xref: /dpdk/drivers/net/virtio/virtio_user/virtio_user_dev.c (revision 4ad4b20a79052d9c8062b64eaf0170c16a333ff8)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright(c) 2010-2016 Intel Corporation
3  */
4 
5 #include <stdint.h>
6 #include <stdio.h>
7 #include <fcntl.h>
8 #include <string.h>
9 #include <errno.h>
10 #include <sys/mman.h>
11 #include <unistd.h>
12 #include <sys/eventfd.h>
13 #include <sys/types.h>
14 #include <sys/stat.h>
15 
16 #include <rte_alarm.h>
17 #include <rte_string_fns.h>
18 #include <rte_eal_memconfig.h>
19 
20 #include "vhost.h"
21 #include "virtio_user_dev.h"
22 #include "../virtio_ethdev.h"
23 
24 #define VIRTIO_USER_MEM_EVENT_CLB_NAME "virtio_user_mem_event_clb"
25 
26 const char * const virtio_user_backend_strings[] = {
27 	[VIRTIO_USER_BACKEND_UNKNOWN] = "VIRTIO_USER_BACKEND_UNKNOWN",
28 	[VIRTIO_USER_BACKEND_VHOST_USER] = "VHOST_USER",
29 	[VIRTIO_USER_BACKEND_VHOST_KERNEL] = "VHOST_NET",
30 	[VIRTIO_USER_BACKEND_VHOST_VDPA] = "VHOST_VDPA",
31 };
32 
33 static int
34 virtio_user_create_queue(struct virtio_user_dev *dev, uint32_t queue_sel)
35 {
36 	/* Of all per virtqueue MSGs, make sure VHOST_SET_VRING_CALL come
37 	 * firstly because vhost depends on this msg to allocate virtqueue
38 	 * pair.
39 	 */
40 	struct vhost_vring_file file;
41 	int ret;
42 
43 	file.index = queue_sel;
44 	file.fd = dev->callfds[queue_sel];
45 	ret = dev->ops->set_vring_call(dev, &file);
46 	if (ret < 0) {
47 		PMD_INIT_LOG(ERR, "(%s) Failed to create queue %u\n", dev->path, queue_sel);
48 		return -1;
49 	}
50 
51 	return 0;
52 }
53 
54 static int
55 virtio_user_kick_queue(struct virtio_user_dev *dev, uint32_t queue_sel)
56 {
57 	int ret;
58 	struct vhost_vring_file file;
59 	struct vhost_vring_state state;
60 	struct vring *vring = &dev->vrings[queue_sel];
61 	struct vring_packed *pq_vring = &dev->packed_vrings[queue_sel];
62 	struct vhost_vring_addr addr = {
63 		.index = queue_sel,
64 		.log_guest_addr = 0,
65 		.flags = 0, /* disable log */
66 	};
67 
68 	if (dev->features & (1ULL << VIRTIO_F_RING_PACKED)) {
69 		addr.desc_user_addr =
70 			(uint64_t)(uintptr_t)pq_vring->desc;
71 		addr.avail_user_addr =
72 			(uint64_t)(uintptr_t)pq_vring->driver;
73 		addr.used_user_addr =
74 			(uint64_t)(uintptr_t)pq_vring->device;
75 	} else {
76 		addr.desc_user_addr = (uint64_t)(uintptr_t)vring->desc;
77 		addr.avail_user_addr = (uint64_t)(uintptr_t)vring->avail;
78 		addr.used_user_addr = (uint64_t)(uintptr_t)vring->used;
79 	}
80 
81 	state.index = queue_sel;
82 	state.num = vring->num;
83 	ret = dev->ops->set_vring_num(dev, &state);
84 	if (ret < 0)
85 		goto err;
86 
87 	state.index = queue_sel;
88 	state.num = 0; /* no reservation */
89 	if (dev->features & (1ULL << VIRTIO_F_RING_PACKED))
90 		state.num |= (1 << 15);
91 	ret = dev->ops->set_vring_base(dev, &state);
92 	if (ret < 0)
93 		goto err;
94 
95 	ret = dev->ops->set_vring_addr(dev, &addr);
96 	if (ret < 0)
97 		goto err;
98 
99 	/* Of all per virtqueue MSGs, make sure VHOST_USER_SET_VRING_KICK comes
100 	 * lastly because vhost depends on this msg to judge if
101 	 * virtio is ready.
