xref: /dpdk/examples/vhost/main.c (revision 16ae8abe1cb8b545d4ca5cdee0e3db32c6b8b498)
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 MBUF_TABLE_DRAIN_TSC	((rte_get_tsc_hz() + US_PER_S - 1) \
216 				 / US_PER_S * BURST_TX_DRAIN_US)
217 #define VLAN_HLEN       4
218 
219 /*
220  * Builds up the correct configuration for VMDQ VLAN pool map
221  * according to the pool & queue limits.
222  */
223 static inline int
224 get_eth_conf(struct rte_eth_conf *eth_conf, uint32_t num_devices)
225 {
226 	struct rte_eth_vmdq_rx_conf conf;
227 	struct rte_eth_vmdq_rx_conf *def_conf =
228 		&vmdq_conf_default.rx_adv_conf.vmdq_rx_conf;
229 	unsigned i;
230 
231 	memset(&conf, 0, sizeof(conf));
232 	conf.nb_queue_pools = (enum rte_eth_nb_pools)num_devices;
233 	conf.nb_pool_maps = num_devices;
234 	conf.enable_loop_back = def_conf->enable_loop_back;
235 	conf.rx_mode = def_conf->rx_mode;
236 
237 	for (i = 0; i < conf.nb_pool_maps; i++) {
238 		conf.pool_map[i].vlan_id = vlan_tags[ i ];
239 		conf.pool_map[i].pools = (1UL << i);
240 	}
241 
242 	(void)(rte_memcpy(eth_conf, &vmdq_conf_default, sizeof(*eth_conf)));
243 	(void)(rte_memcpy(&eth_conf->rx_adv_conf.vmdq_rx_conf, &conf,
244 		   sizeof(eth_conf->rx_adv_conf.vmdq_rx_conf)));
245 	return 0;
246 }
247 
248 /*
249  * Validate the device number according to the max pool number gotten form
250  * dev_info. If the device number is invalid, give the error message and
251  * return -1. Each device must have its own pool.
252  */
253 static inline int
254 validate_num_devices(uint32_t max_nb_devices)
255 {
256 	if (num_devices > max_nb_devices) {
257 		RTE_LOG(ERR, VHOST_PORT, "invalid number of devices\n");
258 		return -1;
259 	}
260 	return 0;
261 }
262 
263 /*
264  * Initialises a given port using global settings and with the rx buffers
265  * coming from the mbuf_pool passed as parameter
266  */
267 static inline int
268 port_init(uint8_t port)
269 {
270 	struct rte_eth_dev_info dev_info;
271 	struct rte_eth_conf port_conf;
272 	struct rte_eth_rxconf *rxconf;
273 	struct rte_eth_txconf *txconf;
274 	int16_t rx_rings, tx_rings;
275 	uint16_t rx_ring_size, tx_ring_size;
276 	int retval;
277 	uint16_t q;
278 
279 	/* The max pool number from dev_info will be used to validate the pool number specified in cmd line */
280 	rte_eth_dev_info_get (port, &dev_info);
281 
282 	if (dev_info.max_rx_queues > MAX_QUEUES) {
283 		rte_exit(EXIT_FAILURE,
284 			"please define MAX_QUEUES no less than %u in %s\n",
285 			dev_info.max_rx_queues, __FILE__);
286 	}
287 
288 	rxconf = &dev_info.default_rxconf;
289 	txconf = &dev_info.default_txconf;
290 	rxconf->rx_drop_en = 1;
291 
292 	/* Enable vlan offload */
293 	txconf->txq_flags &= ~ETH_TXQ_FLAGS_NOVLANOFFL;
294 
295 	/*configure the number of supported virtio devices based on VMDQ limits */
296 	num_devices = dev_info.max_vmdq_pools;
297 
298 	rx_ring_size = RTE_TEST_RX_DESC_DEFAULT;
299 	tx_ring_size = RTE_TEST_TX_DESC_DEFAULT;
300 	tx_rings = (uint16_t)rte_lcore_count();
301 
302 	retval = validate_num_devices(MAX_DEVICES);
303 	if (retval < 0)
304 		return retval;
305 
306 	/* Get port configuration. */
307 	retval = get_eth_conf(&port_conf, num_devices);
308 	if (retval < 0)
309 		return retval;
310 	/* NIC queues are divided into pf queues and vmdq queues.  */
311 	num_pf_queues = dev_info.max_rx_queues - dev_info.vmdq_queue_num;
312 	queues_per_pool = dev_info.vmdq_queue_num / dev_info.max_vmdq_pools;
313 	num_vmdq_queues = num_devices * queues_per_pool;
314 	num_queues = num_pf_queues + num_vmdq_queues;
315 	vmdq_queue_base = dev_info.vmdq_queue_base;
316 	vmdq_pool_base  = dev_info.vmdq_pool_base;
317 	printf("pf queue num: %u, configured vmdq pool num: %u, each vmdq pool has %u queues\n",
318 		num_pf_queues, num_devices, queues_per_pool);
319 
320 	if (port >= rte_eth_dev_count()) return -1;
321 
322 	if (enable_tx_csum == 0)
323 		rte_vhost_feature_disable(1ULL << VIRTIO_NET_F_CSUM);
324 
325 	if (enable_tso == 0) {
326 		rte_vhost_feature_disable(1ULL << VIRTIO_NET_F_HOST_TSO4);
327 		rte_vhost_feature_disable(1ULL << VIRTIO_NET_F_HOST_TSO6);
328 	}
329 
330 	rx_rings = (uint16_t)dev_info.max_rx_queues;
331 	/* Configure ethernet device. */
332 	retval = rte_eth_dev_configure(port, rx_rings, tx_rings, &port_conf);
333 	if (retval != 0)
334 		return retval;
335 
336 	/* Setup the queues. */
337 	for (q = 0; q < rx_rings; q ++) {
338 		retval = rte_eth_rx_queue_setup(port, q, rx_ring_size,
339 						rte_eth_dev_socket_id(port),
340 						rxconf,
341 						mbuf_pool);
342 		if (retval < 0)
343 			return retval;
344 	}
345 	for (q = 0; q < tx_rings; q ++) {
346 		retval = rte_eth_tx_queue_setup(port, q, tx_ring_size,
347 						rte_eth_dev_socket_id(port),
348 						txconf);
349 		if (retval < 0)
350 			return retval;
351 	}
352 
353 	/* Start the device. */
354 	retval  = rte_eth_dev_start(port);
355 	if (retval < 0) {
356 		RTE_LOG(ERR, VHOST_DATA, "Failed to start the device.\n");
357 		return retval;
358 	}
359 
360 	if (promiscuous)
361 		rte_eth_promiscuous_enable(port);
362 
363 	rte_eth_macaddr_get(port, &vmdq_ports_eth_addr[port]);
364 	RTE_LOG(INFO, VHOST_PORT, "Max virtio devices supported: %u\n", num_devices);
365 	RTE_LOG(INFO, VHOST_PORT, "Port %u MAC: %02"PRIx8" %02"PRIx8" %02"PRIx8
366 			" %02"PRIx8" %02"PRIx8" %02"PRIx8"\n",
367 			(unsigned)port,
368 			vmdq_ports_eth_addr[port].addr_bytes[0],
369 			vmdq_ports_eth_addr[port].addr_bytes[1],
370 			vmdq_ports_eth_addr[port].addr_bytes[2],
371 			vmdq_ports_eth_addr[port].addr_bytes[3],
372 			vmdq_ports_eth_addr[port].addr_bytes[4],
373 			vmdq_ports_eth_addr[port].addr_bytes[5]);
374 
375 	return 0;
376 }
377 
378 /*
379  * Set character device basename.
