xref: /dpdk/drivers/net/virtio/virtio_user/virtio_user_dev.c (revision 68a03efeed657e6e05f281479b33b51102797e15)
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_string_fns.h>
17 #include <rte_eal_memconfig.h>
18 
19 #include "vhost.h"
20 #include "virtio_user_dev.h"
21 #include "../virtio_ethdev.h"
22 
23 #define VIRTIO_USER_MEM_EVENT_CLB_NAME "virtio_user_mem_event_clb"
24 
25 const char * const virtio_user_backend_strings[] = {
26 	[VIRTIO_USER_BACKEND_UNKNOWN] = "VIRTIO_USER_BACKEND_UNKNOWN",
27 	[VIRTIO_USER_BACKEND_VHOST_USER] = "VHOST_USER",
28 	[VIRTIO_USER_BACKEND_VHOST_KERNEL] = "VHOST_NET",
29 	[VIRTIO_USER_BACKEND_VHOST_VDPA] = "VHOST_VDPA",
30 };
31 
32 static int
33 virtio_user_create_queue(struct virtio_user_dev *dev, uint32_t queue_sel)
34 {
35 	/* Of all per virtqueue MSGs, make sure VHOST_SET_VRING_CALL come
36 	 * firstly because vhost depends on this msg to allocate virtqueue
37 	 * pair.
38 	 */
39 	struct vhost_vring_file file;
40 	int ret;
41 
42 	file.index = queue_sel;
43 	file.fd = dev->callfds[queue_sel];
44 	ret = dev->ops->set_vring_call(dev, &file);
45 	if (ret < 0) {
46 		PMD_INIT_LOG(ERR, "(%s) Failed to create queue %u\n", dev->path, queue_sel);
47 		return -1;
48 	}
49 
50 	return 0;
51 }
52 
53 static int
54 virtio_user_kick_queue(struct virtio_user_dev *dev, uint32_t queue_sel)
55 {
56 	int ret;
57 	struct vhost_vring_file file;
58 	struct vhost_vring_state state;
59 	struct vring *vring = &dev->vrings[queue_sel];
60 	struct vring_packed *pq_vring = &dev->packed_vrings[queue_sel];
61 	struct vhost_vring_addr addr = {
62 		.index = queue_sel,
63 		.log_guest_addr = 0,
64 		.flags = 0, /* disable log */
65 	};
66 
67 	if (dev->features & (1ULL << VIRTIO_F_RING_PACKED)) {
68 		addr.desc_user_addr =
69 			(uint64_t)(uintptr_t)pq_vring->desc;
70 		addr.avail_user_addr =
71 			(uint64_t)(uintptr_t)pq_vring->driver;
72 		addr.used_user_addr =
73 			(uint64_t)(uintptr_t)pq_vring->device;
74 	} else {
75 		addr.desc_user_addr = (uint64_t)(uintptr_t)vring->desc;
76 		addr.avail_user_addr = (uint64_t)(uintptr_t)vring->avail;
77 		addr.used_user_addr = (uint64_t)(uintptr_t)vring->used;
78 	}
79 
80 	state.index = queue_sel;
81 	state.num = vring->num;
82 	ret = dev->ops->set_vring_num(dev, &state);
83 	if (ret < 0)
84 		goto err;
85 
86 	state.index = queue_sel;
87 	state.num = 0; /* no reservation */
88 	if (dev->features & (1ULL << VIRTIO_F_RING_PACKED))
89 		state.num |= (1 << 15);
90 	ret = dev->ops->set_vring_base(dev, &state);
91 	if (ret < 0)
92 		goto err;
93 
94 	ret = dev->ops->set_vring_addr(dev, &addr);
95 	if (ret < 0)
96 		goto err;
97 
98 	/* Of all per virtqueue MSGs, make sure VHOST_USER_SET_VRING_KICK comes
99 	 * lastly because vhost depends on this msg to judge if
100 	 * virtio is ready.