102 	 */
103 	file.index = queue_sel;
104 	file.fd = dev->kickfds[queue_sel];
105 	ret = dev->ops->set_vring_kick(dev, &file);
106 	if (ret < 0)
107 		goto err;
108 
109 	return 0;
110 err:
111 	PMD_INIT_LOG(ERR, "(%s) Failed to kick queue %u\n", dev->path, queue_sel);
112 
113 	return -1;
114 }
115 
116 static int
117 virtio_user_queue_setup(struct virtio_user_dev *dev,
118 			int (*fn)(struct virtio_user_dev *, uint32_t))
119 {
120 	uint32_t i, queue_sel;
121 
122 	for (i = 0; i < dev->max_queue_pairs; ++i) {
123 		queue_sel = 2 * i + VTNET_SQ_RQ_QUEUE_IDX;
124 		if (fn(dev, queue_sel) < 0) {
125 			PMD_DRV_LOG(ERR, "(%s) setup rx vq %u failed", dev->path, i);
126 			return -1;
127 		}
128 	}
129 	for (i = 0; i < dev->max_queue_pairs; ++i) {
130 		queue_sel = 2 * i + VTNET_SQ_TQ_QUEUE_IDX;
131 		if (fn(dev, queue_sel) < 0) {
132 			PMD_DRV_LOG(INFO, "(%s) setup tx vq %u failed", dev->path, i);
133 			return -1;
134 		}
135 	}
136 
137 	return 0;
138 }
139 
140 int
141 virtio_user_dev_set_features(struct virtio_user_dev *dev)
142 {
143 	uint64_t features;
144 	int ret = -1;
145 
146 	pthread_mutex_lock(&dev->mutex);
147 
148 	/* Step 0: tell vhost to create queues */
149 	if (virtio_user_queue_setup(dev, virtio_user_create_queue) < 0)
150 		goto error;
151 
152 	features = dev->features;
153 
154 	/* Strip VIRTIO_NET_F_MAC, as MAC address is handled in vdev init */
155 	features &= ~(1ull << VIRTIO_NET_F_MAC);
156 	/* Strip VIRTIO_NET_F_CTRL_VQ, as devices do not really need to know */
157 	features &= ~(1ull << VIRTIO_NET_F_CTRL_VQ);
158 	features &= ~(1ull << VIRTIO_NET_F_STATUS);
159 	ret = dev->ops->set_features(dev, features);
160 	if (ret < 0)
161 		goto error;
162 	PMD_DRV_LOG(INFO, "(%s) set features: 0x%" PRIx64, dev->path, features);
163 error:
164 	pthread_mutex_unlock(&dev->mutex);
165 
166 	return ret;
167 }
168 
169 int
170 virtio_user_start_device(struct virtio_user_dev *dev)
171 {
172 	int ret;
173 
174 	/*
175 	 * XXX workaround!
176 	 *
177 	 * We need to make sure that the locks will be
178 	 * taken in the correct order to avoid deadlocks.
179 	 *
180 	 * Before releasing this lock, this thread should
181 	 * not trigger any memory hotplug events.
182 	 *
183 	 * This is a temporary workaround, and should be
184 	 * replaced when we get proper supports from the
185 	 * memory subsystem in the future.
186 	 */
187 	rte_mcfg_mem_read_lock();
188 	pthread_mutex_lock(&dev->mutex);
189 
190 	/* Step 2: share memory regions */
191 	ret = dev->ops->set_memory_table(dev);
192 	if (ret < 0)
193 		goto error;
194 
195 	/* Step 3: kick queues */
196 	ret = virtio_user_queue_setup(dev, virtio_user_kick_queue);
197 	if (ret < 0)
198 		goto error;
199 
200 	/* Step 4: enable queues
201 	 * we enable the 1st queue pair by default.
202 	 */
203 	ret = dev->ops->enable_qp(dev, 0, 1);
204 	if (ret < 0)
205 		goto error;
206 
207 	dev->started = true;
208 
209 	pthread_mutex_unlock(&dev->mutex);
210 	rte_mcfg_mem_read_unlock();
211 
212 	return 0;
213 error:
214 	pthread_mutex_unlock(&dev->mutex);
215 	rte_mcfg_mem_read_unlock();
216 
217 	PMD_INIT_LOG(ERR, "(%s) Failed to start device\n", dev->path);
218 
219 	/* TODO: free resource here or caller to check */
220 	return -1;
221 }
222 
223 int virtio_user_stop_device(struct virtio_user_dev *dev)
224 {
225 	struct vhost_vring_state state;
226 	uint32_t i;
227 	int ret;
228 
229 	pthread_mutex_lock(&dev->mutex);
230 	if (!dev->started)
231 		goto out;
232 
233 	for (i = 0; i < dev->max_queue_pairs; ++i) {
234 		ret = dev->ops->enable_qp(dev, i, 0);
235 		if (ret < 0)
236 			goto err;
237 	}
238 
239 	/* Stop the backend. */
240 	for (i = 0; i < dev->max_queue_pairs * 2; ++i) {
241 		state.index = i;
242 		ret = dev->ops->get_vring_base(dev, &state);
243 		if (ret < 0) {
244 			PMD_DRV_LOG(ERR, "(%s) get_vring_base failed, index=%u", dev->path, i);
245 			goto err;
246 		}
247 	}
248 
249 	dev->started = false;
250 
251 out:
252 	pthread_mutex_unlock(&dev->mutex);
253 
254 	return 0;
255 err:
256 	pthread_mutex_unlock(&dev->mutex);
257 
258 	PMD_INIT_LOG(ERR, "(%s) Failed to stop device\n", dev->path);
259 
260 	return -1;
261 }
262 
263 static inline void
264 parse_mac(struct virtio_user_dev *dev, const char *mac)
265 {
266 	struct rte_ether_addr tmp;
267 
268 	if (!mac)
269 		return;
270 
271 	if (rte_ether_unformat_addr(mac, &tmp) == 0) {
272 		memcpy(dev->mac_addr, &tmp, RTE_ETHER_ADDR_LEN);
273 		dev->mac_specified = 1;
274 	} else {
275 		/* ignore the wrong mac, use random mac */
276 		PMD_DRV_LOG(ERR, "wrong format of mac: %s", mac);
277 	}
278 }
279 
280 static int
281 virtio_user_dev_init_notify(struct virtio_user_dev *dev)
282 {
283 	uint32_t i, j;
284 	int callfd;
285 	int kickfd;
286 
287 	for (i = 0; i < dev->max_queue_pairs * 2; i++) {
288 		/* May use invalid flag, but some backend uses kickfd and
289 		 * callfd as criteria to judge if dev is alive. so finally we
290 		 * use real event_fd.