380  */
381 static int
382 us_vhost_parse_basename(const char *q_arg)
383 {
384 	/* parse number string */
385 
386 	if (strnlen(q_arg, MAX_BASENAME_SZ) > MAX_BASENAME_SZ)
387 		return -1;
388 	else
389 		snprintf((char*)&dev_basename, MAX_BASENAME_SZ, "%s", q_arg);
390 
391 	return 0;
392 }
393 
394 /*
395  * Parse the portmask provided at run time.
396  */
397 static int
398 parse_portmask(const char *portmask)
399 {
400 	char *end = NULL;
401 	unsigned long pm;
402 
403 	errno = 0;
404 
405 	/* parse hexadecimal string */
406 	pm = strtoul(portmask, &end, 16);
407 	if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0') || (errno != 0))
408 		return -1;
409 
410 	if (pm == 0)
411 		return -1;
412 
413 	return pm;
414 
415 }
416 
417 /*
418  * Parse num options at run time.
419  */
420 static int
421 parse_num_opt(const char *q_arg, uint32_t max_valid_value)
422 {
423 	char *end = NULL;
424 	unsigned long num;
425 
426 	errno = 0;
427 
428 	/* parse unsigned int string */
429 	num = strtoul(q_arg, &end, 10);
430 	if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0') || (errno != 0))
431 		return -1;
432 
433 	if (num > max_valid_value)
434 		return -1;
435 
436 	return num;
437 
438 }
439 
440 /*
441  * Display usage
442  */
443 static void
444 us_vhost_usage(const char *prgname)
445 {
446 	RTE_LOG(INFO, VHOST_CONFIG, "%s [EAL options] -- -p PORTMASK\n"
447 	"		--vm2vm [0|1|2]\n"
448 	"		--rx_retry [0|1] --mergeable [0|1] --stats [0-N]\n"
449 	"		--dev-basename <name>\n"
450 	"		--nb-devices ND\n"
451 	"		-p PORTMASK: Set mask for ports to be used by application\n"
452 	"		--vm2vm [0|1|2]: disable/software(default)/hardware vm2vm comms\n"
453 	"		--rx-retry [0|1]: disable/enable(default) retries on rx. Enable retry if destintation queue is full\n"
454 	"		--rx-retry-delay [0-N]: timeout(in usecond) between retries on RX. This makes effect only if retries on rx enabled\n"
455 	"		--rx-retry-num [0-N]: the number of retries on rx. This makes effect only if retries on rx enabled\n"
456 	"		--mergeable [0|1]: disable(default)/enable RX mergeable buffers\n"
457 	"		--vlan-strip [0|1]: disable/enable(default) RX VLAN strip on host\n"
458 	"		--stats [0-N]: 0: Disable stats, N: Time in seconds to print stats\n"
459 	"		--dev-basename: The basename to be used for the character device.\n"
460 	"		--tx-csum [0|1] disable/enable TX checksum offload.\n"
461 	"		--tso [0|1] disable/enable TCP segment offload.\n",
462 	       prgname);
463 }
464 
465 /*
466  * Parse the arguments given in the command line of the application.
467  */
468 static int
469 us_vhost_parse_args(int argc, char **argv)
470 {
471 	int opt, ret;
472 	int option_index;
473 	unsigned i;
474 	const char *prgname = argv[0];
475 	static struct option long_option[] = {
476 		{"vm2vm", required_argument, NULL, 0},
477 		{"rx-retry", required_argument, NULL, 0},
478 		{"rx-retry-delay", required_argument, NULL, 0},
479 		{"rx-retry-num", required_argument, NULL, 0},
480 		{"mergeable", required_argument, NULL, 0},
481 		{"vlan-strip", required_argument, NULL, 0},
482 		{"stats", required_argument, NULL, 0},
483 		{"dev-basename", required_argument, NULL, 0},
484 		{"tx-csum", required_argument, NULL, 0},
485 		{"tso", required_argument, NULL, 0},
486 		{NULL, 0, 0, 0},
487 	};
488 
489 	/* Parse command line */
490 	while ((opt = getopt_long(argc, argv, "p:P",
491 			long_option, &option_index)) != EOF) {
492 		switch (opt) {
493 		/* Portmask */
494 		case 'p':
495 			enabled_port_mask = parse_portmask(optarg);
496 			if (enabled_port_mask == 0) {
497 				RTE_LOG(INFO, VHOST_CONFIG, "Invalid portmask\n");
498 				us_vhost_usage(prgname);
499 				return -1;
500 			}
501 			break;
502 
503 		case 'P':
504 			promiscuous = 1;
505 			vmdq_conf_default.rx_adv_conf.vmdq_rx_conf.rx_mode =
506 				ETH_VMDQ_ACCEPT_BROADCAST |
507 				ETH_VMDQ_ACCEPT_MULTICAST;
508 			rte_vhost_feature_enable(1ULL << VIRTIO_NET_F_CTRL_RX);
509 
510 			break;
511 
512 		case 0:
513 			/* Enable/disable vm2vm comms. */
514 			if (!strncmp(long_option[option_index].name, "vm2vm",
515 				MAX_LONG_OPT_SZ)) {
516 				ret = parse_num_opt(optarg, (VM2VM_LAST - 1));
517 				if (ret == -1) {
518 					RTE_LOG(INFO, VHOST_CONFIG,
519 						"Invalid argument for "
520 						"vm2vm [0|1|2]\n");
521 					us_vhost_usage(prgname);
522 					return -1;
523 				} else {
524 					vm2vm_mode = (vm2vm_type)ret;
525 				}
526 			}
527 
528 			/* Enable/disable retries on RX. */
529 			if (!strncmp(long_option[option_index].name, "rx-retry", MAX_LONG_OPT_SZ)) {
530 				ret = parse_num_opt(optarg, 1);
531 				if (ret == -1) {
532 					RTE_LOG(INFO, VHOST_CONFIG, "Invalid argument for rx-retry [0|1]\n");
533 					us_vhost_usage(prgname);
534 					return -1;
535 				} else {
536 					enable_retry = ret;
537 				}
538 			}
539 
540 			/* Enable/disable TX checksum offload. */
541 			if (!strncmp(long_option[option_index].name, "tx-csum", MAX_LONG_OPT_SZ)) {
542 				ret = parse_num_opt(optarg, 1);
543 				if (ret == -1) {
544 					RTE_LOG(INFO, VHOST_CONFIG, "Invalid argument for tx-csum [0|1]\n");