101 	 */
102 	file.index = queue_sel;
103 	file.fd = dev->kickfds[queue_sel];
104 	ret = dev->ops->set_vring_kick(dev, &file);
105 	if (ret < 0)
106 		goto err;
107 
108 	return 0;
109 err:
110 	PMD_INIT_LOG(ERR, "(%s) Failed to kick queue %u\n", dev->path, queue_sel);
111 
112 	return -1;
113 }
114 
115 static int
116 virtio_user_queue_setup(struct virtio_user_dev *dev,
117 			int (*fn)(struct virtio_user_dev *, uint32_t))
118 {
119 	uint32_t i, queue_sel;
120 
121 	for (i = 0; i < dev->max_queue_pairs; ++i) {
122 		queue_sel = 2 * i + VTNET_SQ_RQ_QUEUE_IDX;
123 		if (fn(dev, queue_sel) < 0) {
124 			PMD_DRV_LOG(ERR, "(%s) setup rx vq %u failed", dev->path, i);
125 			return -1;
126 		}
127 	}
128 	for (i = 0; i < dev->max_queue_pairs; ++i) {
129 		queue_sel = 2 * i + VTNET_SQ_TQ_QUEUE_IDX;
130 		if (fn(dev, queue_sel) < 0) {
131 			PMD_DRV_LOG(INFO, "(%s) setup tx vq %u failed", dev->path, i);
132 			return -1;
133 		}
134 	}
135 
136 	return 0;
137 }
138 
139 int
140 virtio_user_dev_set_features(struct virtio_user_dev *dev)
141 {
142 	uint64_t features;
143 	int ret = -1;
144 
145 	pthread_mutex_lock(&dev->mutex);
146 
147 	/* Step 0: tell vhost to create queues */
148 	if (virtio_user_queue_setup(dev, virtio_user_create_queue) < 0)
149 		goto error;
150 
151 	features = dev->features;
152 
153 	/* Strip VIRTIO_NET_F_MAC, as MAC address is handled in vdev init */
154 	features &= ~(1ull << VIRTIO_NET_F_MAC);
155 	/* Strip VIRTIO_NET_F_CTRL_VQ, as devices do not really need to know */
156 	features &= ~(1ull << VIRTIO_NET_F_CTRL_VQ);
157 	features &= ~(1ull << VIRTIO_NET_F_STATUS);
158 	ret = dev->ops->set_features(dev, features);
159 	if (ret < 0)
160 		goto error;
161 	PMD_DRV_LOG(INFO, "(%s) set features: 0x%" PRIx64, dev->path, features);
162 error:
163 	pthread_mutex_unlock(&dev->mutex);
164 
165 	return ret;
166 }
167 
168 int
169 virtio_user_start_device(struct virtio_user_dev *dev)
170 {
171 	int ret;
172 
173 	/*
174 	 * XXX workaround!
175 	 *
176 	 * We need to make sure that the locks will be
177 	 * taken in the correct order to avoid deadlocks.
178 	 *
179 	 * Before releasing this lock, this thread should
180 	 * not trigger any memory hotplug events.
181 	 *
182 	 * This is a temporary workaround, and should be
183 	 * replaced when we get proper supports from the
184 	 * memory subsystem in the future.
185 	 */
186 	rte_mcfg_mem_read_lock();
187 	pthread_mutex_lock(&dev->mutex);
188 
189 	/* Step 2: share memory regions */
190 	ret = dev->ops->set_memory_table(dev);
191 	if (ret < 0)
192 		goto error;
193 
194 	/* Step 3: kick queues */
195 	ret = virtio_user_queue_setup(dev, virtio_user_kick_queue);
196 	if (ret < 0)
197 		goto error;
198 
199 	/* Step 4: enable queues
200 	 * we enable the 1st queue pair by default.
201 	 */
202 	ret = dev->ops->enable_qp(dev, 0, 1);
203 	if (ret < 0)
204 		goto error;
205 
206 	dev->started = true;
207 
208 	pthread_mutex_unlock(&dev->mutex);
209 	rte_mcfg_mem_read_unlock();
210 
211 	return 0;
212 error:
213 	pthread_mutex_unlock(&dev->mutex);
214 	rte_mcfg_mem_read_unlock();
215 
216 	PMD_INIT_LOG(ERR, "(%s) Failed to start device\n", dev->path);
217 
218 	/* TODO: free resource here or caller to check */
219 	return -1;
220 }
221 
222 int virtio_user_stop_device(struct virtio_user_dev *dev)
223 {
224 	struct vhost_vring_state state;
225 	uint32_t i;
226 	int ret;
227 
228 	pthread_mutex_lock(&dev->mutex);
229 	if (!dev->started)
230 		goto out;
231 
232 	for (i = 0; i < dev->max_queue_pairs; ++i) {
233 		ret = dev->ops->enable_qp(dev, i, 0);
234 		if (ret < 0)
235 			goto err;
236 	}
237 
238 	/* Stop the backend. */
239 	for (i = 0; i < dev->max_queue_pairs * 2; ++i) {
240 		state.index = i;
241 		ret = dev->ops->get_vring_base(dev, &state);
242 		if (ret < 0) {
243 			PMD_DRV_LOG(ERR, "(%s) get_vring_base failed, index=%u", dev->path, i);
244 			goto err;
245 		}
246 	}
247 
248 	dev->started = false;
249 
250 out:
251 	pthread_mutex_unlock(&dev->mutex);
252 
253 	return 0;
254 err:
255 	pthread_mutex_unlock(&dev->mutex);
256 
257 	PMD_INIT_LOG(ERR, "(%s) Failed to stop device\n", dev->path);
258 
259 	return -1;
260 }
261 
262 static inline void
263 parse_mac(struct virtio_user_dev *dev, const char *mac)
264 {
265 	struct rte_ether_addr tmp;
266 
267 	if (!mac)
268 		return;
269 
270 	if (rte_ether_unformat_addr(mac, &tmp) == 0) {
271 		memcpy(dev->mac_addr, &tmp, RTE_ETHER_ADDR_LEN);
272 		dev->mac_specified = 1;
273 	} else {
274 		/* ignore the wrong mac, use random mac */
275 		PMD_DRV_LOG(ERR, "wrong format of mac: %s", mac);
276 	}
277 }
278 
279 static int
280 virtio_user_dev_init_notify(struct virtio_user_dev *dev)
281 {
282 	uint32_t i, j;
283 	int callfd;
284 	int kickfd;
285 
286 	for (i = 0; i < dev->max_queue_pairs * 2; i++) {
287 		/* May use invalid flag, but some backend uses kickfd and
288 		 * callfd as criteria to judge if dev is alive. so finally we
289 		 * use real event_fd.