291 		 */
292 		callfd = eventfd(0, EFD_CLOEXEC | EFD_NONBLOCK);
293 		if (callfd < 0) {
294 			PMD_DRV_LOG(ERR, "(%s) callfd error, %s", dev->path, strerror(errno));
295 			goto err;
296 		}
297 		kickfd = eventfd(0, EFD_CLOEXEC | EFD_NONBLOCK);
298 		if (kickfd < 0) {
299 			close(callfd);
300 			PMD_DRV_LOG(ERR, "(%s) kickfd error, %s", dev->path, strerror(errno));
301 			goto err;
302 		}
303 		dev->callfds[i] = callfd;
304 		dev->kickfds[i] = kickfd;
305 	}
306 
307 	return 0;
308 err:
309 	for (j = 0; j < i; j++) {
310 		if (dev->kickfds[j] >= 0) {
311 			close(dev->kickfds[j]);
312 			dev->kickfds[j] = -1;
313 		}
314 		if (dev->callfds[j] >= 0) {
315 			close(dev->callfds[j]);
316 			dev->callfds[j] = -1;
317 		}
318 	}
319 
320 	return -1;
321 }
322 
323 static void
324 virtio_user_dev_uninit_notify(struct virtio_user_dev *dev)
325 {
326 	uint32_t i;
327 
328 	for (i = 0; i < dev->max_queue_pairs * 2; ++i) {
329 		if (dev->kickfds[i] >= 0) {
330 			close(dev->kickfds[i]);
331 			dev->kickfds[i] = -1;
332 		}
333 		if (dev->callfds[i] >= 0) {
334 			close(dev->callfds[i]);
335 			dev->callfds[i] = -1;
336 		}
337 	}
338 }
339 
340 static int
341 virtio_user_fill_intr_handle(struct virtio_user_dev *dev)
342 {
343 	uint32_t i;
344 	struct rte_eth_dev *eth_dev = &rte_eth_devices[dev->hw.port_id];
345 
346 	if (!eth_dev->intr_handle) {
347 		eth_dev->intr_handle = malloc(sizeof(*eth_dev->intr_handle));
348 		if (!eth_dev->intr_handle) {
349 			PMD_DRV_LOG(ERR, "(%s) failed to allocate intr_handle", dev->path);
350 			return -1;
351 		}
352 		memset(eth_dev->intr_handle, 0, sizeof(*eth_dev->intr_handle));
353 	}
354 
355 	for (i = 0; i < dev->max_queue_pairs; ++i)
356 		eth_dev->intr_handle->efds[i] = dev->callfds[i];
357 	eth_dev->intr_handle->nb_efd = dev->max_queue_pairs;
358 	eth_dev->intr_handle->max_intr = dev->max_queue_pairs + 1;
359 	eth_dev->intr_handle->type = RTE_INTR_HANDLE_VDEV;
360 	/* For virtio vdev, no need to read counter for clean */
361 	eth_dev->intr_handle->efd_counter_size = 0;
362 	eth_dev->intr_handle->fd = dev->ops->get_intr_fd(dev);
363 
364 	return 0;
365 }
366 
367 static void
368 virtio_user_mem_event_cb(enum rte_mem_event type __rte_unused,
369 			 const void *addr,
370 			 size_t len __rte_unused,
371 			 void *arg)
372 {
373 	struct virtio_user_dev *dev = arg;
374 	struct rte_memseg_list *msl;
375 	uint16_t i;
376 	int ret = 0;
377 
378 	/* ignore externally allocated memory */
379 	msl = rte_mem_virt2memseg_list(addr);
380 	if (msl->external)
381 		return;
382 
383 	pthread_mutex_lock(&dev->mutex);
384 
385 	if (dev->started == false)
386 		goto exit;
387 
388 	/* Step 1: pause the active queues */
389 	for (i = 0; i < dev->queue_pairs; i++) {
390 		ret = dev->ops->enable_qp(dev, i, 0);
391 		if (ret < 0)
392 			goto exit;
393 	}
394 
395 	/* Step 2: update memory regions */
396 	ret = dev->ops->set_memory_table(dev);
397 	if (ret < 0)
398 		goto exit;
399 
400 	/* Step 3: resume the active queues */
401 	for (i = 0; i < dev->queue_pairs; i++) {
402 		ret = dev->ops->enable_qp(dev, i, 1);
403 		if (ret < 0)
404 			goto exit;
405 	}
406 
407 exit:
408 	pthread_mutex_unlock(&dev->mutex);
409 
410 	if (ret < 0)
411 		PMD_DRV_LOG(ERR, "(%s) Failed to update memory table\n", dev->path);
412 }
413 
414 static int
415 virtio_user_dev_setup(struct virtio_user_dev *dev)
416 {