545 					us_vhost_usage(prgname);
546 					return -1;
547 				} else
548 					enable_tx_csum = ret;
549 			}
550 
551 			/* Enable/disable TSO offload. */
552 			if (!strncmp(long_option[option_index].name, "tso", MAX_LONG_OPT_SZ)) {
553 				ret = parse_num_opt(optarg, 1);
554 				if (ret == -1) {
555 					RTE_LOG(INFO, VHOST_CONFIG, "Invalid argument for tso [0|1]\n");
556 					us_vhost_usage(prgname);
557 					return -1;
558 				} else
559 					enable_tso = ret;
560 			}
561 
562 			/* Specify the retries delay time (in useconds) on RX. */
563 			if (!strncmp(long_option[option_index].name, "rx-retry-delay", MAX_LONG_OPT_SZ)) {
564 				ret = parse_num_opt(optarg, INT32_MAX);
565 				if (ret == -1) {
566 					RTE_LOG(INFO, VHOST_CONFIG, "Invalid argument for rx-retry-delay [0-N]\n");
567 					us_vhost_usage(prgname);
568 					return -1;
569 				} else {
570 					burst_rx_delay_time = ret;
571 				}
572 			}
573 
574 			/* Specify the retries number on RX. */
575 			if (!strncmp(long_option[option_index].name, "rx-retry-num", MAX_LONG_OPT_SZ)) {
576 				ret = parse_num_opt(optarg, INT32_MAX);
577 				if (ret == -1) {
578 					RTE_LOG(INFO, VHOST_CONFIG, "Invalid argument for rx-retry-num [0-N]\n");
579 					us_vhost_usage(prgname);
580 					return -1;
581 				} else {
582 					burst_rx_retry_num = ret;
583 				}
584 			}
585 
586 			/* Enable/disable RX mergeable buffers. */
587 			if (!strncmp(long_option[option_index].name, "mergeable", MAX_LONG_OPT_SZ)) {
588 				ret = parse_num_opt(optarg, 1);
589 				if (ret == -1) {
590 					RTE_LOG(INFO, VHOST_CONFIG, "Invalid argument for mergeable [0|1]\n");
591 					us_vhost_usage(prgname);
592 					return -1;
593 				} else {
594 					mergeable = !!ret;
595 					if (ret) {
596 						vmdq_conf_default.rxmode.jumbo_frame = 1;
597 						vmdq_conf_default.rxmode.max_rx_pkt_len
598 							= JUMBO_FRAME_MAX_SIZE;
599 					}
600 				}
601 			}
602 
603 			/* Enable/disable RX VLAN strip on host. */
604 			if (!strncmp(long_option[option_index].name,
605 				"vlan-strip", MAX_LONG_OPT_SZ)) {
606 				ret = parse_num_opt(optarg, 1);
607 				if (ret == -1) {
608 					RTE_LOG(INFO, VHOST_CONFIG,
609 						"Invalid argument for VLAN strip [0|1]\n");
610 					us_vhost_usage(prgname);
611 					return -1;
612 				} else {
613 					vlan_strip = !!ret;
614 					vmdq_conf_default.rxmode.hw_vlan_strip =
615 						vlan_strip;
616 				}
617 			}
618 
619 			/* Enable/disable stats. */
620 			if (!strncmp(long_option[option_index].name, "stats", MAX_LONG_OPT_SZ)) {
621 				ret = parse_num_opt(optarg, INT32_MAX);
622 				if (ret == -1) {
623 					RTE_LOG(INFO, VHOST_CONFIG, "Invalid argument for stats [0..N]\n");
624 					us_vhost_usage(prgname);
625 					return -1;
626 				} else {
627 					enable_stats = ret;
628 				}
629 			}
630 
631 			/* Set character device basename. */
632 			if (!strncmp(long_option[option_index].name, "dev-basename", MAX_LONG_OPT_SZ)) {
633 				if (us_vhost_parse_basename(optarg) == -1) {
634 					RTE_LOG(INFO, VHOST_CONFIG, "Invalid argument for character device basename (Max %d characters)\n", MAX_BASENAME_SZ);
635 					us_vhost_usage(prgname);
636 					return -1;
637 				}
638 			}
639 
640 			break;
641 
642 			/* Invalid option - print options. */
643 		default:
644 			us_vhost_usage(prgname);
645 			return -1;
646 		}
647 	}
648 
649 	for (i = 0; i < RTE_MAX_ETHPORTS; i++) {
650 		if (enabled_port_mask & (1 << i))
651 			ports[num_ports++] = (uint8_t)i;
652 	}
653 
654 	if ((num_ports ==  0) || (num_ports > MAX_SUP_PORTS)) {
655 		RTE_LOG(INFO, VHOST_PORT, "Current enabled port number is %u,"
656 			"but only %u port can be enabled\n",num_ports, MAX_SUP_PORTS);
657 		return -1;
658 	}
659 
660 	return 0;
661 }
662 
663 /*
664  * Update the global var NUM_PORTS and array PORTS according to system ports number
665  * and return valid ports number
666  */
667 static unsigned check_ports_num(unsigned nb_ports)
668 {
669 	unsigned valid_num_ports = num_ports;
670 	unsigned portid;
671 
672 	if (num_ports > nb_ports) {
673 		RTE_LOG(INFO, VHOST_PORT, "\nSpecified port number(%u) exceeds total system port number(%u)\n",
674 			num_ports, nb_ports);
675 		num_ports = nb_ports;
676 	}
677 
678 	for (portid = 0; portid < num_ports; portid ++) {
679 		if (ports[portid] >= nb_ports) {
680 			RTE_LOG(INFO, VHOST_PORT, "\nSpecified port ID(%u) exceeds max system port ID(%u)\n",
681 				ports[portid], (nb_ports - 1));
682 			ports[portid] = INVALID_PORT_ID;
683 			valid_num_ports--;
684 		}
685 	}
686 	return valid_num_ports;
687 }
688 
689 static inline struct vhost_dev *__attribute__((always_inline))
690 find_vhost_dev(struct ether_addr *mac)
691 {
692 	struct vhost_dev *vdev;
693 
694 	TAILQ_FOREACH(vdev, &vhost_dev_list, global_vdev_entry) {
695 		if (vdev->ready == DEVICE_RX &&
696 		    is_same_ether_addr(mac, &vdev->mac_address))
697 			return vdev;
698 	}
699 
700 	return NULL;
701 }
702 
703 /*
704  * This function learns the MAC address of the device and registers this along with a
705  * vlan tag to a VMDQ.
