,
1381 					    rte_socket_id());
1382 	if (mbuf_pool == NULL)
1383 		rte_exit(EXIT_FAILURE, "Cannot create mbuf pool\n");
1384 }
1385 
1386 /*
1387  * Main function, does initialisation and calls the per-lcore functions. The CUSE
1388  * device is also registered here to handle the IOCTLs.
1389  */
1390 int
1391 main(int argc, char *argv[])
1392 {
1393 	unsigned lcore_id, core_id = 0;
1394 	unsigned nb_ports, valid_num_ports;
1395 	int ret;
1396 	uint8_t portid;
1397 	static pthread_t tid;
1398 	char thread_name[RTE_MAX_THREAD_NAME_LEN];
1399 
1400 	signal(SIGINT, sigint_handler);
1401 
1402 	/* init EAL */
1403 	ret = rte_eal_init(argc, argv);
1404 	if (ret < 0)
1405 		rte_exit(EXIT_FAILURE, "Error with EAL initialization\n");
1406 	argc -= ret;
1407 	argv += ret;
1408 
1409 	/* parse app arguments */
1410 	ret = us_vhost_parse_args(argc, argv);
1411 	if (ret < 0)
1412 		rte_exit(EXIT_FAILURE, "Invalid argument\n");
1413 
1414 	for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id ++)
1415 		TAILQ_INIT(&lcore_info[lcore_id].vdev_list);
1416 
1417 		if (rte_lcore_is_enabled(lcore_id))
1418 			lcore_ids[core_id ++] = lcore_id;
1419 
1420 	if (rte_lcore_count() > RTE_MAX_LCORE)
1421 		rte_exit(EXIT_FAILURE,"Not enough cores\n");
1422 
1423 	/* Get the number of physical ports. */
1424 	nb_ports = rte_eth_dev_count();
1425 	if (nb_ports > RTE_MAX_ETHPORTS)
1426 		nb_ports = RTE_MAX_ETHPORTS;
1427 
1428 	/*
1429 	 * Update the global var NUM_PORTS and global array PORTS
1430 	 * and get value of var VALID_NUM_PORTS according to system ports number
1431 	 */
1432 	valid_num_ports = check_ports_num(nb_ports);
1433 
1434 	if ((valid_num_ports ==  0) || (valid_num_ports > MAX_SUP_PORTS)) {
1435 		RTE_LOG(INFO, VHOST_PORT, "Current enabled port number is %u,"
1436 			"but only %u port can be enabled\n",num_ports, MAX_SUP_PORTS);
1437 		return -1;
1438 	}
1439 
1440 	/*
1441 	 * FIXME: here we are trying to allocate mbufs big enough for
1442 	 * @MAX_QUEUES, but the truth is we're never going to use that
1443 	 * many queues here. We probably should only do allocation for
1444 	 * those queues we are going to use.
1445 	 */
1446 	create_mbuf_pool(valid_num_ports, rte_lcore_count() - 1, MBUF_DATA_SIZE,
1447 			 MAX_QUEUES, RTE_TEST_RX_DESC_DEFAULT, MBUF_CACHE_SIZE);
1448 
1449 	if (vm2vm_mode == VM2VM_HARDWARE) {
1450 		/* Enable VT loop back to let L2 switch to do it. */
1451 		vmdq_conf_default.rx_adv_conf.vmdq_rx_conf.enable_loop_back = 1;
1452 		RTE_LOG(DEBUG, VHOST_CONFIG,
1453 			"Enable loop back for L2 switch in vmdq.\n");
1454 	}
1455 
1456 	/* initialize all ports */
1457 	for (portid = 0; portid < nb_ports; portid++) {
1458 		/* skip ports that are not enabled */
1459 		if ((enabled_port_mask & (1 << portid)) == 0) {
1460 			RTE_LOG(INFO, VHOST_PORT,
1461 				"Skipping disabled port %d\n", portid);
1462 			continue;
1463 		}
1464 		if (port_init(portid) != 0)
1465 			rte_exit(EXIT_FAILURE,
1466 				"Cannot initialize network ports\n");
1467 	}
1468 
1469 	/* Initialize device stats */
1470 	memset(&dev_statistics, 0, sizeof(dev_statistics));
1471 
1472 	/* Enable stats if the user option is set. */
1473 	if (enable_stats) {
1474 		ret = pthread_create(&tid, NULL, (void *)print_stats, NULL);
1475 		if (ret != 0)
1476 			rte_exit(EXIT_FAILURE,
1477 				"Cannot create print-stats thread\n");
1478 
1479 		/* Set thread_name for aid in debugging.  */
1480 		snprintf(thread_name, RTE_MAX_THREAD_NAME_LEN, "print-stats");
1481 		ret = rte_thread_setname(tid, thread_name);
1482 		if (ret != 0)
1483 			RTE_LOG(ERR, VHOST_CONFIG,
1484 				"Cannot set print-stats name\n");
1485 	}
1486 
1487 	/* Launch all data cores. */
1488 	RTE_LCORE_FOREACH_SLAVE(lcore_id)
1489 		rte_eal_remote_launch(switch_worker, NULL, lcore_id);
1490 
1491 	if (mergeable == 0)
1492 		rte_vhost_feature_disable(1ULL << VIRTIO_NET_F_MRG_RXBUF);
1493 
1494 	/* Register vhost(cuse or user) driver to handle vhost messages. */
1495 	ret = rte_vhost_driver_register((char *)&dev_basename);
1496 	if (ret != 0)
1497 		rte_exit(EXIT_FAILURE, "vhost driver register failure.\n");
1498 
1499 	rte_vhost_driver_callback_register(&virtio_net_device_ops);
1500 
1501 	/* Start CUSE session. */
1502 	rte_vhost_driver_session_start();
1503 	return 0;
1504 
1505 }
1506