xref: /dpdk/drivers/net/virtio/virtio_user/virtio_user_dev.c (revision e9fd1ebf981f361844aea9ec94e17f4bda5e1479)
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 <stdlib.h>
8 #include <fcntl.h>
9 #include <string.h>
10 #include <errno.h>
11 #include <sys/mman.h>
12 #include <unistd.h>
13 #include <sys/eventfd.h>
14 #include <sys/types.h>
15 #include <sys/stat.h>
16 
17 #include <rte_alarm.h>
18 #include <rte_string_fns.h>
19 #include <rte_eal_memconfig.h>
20 #include <rte_malloc.h>
21 #include <rte_io.h>
22 
23 #include "vhost.h"
24 #include "virtio_user_dev.h"
25 #include "../virtio_ethdev.h"
26 
27 #define VIRTIO_USER_MEM_EVENT_CLB_NAME "virtio_user_mem_event_clb"
28 
29 const char * const virtio_user_backend_strings[] = {
30 	[VIRTIO_USER_BACKEND_UNKNOWN] = "VIRTIO_USER_BACKEND_UNKNOWN",
31 	[VIRTIO_USER_BACKEND_VHOST_USER] = "VHOST_USER",
32 	[VIRTIO_USER_BACKEND_VHOST_KERNEL] = "VHOST_NET",
33 	[VIRTIO_USER_BACKEND_VHOST_VDPA] = "VHOST_VDPA",
34 };
35 
36 static int
37 virtio_user_create_queue(struct virtio_user_dev *dev, uint32_t queue_sel)
38 {
39 	/* Of all per virtqueue MSGs, make sure VHOST_SET_VRING_CALL come
40 	 * firstly because vhost depends on this msg to allocate virtqueue
41 	 * pair.
42 	 */
43 	struct vhost_vring_file file;
44 	int ret;
45 
46 	file.index = queue_sel;
47 	file.fd = dev->callfds[queue_sel];
48 	ret = dev->ops->set_vring_call(dev, &file);
49 	if (ret < 0) {
50 		PMD_INIT_LOG(ERR, "(%s) Failed to create queue %u", dev->path, queue_sel);
51 		return -1;
52 	}
53 
54 	return 0;
55 }
56 
57 static int
58 virtio_user_kick_queue(struct virtio_user_dev *dev, uint32_t queue_sel)
59 {
60 	int ret;
61 	struct vhost_vring_file file;
62 	struct vhost_vring_state state;
63 	struct vring *vring = &dev->vrings.split[queue_sel];
64 	struct vring_packed *pq_vring = &dev->vrings.packed[queue_sel];
65 	struct vhost_vring_addr addr = {
66 		.index = queue_sel,
67 		.log_guest_addr = 0,
68 		.flags = 0, /* disable log */
69 	};
70 
71 	if (queue_sel == dev->max_queue_pairs * 2) {
72 		if (!dev->scvq) {
73 			PMD_INIT_LOG(ERR, "(%s) Shadow control queue expected but missing",
74 					dev->path);
75 			goto err;
76 		}
77 
78 		/* Use shadow control queue information */
79 		vring = &dev->scvq->vq_split.ring;
80 		pq_vring = &dev->scvq->vq_packed.ring;
81 	}
82 
83 	if (dev->features & (1ULL << VIRTIO_F_RING_PACKED)) {
84 		addr.desc_user_addr =
85 			(uint64_t)(uintptr_t)pq_vring->desc;
86 		addr.avail_user_addr =
87 			(uint64_t)(uintptr_t)pq_vring->driver;
88 		addr.used_user_addr =
89 			(uint64_t)(uintptr_t)pq_vring->device;
90 	} else {
91 		addr.desc_user_addr = (uint64_t)(uintptr_t)vring->desc;
92 		addr.avail_user_addr = (uint64_t)(uintptr_t)vring->avail;
93 		addr.used_user_addr = (uint64_t)(uintptr_t)vring->used;
94 	}
95 
96 	state.index = queue_sel;
97 	state.num = vring->num;
98 	ret = dev->ops->set_vring_num(dev, &state);
99 	if (ret < 0)
100 		goto err;
101 
102 	state.index = queue_sel;
103 	state.num = 0; /* no reservation */
104 	if (dev->features & (1ULL << VIRTIO_F_RING_PACKED))
105 		state.num |= (1 << 15);
106 	ret = dev->ops->set_vring_base(dev, &state);
107 	if (ret < 0)
108 		goto err;
109 
110 	ret = dev->ops->set_vring_addr(dev, &addr);
111 	if (ret < 0)
112 		goto err;
113 
114 	/* Of all per virtqueue MSGs, make sure VHOST_USER_SET_VRING_KICK comes
115 	 * lastly because vhost depends on this msg to judge if
116 	 * virtio is ready.