706  */
707 static int
708 link_vmdq(struct vhost_dev *vdev, struct rte_mbuf *m)
709 {
710 	struct ether_hdr *pkt_hdr;
711 	int i, ret;
712 
713 	/* Learn MAC address of guest device from packet */
714 	pkt_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
715 
716 	if (find_vhost_dev(&pkt_hdr->s_addr)) {
717 		RTE_LOG(ERR, VHOST_DATA,
718 			"(%d) device is using a registered MAC!\n",
719 			vdev->vid);
720 		return -1;
721 	}
722 
723 	for (i = 0; i < ETHER_ADDR_LEN; i++)
724 		vdev->mac_address.addr_bytes[i] = pkt_hdr->s_addr.addr_bytes[i];
725 
726 	/* vlan_tag currently uses the device_id. */
727 	vdev->vlan_tag = vlan_tags[vdev->vid];
728 
729 	/* Print out VMDQ registration info. */
730 	RTE_LOG(INFO, VHOST_DATA,
731 		"(%d) mac %02x:%02x:%02x:%02x:%02x:%02x and vlan %d registered\n",
732 		vdev->vid,
733 		vdev->mac_address.addr_bytes[0], vdev->mac_address.addr_bytes[1],
734 		vdev->mac_address.addr_bytes[2], vdev->mac_address.addr_bytes[3],
735 		vdev->mac_address.addr_bytes[4], vdev->mac_address.addr_bytes[5],
736 		vdev->vlan_tag);
737 
738 	/* Register the MAC address. */
739 	ret = rte_eth_dev_mac_addr_add(ports[0], &vdev->mac_address,
740 				(uint32_t)vdev->vid + vmdq_pool_base);
741 	if (ret)
742 		RTE_LOG(ERR, VHOST_DATA,
743 			"(%d) failed to add device MAC address to VMDQ\n",
744 			vdev->vid);
745 
746 	/* Enable stripping of the vlan tag as we handle routing. */
747 	if (vlan_strip)
748 		rte_eth_dev_set_vlan_strip_on_queue(ports[0],
749 			(uint16_t)vdev->vmdq_rx_q, 1);
750 
751 	/* Set device as ready for RX. */
752 	vdev->ready = DEVICE_RX;
753 
754 	return 0;
755 }
756 
757 /*
758  * Removes MAC address and vlan tag from VMDQ. Ensures that nothing is adding buffers to the RX
759  * queue before disabling RX on the device.
760  */
761 static inline void
762 unlink_vmdq(struct vhost_dev *vdev)
763 {
764 	unsigned i = 0;
765 	unsigned rx_count;
766 	struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
767 
768 	if (vdev->ready == DEVICE_RX) {
769 		/*clear MAC and VLAN settings*/
770 		rte_eth_dev_mac_addr_remove(ports[0], &vdev->mac_address);
771 		for (i = 0; i < 6; i++)
772 			vdev->mac_address.addr_bytes[i] = 0;
773 
774 		vdev->vlan_tag = 0;
775 
776 		/*Clear out the receive buffers*/
777 		rx_count = rte_eth_rx_burst(ports[0],
778 					(uint16_t)vdev->vmdq_rx_q, pkts_burst, MAX_PKT_BURST);
779 
780 		while (rx_count) {
781 			for (i = 0; i < rx_count; i++)
782 				rte_pktmbuf_free(pkts_burst[i]);
783 
784 			rx_count = rte_eth_rx_burst(ports[0],
785 					(uint16_t)vdev->vmdq_rx_q, pkts_burst, MAX_PKT_BURST);
786 		}
787 
788 		vdev->ready = DEVICE_MAC_LEARNING;
789 	}
790 }
791 
792 static inline void __attribute__((always_inline))
793 virtio_xmit(struct vhost_dev *dst_vdev, struct vhost_dev *src_vdev,
794 	    struct rte_mbuf *m)
795 {
796 	uint16_t ret;
797 
798 	ret = rte_vhost_enqueue_burst(dst_vdev->dev, VIRTIO_RXQ, &m, 1);
799 	if (enable_stats) {
800 		rte_atomic64_inc(&dst_vdev->stats.rx_total_atomic);
801 		rte_atomic64_add(&dst_vdev->stats.rx_atomic, ret);
802 		src_vdev->stats.tx_total++;
803 		src_vdev->stats.tx += ret;
804 	}
805 }
806 
807 /*
808  * Check if the packet destination MAC address is for a local device. If so then put
809  * the packet on that devices RX queue. If not then return.
810  */
811 static inline int __attribute__((always_inline))
812 virtio_tx_local(struct vhost_dev *vdev, struct rte_mbuf *m)
813 {
814 	struct ether_hdr *pkt_hdr;
815 	struct vhost_dev *dst_vdev;
816 
817 	pkt_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
818 
819 	dst_vdev = find_vhost_dev(&pkt_hdr->d_addr);
820 	if (!dst_vdev)
821 		return -1;
822 
823 	if (vdev->vid == dst_vdev->vid) {
824 		RTE_LOG(DEBUG, VHOST_DATA,
825 			"(%d) TX: src and dst MAC is same. Dropping packet.\n",
826 			vdev->vid);
827 		return 0;
828 	}
829 
830 	RTE_LOG(DEBUG, VHOST_DATA,
831 		"(%d) TX: MAC address is local\n", dst_vdev->vid);
832 
833 	if (unlikely(dst_vdev->remove)) {
834 		RTE_LOG(DEBUG, VHOST_DATA,
835 			"(%d) device is marked for removal\n", dst_vdev->vid);
836 		return 0;
837 	}
838 
839 	virtio_xmit(dst_vdev, vdev, m);
840 	return 0;
841 }
842 
843 /*
844  * Check if the destination MAC of a packet is one local VM,
845  * and get its vlan tag, and offset if it is.
846  */
847 static inline int __attribute__((always_inline))
848 find_local_dest(struct vhost_dev *vdev, struct rte_mbuf *m,
849 	uint32_t *offset, uint16_t *vlan_tag)
850 {
851 	struct vhost_dev *dst_vdev;
852 	struct ether_hdr *pkt_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
853 
854 	dst_vdev = find_vhost_dev(&pkt_hdr->d_addr);
855 	if (!dst_vdev)
856 		return 0;
857 
858 	if (vdev->vid == dst_vdev->vid) {
859 		RTE_LOG(DEBUG, VHOST_DATA,
860 			"(%d) TX: src and dst MAC is same. Dropping packet.\n",
861 			vdev->vid);
862 		return -1;
863 	}
864 
865 	/*
866 	 * HW vlan strip will reduce the packet length
867 	 * by minus length of vlan tag, so need restore
868 	 * the packet length by plus it.