290 		 */
291 		callfd = eventfd(0, EFD_CLOEXEC | EFD_NONBLOCK);
292 		if (callfd < 0) {
293 			PMD_DRV_LOG(ERR, "(%s) callfd error, %s", dev->path, strerror(errno));
294 			goto err;
295 		}
296 		kickfd = eventfd(0, EFD_CLOEXEC | EFD_NONBLOCK);
297 		if (kickfd < 0) {
298 			close(callfd);
299 			PMD_DRV_LOG(ERR, "(%s) kickfd error, %s", dev->path, strerror(errno));
300 			goto err;
301 		}
302 		dev->callfds[i] = callfd;
303 		dev->kickfds[i] = kickfd;
304 	}
305 
306 	return 0;
307 err:
308 	for (j = 0; j < i; j++) {
309 		if (dev->kickfds[j] >= 0) {
310 			close(dev->kickfds[j]);
311 			dev->kickfds[j] = -1;
312 		}
313 		if (dev->callfds[j] >= 0) {
314 			close(dev->callfds[j]);
315 			dev->callfds[j] = -1;
316 		}
317 	}
318 
319 	return -1;
320 }
321 
322 static void
323 virtio_user_dev_uninit_notify(struct virtio_user_dev *dev)
324 {
325 	uint32_t i;
326 
327 	for (i = 0; i < dev->max_queue_pairs * 2; ++i) {
328 		if (dev->kickfds[i] >= 0) {
329 			close(dev->kickfds[i]);
330 			dev->kickfds[i] = -1;
331 		}
332 		if (dev->callfds[i] >= 0) {
333 			close(dev->callfds[i]);
334 			dev->callfds[i] = -1;
335 		}
336 	}
337 }
338 
339 static int
340 virtio_user_fill_intr_handle(struct virtio_user_dev *dev)
341 {
342 	uint32_t i;
343 	struct rte_eth_dev *eth_dev = &rte_eth_devices[dev->hw.port_id];
344 
345 	if (!eth_dev->intr_handle) {
346 		eth_dev->intr_handle = malloc(sizeof(*eth_dev->intr_handle));
347 		if (!eth_dev->intr_handle) {
348 			PMD_DRV_LOG(ERR, "(%s) failed to allocate intr_handle", dev->path);
349 			return -1;
350 		}
351 		memset(eth_dev->intr_handle, 0, sizeof(*eth_dev->intr_handle));
352 	}
353 
354 	for (i = 0; i < dev->max_queue_pairs; ++i)
355 		eth_dev->intr_handle->efds[i] = dev->callfds[i];
356 	eth_dev->intr_handle->nb_efd = dev->max_queue_pairs;
357 	eth_dev->intr_handle->max_intr = dev->max_queue_pairs + 1;
358 	eth_dev->intr_handle->type = RTE_INTR_HANDLE_VDEV;
359 	/* For virtio vdev, no need to read counter for clean */
360 	eth_dev->intr_handle->efd_counter_size = 0;
361 	eth_dev->intr_handle->fd = dev->ops->get_intr_fd(dev);
362 
363 	return 0;
364 }
365 
366 static void
367 virtio_user_mem_event_cb(enum rte_mem_event type __rte_unused,
368 			 const void *addr,
369 			 size_t len __rte_unused,
370 			 void *arg)
371 {
372 	struct virtio_user_dev *dev = arg;
373 	struct rte_memseg_list *msl;
374 	uint16_t i;
375 	int ret = 0;
376 
377 	/* ignore externally allocated memory */
378 	msl = rte_mem_virt2memseg_list(addr);
379 	if (msl->external)
380 		return;
381 
382 	pthread_mutex_lock(&dev->mutex);
383 
384 	if (dev->started == false)
385 		goto exit;
386 
387 	/* Step 1: pause the active queues */
388 	for (i = 0; i < dev->queue_pairs; i++) {
389 		ret = dev->ops->enable_qp(dev, i, 0);
390 		if (ret < 0)
391 			goto exit;
392 	}
393 
394 	/* Step 2: update memory regions */
395 	ret = dev->ops->set_memory_table(dev);
396 	if (ret < 0)
397 		goto exit;
398 
399 	/* Step 3: resume the active queues */
400 	for (i = 0; i < dev->queue_pairs; i++) {
401 		ret = dev->ops->enable_qp(dev, i, 1);
402 		if (ret < 0)
403 			goto exit;
404 	}
405 
406 exit:
407 	pthread_mutex_unlock(&dev->mutex);
408 
409 	if (ret < 0)
410 		PMD_DRV_LOG(ERR, "(%s) Failed to update memory table\n", dev->path);
411 }
412 
413 static int
414 virtio_user_dev_setup(struct virtio_user_dev *dev)
415 {