117 	 */
118 	file.index = queue_sel;
119 	file.fd = dev->kickfds[queue_sel];
120 	ret = dev->ops->set_vring_kick(dev, &file);
121 	if (ret < 0)
122 		goto err;
123 
124 	return 0;
125 err:
126 	PMD_INIT_LOG(ERR, "(%s) Failed to kick queue %u", dev->path, queue_sel);
127 
128 	return -1;
129 }
130 
131 static int
132 virtio_user_queue_setup(struct virtio_user_dev *dev,
133 			int (*fn)(struct virtio_user_dev *, uint32_t))
134 {
135 	uint32_t i, nr_vq;
136 
137 	nr_vq = dev->max_queue_pairs * 2;
138 	if (dev->hw_cvq)
139 		nr_vq++;
140 
141 	for (i = 0; i < nr_vq; i++) {
142 		if (fn(dev, i) < 0) {
143 			PMD_DRV_LOG(ERR, "(%s) setup VQ %u failed", dev->path, i);
144 			return -1;
145 		}
146 	}
147 
148 	return 0;
149 }
150 
151 int
152 virtio_user_dev_set_features(struct virtio_user_dev *dev)
153 {
154 	uint64_t features;
155 	int ret = -1;
156 
157 	pthread_mutex_lock(&dev->mutex);
158 
159 	/* Step 0: tell vhost to create queues */
160 	if (virtio_user_queue_setup(dev, virtio_user_create_queue) < 0)
161 		goto error;
162 
163 	features = dev->features;
164 
165 	/* Strip VIRTIO_NET_F_MAC, as MAC address is handled in vdev init */
166 	features &= ~(1ull << VIRTIO_NET_F_MAC);
167 	/* Strip VIRTIO_NET_F_CTRL_VQ if the devices does not really support control VQ */
168 	if (!dev->hw_cvq)
169 		features &= ~(1ull << VIRTIO_NET_F_CTRL_VQ);
170 	features &= ~(1ull << VIRTIO_NET_F_STATUS);
171 	ret = dev->ops->set_features(dev, features);
172 	if (ret < 0)
173 		goto error;
174 	PMD_DRV_LOG(INFO, "(%s) set features: 0x%" PRIx64, dev->path, features);
175 error:
176 	pthread_mutex_unlock(&dev->mutex);
177 
178 	return ret;
179 }
180 
181 int
182 virtio_user_start_device(struct virtio_user_dev *dev)
183 {
184 	int ret;
185 
186 	/*
187 	 * XXX workaround!
188 	 *
189 	 * We need to make sure that the locks will be
190 	 * taken in the correct order to avoid deadlocks.
191 	 *
192 	 * Before releasing this lock, this thread should
193 	 * not trigger any memory hotplug events.
194 	 *
195 	 * This is a temporary workaround, and should be
196 	 * replaced when we get proper supports from the
197 	 * memory subsystem in the future.
198 	 */
199 	rte_mcfg_mem_read_lock();
200 	pthread_mutex_lock(&dev->mutex);
201 
202 	/* Step 2: share memory regions */
203 	ret = dev->ops->set_memory_table(dev);
204 	if (ret < 0)
205 		goto error;
206 
207 	/* Step 3: kick queues */
208 	ret = virtio_user_queue_setup(dev, virtio_user_kick_queue);
209 	if (ret < 0)
210 		goto error;
211 
212 	/* Step 4: enable queues
213 	 * we enable the 1st queue pair by default.