417 	if (dev->is_server) {
418 		if (dev->backend_type != VIRTIO_USER_BACKEND_VHOST_USER) {
419 			PMD_DRV_LOG(ERR, "Server mode only supports vhost-user!");
420 			return -1;
421 		}
422 	}
423 
424 	switch (dev->backend_type) {
425 	case VIRTIO_USER_BACKEND_VHOST_USER:
426 		dev->ops = &virtio_ops_user;
427 		break;
428 	case VIRTIO_USER_BACKEND_VHOST_KERNEL:
429 		dev->ops = &virtio_ops_kernel;
430 		break;
431 	case VIRTIO_USER_BACKEND_VHOST_VDPA:
432 		dev->ops = &virtio_ops_vdpa;
433 		break;
434 	default:
435 		PMD_DRV_LOG(ERR, "(%s) Unknown backend type", dev->path);
436 		return -1;
437 	}
438 
439 	if (dev->ops->setup(dev) < 0) {
440 		PMD_INIT_LOG(ERR, "(%s) Failed to setup backend\n", dev->path);
441 		return -1;
442 	}
443 
444 	if (virtio_user_dev_init_notify(dev) < 0) {
445 		PMD_INIT_LOG(ERR, "(%s) Failed to init notifiers\n", dev->path);
446 		goto destroy;
447 	}
448 
449 	if (virtio_user_fill_intr_handle(dev) < 0) {
450 		PMD_INIT_LOG(ERR, "(%s) Failed to init interrupt handler\n", dev->path);
451 		goto uninit;
452 	}
453 
454 	return 0;
455 
456 uninit:
457 	virtio_user_dev_uninit_notify(dev);
458 destroy:
459 	dev->ops->destroy(dev);
460 
461 	return -1;
462 }
463 
464 /* Use below macro to filter features from vhost backend */
465 #define VIRTIO_USER_SUPPORTED_FEATURES			\
466 	(1ULL << VIRTIO_NET_F_MAC		|	\
467 	 1ULL << VIRTIO_NET_F_STATUS		|	\
468 	 1ULL << VIRTIO_NET_F_MQ		|	\
469 	 1ULL << VIRTIO_NET_F_CTRL_MAC_ADDR	|	\
470 	 1ULL << VIRTIO_NET_F_CTRL_VQ		|	\
471 	 1ULL << VIRTIO_NET_F_CTRL_RX		|	\
472 	 1ULL << VIRTIO_NET_F_CTRL_VLAN		|	\
473 	 1ULL << VIRTIO_NET_F_CSUM		|	\
474 	 1ULL << VIRTIO_NET_F_HOST_TSO4		|	\
475 	 1ULL << VIRTIO_NET_F_HOST_TSO6		|	\
476 	 1ULL << VIRTIO_NET_F_MRG_RXBUF		|	\
477 	 1ULL << VIRTIO_RING_F_INDIRECT_DESC	|	\
478 	 1ULL << VIRTIO_NET_F_GUEST_CSUM	|	\
479 	 1ULL << VIRTIO_NET_F_GUEST_TSO4	|	\
480 	 1ULL << VIRTIO_NET_F_GUEST_TSO6	|	\
481 	 1ULL << VIRTIO_F_IN_ORDER		|	\
482 	 1ULL << VIRTIO_F_VERSION_1		|	\
483 	 1ULL << VIRTIO_F_RING_PACKED)
484 
485 int
486 virtio_user_dev_init(struct virtio_user_dev *dev, char *path, int queues,
487 		     int cq, int queue_size, const char *mac, char **ifname,
488 		     int server, int mrg_rxbuf, int in_order, int packed_vq,
489 		     enum virtio_user_backend_type backend_type)
490 {
491 	uint64_t backend_features;
492 	int i;
493 
494 	pthread_mutex_init(&dev->mutex, NULL);
495 	strlcpy(dev->path, path, PATH_MAX);
496 
497 	for (i = 0; i < VIRTIO_MAX_VIRTQUEUES; i++) {
498 		dev->kickfds[i] = -1;
499 		dev->callfds[i] = -1;
500 	}
501 
502 	dev->started = 0;
503 	dev->max_queue_pairs = queues;
504 	dev->queue_pairs = 1; /* mq disabled by default */
505 	dev->queue_size = queue_size;
506 	dev->is_server = server;
507 	dev->mac_specified = 0;
508 	dev->frontend_features = 0;
509 	dev->unsupported_features = 0;
510 	dev->backend_type = backend_type;
511 
512 	parse_mac(dev, mac);
513 
514 	if (*ifname) {
515 		dev->ifname = *ifname;
516 		*ifname = NULL;
517 	}
518 
519 	if (virtio_user_dev_setup(dev) < 0) {
520 		PMD_INIT_LOG(ERR, "(%s) backend set up fails", dev->path);
521 		return -1;
522 	}
523 
524 	if (dev->ops->set_owner(dev) < 0) {
525 		PMD_INIT_LOG(ERR, "(%s) Failed to set backend owner", dev->path);
526 		return -1;
527 	}
528 
529 	if (dev->ops->get_backend_features(&backend_features) < 0) {