416 	if (dev->is_server) {
417 		if (dev->backend_type != VIRTIO_USER_BACKEND_VHOST_USER) {
418 			PMD_DRV_LOG(ERR, "Server mode only supports vhost-user!");
419 			return -1;
420 		}
421 	}
422 
423 	switch (dev->backend_type) {
424 	case VIRTIO_USER_BACKEND_VHOST_USER:
425 		dev->ops = &virtio_ops_user;
426 		break;
427 	case VIRTIO_USER_BACKEND_VHOST_KERNEL:
428 		dev->ops = &virtio_ops_kernel;
429 		break;
430 	case VIRTIO_USER_BACKEND_VHOST_VDPA:
431 		dev->ops = &virtio_ops_vdpa;
432 		break;
433 	default:
434 		PMD_DRV_LOG(ERR, "(%s) Unknown backend type", dev->path);
435 		return -1;
436 	}
437 
438 	if (dev->ops->setup(dev) < 0) {
439 		PMD_INIT_LOG(ERR, "(%s) Failed to setup backend\n", dev->path);
440 		return -1;
441 	}
442 
443 	if (virtio_user_dev_init_notify(dev) < 0) {
444 		PMD_INIT_LOG(ERR, "(%s) Failed to init notifiers\n", dev->path);
445 		goto destroy;
446 	}
447 
448 	if (virtio_user_fill_intr_handle(dev) < 0) {
449 		PMD_INIT_LOG(ERR, "(%s) Failed to init interrupt handler\n", dev->path);
450 		goto uninit;
451 	}
452 
453 	return 0;
454 
455 uninit:
456 	virtio_user_dev_uninit_notify(dev);
457 destroy:
458 	dev->ops->destroy(dev);
459 
460 	return -1;
461 }
462 
463 /* Use below macro to filter features from vhost backend */
464 #define VIRTIO_USER_SUPPORTED_FEATURES			\
465 	(1ULL << VIRTIO_NET_F_MAC		|	\
466 	 1ULL << VIRTIO_NET_F_STATUS		|	\
467 	 1ULL << VIRTIO_NET_F_MQ		|	\
468 	 1ULL << VIRTIO_NET_F_CTRL_MAC_ADDR	|	\
469 	 1ULL << VIRTIO_NET_F_CTRL_VQ		|	\
470 	 1ULL << VIRTIO_NET_F_CTRL_RX		|	\
471 	 1ULL << VIRTIO_NET_F_CTRL_VLAN		|	\
472 	 1ULL << VIRTIO_NET_F_CSUM		|	\
473 	 1ULL << VIRTIO_NET_F_HOST_TSO4		|	\
474 	 1ULL << VIRTIO_NET_F_HOST_TSO6		|	\
475 	 1ULL << VIRTIO_NET_F_MRG_RXBUF		|	\
476 	 1ULL << VIRTIO_RING_F_INDIRECT_DESC	|	\
477 	 1ULL << VIRTIO_NET_F_GUEST_CSUM	|	\
478 	 1ULL << VIRTIO_NET_F_GUEST_TSO4	|	\
479 	 1ULL << VIRTIO_NET_F_GUEST_TSO6	|	\
480 	 1ULL << VIRTIO_F_IN_ORDER		|	\
481 	 1ULL << VIRTIO_F_VERSION_1		|	\
482 	 1ULL << VIRTIO_F_RING_PACKED)
483 
484 int
485 virtio_user_dev_init(struct virtio_user_dev *dev, char *path, int queues,
486 		     int cq, int queue_size, const char *mac, char **ifname,
487 		     int server, int mrg_rxbuf, int in_order, int packed_vq,
488 		     enum virtio_user_backend_type backend_type)
489 {
490 	uint64_t backend_features;
491 	int i;
492 
493 	pthread_mutex_init(&dev->mutex, NULL);
494 	strlcpy(dev->path, path, PATH_MAX);
495 
496 	for (i = 0; i < VIRTIO_MAX_VIRTQUEUES; i++) {
497 		dev->kickfds[i] = -1;
498 		dev->callfds[i] = -1;
499 	}
500 
501 	dev->started = 0;
502 	dev->max_queue_pairs = queues;
503 	dev->queue_pairs = 1; /* mq disabled by default */
504 	dev->queue_size = queue_size;
505 	dev->is_server = server;
506 	dev->mac_specified = 0;
507 	dev->frontend_features = 0;
508 	dev->unsupported_features = 0;
509 	dev->backend_type = backend_type;
510 
511 	parse_mac(dev, mac);
512 
513 	if (*ifname) {
514 		dev->ifname = *ifname;
515 		*ifname = NULL;
516 	}
517 
518 	if (virtio_user_dev_setup(dev) < 0) {
519 		PMD_INIT_LOG(ERR, "(%s) backend set up fails", dev->path);
520 		return -1;
521 	}
522 
523 	if (dev->ops->set_owner(dev) < 0) {
524 		PMD_INIT_LOG(ERR, "(%s) Failed to set backend owner", dev->path);
525 		return -1;
526 	}
527 
528 	if (dev->ops->get_backend_features(&backend_features) < 0) {