214 	 */
215 	ret = dev->ops->enable_qp(dev, 0, 1);
216 	if (ret < 0)
217 		goto error;
218 
219 	dev->started = true;
220 
221 	pthread_mutex_unlock(&dev->mutex);
222 	rte_mcfg_mem_read_unlock();
223 
224 	return 0;
225 error:
226 	pthread_mutex_unlock(&dev->mutex);
227 	rte_mcfg_mem_read_unlock();
228 
229 	PMD_INIT_LOG(ERR, "(%s) Failed to start device", dev->path);
230 
231 	/* TODO: free resource here or caller to check */
232 	return -1;
233 }
234 
235 int virtio_user_stop_device(struct virtio_user_dev *dev)
236 {
237 	struct vhost_vring_state state;
238 	uint32_t i;
239 	int ret;
240 
241 	pthread_mutex_lock(&dev->mutex);
242 	if (!dev->started)
243 		goto out;
244 
245 	for (i = 0; i < dev->max_queue_pairs; ++i) {
246 		ret = dev->ops->enable_qp(dev, i, 0);
247 		if (ret < 0)
248 			goto err;
249 	}
250 
251 	/* Stop the backend. */
252 	for (i = 0; i < dev->max_queue_pairs * 2; ++i) {
253 		state.index = i;
254 		ret = dev->ops->get_vring_base(dev, &state);
255 		if (ret < 0) {
256 			PMD_DRV_LOG(ERR, "(%s) get_vring_base failed, index=%u", dev->path, i);
257 			goto err;
258 		}
259 	}
260 
261 	dev->started = false;
262 
263 out:
264 	pthread_mutex_unlock(&dev->mutex);
265 
266 	return 0;
267 err:
268 	pthread_mutex_unlock(&dev->mutex);
269 
270 	PMD_INIT_LOG(ERR, "(%s) Failed to stop device", dev->path);
271 
272 	return -1;
273 }
274 
275 static int
276 virtio_user_dev_init_max_queue_pairs(struct virtio_user_dev *dev, uint32_t user_max_qp)
277 {
278 	int ret;
279 
280 	if (!(dev->device_features & (1ULL << VIRTIO_NET_F_MQ))) {
281 		dev->max_queue_pairs = 1;
282 		return 0;
283 	}
284 
285 	if (!dev->ops->get_config) {
286 		dev->max_queue_pairs = user_max_qp;
287 		return 0;
288 	}
289 
290 	ret = dev->ops->get_config(dev, (uint8_t *)&dev->max_queue_pairs,
291 			offsetof(struct virtio_net_config, max_virtqueue_pairs),
292 			sizeof(uint16_t));
293 	if (ret) {
294 		/*
295 		 * We need to know the max queue pair from the device so that
296 		 * the control queue gets the right index.
297 		 */
298 		dev->max_queue_pairs = 1;
299 		PMD_DRV_LOG(ERR, "(%s) Failed to get max queue pairs from device", dev->path);
300 
301 		return ret;
302 	}
303 
304 	return 0;
305 }
306 
307 int
308 virtio_user_dev_get_rss_config(struct virtio_user_dev *dev, void *dst, size_t offset, int length)
309 {
310 	int ret = 0;
311 
312 	if (!(dev->device_features & (1ULL << VIRTIO_NET_F_RSS)))
313 		return -ENOTSUP;
314 
315 	if (!dev->ops->get_config)
316 		return -ENOTSUP;
317 
318 	ret = dev->ops->get_config(dev, dst, offset, length);
319 	if (ret)
320 		PMD_DRV_LOG(ERR, "(%s) Failed to get rss config in device", dev->path);
321 
322 	return ret;
323 }
324 
325 int
326 virtio_user_dev_set_mac(struct virtio_user_dev *dev)
327 {
328 	int ret = 0;
329 
330 	if (!(dev->device_features & (1ULL << VIRTIO_NET_F_MAC)))
331 		return -ENOTSUP;
332 
333 	if (!dev->ops->set_config)
334 		return -ENOTSUP;
335 
336 	ret = dev->ops->set_config(dev, dev->mac_addr,
337 			offsetof(struct virtio_net_config, mac),
338 			RTE_ETHER_ADDR_LEN);
339 	if (ret)
340 		PMD_DRV_LOG(ERR, "(%s) Failed to set MAC address in device", dev->path);
341 
342 	return ret;
343 }
344 
345 int
346 virtio_user_dev_get_mac(struct virtio_user_dev *dev)
347 {
348 	int ret = 0;
349 
350 	if (!(dev->device_features & (1ULL << VIRTIO_NET_F_MAC)))
351 		return -ENOTSUP;
352 
353 	if (!dev->ops->get_config)
354 		return -ENOTSUP;
355 
356 	ret = dev->ops->get_config(dev, dev->mac_addr,
357 			offsetof(struct virtio_net_config, mac),
358 			RTE_ETHER_ADDR_LEN);
359 	if (ret)
360 		PMD_DRV_LOG(ERR, "(%s) Failed to get MAC address from device", dev->path);
361 
362 	return ret;
363 }
364 
365 static void
366 virtio_user_dev_init_mac(struct virtio_user_dev *dev, const char *mac)
367 {
368 	struct rte_ether_addr cmdline_mac;
369 	char buf[RTE_ETHER_ADDR_FMT_SIZE];
370 	int ret;
371 
372 	if (mac && rte_ether_unformat_addr(mac, &cmdline_mac) == 0) {
373 		/*
374 		 * MAC address was passed from command-line, try to store
375 		 * it in the device if it supports it. Otherwise try to use
376 		 * the device one.