530 		PMD_INIT_LOG(ERR, "(%s) Failed to get backend features", dev->path);
531 		return -1;
532 	}
533 
534 	dev->unsupported_features = ~(VIRTIO_USER_SUPPORTED_FEATURES | backend_features);
535 
536 	if (dev->ops->get_features(dev, &dev->device_features) < 0) {
537 		PMD_INIT_LOG(ERR, "(%s) Failed to get device features", dev->path);
538 		return -1;
539 	}
540 
541 	if (!mrg_rxbuf)
542 		dev->unsupported_features |= (1ull << VIRTIO_NET_F_MRG_RXBUF);
543 
544 	if (!in_order)
545 		dev->unsupported_features |= (1ull << VIRTIO_F_IN_ORDER);
546 
547 	if (!packed_vq)
548 		dev->unsupported_features |= (1ull << VIRTIO_F_RING_PACKED);
549 
550 	if (dev->mac_specified)
551 		dev->frontend_features |= (1ull << VIRTIO_NET_F_MAC);
552 	else
553 		dev->unsupported_features |= (1ull << VIRTIO_NET_F_MAC);
554 
555 	if (cq) {
556 		/* device does not really need to know anything about CQ,
557 		 * so if necessary, we just claim to support CQ
558 		 */
559 		dev->frontend_features |= (1ull << VIRTIO_NET_F_CTRL_VQ);
560 	} else {
561 		dev->unsupported_features |= (1ull << VIRTIO_NET_F_CTRL_VQ);
562 		/* Also disable features that depend on VIRTIO_NET_F_CTRL_VQ */
563 		dev->unsupported_features |= (1ull << VIRTIO_NET_F_CTRL_RX);
564 		dev->unsupported_features |= (1ull << VIRTIO_NET_F_CTRL_VLAN);
565 		dev->unsupported_features |=
566 			(1ull << VIRTIO_NET_F_GUEST_ANNOUNCE);
567 		dev->unsupported_features |= (1ull << VIRTIO_NET_F_MQ);
568 		dev->unsupported_features |=
569 			(1ull << VIRTIO_NET_F_CTRL_MAC_ADDR);
570 	}
571 
572 	/* The backend will not report this feature, we add it explicitly */
573 	if (dev->backend_type == VIRTIO_USER_BACKEND_VHOST_USER)
574 		dev->frontend_features |= (1ull << VIRTIO_NET_F_STATUS);
575 
576 	/*
577 	 * Device features =
578 	 *     (frontend_features | backend_features) & ~unsupported_features;
579 	 */
580 	dev->device_features |= dev->frontend_features;
581 	dev->device_features &= ~dev->unsupported_features;
582 
583 	if (rte_mem_event_callback_register(VIRTIO_USER_MEM_EVENT_CLB_NAME,
584 				virtio_user_mem_event_cb, dev)) {
585 		if (rte_errno != ENOTSUP) {
586 			PMD_INIT_LOG(ERR, "(%s) Failed to register mem event callback\n",
587 					dev->path);
588 			return -1;
589 		}
590 	}
591 
592 	return 0;
593 }
594 
595 void
596 virtio_user_dev_uninit(struct virtio_user_dev *dev)
597 {
598 	virtio_user_stop_device(dev);
599 
600 	rte_mem_event_callback_unregister(VIRTIO_USER_MEM_EVENT_CLB_NAME, dev);
601 
602 	virtio_user_dev_uninit_notify(dev);
603 
604 	free(dev->ifname);
605 
606 	if (dev->is_server)
607 		unlink(dev->path);
608 
609 	dev->ops->destroy(dev);
610 }
611 
612 uint8_t
613 virtio_user_handle_mq(struct virtio_user_dev *dev, uint16_t q_pairs)
614 {
615 	uint16_t i;
616 	uint8_t ret = 0;
617 
618 	if (q_pairs > dev->max_queue_pairs) {
619 		PMD_INIT_LOG(ERR, "(%s) multi-q config %u, but only %u supported",
620 			     dev->path, q_pairs, dev->max_queue_pairs);
621 		return -1;
622 	}
623 
624 	for (i = 0; i < q_pairs; ++i)
625 		ret |= dev->ops->enable_qp(dev, i, 1);
626 	for (i = q_pairs; i < dev->max_queue_pairs; ++i)
627 		ret |= dev->ops->enable_qp(dev, i, 0);
628 
629 	dev->queue_pairs = q_pairs;
630 
631 	return ret;
632 }
633 
634 static uint32_t
635 virtio_user_handle_ctrl_msg(struct virtio_user_dev *dev, struct vring *vring,
636 			    uint16_t idx_hdr)
637 {
638 	struct virtio_net_ctrl_hdr *hdr;