529 		PMD_INIT_LOG(ERR, "(%s) Failed to get backend features", dev->path);
530 		return -1;
531 	}
532 
533 	dev->unsupported_features = ~(VIRTIO_USER_SUPPORTED_FEATURES | backend_features);
534 
535 	if (dev->ops->get_features(dev, &dev->device_features) < 0) {
536 		PMD_INIT_LOG(ERR, "(%s) Failed to get device features", dev->path);
537 		return -1;
538 	}
539 
540 	if (!mrg_rxbuf)
541 		dev->unsupported_features |= (1ull << VIRTIO_NET_F_MRG_RXBUF);
542 
543 	if (!in_order)
544 		dev->unsupported_features |= (1ull << VIRTIO_F_IN_ORDER);
545 
546 	if (!packed_vq)
547 		dev->unsupported_features |= (1ull << VIRTIO_F_RING_PACKED);
548 
549 	if (dev->mac_specified)
550 		dev->frontend_features |= (1ull << VIRTIO_NET_F_MAC);
551 	else
552 		dev->unsupported_features |= (1ull << VIRTIO_NET_F_MAC);
553 
554 	if (cq) {
555 		/* device does not really need to know anything about CQ,
556 		 * so if necessary, we just claim to support CQ
557 		 */
558 		dev->frontend_features |= (1ull << VIRTIO_NET_F_CTRL_VQ);
559 	} else {
560 		dev->unsupported_features |= (1ull << VIRTIO_NET_F_CTRL_VQ);
561 		/* Also disable features that depend on VIRTIO_NET_F_CTRL_VQ */
562 		dev->unsupported_features |= (1ull << VIRTIO_NET_F_CTRL_RX);
563 		dev->unsupported_features |= (1ull << VIRTIO_NET_F_CTRL_VLAN);
564 		dev->unsupported_features |=
565 			(1ull << VIRTIO_NET_F_GUEST_ANNOUNCE);
566 		dev->unsupported_features |= (1ull << VIRTIO_NET_F_MQ);
567 		dev->unsupported_features |=
568 			(1ull << VIRTIO_NET_F_CTRL_MAC_ADDR);
569 	}
570 
571 	/* The backend will not report this feature, we add it explicitly */
572 	if (dev->backend_type == VIRTIO_USER_BACKEND_VHOST_USER)
573 		dev->frontend_features |= (1ull << VIRTIO_NET_F_STATUS);
574 
575 	/*
576 	 * Device features =
577 	 *     (frontend_features | backend_features) & ~unsupported_features;
578 	 */
579 	dev->device_features |= dev->frontend_features;
580 	dev->device_features &= ~dev->unsupported_features;
581 
582 	if (rte_mem_event_callback_register(VIRTIO_USER_MEM_EVENT_CLB_NAME,
583 				virtio_user_mem_event_cb, dev)) {
584 		if (rte_errno != ENOTSUP) {
585 			PMD_INIT_LOG(ERR, "(%s) Failed to register mem event callback\n",
586 					dev->path);
587 			return -1;
588 		}
589 	}
590 
591 	return 0;
592 }
593 
594 void
595 virtio_user_dev_uninit(struct virtio_user_dev *dev)
596 {
597 	virtio_user_stop_device(dev);
598 
599 	rte_mem_event_callback_unregister(VIRTIO_USER_MEM_EVENT_CLB_NAME, dev);
600 
601 	virtio_user_dev_uninit_notify(dev);
602 
603 	free(dev->ifname);
604 
605 	if (dev->is_server)
606 		unlink(dev->path);
607 
608 	dev->ops->destroy(dev);
609 }
610 
611 uint8_t
612 virtio_user_handle_mq(struct virtio_user_dev *dev, uint16_t q_pairs)
613 {
614 	uint16_t i;
615 	uint8_t ret = 0;
616 
617 	if (q_pairs > dev->max_queue_pairs) {
618 		PMD_INIT_LOG(ERR, "(%s) multi-q config %u, but only %u supported",
619 			     dev->path, q_pairs, dev->max_queue_pairs);
620 		return -1;
621 	}
622 
623 	for (i = 0; i < q_pairs; ++i)
624 		ret |= dev->ops->enable_qp(dev, i, 1);
625 	for (i = q_pairs; i < dev->max_queue_pairs; ++i)
626 		ret |= dev->ops->enable_qp(dev, i, 0);
627 
628 	dev->queue_pairs = q_pairs;
629 
630 	return ret;
631 }
632 
633 static uint32_t
634 virtio_user_handle_ctrl_msg(struct virtio_user_dev *dev, struct vring *vring,
635 			    uint16_t idx_hdr)
636 {
637 	struct virtio_net_ctrl_hdr *hdr;