377 		 */
378 		memcpy(dev->mac_addr, &cmdline_mac, RTE_ETHER_ADDR_LEN);
379 		dev->mac_specified = 1;
380 
381 		/* Setting MAC may fail, continue to get the device one in this case */
382 		virtio_user_dev_set_mac(dev);
383 		ret = virtio_user_dev_get_mac(dev);
384 		if (ret == -ENOTSUP)
385 			goto out;
386 
387 		if (memcmp(&cmdline_mac, dev->mac_addr, RTE_ETHER_ADDR_LEN))
388 			PMD_DRV_LOG(INFO, "(%s) Device MAC update failed", dev->path);
389 	} else {
390 		ret = virtio_user_dev_get_mac(dev);
391 		if (ret) {
392 			PMD_DRV_LOG(ERR, "(%s) No valid MAC in devargs or device, use random",
393 					dev->path);
394 			return;
395 		}
396 
397 		dev->mac_specified = 1;
398 	}
399 out:
400 	rte_ether_format_addr(buf, RTE_ETHER_ADDR_FMT_SIZE,
401 			(struct rte_ether_addr *)dev->mac_addr);
402 	PMD_DRV_LOG(INFO, "(%s) MAC %s specified", dev->path, buf);
403 }
404 
405 static int
406 virtio_user_dev_init_notify(struct virtio_user_dev *dev)
407 {
408 	uint32_t i, j, nr_vq;
409 	int callfd;
410 	int kickfd;
411 
412 	nr_vq = dev->max_queue_pairs * 2;
413 	if (dev->hw_cvq)
414 		nr_vq++;
415 
416 	for (i = 0; i < nr_vq; i++) {
417 		/* May use invalid flag, but some backend uses kickfd and
418 		 * callfd as criteria to judge if dev is alive. so finally we
419 		 * use real event_fd.
420 		 */
421 		callfd = eventfd(0, EFD_CLOEXEC | EFD_NONBLOCK);
422 		if (callfd < 0) {
423 			PMD_DRV_LOG(ERR, "(%s) callfd error, %s", dev->path, strerror(errno));
424 			goto err;
425 		}
426 		kickfd = eventfd(0, EFD_CLOEXEC | EFD_NONBLOCK);
427 		if (kickfd < 0) {
428 			close(callfd);
429 			PMD_DRV_LOG(ERR, "(%s) kickfd error, %s", dev->path, strerror(errno));
430 			goto err;
431 		}
432 		dev->callfds[i] = callfd;
433 		dev->kickfds[i] = kickfd;
434 	}
435 
436 	if (dev->ops->map_notification_area)
437 		if (dev->ops->map_notification_area(dev))
438 			goto err;
439 
440 	return 0;
441 err:
442 	for (j = 0; j < i; j++) {
443 		if (dev->kickfds[j] >= 0) {
444 			close(dev->kickfds[j]);
445 			dev->kickfds[j] = -1;
446 		}
447 		if (dev->callfds[j] >= 0) {
448 			close(dev->callfds[j]);
449 			dev->callfds[j] = -1;
450 		}
451 	}
452 
453 	return -1;
454 }
455 
456 static void
457 virtio_user_dev_uninit_notify(struct virtio_user_dev *dev)
458 {
459 	uint32_t i;
460 
461 	for (i = 0; i < dev->max_queue_pairs * 2; ++i) {
462 		if (dev->kickfds[i] >= 0) {
463 			close(dev->kickfds[i]);
464 			dev->kickfds[i] = -1;
465 		}
466 		if (dev->callfds[i] >= 0) {
467 			close(dev->callfds[i]);
468 			dev->callfds[i] = -1;
469 		}
470 	}
471 	if (dev->ops->unmap_notification_area && dev->notify_area)
472 		dev->ops->unmap_notification_area(dev);
473 }
474 
475 static int
476 virtio_user_fill_intr_handle(struct virtio_user_dev *dev)
477 {
478 	uint32_t i;
479 	struct rte_eth_dev *eth_dev = &rte_eth_devices[dev->hw.port_id];
480 
481 	if (eth_dev->intr_handle == NULL) {
482 		eth_dev->intr_handle =
483 			rte_intr_instance_alloc(RTE_INTR_INSTANCE_F_PRIVATE);
484 		if (eth_dev->intr_handle == NULL) {
485 			PMD_DRV_LOG(ERR, "(%s) failed to allocate intr_handle", dev->path);
486 			return -1;
487 		}
488 	}
489 
490 	for (i = 0; i < dev->max_queue_pairs; ++i) {
491 		if (rte_intr_efds_index_set(eth_dev->intr_handle, i,
492 				dev->callfds[2 * i + VTNET_SQ_RQ_QUEUE_IDX]))
493 			return -rte_errno;