869 	 */
870 	*offset  = VLAN_HLEN;
871 	*vlan_tag = vlan_tags[vdev->vid];
872 
873 	RTE_LOG(DEBUG, VHOST_DATA,
874 		"(%d) TX: pkt to local VM device id: (%d), vlan tag: %u.\n",
875 		vdev->vid, dst_vdev->vid, *vlan_tag);
876 
877 	return 0;
878 }
879 
880 static uint16_t
881 get_psd_sum(void *l3_hdr, uint64_t ol_flags)
882 {
883 	if (ol_flags & PKT_TX_IPV4)
884 		return rte_ipv4_phdr_cksum(l3_hdr, ol_flags);
885 	else /* assume ethertype == ETHER_TYPE_IPv6 */
886 		return rte_ipv6_phdr_cksum(l3_hdr, ol_flags);
887 }
888 
889 static void virtio_tx_offload(struct rte_mbuf *m)
890 {
891 	void *l3_hdr;
892 	struct ipv4_hdr *ipv4_hdr = NULL;
893 	struct tcp_hdr *tcp_hdr = NULL;
894 	struct ether_hdr *eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
895 
896 	l3_hdr = (char *)eth_hdr + m->l2_len;
897 
898 	if (m->ol_flags & PKT_TX_IPV4) {
899 		ipv4_hdr = l3_hdr;
900 		ipv4_hdr->hdr_checksum = 0;
901 		m->ol_flags |= PKT_TX_IP_CKSUM;
902 	}
903 
904 	tcp_hdr = (struct tcp_hdr *)((char *)l3_hdr + m->l3_len);
905 	tcp_hdr->cksum = get_psd_sum(l3_hdr, m->ol_flags);
906 }
907 
908 static inline void
909 free_pkts(struct rte_mbuf **pkts, uint16_t n)
910 {
911 	while (n--)
912 		rte_pktmbuf_free(pkts[n]);
913 }
914 
915 static inline void __attribute__((always_inline))
916 do_drain_mbuf_table(struct mbuf_table *tx_q)
917 {
918 	uint16_t count;
919 
920 	count = rte_eth_tx_burst(ports[0], tx_q->txq_id,
921 				 tx_q->m_table, tx_q->len);
922 	if (unlikely(count < tx_q->len))
923 		free_pkts(&tx_q->m_table[count], tx_q->len - count);
924 
925 	tx_q->len = 0;
926 }
927 
928 /*
929  * This function routes the TX packet to the correct interface. This
930  * may be a local device or the physical port.
931  */
932 static inline void __attribute__((always_inline))
933 virtio_tx_route(struct vhost_dev *vdev, struct rte_mbuf *m, uint16_t vlan_tag)
934 {
935 	struct mbuf_table *tx_q;
936 	unsigned offset = 0;
937 	const uint16_t lcore_id = rte_lcore_id();
938 	struct ether_hdr *nh;
939 
940 
941 	nh = rte_pktmbuf_mtod(m, struct ether_hdr *);
942 	if (unlikely(is_broadcast_ether_addr(&nh->d_addr))) {
943 		struct vhost_dev *vdev2;
944 
945 		TAILQ_FOREACH(vdev2, &vhost_dev_list, global_vdev_entry) {
946 			virtio_xmit(vdev2, vdev, m);
947 		}
948 		goto queue2nic;
949 	}
950 
951 	/*check if destination is local VM*/
952 	if ((vm2vm_mode == VM2VM_SOFTWARE) && (virtio_tx_local(vdev, m) == 0)) {
953 		rte_pktmbuf_free(m);
954 		return;
955 	}
956 
957 	if (unlikely(vm2vm_mode == VM2VM_HARDWARE)) {
958 		if (unlikely(find_local_dest(vdev, m, &offset,
959 					     &vlan_tag) != 0)) {
960 			rte_pktmbuf_free(m);
961 			return;
962 		}
963 	}
964 
965 	RTE_LOG(DEBUG, VHOST_DATA,
966 		"(%d) TX: MAC address is external\n", vdev->vid);
967 
968 queue2nic:
969 
970 	/*Add packet to the port tx queue*/
971 	tx_q = &lcore_tx_queue[lcore_id];
972 
973 	nh = rte_pktmbuf_mtod(m, struct ether_hdr *);
974 	if (unlikely(nh->ether_type == rte_cpu_to_be_16(ETHER_TYPE_VLAN))) {
975 		/* Guest has inserted the vlan tag. */
976 		struct vlan_hdr *vh = (struct vlan_hdr *) (nh + 1);
977 		uint16_t vlan_tag_be = rte_cpu_to_be_16(vlan_tag);
978 		if ((vm2vm_mode == VM2VM_HARDWARE) &&
979 			(vh->vlan_tci != vlan_tag_be))
980 			vh->vlan_tci = vlan_tag_be;
981 	} else {
982 		m->ol_flags |= PKT_TX_VLAN_PKT;
983 
984 		/*
985 		 * Find the right seg to adjust the data len when offset is
986 		 * bigger than tail room size.
987 		 */
988 		if (unlikely(vm2vm_mode == VM2VM_HARDWARE)) {
989 			if (likely(offset <= rte_pktmbuf_tailroom(m)))
990 				m->data_len += offset;
991 			else {
992 				struct rte_mbuf *seg = m;
993 
994 				while ((seg->next != NULL) &&
995 					(offset > rte_pktmbuf_tailroom(seg)))
996 					seg = seg->next;
997 
998 				seg->data_len += offset;
999 			}
1000 			m->pkt_len += offset;
1001 		}
1002 
1003 		m->vlan_tci = vlan_tag;
1004 	}
1005 
1006 	if (m->ol_flags & PKT_TX_TCP_SEG)
1007 		virtio_tx_offload(m);
1008 
1009 	tx_q->m_table[tx_q->len++] = m;
1010 	if (enable_stats) {
1011 		vdev->stats.tx_total++;
1012 		vdev->stats.tx++;
1013 	}
1014 
1015 	if (unlikely(tx_q->len == MAX_PKT_BURST))
1016 		do_drain_mbuf_table(tx_q);
1017 }
1018 
1019 
1020 static inline void __attribute__((always_inline))
1021 drain_mbuf_table(struct mbuf_table *tx_q)
1022 {
1023 	static uint64_t prev_tsc;
1024 	uint64_t cur_tsc;
1025 
1026 	if (tx_q->len == 0)
1027 		return;
1028 
1029 	cur_tsc = rte_rdtsc();
1030 	if (unlikely(cur_tsc - prev_tsc > MBUF_TABLE_DRAIN_TSC)) {
1031 		prev_tsc = cur_tsc;
1032 
1033 		RTE_LOG(DEBUG, VHOST_DATA,
1034 			"TX queue drained after timeout with burst size %u\n",
1035 			tx_q->len);
1036 		do_drain_mbuf_table(tx_q);
1037 	}
1038 }
1039 
1040 static inline void __attribute__((always_inline))
1041 drain_eth_rx(struct vhost_dev *vdev)
1042 {
1043 	uint16_t rx_count, enqueue_count;
1044 	struct virtio_net *dev = vdev->dev;
1045 	struct rte_mbuf *pkts[MAX_PKT_BURST];
1046 
1047 	rx_count = rte_eth_rx_burst(ports[0], vdev->vmdq_rx_q,
1048 				    pkts, MAX_PKT_BURST);
1049 	if (!rx_count)
1050 		return;
1051 
1052 	/*
1053 	 * When "enable_retry" is set, here we wait and retry when there
1054 	 * is no enough free slots in the queue to hold @rx_count packets,
1055 	 * to diminish packet loss.