639 	virtio_net_ctrl_ack status = ~0;
640 	uint16_t i, idx_data, idx_status;
641 	uint32_t n_descs = 0;
642 
643 	/* locate desc for header, data, and status */
644 	idx_data = vring->desc[idx_hdr].next;
645 	n_descs++;
646 
647 	i = idx_data;
648 	while (vring->desc[i].flags == VRING_DESC_F_NEXT) {
649 		i = vring->desc[i].next;
650 		n_descs++;
651 	}
652 
653 	/* locate desc for status */
654 	idx_status = i;
655 	n_descs++;
656 
657 	hdr = (void *)(uintptr_t)vring->desc[idx_hdr].addr;
658 	if (hdr->class == VIRTIO_NET_CTRL_MQ &&
659 	    hdr->cmd == VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET) {
660 		uint16_t queues;
661 
662 		queues = *(uint16_t *)(uintptr_t)vring->desc[idx_data].addr;
663 		status = virtio_user_handle_mq(dev, queues);
664 	} else if (hdr->class == VIRTIO_NET_CTRL_RX  ||
665 		   hdr->class == VIRTIO_NET_CTRL_MAC ||
666 		   hdr->class == VIRTIO_NET_CTRL_VLAN) {
667 		status = 0;
668 	}
669 
670 	/* Update status */
671 	*(virtio_net_ctrl_ack *)(uintptr_t)vring->desc[idx_status].addr = status;
672 
673 	return n_descs;
674 }
675 
676 static inline int
677 desc_is_avail(struct vring_packed_desc *desc, bool wrap_counter)
678 {
679 	uint16_t flags = __atomic_load_n(&desc->flags, __ATOMIC_ACQUIRE);
680 
681 	return wrap_counter == !!(flags & VRING_PACKED_DESC_F_AVAIL) &&
682 		wrap_counter != !!(flags & VRING_PACKED_DESC_F_USED);
683 }
684 
685 static uint32_t
686 virtio_user_handle_ctrl_msg_packed(struct virtio_user_dev *dev,
687 				   struct vring_packed *vring,
688 				   uint16_t idx_hdr)
689 {
690 	struct virtio_net_ctrl_hdr *hdr;
691 	virtio_net_ctrl_ack status = ~0;
692 	uint16_t idx_data, idx_status;
693 	/* initialize to one, header is first */
694 	uint32_t n_descs = 1;
695 
696 	/* locate desc for header, data, and status */
697 	idx_data = idx_hdr + 1;
698 	if (idx_data >= dev->queue_size)
699 		idx_data -= dev->queue_size;
700 
701 	n_descs++;
702 
703 	idx_status = idx_data;
704 	while (vring->desc[idx_status].flags & VRING_DESC_F_NEXT) {
705 		idx_status++;
706 		if (idx_status >= dev->queue_size)
707 			idx_status -= dev->queue_size;
708 		n_descs++;
709 	}
710 
711 	hdr = (void *)(uintptr_t)vring->desc[idx_hdr].addr;
712 	if (hdr->class == VIRTIO_NET_CTRL_MQ &&
713 	    hdr->cmd == VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET) {
714 		uint16_t queues;
715 
716 		queues = *(uint16_t *)(uintptr_t)
717 				vring->desc[idx_data].addr;
718 		status = virtio_user_handle_mq(dev, queues);
719 	} else if (hdr->class == VIRTIO_NET_CTRL_RX  ||
720 		   hdr->class == VIRTIO_NET_CTRL_MAC ||
721 		   hdr->class == VIRTIO_NET_CTRL_VLAN) {
722 		status = 0;
723 	}
724 
725 	/* Update status */
726 	*(virtio_net_ctrl_ack *)(uintptr_t)
727 		vring->desc[idx_status].addr = status;
728 
729 	/* Update used descriptor */
730 	vring->desc[idx_hdr].id = vring->desc[idx_status].id;
731 	vring->desc[idx_hdr].len = sizeof(status);
732 
733 	return n_descs;
734 }
735 
736 void
737 virtio_user_handle_cq_packed(struct virtio_user_dev *dev, uint16_t queue_idx)
738 {
739 	struct virtio_user_queue *vq = &dev->packed_queues[queue_idx];
740 	struct vring_packed *vring = &dev->packed_vrings[queue_idx];
741 	uint16_t n_descs, flags;
742 
743 	/* Perform a load-acquire barrier in desc_is_avail to
744 	 * enforce the ordering between desc flags and desc
745 	 * content.