494 	}
495 
496 	if (rte_intr_nb_efd_set(eth_dev->intr_handle, dev->max_queue_pairs))
497 		return -rte_errno;
498 
499 	if (rte_intr_max_intr_set(eth_dev->intr_handle,
500 			dev->max_queue_pairs + 1))
501 		return -rte_errno;
502 
503 	if (rte_intr_type_set(eth_dev->intr_handle, RTE_INTR_HANDLE_VDEV))
504 		return -rte_errno;
505 
506 	/* For virtio vdev, no need to read counter for clean */
507 	if (rte_intr_efd_counter_size_set(eth_dev->intr_handle, 0))
508 		return -rte_errno;
509 
510 	if (rte_intr_fd_set(eth_dev->intr_handle, dev->ops->get_intr_fd(dev)))
511 		return -rte_errno;
512 
513 	return 0;
514 }
515 
516 static void
517 virtio_user_mem_event_cb(enum rte_mem_event type __rte_unused,
518 			 const void *addr,
519 			 size_t len __rte_unused,
520 			 void *arg)
521 {
522 	struct virtio_user_dev *dev = arg;
523 	struct rte_memseg_list *msl;
524 	uint16_t i;
525 	int ret = 0;
526 
527 	/* ignore externally allocated memory */
528 	msl = rte_mem_virt2memseg_list(addr);
529 	if (msl->external)
530 		return;
531 
532 	pthread_mutex_lock(&dev->mutex);
533 
534 	if (dev->started == false)
535 		goto exit;
536 
537 	/* Step 1: pause the active queues */
538 	for (i = 0; i < dev->queue_pairs; i++) {
539 		ret = dev->ops->enable_qp(dev, i, 0);
540 		if (ret < 0)
541 			goto exit;
542 	}
543 
544 	/* Step 2: update memory regions */
545 	ret = dev->ops->set_memory_table(dev);
546 	if (ret < 0)
547 		goto exit;
548 
549 	/* Step 3: resume the active queues */
550 	for (i = 0; i < dev->queue_pairs; i++) {
551 		ret = dev->ops->enable_qp(dev, i, 1);
552 		if (ret < 0)
553 			goto exit;
554 	}
555 
556 exit:
557 	pthread_mutex_unlock(&dev->mutex);
558 
559 	if (ret < 0)
560 		PMD_DRV_LOG(ERR, "(%s) Failed to update memory table", dev->path);
561 }
562 
563 static int
564 virtio_user_dev_setup(struct virtio_user_dev *dev)
565 {
566 	if (dev->is_server) {
567 		if (dev->backend_type != VIRTIO_USER_BACKEND_VHOST_USER) {
568 			PMD_DRV_LOG(ERR, "Server mode only supports vhost-user!");
569 			return -1;
570 		}
571 	}
572 
573 	switch (dev->backend_type) {
574 	case VIRTIO_USER_BACKEND_VHOST_USER:
575 		dev->ops = &virtio_ops_user;
576 		break;
577 	case VIRTIO_USER_BACKEND_VHOST_KERNEL:
578 		dev->ops = &virtio_ops_kernel;
579 		break;
580 	case VIRTIO_USER_BACKEND_VHOST_VDPA:
581 		dev->ops = &virtio_ops_vdpa;
582 		break;
583 	default:
584 		PMD_DRV_LOG(ERR, "(%s) Unknown backend type", dev->path);
585 		return -1;
586 	}
587 
588 	if (dev->ops->setup(dev) < 0) {
589 		PMD_INIT_LOG(ERR, "(%s) Failed to setup backend", dev->path);
590 		return -1;
591 	}
592 
593 	return 0;
594 }
595 
596 static int
597 virtio_user_alloc_vrings(struct virtio_user_dev *dev)
598 {
599 	int i, size, nr_vrings;
600 	bool packed_ring = !!(dev->device_features & (1ull << VIRTIO_F_RING_PACKED));
601 
602 	nr_vrings = dev->max_queue_pairs * 2;
603 	if (dev->device_features & (1ull << VIRTIO_NET_F_MQ))
604 		nr_vrings++;
605 
606 	dev->callfds = rte_zmalloc("virtio_user_dev", nr_vrings * sizeof(*dev->callfds), 0);
607 	if (!dev->callfds) {
608 		PMD_INIT_LOG(ERR, "(%s) Failed to alloc callfds", dev->path);
609 		return -1;
610 	}
611 
612 	dev->kickfds = rte_zmalloc("virtio_user_dev", nr_vrings * sizeof(*dev->kickfds), 0);
613 	if (!dev->kickfds) {
614 		PMD_INIT_LOG(ERR, "(%s) Failed to alloc kickfds", dev->path);