1056 	 */
1057 	if (enable_retry &&
1058 	    unlikely(rx_count > rte_vhost_avail_entries(dev->vid,
1059 			VIRTIO_RXQ))) {
1060 		uint32_t retry;
1061 
1062 		for (retry = 0; retry < burst_rx_retry_num; retry++) {
1063 			rte_delay_us(burst_rx_delay_time);
1064 			if (rx_count <= rte_vhost_avail_entries(dev->vid,
1065 					VIRTIO_RXQ))
1066 				break;
1067 		}
1068 	}
1069 
1070 	enqueue_count = rte_vhost_enqueue_burst(dev, VIRTIO_RXQ,
1071 						pkts, rx_count);
1072 	if (enable_stats) {
1073 		rte_atomic64_add(&vdev->stats.rx_total_atomic, rx_count);
1074 		rte_atomic64_add(&vdev->stats.rx_atomic, enqueue_count);
1075 	}
1076 
1077 	free_pkts(pkts, rx_count);
1078 }
1079 
1080 static inline void __attribute__((always_inline))
1081 drain_virtio_tx(struct vhost_dev *vdev)
1082 {
1083 	struct rte_mbuf *pkts[MAX_PKT_BURST];
1084 	uint16_t count;
1085 	uint16_t i;
1086 
1087 	count = rte_vhost_dequeue_burst(vdev->dev, VIRTIO_TXQ, mbuf_pool,
1088 					pkts, MAX_PKT_BURST);
1089 
1090 	/* setup VMDq for the first packet */
1091 	if (unlikely(vdev->ready == DEVICE_MAC_LEARNING) && count) {
1092 		if (vdev->remove || link_vmdq(vdev, pkts[0]) == -1)
1093 			free_pkts(pkts, count);
1094 	}
1095 
1096 	for (i = 0; i < count; ++i)
1097 		virtio_tx_route(vdev, pkts[i], vlan_tags[vdev->vid]);
1098 }
1099 
1100 /*
1101  * Main function of vhost-switch. It basically does:
1102  *
1103  * for each vhost device {
1104  *    - drain_eth_rx()
1105  *
1106  *      Which drains the host eth Rx queue linked to the vhost device,
1107  *      and deliver all of them to guest virito Rx ring associated with
1108  *      this vhost device.
1109  *
1110  *    - drain_virtio_tx()
1111  *
1112  *      Which drains the guest virtio Tx queue and deliver all of them
1113  *      to the target, which could be another vhost device, or the
1114  *      physical eth dev. The route is done in function "virtio_tx_route".
1115  * }
1116  */
1117 static int
1118 switch_worker(void *arg __rte_unused)
1119 {
1120 	unsigned i;
1121 	unsigned lcore_id = rte_lcore_id();
1122 	struct vhost_dev *vdev;
1123 	struct mbuf_table *tx_q;
1124 
1125 	RTE_LOG(INFO, VHOST_DATA, "Procesing on Core %u started\n", lcore_id);
1126 
1127 	tx_q = &lcore_tx_queue[lcore_id];
1128 	for (i = 0; i < rte_lcore_count(); i++) {
1129 		if (lcore_ids[i] == lcore_id) {
1130 			tx_q->txq_id = i;
1131 			break;
1132 		}
1133 	}
1134 
1135 	while(1) {
1136 		drain_mbuf_table(tx_q);
1137 
1138 		/*
1139 		 * Inform the configuration core that we have exited the
1140 		 * linked list and that no devices are in use if requested.
1141 		 */
1142 		if (lcore_info[lcore_id].dev_removal_flag == REQUEST_DEV_REMOVAL)
1143 			lcore_info[lcore_id].dev_removal_flag = ACK_DEV_REMOVAL;
1144 
1145 		/*
1146 		 * Process vhost devices
1147 		 */
1148 		TAILQ_FOREACH(vdev, &lcore_info[lcore_id].vdev_list,
1149 			      lcore_vdev_entry) {
1150 			if (unlikely(vdev->remove)) {
1151 				unlink_vmdq(vdev);
1152 				vdev->ready = DEVICE_SAFE_REMOVE;
1153 				continue;
1154 			}
1155 
1156 			if (likely(vdev->ready == DEVICE_RX))
1157 				drain_eth_rx(vdev);
1158 
1159 			if (likely(!vdev->remove))
1160 				drain_virtio_tx(vdev);
1161 		}
1162 	}
1163 
1164 	return 0;
1165 }
1166 
1167 /*
1168  * Remove a device from the specific data core linked list and from the
1169  * main linked list. Synchonization  occurs through the use of the
1170  * lcore dev_removal_flag. Device is made volatile here to avoid re-ordering
1171  * of dev->remove=1 which can cause an infinite loop in the rte_pause loop.
1172  */
1173 static void
1174 destroy_device (volatile struct virtio_net *dev)
1175 {
1176 	struct vhost_dev *vdev = NULL;
1177 	int lcore;
1178 
1179 	TAILQ_FOREACH(vdev, &vhost_dev_list, global_vdev_entry) {
1180 		if (vdev->vid == dev->vid)
1181 			break;
1182 	}
1183 	if (!vdev)
1184 		return;
1185 	/*set the remove flag. */
1186 	vdev->remove = 1;
1187 	while(vdev->ready != DEVICE_SAFE_REMOVE) {
1188 		rte_pause();
1189 	}
1190 
1191 	TAILQ_REMOVE(&lcore_info[vdev->coreid].vdev_list, vdev,
1192 		     lcore_vdev_entry);
1193 	TAILQ_REMOVE(&vhost_dev_list, vdev, global_vdev_entry);
1194 
1195 
1196 	/* Set the dev_removal_flag on each lcore. */
1197 	RTE_LCORE_FOREACH_SLAVE(lcore)
1198 		lcore_info[lcore].dev_removal_flag = REQUEST_DEV_REMOVAL;
1199 
1200 	/*
1201 	 * Once each core has set the dev_removal_flag to ACK_DEV_REMOVAL
1202 	 * we can be sure that they can no longer access the device removed
1203 	 * from the linked lists and that the devices are no longer in use.