746 	 */
747 	while (desc_is_avail(&vring->desc[vq->used_idx],
748 			     vq->used_wrap_counter)) {
749 
750 		n_descs = virtio_user_handle_ctrl_msg_packed(dev, vring,
751 				vq->used_idx);
752 
753 		flags = VRING_DESC_F_WRITE;
754 		if (vq->used_wrap_counter)
755 			flags |= VRING_PACKED_DESC_F_AVAIL_USED;
756 
757 		__atomic_store_n(&vring->desc[vq->used_idx].flags, flags,
758 				 __ATOMIC_RELEASE);
759 
760 		vq->used_idx += n_descs;
761 		if (vq->used_idx >= dev->queue_size) {
762 			vq->used_idx -= dev->queue_size;
763 			vq->used_wrap_counter ^= 1;
764 		}
765 	}
766 }
767 
768 void
769 virtio_user_handle_cq(struct virtio_user_dev *dev, uint16_t queue_idx)
770 {
771 	uint16_t avail_idx, desc_idx;
772 	struct vring_used_elem *uep;
773 	uint32_t n_descs;
774 	struct vring *vring = &dev->vrings[queue_idx];
775 
776 	/* Consume avail ring, using used ring idx as first one */
777 	while (__atomic_load_n(&vring->used->idx, __ATOMIC_RELAXED)
778 	       != vring->avail->idx) {
779 		avail_idx = __atomic_load_n(&vring->used->idx, __ATOMIC_RELAXED)
780 			    & (vring->num - 1);
781 		desc_idx = vring->avail->ring[avail_idx];
782 
783 		n_descs = virtio_user_handle_ctrl_msg(dev, vring, desc_idx);
784 
785 		/* Update used ring */
786 		uep = &vring->used->ring[avail_idx];
787 		uep->id = desc_idx;
788 		uep->len = n_descs;
789 
790 		__atomic_add_fetch(&vring->used->idx, 1, __ATOMIC_RELAXED);
791 	}
792 }
793 
794 int
795 virtio_user_dev_set_status(struct virtio_user_dev *dev, uint8_t status)
796 {
797 	int ret;
798 
799 	pthread_mutex_lock(&dev->mutex);
800 	dev->status = status;
801 	ret = dev->ops->set_status(dev, status);
802 	if (ret && ret != -ENOTSUP)
803 		PMD_INIT_LOG(ERR, "(%s) Failed to set backend status\n", dev->path);
804 
805 	pthread_mutex_unlock(&dev->mutex);
806 	return ret;
807 }
808 
809 int
810 virtio_user_dev_update_status(struct virtio_user_dev *dev)
811 {
812 	int ret;
813 	uint8_t status;
814 
815 	pthread_mutex_lock(&dev->mutex);
816 
817 	ret = dev->ops->get_status(dev, &status);
818 	if (!ret) {
819 		dev->status = status;
820 		PMD_INIT_LOG(DEBUG, "Updated Device Status(0x%08x):\n"
821 			"\t-RESET: %u\n"
822 			"\t-ACKNOWLEDGE: %u\n"
823 			"\t-DRIVER: %u\n"
824 			"\t-DRIVER_OK: %u\n"
825 			"\t-FEATURES_OK: %u\n"
826 			"\t-DEVICE_NEED_RESET: %u\n"
827 			"\t-FAILED: %u\n",
828 			dev->status,
829 			(dev->status == VIRTIO_CONFIG_STATUS_RESET),
830 			!!(dev->status & VIRTIO_CONFIG_STATUS_ACK),
831 			!!(dev->status & VIRTIO_CONFIG_STATUS_DRIVER),
832 			!!(dev->status & VIRTIO_CONFIG_STATUS_DRIVER_OK),
833 			!!(dev->status & VIRTIO_CONFIG_STATUS_FEATURES_OK),
834 			!!(dev->status & VIRTIO_CONFIG_STATUS_DEV_NEED_RESET),
835 			!!(dev->status & VIRTIO_CONFIG_STATUS_FAILED));
836 	} else if (ret != -ENOTSUP) {
837 		PMD_INIT_LOG(ERR, "(%s) Failed to get backend status\n", dev->path);
838 	}
839 
840 	pthread_mutex_unlock(&dev->mutex);
841 	return ret;
842 }
843 
844 int
845 virtio_user_dev_update_link_state(struct virtio_user_dev *dev)
846 {
847 	if (dev->ops->update_link_state)
848 		return dev->ops->update_link_state(dev);
849 
850 	return 0;
851 }
852 
853 static void
854 virtio_user_dev_reset_queues_packed(struct rte_eth_dev *eth_dev)
855 {
856 	struct virtio_user_dev *dev = eth_dev->data->dev_private;
857 	struct virtio_hw *hw = &dev->hw;
858 	struct virtnet_rx *rxvq;
859 	struct virtnet_tx *txvq;
860 	uint16_t i;
861 
862 	/* Add lock to avoid queue contention. */
863 	rte_spinlock_lock(&hw->state_lock);
864 	hw->started = 0;
865 
866 	/*
867 	 * Waiting for datapath to complete before resetting queues.
868 	 * 1 ms should be enough for the ongoing Tx/Rx function to finish.