615 		goto free_callfds;
616 	}
617 
618 	for (i = 0; i < nr_vrings; i++) {
619 		dev->callfds[i] = -1;
620 		dev->kickfds[i] = -1;
621 	}
622 
623 	if (packed_ring)
624 		size = sizeof(*dev->vrings.packed);
625 	else
626 		size = sizeof(*dev->vrings.split);
627 	dev->vrings.ptr = rte_zmalloc("virtio_user_dev", nr_vrings * size, 0);
628 	if (!dev->vrings.ptr) {
629 		PMD_INIT_LOG(ERR, "(%s) Failed to alloc vrings metadata", dev->path);
630 		goto free_kickfds;
631 	}
632 
633 	if (packed_ring) {
634 		dev->packed_queues = rte_zmalloc("virtio_user_dev",
635 				nr_vrings * sizeof(*dev->packed_queues), 0);
636 		if (!dev->packed_queues) {
637 			PMD_INIT_LOG(ERR, "(%s) Failed to alloc packed queues metadata",
638 					dev->path);
639 			goto free_vrings;
640 		}
641 	}
642 
643 	dev->qp_enabled = rte_zmalloc("virtio_user_dev",
644 			dev->max_queue_pairs * sizeof(*dev->qp_enabled), 0);
645 	if (!dev->qp_enabled) {
646 		PMD_INIT_LOG(ERR, "(%s) Failed to alloc QP enable states", dev->path);
647 		goto free_packed_queues;
648 	}
649 
650 	return 0;
651 
652 free_packed_queues:
653 	rte_free(dev->packed_queues);
654 	dev->packed_queues = NULL;
655 free_vrings:
656 	rte_free(dev->vrings.ptr);
657 	dev->vrings.ptr = NULL;
658 free_kickfds:
659 	rte_free(dev->kickfds);
660 	dev->kickfds = NULL;
661 free_callfds:
662 	rte_free(dev->callfds);
663 	dev->callfds = NULL;
664 
665 	return -1;
666 }
667 
668 static void
669 virtio_user_free_vrings(struct virtio_user_dev *dev)
670 {
671 	rte_free(dev->qp_enabled);
672 	dev->qp_enabled = NULL;
673 	rte_free(dev->packed_queues);
674 	dev->packed_queues = NULL;
675 	rte_free(dev->vrings.ptr);
676 	dev->vrings.ptr = NULL;
677 	rte_free(dev->kickfds);
678 	dev->kickfds = NULL;
679 	rte_free(dev->callfds);
680 	dev->callfds = NULL;
681 }
682 
683 /* Use below macro to filter features from vhost backend */
684 #define VIRTIO_USER_SUPPORTED_FEATURES			\
685 	(1ULL << VIRTIO_NET_F_MAC		|	\
686 	 1ULL << VIRTIO_NET_F_STATUS		|	\
687 	 1ULL << VIRTIO_NET_F_MQ		|	\
688 	 1ULL << VIRTIO_NET_F_CTRL_MAC_ADDR	|	\
689 	 1ULL << VIRTIO_NET_F_CTRL_VQ		|	\
690 	 1ULL << VIRTIO_NET_F_CTRL_RX		|	\
691 	 1ULL << VIRTIO_NET_F_CTRL_VLAN		|	\
692 	 1ULL << VIRTIO_NET_F_CSUM		|	\
693 	 1ULL << VIRTIO_NET_F_HOST_TSO4		|	\
694 	 1ULL << VIRTIO_NET_F_HOST_TSO6		|	\
695 	 1ULL << VIRTIO_NET_F_MRG_RXBUF		|	\
696 	 1ULL << VIRTIO_RING_F_INDIRECT_DESC	|	\
697 	 1ULL << VIRTIO_NET_F_GUEST_CSUM	|	\
698 	 1ULL << VIRTIO_NET_F_GUEST_TSO4	|	\
699 	 1ULL << VIRTIO_NET_F_GUEST_TSO6	|	\
700 	 1ULL << VIRTIO_F_IN_ORDER		|	\
701 	 1ULL << VIRTIO_F_VERSION_1		|	\
702 	 1ULL << VIRTIO_F_RING_PACKED		|	\
703 	 1ULL << VIRTIO_F_NOTIFICATION_DATA	|	\
704 	 1ULL << VIRTIO_NET_F_RSS)
705 
706 int
707 virtio_user_dev_init(struct virtio_user_dev *dev, char *path, uint16_t queues,
708 		     int cq, int queue_size, const char *mac, char **ifname,
709 		     int server, int mrg_rxbuf, int in_order, int packed_vq,
710 		     enum virtio_user_backend_type backend_type)
711 {
712 	uint64_t backend_features;
713 
714 	pthread_mutex_init(&dev->mutex, NULL);
715 	strlcpy(dev->path, path, PATH_MAX);
716 
717 	dev->started = 0;
718 	dev->queue_pairs = 1; /* mq disabled by default */
719 	dev->max_queue_pairs = queues; /* initialize to user requested value for kernel backend */