638 	virtio_net_ctrl_ack status = ~0;
639 	uint16_t i, idx_data, idx_status;
640 	uint32_t n_descs = 0;
641 
642 	/* locate desc for header, data, and status */
643 	idx_data = vring->desc[idx_hdr].next;
644 	n_descs++;
645 
646 	i = idx_data;
647 	while (vring->desc[i].flags == VRING_DESC_F_NEXT) {
648 		i = vring->desc[i].next;
649 		n_descs++;
650 	}
651 
652 	/* locate desc for status */
653 	idx_status = i;
654 	n_descs++;
655 
656 	hdr = (void *)(uintptr_t)vring->desc[idx_hdr].addr;
657 	if (hdr->class == VIRTIO_NET_CTRL_MQ &&
658 	    hdr->cmd == VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET) {
659 		uint16_t queues;
660 
661 		queues = *(uint16_t *)(uintptr_t)vring->desc[idx_data].addr;
662 		status = virtio_user_handle_mq(dev, queues);
663 	} else if (hdr->class == VIRTIO_NET_CTRL_RX  ||
664 		   hdr->class == VIRTIO_NET_CTRL_MAC ||
665 		   hdr->class == VIRTIO_NET_CTRL_VLAN) {
666 		status = 0;
667 	}
668 
669 	/* Update status */
670 	*(virtio_net_ctrl_ack *)(uintptr_t)vring->desc[idx_status].addr = status;
671 
672 	return n_descs;
673 }
674 
675 static inline int
676 desc_is_avail(struct vring_packed_desc *desc, bool wrap_counter)
677 {
678 	uint16_t flags = __atomic_load_n(&desc->flags, __ATOMIC_ACQUIRE);
679 
680 	return wrap_counter == !!(flags & VRING_PACKED_DESC_F_AVAIL) &&
681 		wrap_counter != !!(flags & VRING_PACKED_DESC_F_USED);
682 }
683 
684 static uint32_t
685 virtio_user_handle_ctrl_msg_packed(struct virtio_user_dev *dev,
686 				   struct vring_packed *vring,
687 				   uint16_t idx_hdr)
688 {
689 	struct virtio_net_ctrl_hdr *hdr;
690 	virtio_net_ctrl_ack status = ~0;
691 	uint16_t idx_data, idx_status;
692 	/* initialize to one, header is first */
693 	uint32_t n_descs = 1;
694 
695 	/* locate desc for header, data, and status */
696 	idx_data = idx_hdr + 1;
697 	if (idx_data >= dev->queue_size)
698 		idx_data -= dev->queue_size;
699 
700 	n_descs++;
701 
702 	idx_status = idx_data;
703 	while (vring->desc[idx_status].flags & VRING_DESC_F_NEXT) {
704 		idx_status++;
705 		if (idx_status >= dev->queue_size)
706 			idx_status -= dev->queue_size;
707 		n_descs++;
708 	}
709 
710 	hdr = (void *)(uintptr_t)vring->desc[idx_hdr].addr;
711 	if (hdr->class == VIRTIO_NET_CTRL_MQ &&
712 	    hdr->cmd == VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET) {
713 		uint16_t queues;
714 
715 		queues = *(uint16_t *)(uintptr_t)
716 				vring->desc[idx_data].addr;
717 		status = virtio_user_handle_mq(dev, queues);
718 	} else if (hdr->class == VIRTIO_NET_CTRL_RX  ||
719 		   hdr->class == VIRTIO_NET_CTRL_MAC ||
720 		   hdr->class == VIRTIO_NET_CTRL_VLAN) {
721 		status = 0;
722 	}
723 
724 	/* Update status */
725 	*(virtio_net_ctrl_ack *)(uintptr_t)
726 		vring->desc[idx_status].addr = status;
727 
728 	/* Update used descriptor */
729 	vring->desc[idx_hdr].id = vring->desc[idx_status].id;
730 	vring->desc[idx_hdr].len = sizeof(status);
731 
732 	return n_descs;
733 }
734 
735 void
736 virtio_user_handle_cq_packed(struct virtio_user_dev *dev, uint16_t queue_idx)
737 {
738 	struct virtio_user_queue *vq = &dev->packed_queues[queue_idx];
739 	struct vring_packed *vring = &dev->packed_vrings[queue_idx];
740 	uint16_t n_descs, flags;
741 
742 	/* Perform a load-acquire barrier in desc_is_avail to
743 	 * enforce the ordering between desc flags and desc
744 	 * content.