1204 	 */
1205 	RTE_LCORE_FOREACH_SLAVE(lcore) {
1206 		while (lcore_info[lcore].dev_removal_flag != ACK_DEV_REMOVAL)
1207 			rte_pause();
1208 	}
1209 
1210 	lcore_info[vdev->coreid].device_num--;
1211 
1212 	RTE_LOG(INFO, VHOST_DATA,
1213 		"(%d) device has been removed from data core\n",
1214 		vdev->vid);
1215 
1216 	rte_free(vdev);
1217 }
1218 
1219 /*
1220  * A new device is added to a data core. First the device is added to the main linked list
1221  * and the allocated to a specific data core.
1222  */
1223 static int
1224 new_device (struct virtio_net *dev)
1225 {
1226 	int lcore, core_add = 0;
1227 	uint32_t device_num_min = num_devices;
1228 	struct vhost_dev *vdev;
1229 	int vid = dev->vid;
1230 
1231 	vdev = rte_zmalloc("vhost device", sizeof(*vdev), RTE_CACHE_LINE_SIZE);
1232 	if (vdev == NULL) {
1233 		RTE_LOG(INFO, VHOST_DATA,
1234 			"(%d) couldn't allocate memory for vhost dev\n",
1235 			vid);
1236 		return -1;
1237 	}
1238 	vdev->dev = dev;
1239 	vdev->vid = vid;
1240 
1241 	TAILQ_INSERT_TAIL(&vhost_dev_list, vdev, global_vdev_entry);
1242 	vdev->vmdq_rx_q = vid * queues_per_pool + vmdq_queue_base;
1243 
1244 	/*reset ready flag*/
1245 	vdev->ready = DEVICE_MAC_LEARNING;
1246 	vdev->remove = 0;
1247 
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;
1252 			core_add = lcore;
1253 		}
1254 	}
1255 	vdev->coreid = core_add;
1256 
1257 	TAILQ_INSERT_TAIL(&lcore_info[vdev->coreid].vdev_list, vdev,
1258 			  lcore_vdev_entry);
1259 	lcore_info[vdev->coreid].device_num++;
1260 
1261 	/* Disable notifications. */
1262 	rte_vhost_enable_guest_notification(dev, VIRTIO_RXQ, 0);
1263 	rte_vhost_enable_guest_notification(dev, VIRTIO_TXQ, 0);
1264 
1265 	RTE_LOG(INFO, VHOST_DATA,
1266 		"(%d) device has been added to data core %d\n",
1267 		vid, vdev->coreid);
1268 
1269 	return 0;
1270 }
1271 
1272 /*
1273  * These callback allow devices to be added to the data core when configuration
1274  * has been fully complete.
1275  */
1276 static const struct virtio_net_device_ops virtio_net_device_ops =
1277 {
1278 	.new_device =  new_device,
1279 	.destroy_device = destroy_device,
1280 };
1281 
1282 /*
1283  * This is a thread will wake up after a period to print stats if the user has
1284  * enabled them.
1285  */
1286 static void
1287 print_stats(void)
1288 {
1289 	struct vhost_dev *vdev;
1290 	uint64_t tx_dropped, rx_dropped;
1291 	uint64_t tx, tx_total, rx, rx_total;
1292 	const char clr[] = { 27, '[', '2', 'J', '\0' };
1293 	const char top_left[] = { 27, '[', '1', ';', '1', 'H','\0' };
1294 
1295 	while(1) {
1296 		sleep(enable_stats);
1297 
1298 		/* Clear screen and move to top left */
1299 		printf("%s%s\n", clr, top_left);
1300 		printf("Device statistics =================================\n");
1301 
1302 		TAILQ_FOREACH(vdev, &vhost_dev_list, global_vdev_entry) {
1303 			tx_total   = vdev->stats.tx_total;
1304 			tx         = vdev->stats.tx;
1305 			tx_dropped = tx_total - tx;
1306 
1307 			rx_total   = rte_atomic64_read(&vdev->stats.rx_total_atomic);
1308 			rx         = rte_atomic64_read(&vdev->stats.rx_atomic);
1309 			rx_dropped = rx_total - rx;
1310 
1311 			printf("Statistics for device %d\n"
1312 				"-----------------------\n"
1313 				"TX total:              %" PRIu64 "\n"
1314 				"TX dropped:            %" PRIu64 "\n"
1315 				"TX successful:         %" PRIu64 "\n"
1316 				"RX total:              %" PRIu64 "\n"
1317 				"RX dropped:            %" PRIu64 "\n"
1318 				"RX successful:         %" PRIu64 "\n",
1319 				vdev->dev->vid,
1320 				tx_total, tx_dropped, tx,
1321 				rx_total, rx_dropped, rx);
1322 		}
1323 
1324 		printf("===================================================\n");
1325 	}
1326 }
1327 
1328 /* When we receive a INT signal, unregister vhost driver */
1329 static void
1330 sigint_handler(__rte_unused int signum)
1331 {
1332 	/* Unregister vhost driver. */
1333 	int ret = rte_vhost_driver_unregister((char *)&dev_basename);
1334 	if (ret != 0)
1335 		rte_exit(EXIT_FAILURE, "vhost driver unregister failure.\n");
1336 	exit(0);
1337 }
1338 
1339 /*
1340  * While creating an mbuf pool, one key thing is to figure out how
1341  * many mbuf entries is enough for our use. FYI, here are some
1342  * guidelines:
1343  *
1344  * - Each rx queue would reserve @nr_rx_desc mbufs at queue setup stage
1345  *
1346  * - For each switch core (A CPU core does the packet switch), we need
1347  *   also make some reservation for receiving the packets from virtio
1348  *   Tx queue. How many is enough depends on the usage. It's normally
1349  *   a simple calculation like following:
1350  *
1351  *       MAX_PKT_BURST * max packet size / mbuf size
1352  *
1353  *   So, we definitely need allocate more mbufs when TSO is enabled.
1354  *
1355  * - Similarly, for each switching core, we should serve @nr_rx_desc
1356  *   mbufs for receiving the packets from physical NIC device.
1357  *
1358  * - We also need make sure, for each switch core, we have allocated
1359  *   enough mbufs to fill up the mbuf cache.