869 	 */
870 	rte_delay_ms(1);
871 
872 	/* Vring reset for each Tx queue and Rx queue. */
873 	for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
874 		rxvq = eth_dev->data->rx_queues[i];
875 		virtqueue_rxvq_reset_packed(virtnet_rxq_to_vq(rxvq));
876 		virtio_dev_rx_queue_setup_finish(eth_dev, i);
877 	}
878 
879 	for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
880 		txvq = eth_dev->data->tx_queues[i];
881 		virtqueue_txvq_reset_packed(virtnet_txq_to_vq(txvq));
882 	}
883 
884 	hw->started = 1;
885 	rte_spinlock_unlock(&hw->state_lock);
886 }
887 
888 void
889 virtio_user_dev_delayed_disconnect_handler(void *param)
890 {
891 	struct virtio_user_dev *dev = param;
892 	struct rte_eth_dev *eth_dev = &rte_eth_devices[dev->hw.port_id];
893 
894 	if (rte_intr_disable(eth_dev->intr_handle) < 0) {
895 		PMD_DRV_LOG(ERR, "interrupt disable failed");
896 		return;
897 	}
898 	PMD_DRV_LOG(DEBUG, "Unregistering intr fd: %d",
899 		    eth_dev->intr_handle->fd);
900 	if (rte_intr_callback_unregister(eth_dev->intr_handle,
901 					 virtio_interrupt_handler,
902 					 eth_dev) != 1)
903 		PMD_DRV_LOG(ERR, "interrupt unregister failed");
904 
905 	if (dev->is_server) {
906 		if (dev->ops->server_disconnect)
907 			dev->ops->server_disconnect(dev);
908 
909 		eth_dev->intr_handle->fd = dev->ops->get_intr_fd(dev);
910 
911 		PMD_DRV_LOG(DEBUG, "Registering intr fd: %d",
912 			    eth_dev->intr_handle->fd);
913 
914 		if (rte_intr_callback_register(eth_dev->intr_handle,
915 					       virtio_interrupt_handler,
916 					       eth_dev))
917 			PMD_DRV_LOG(ERR, "interrupt register failed");
918 
919 		if (rte_intr_enable(eth_dev->intr_handle) < 0) {
920 			PMD_DRV_LOG(ERR, "interrupt enable failed");
921 			return;
922 		}
923 	}
924 }
925 
926 static void
927 virtio_user_dev_delayed_intr_reconfig_handler(void *param)
928 {
929 	struct virtio_user_dev *dev = param;
930 	struct rte_eth_dev *eth_dev = &rte_eth_devices[dev->hw.port_id];
931 
932 	PMD_DRV_LOG(DEBUG, "Unregistering intr fd: %d",
933 		    eth_dev->intr_handle->fd);
934 
935 	if (rte_intr_callback_unregister(eth_dev->intr_handle,
936 					 virtio_interrupt_handler,
937 					 eth_dev) != 1)
938 		PMD_DRV_LOG(ERR, "interrupt unregister failed");
939 
940 	eth_dev->intr_handle->fd = dev->ops->get_intr_fd(dev);
941 
942 	PMD_DRV_LOG(DEBUG, "Registering intr fd: %d", eth_dev->intr_handle->fd);
943 
944 	if (rte_intr_callback_register(eth_dev->intr_handle,
945 				       virtio_interrupt_handler, eth_dev))
946 		PMD_DRV_LOG(ERR, "interrupt register failed");
947 
948 	if (rte_intr_enable(eth_dev->intr_handle) < 0)
949 		PMD_DRV_LOG(ERR, "interrupt enable failed");
950 }
951 
952 int
953 virtio_user_dev_server_reconnect(struct virtio_user_dev *dev)
954 {
955 	int ret, old_status;
956 	struct rte_eth_dev *eth_dev = &rte_eth_devices[dev->hw.port_id];
957 	struct virtio_hw *hw = &dev->hw;
958 
959 	if (!dev->ops->server_reconnect) {
960 		PMD_DRV_LOG(ERR, "(%s) Missing server reconnect callback", dev->path);
961 		return -1;
962 	}
963 
964 	if (dev->ops->server_reconnect(dev)) {
965 		PMD_DRV_LOG(ERR, "(%s) Reconnect callback call failed", dev->path);
966 		return -1;
967 	}
968 
969 	old_status = dev->status;
970 
971 	virtio_reset(hw);
972 
973 	virtio_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
974 
975 	virtio_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
976 
977 	if (dev->ops->get_features(dev, &dev->device_features) < 0) {
978 		PMD_INIT_LOG(ERR, "get_features failed: %s",
979 			     strerror(errno));
980 		return -1;
981 	}
982 
983 	dev->device_features |= dev->frontend_features;
984 
985 	/* unmask vhost-user unsupported features */
986 	dev->device_features &= ~(dev->unsupported_features);
987 
988 	dev->features &= dev->device_features;
989 
990 	/* For packed ring, resetting queues is required in reconnection. */
991 	if (virtio_with_packed_queue(hw) &&
992 	   (old_status & VIRTIO_CONFIG_STATUS_DRIVER_OK)) {
993 		PMD_INIT_LOG(NOTICE, "Packets on the fly will be dropped"
994 				" when packed ring reconnecting.");
995 		virtio_user_dev_reset_queues_packed(eth_dev);
996 	}
997 
998 	virtio_set_status(hw, VIRTIO_CONFIG_STATUS_FEATURES_OK);
999 
1000 	/* Start the device */
1001 	virtio_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER_OK);
1002 	if (!dev->started)
1003 		return -1;
1004 
1005 	if (dev->queue_pairs > 1) {
1006 		ret = virtio_user_handle_mq(dev, dev->queue_pairs);
1007 		if (ret != 0) {
1008 			PMD_INIT_LOG(ERR, "Fails to enable multi-queue pairs!");
1009 			return -1;
1010 		}
1011 	}
1012 	if (eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC) {
1013 		if (rte_intr_disable(eth_dev->intr_handle) < 0) {
1014 			PMD_DRV_LOG(ERR, "interrupt disable failed");
1015 			return -1;
1016 		}
1017 		/*
1018 		 * This function can be called from the interrupt handler, so
1019 		 * we can't unregister interrupt handler here.  Setting
1020 		 * alarm to do that later.
1021 		 */
1022 		rte_eal_alarm_set(1,
1023 			virtio_user_dev_delayed_intr_reconfig_handler,
1024 			(void *)dev);
1025 	}
1026 	PMD_INIT_LOG(NOTICE, "server mode virtio-user reconnection succeeds!");
1027 	return 0;
1028 }
1029