745 	 */
746 	while (desc_is_avail(&vring->desc[vq->used_idx],
747 			     vq->used_wrap_counter)) {
748 
749 		n_descs = virtio_user_handle_ctrl_msg_packed(dev, vring,
750 				vq->used_idx);
751 
752 		flags = VRING_DESC_F_WRITE;
753 		if (vq->used_wrap_counter)
754 			flags |= VRING_PACKED_DESC_F_AVAIL_USED;
755 
756 		__atomic_store_n(&vring->desc[vq->used_idx].flags, flags,
757 				 __ATOMIC_RELEASE);
758 
759 		vq->used_idx += n_descs;
760 		if (vq->used_idx >= dev->queue_size) {
761 			vq->used_idx -= dev->queue_size;
762 			vq->used_wrap_counter ^= 1;
763 		}
764 	}
765 }
766 
767 void
768 virtio_user_handle_cq(struct virtio_user_dev *dev, uint16_t queue_idx)
769 {
770 	uint16_t avail_idx, desc_idx;
771 	struct vring_used_elem *uep;
772 	uint32_t n_descs;
773 	struct vring *vring = &dev->vrings[queue_idx];
774 
775 	/* Consume avail ring, using used ring idx as first one */
776 	while (__atomic_load_n(&vring->used->idx, __ATOMIC_RELAXED)
777 	       != vring->avail->idx) {
778 		avail_idx = __atomic_load_n(&vring->used->idx, __ATOMIC_RELAXED)
779 			    & (vring->num - 1);
780 		desc_idx = vring->avail->ring[avail_idx];
781 
782 		n_descs = virtio_user_handle_ctrl_msg(dev, vring, desc_idx);
783 
784 		/* Update used ring */
785 		uep = &vring->used->ring[avail_idx];
786 		uep->id = desc_idx;
787 		uep->len = n_descs;
788 
789 		__atomic_add_fetch(&vring->used->idx, 1, __ATOMIC_RELAXED);
790 	}
791 }
792 
793 int
794 virtio_user_dev_set_status(struct virtio_user_dev *dev, uint8_t status)
795 {
796 	int ret;
797 
798 	pthread_mutex_lock(&dev->mutex);
799 	dev->status = status;
800 	ret = dev->ops->set_status(dev, status);
801 	if (ret && ret != -ENOTSUP)
802 		PMD_INIT_LOG(ERR, "(%s) Failed to set backend status\n", dev->path);
803 
804 	pthread_mutex_unlock(&dev->mutex);
805 	return ret;
806 }
807 
808 int
809 virtio_user_dev_update_status(struct virtio_user_dev *dev)
810 {
811 	int ret;
812 	uint8_t status;
813 
814 	pthread_mutex_lock(&dev->mutex);
815 
816 	ret = dev->ops->get_status(dev, &status);
817 	if (!ret) {
818 		dev->status = status;
819 		PMD_INIT_LOG(DEBUG, "Updated Device Status(0x%08x):\n"
820 			"\t-RESET: %u\n"
821 			"\t-ACKNOWLEDGE: %u\n"
822 			"\t-DRIVER: %u\n"
823 			"\t-DRIVER_OK: %u\n"
824 			"\t-FEATURES_OK: %u\n"
825 			"\t-DEVICE_NEED_RESET: %u\n"
826 			"\t-FAILED: %u\n",
827 			dev->status,
828 			(dev->status == VIRTIO_CONFIG_STATUS_RESET),
829 			!!(dev->status & VIRTIO_CONFIG_STATUS_ACK),
830 			!!(dev->status & VIRTIO_CONFIG_STATUS_DRIVER),
831 			!!(dev->status & VIRTIO_CONFIG_STATUS_DRIVER_OK),
832 			!!(dev->status & VIRTIO_CONFIG_STATUS_FEATURES_OK),
833 			!!(dev->status & VIRTIO_CONFIG_STATUS_DEV_NEED_RESET),
834 			!!(dev->status & VIRTIO_CONFIG_STATUS_FAILED));
835 	} else if (ret != -ENOTSUP) {
836 		PMD_INIT_LOG(ERR, "(%s) Failed to get backend status\n", dev->path);
837 	}
838 
839 	pthread_mutex_unlock(&dev->mutex);
840 	return ret;
841 }
842 
843 int
844 virtio_user_dev_update_link_state(struct virtio_user_dev *dev)
845 {
846 	if (dev->ops->update_link_state)
847 		return dev->ops->update_link_state(dev);
848 
849 	return 0;
850 }
851 
852 static void
853 virtio_user_dev_reset_queues_packed(struct rte_eth_dev *eth_dev)
854 {
855 	struct virtio_user_dev *dev = eth_dev->data->dev_private;
856 	struct virtio_hw *hw = &dev->hw;
857 	struct virtnet_rx *rxvq;
858 	struct virtnet_tx *txvq;
859 	uint16_t i;
860 
861 	/* Add lock to avoid queue contention. */
862 	rte_spinlock_lock(&hw->state_lock);
863 	hw->started = 0;
864 
865 	/*
866 	 * Waiting for datapath to complete before resetting queues.
867 	 * 1 ms should be enough for the ongoing Tx/Rx function to finish.