720 	dev->queue_size = queue_size;
721 	dev->is_server = server;
722 	dev->mac_specified = 0;
723 	dev->frontend_features = 0;
724 	dev->unsupported_features = 0;
725 	dev->backend_type = backend_type;
726 	dev->ifname = *ifname;
727 
728 	if (virtio_user_dev_setup(dev) < 0) {
729 		PMD_INIT_LOG(ERR, "(%s) backend set up fails", dev->path);
730 		return -1;
731 	}
732 
733 	if (dev->ops->set_owner(dev) < 0) {
734 		PMD_INIT_LOG(ERR, "(%s) Failed to set backend owner", dev->path);
735 		goto destroy;
736 	}
737 
738 	if (dev->ops->get_backend_features(&backend_features) < 0) {
739 		PMD_INIT_LOG(ERR, "(%s) Failed to get backend features", dev->path);
740 		goto destroy;
741 	}
742 
743 	dev->unsupported_features = ~(VIRTIO_USER_SUPPORTED_FEATURES | backend_features);
744 
745 	if (dev->ops->get_features(dev, &dev->device_features) < 0) {
746 		PMD_INIT_LOG(ERR, "(%s) Failed to get device features", dev->path);
747 		goto destroy;
748 	}
749 
750 	virtio_user_dev_init_mac(dev, mac);
751 
752 	if (virtio_user_dev_init_max_queue_pairs(dev, queues))
753 		dev->unsupported_features |= (1ull << VIRTIO_NET_F_MQ);
754 
755 	if (dev->max_queue_pairs > 1)
756 		cq = 1;
757 
758 	if (!mrg_rxbuf)
759 		dev->unsupported_features |= (1ull << VIRTIO_NET_F_MRG_RXBUF);
760 
761 	if (!in_order)
762 		dev->unsupported_features |= (1ull << VIRTIO_F_IN_ORDER);
763 
764 	if (!packed_vq)
765 		dev->unsupported_features |= (1ull << VIRTIO_F_RING_PACKED);
766 
767 	if (dev->mac_specified)
768 		dev->frontend_features |= (1ull << VIRTIO_NET_F_MAC);
769 	else
770 		dev->unsupported_features |= (1ull << VIRTIO_NET_F_MAC);
771 
772 	if (cq) {
773 		/* device does not really need to know anything about CQ,
774 		 * so if necessary, we just claim to support CQ
775 		 */
776 		dev->frontend_features |= (1ull << VIRTIO_NET_F_CTRL_VQ);
777 	} else {
778 		dev->unsupported_features |= (1ull << VIRTIO_NET_F_CTRL_VQ);
779 		/* Also disable features that depend on VIRTIO_NET_F_CTRL_VQ */
780 		dev->unsupported_features |= (1ull << VIRTIO_NET_F_CTRL_RX);
781 		dev->unsupported_features |= (1ull << VIRTIO_NET_F_CTRL_VLAN);
782 		dev->unsupported_features |=
783 			(1ull << VIRTIO_NET_F_GUEST_ANNOUNCE);
784 		dev->unsupported_features |= (1ull << VIRTIO_NET_F_MQ);
785 		dev->unsupported_features |=
786 			(1ull << VIRTIO_NET_F_CTRL_MAC_ADDR);
787 	}
788 
789 	/* The backend will not report this feature, we add it explicitly */
790 	if (dev->backend_type == VIRTIO_USER_BACKEND_VHOST_USER)
791 		dev->frontend_features |= (1ull << VIRTIO_NET_F_STATUS);
792 
793 	dev->frontend_features &= ~dev->unsupported_features;
794 	dev->device_features &= ~dev->unsupported_features;
795 
796 	if (virtio_user_alloc_vrings(dev) < 0) {
797 		PMD_INIT_LOG(ERR, "(%s) Failed to allocate vring metadata", dev->path);
798 		goto destroy;
799 	}
800 
801 	if (virtio_user_dev_init_notify(dev) < 0) {
802 		PMD_INIT_LOG(ERR, "(%s) Failed to init notifiers", dev->path);
803 		goto free_vrings;
804 	}
805 
806 	if (virtio_user_fill_intr_handle(dev) < 0) {
807 		PMD_INIT_LOG(ERR, "(%s) Failed to init interrupt handler", dev->path);
808 		goto notify_uninit;
809 	}
810 
811 	if (rte_mem_event_callback_register(VIRTIO_USER_MEM_EVENT_CLB_NAME,
812 				virtio_user_mem_event_cb, dev)) {
813 		if (rte_errno != ENOTSUP) {