1360  */
1361 static void
1362 create_mbuf_pool(uint16_t nr_port, uint32_t nr_switch_core, uint32_t mbuf_size,
1363 	uint32_t nr_queues, uint32_t nr_rx_desc, uint32_t nr_mbuf_cache)
1364 {
1365 	uint32_t nr_mbufs;
1366 	uint32_t nr_mbufs_per_core;
1367 	uint32_t mtu = 1500;
1368 
1369 	if (mergeable)
1370 		mtu = 9000;
1371 	if (enable_tso)
1372 		mtu = 64 * 1024;
1373 
1374 	nr_mbufs_per_core  = (mtu + mbuf_size) * MAX_PKT_BURST /
1375 			(mbuf_size - RTE_PKTMBUF_HEADROOM) * MAX_PKT_BURST;
1376 	nr_mbufs_per_core += nr_rx_desc;
1377 	nr_mbufs_per_core  = RTE_MAX(nr_mbufs_per_core, nr_mbuf_cache);
1378 
1379 	nr_mbufs  = nr_queues * nr_rx_desc;
1380 	nr_mbufs += nr_mbufs_per_core * nr_switch_core;
1381 	nr_mbufs *= nr_port;
1382 
1383 	mbuf_pool = rte_pktmbuf_pool_create("MBUF_POOL", nr_mbufs,
1384 					    nr_mbuf_cache, 0, mbuf_size,
1385 					    rte_socket_id());
1386 	if (mbuf_pool == NULL)
1387 		rte_exit(EXIT_FAILURE, "Cannot create mbuf pool\n");
1388 }
1389 
1390 /*
1391  * Main function, does initialisation and calls the per-lcore functions. The CUSE
1392  * device is also registered here to handle the IOCTLs.
1393  */
1394 int
1395 main(int argc, char *argv[])
1396 {
1397 	unsigned lcore_id, core_id = 0;
1398 	unsigned nb_ports, valid_num_ports;
1399 	int ret;
1400 	uint8_t portid;
1401 	static pthread_t tid;
1402 	char thread_name[RTE_MAX_THREAD_NAME_LEN];
1403 
1404 	signal(SIGINT, sigint_handler);
1405 
1406 	/* init EAL */
1407 	ret = rte_eal_init(argc, argv);
1408 	if (ret < 0)
1409 		rte_exit(EXIT_FAILURE, "Error with EAL initialization\n");
1410 	argc -= ret;
1411 	argv += ret;
1412 
1413 	/* parse app arguments */
1414 	ret = us_vhost_parse_args(argc, argv);
1415 	if (ret < 0)
1416 		rte_exit(EXIT_FAILURE, "Invalid argument\n");
1417 
1418 	for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id ++)
1419 		TAILQ_INIT(&lcore_info[lcore_id].vdev_list);
1420 
1421 		if (rte_lcore_is_enabled(lcore_id))
1422 			lcore_ids[core_id ++] = lcore_id;
1423 
1424 	if (rte_lcore_count() > RTE_MAX_LCORE)
1425 		rte_exit(EXIT_FAILURE,"Not enough cores\n");
1426 
1427 	/* Get the number of physical ports. */
1428 	nb_ports = rte_eth_dev_count();
1429 
1430 	/*
1431 	 * Update the global var NUM_PORTS and global array PORTS
1432 	 * and get value of var VALID_NUM_PORTS according to system ports number
1433 	 */
1434 	valid_num_ports = check_ports_num(nb_ports);
1435 
1436 	if ((valid_num_ports ==  0) || (valid_num_ports > MAX_SUP_PORTS)) {
1437 		RTE_LOG(INFO, VHOST_PORT, "Current enabled port number is %u,"
1438 			"but only %u port can be enabled\n",num_ports, MAX_SUP_PORTS);
1439 		return -1;
1440 	}
1441 
1442 	/*
1443 	 * FIXME: here we are trying to allocate mbufs big enough for
1444 	 * @MAX_QUEUES, but the truth is we're never going to use that
1445 	 * many queues here. We probably should only do allocation for
1446 	 * those queues we are going to use.
1447 	 */
1448 	create_mbuf_pool(valid_num_ports, rte_lcore_count() - 1, MBUF_DATA_SIZE,
1449 			 MAX_QUEUES, RTE_TEST_RX_DESC_DEFAULT, MBUF_CACHE_SIZE);
1450 
1451 	if (vm2vm_mode == VM2VM_HARDWARE) {
1452 		/* Enable VT loop back to let L2 switch to do it. */
1453 		vmdq_conf_default.rx_adv_conf.vmdq_rx_conf.enable_loop_back = 1;
1454 		RTE_LOG(DEBUG, VHOST_CONFIG,
1455 			"Enable loop back for L2 switch in vmdq.\n");
1456 	}
1457 
1458 	/* initialize all ports */
1459 	for (portid = 0; portid < nb_ports; portid++) {
1460 		/* skip ports that are not enabled */
1461 		if ((enabled_port_mask & (1 << portid)) == 0) {
1462 			RTE_LOG(INFO, VHOST_PORT,
1463 				"Skipping disabled port %d\n", portid);
1464 			continue;
1465 		}
1466 		if (port_init(portid) != 0)
1467 			rte_exit(EXIT_FAILURE,
1468 				"Cannot initialize network ports\n");
1469 	}
1470 
1471 	/* Enable stats if the user option is set. */
1472 	if (enable_stats) {
1473 		ret = pthread_create(&tid, NULL, (void *)print_stats, NULL);
1474 		if (ret != 0)
1475 			rte_exit(EXIT_FAILURE,
1476 				"Cannot create print-stats thread\n");
1477 
1478 		/* Set thread_name for aid in debugging.  */
1479 		snprintf(thread_name, RTE_MAX_THREAD_NAME_LEN, "print-stats");
1480 		ret = rte_thread_setname(tid, thread_name);
1481 		if (ret != 0)
1482 			RTE_LOG(DEBUG, VHOST_CONFIG,
1483 				"Cannot set print-stats name\n");
1484 	}
1485 
1486 	/* Launch all data cores. */
1487 	RTE_LCORE_FOREACH_SLAVE(lcore_id)
1488 		rte_eal_remote_launch(switch_worker, NULL, lcore_id);
1489 
1490 	if (mergeable == 0)
1491 		rte_vhost_feature_disable(1ULL << VIRTIO_NET_F_MRG_RXBUF);
1492 
1493 	/* Register vhost(cuse or user) driver to handle vhost messages. */
1494 	ret = rte_vhost_driver_register((char *)&dev_basename);
1495 	if (ret != 0)
1496 		rte_exit(EXIT_FAILURE, "vhost driver register failure.\n");
1497 
1498 	rte_vhost_driver_callback_register(&virtio_net_device_ops);
1499 
1500 	/* Start CUSE session. */
1501 	rte_vhost_driver_session_start();
1502 	return 0;
1503 
1504 }
1505