868 	 */
869 	rte_delay_ms(1);
870 
871 	/* Vring reset for each Tx queue and Rx queue. */
872 	for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
873 		rxvq = eth_dev->data->rx_queues[i];
874 		virtqueue_rxvq_reset_packed(virtnet_rxq_to_vq(rxvq));
875 		virtio_dev_rx_queue_setup_finish(eth_dev, i);
876 	}
877 
878 	for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
879 		txvq = eth_dev->data->tx_queues[i];
880 		virtqueue_txvq_reset_packed(virtnet_txq_to_vq(txvq));
881 	}
882 
883 	hw->started = 1;
884 	rte_spinlock_unlock(&hw->state_lock);
885 }
886 
887 void
888 virtio_user_dev_delayed_handler(void *param)
889 {
890 	struct virtio_user_dev *dev = param;
891 	struct rte_eth_dev *eth_dev = &rte_eth_devices[dev->hw.port_id];
892 
893 	if (rte_intr_disable(eth_dev->intr_handle) < 0) {
894 		PMD_DRV_LOG(ERR, "interrupt disable failed");
895 		return;
896 	}
897 	rte_intr_callback_unregister(eth_dev->intr_handle,
898 				     virtio_interrupt_handler, eth_dev);
899 	if (dev->is_server) {
900 		if (dev->ops->server_disconnect)
901 			dev->ops->server_disconnect(dev);
902 		eth_dev->intr_handle->fd = dev->ops->get_intr_fd(dev);
903 		rte_intr_callback_register(eth_dev->intr_handle,
904 					   virtio_interrupt_handler, eth_dev);
905 		if (rte_intr_enable(eth_dev->intr_handle) < 0) {
906 			PMD_DRV_LOG(ERR, "interrupt enable failed");
907 			return;
908 		}
909 	}
910 }
911 
912 int
913 virtio_user_dev_server_reconnect(struct virtio_user_dev *dev)
914 {
915 	int ret, old_status;
916 	struct rte_eth_dev *eth_dev = &rte_eth_devices[dev->hw.port_id];
917 	struct virtio_hw *hw = &dev->hw;
918 
919 	if (!dev->ops->server_reconnect) {
920 		PMD_DRV_LOG(ERR, "(%s) Missing server reconnect callback", dev->path);
921 		return -1;
922 	}
923 
924 	if (dev->ops->server_reconnect(dev)) {
925 		PMD_DRV_LOG(ERR, "(%s) Reconnect callback call failed", dev->path);
926 		return -1;
927 	}
928 
929 	old_status = dev->status;
930 
931 	virtio_reset(hw);
932 
933 	virtio_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
934 
935 	virtio_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
936 
937 	if (dev->ops->get_features(dev, &dev->device_features) < 0) {
938 		PMD_INIT_LOG(ERR, "get_features failed: %s",
939 			     strerror(errno));
940 		return -1;
941 	}
942 
943 	dev->device_features |= dev->frontend_features;
944 
945 	/* unmask vhost-user unsupported features */
946 	dev->device_features &= ~(dev->unsupported_features);
947 
948 	dev->features &= dev->device_features;
949 
950 	/* For packed ring, resetting queues is required in reconnection. */
951 	if (virtio_with_packed_queue(hw) &&
952 	   (old_status & VIRTIO_CONFIG_STATUS_DRIVER_OK)) {
953 		PMD_INIT_LOG(NOTICE, "Packets on the fly will be dropped"
954 				" when packed ring reconnecting.");
955 		virtio_user_dev_reset_queues_packed(eth_dev);
956 	}
957 
958 	virtio_set_status(hw, VIRTIO_CONFIG_STATUS_FEATURES_OK);
959 
960 	/* Start the device */
961 	virtio_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER_OK);
962 	if (!dev->started)
963 		return -1;
964 
965 	if (dev->queue_pairs > 1) {
966 		ret = virtio_user_handle_mq(dev, dev->queue_pairs);
967 		if (ret != 0) {
968 			PMD_INIT_LOG(ERR, "Fails to enable multi-queue pairs!");
969 			return -1;
970 		}
971 	}
972 	if (eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC) {
973 		if (rte_intr_disable(eth_dev->intr_handle) < 0) {
974 			PMD_DRV_LOG(ERR, "interrupt disable failed");
975 			return -1;
976 		}
977 		rte_intr_callback_unregister(eth_dev->intr_handle,
978 					     virtio_interrupt_handler,
979 					     eth_dev);
980 
981 		eth_dev->intr_handle->fd = dev->ops->get_intr_fd(dev);
982 		rte_intr_callback_register(eth_dev->intr_handle,
983 					   virtio_interrupt_handler, eth_dev);
984 
985 		if (rte_intr_enable(eth_dev->intr_handle) < 0) {
986 			PMD_DRV_LOG(ERR, "interrupt enable failed");
987 			return -1;
988 		}
989 	}
990 	PMD_INIT_LOG(NOTICE, "server mode virtio-user reconnection succeeds!");
991 	return 0;
992 }
993