814 			PMD_INIT_LOG(ERR, "(%s) Failed to register mem event callback",
815 					dev->path);
816 			goto notify_uninit;
817 		}
818 	}
819 
820 	*ifname = NULL;
821 	return 0;
822 
823 notify_uninit:
824 	virtio_user_dev_uninit_notify(dev);
825 free_vrings:
826 	virtio_user_free_vrings(dev);
827 destroy:
828 	dev->ops->destroy(dev);
829 
830 	return -1;
831 }
832 
833 void
834 virtio_user_dev_uninit(struct virtio_user_dev *dev)
835 {
836 	struct rte_eth_dev *eth_dev = &rte_eth_devices[dev->hw.port_id];
837 
838 	rte_intr_instance_free(eth_dev->intr_handle);
839 	eth_dev->intr_handle = NULL;
840 
841 	virtio_user_stop_device(dev);
842 
843 	rte_mem_event_callback_unregister(VIRTIO_USER_MEM_EVENT_CLB_NAME, dev);
844 
845 	virtio_user_dev_uninit_notify(dev);
846 
847 	virtio_user_free_vrings(dev);
848 
849 	free(dev->ifname);
850 
851 	if (dev->is_server)
852 		unlink(dev->path);
853 
854 	dev->ops->destroy(dev);
855 }
856 
857 static uint8_t
858 virtio_user_handle_mq(struct virtio_user_dev *dev, uint16_t q_pairs)
859 {
860 	uint16_t i;
861 	uint8_t ret = 0;
862 
863 	if (q_pairs > dev->max_queue_pairs) {
864 		PMD_INIT_LOG(ERR, "(%s) multi-q config %u, but only %u supported",
865 			     dev->path, q_pairs, dev->max_queue_pairs);
866 		return -1;
867 	}
868 
869 	for (i = 0; i < q_pairs; ++i)
870 		ret |= dev->ops->enable_qp(dev, i, 1);
871 	for (i = q_pairs; i < dev->max_queue_pairs; ++i)
872 		ret |= dev->ops->enable_qp(dev, i, 0);
873 
874 	if (dev->scvq)
875 		ret |= dev->ops->cvq_enable(dev, 1);
876 
877 	dev->queue_pairs = q_pairs;
878 
879 	return ret;
880 }
881 
882 #define CVQ_MAX_DATA_DESCS 32
883 
884 static uint32_t
885 virtio_user_handle_ctrl_msg_split(struct virtio_user_dev *dev, struct vring *vring,
886 			    uint16_t idx_hdr)
887 {
888 	struct virtio_net_ctrl_hdr *hdr;
889 	virtio_net_ctrl_ack status = ~0;
890 	uint16_t i, idx_data, idx_status;
891 	uint32_t n_descs = 0;
892 	int dlen[CVQ_MAX_DATA_DESCS], nb_dlen = 0;
893 
894 	/* locate desc for header, data, and status */
895 	idx_data = vring->desc[idx_hdr].next;
896 	n_descs++;
897 
898 	i = idx_data;
899 	while (vring->desc[i].flags == VRING_DESC_F_NEXT) {
900 		dlen[nb_dlen++] = vring->desc[i].len;
901 		i = vring->desc[i].next;
902 		n_descs++;
903 	}
904 
905 	/* locate desc for status */
906 	idx_status = i;
907 	n_descs++;
908 
909 	hdr = (void *)(uintptr_t)vring->desc[idx_hdr].addr;
910 	if (hdr->class == VIRTIO_NET_CTRL_MQ &&
911 	    hdr->cmd == VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET) {
912 		uint16_t queues;
913 
914 		queues = *(uint16_t *)(uintptr_t)vring->desc[idx_data].addr;
915 		status = virtio_user_handle_mq(dev, queues);
916 	} else if (hdr->class == VIRTIO_NET_CTRL_MQ && hdr->cmd == VIRTIO_NET_CTRL_MQ_RSS_CONFIG) {
917 		struct virtio_net_ctrl_rss *rss;
918 
919 		rss = (struct virtio_net_ctrl_rss *)(uintptr_t)vring->desc[idx_data].addr;
920 		status = virtio_user_handle_mq(dev, rss->max_tx_vq);
921 	} else if (hdr->class == VIRTIO_NET_CTRL_RX  ||
922 		   hdr->class == VIRTIO_NET_CTRL_MAC ||
923 		   hdr->class == VIRTIO_NET_CTRL_VLAN) {
924 		status = 0;
925 	}
926 
927 	if (!status && dev->scvq)
928 		status = virtio_send_command(&dev->scvq->cq,
929 				(struct virtio_pmd_ctrl *)hdr, dlen, nb_dlen);
930 
931 	/* Update status */
932 	*(virtio_net_ctrl_ack *)(uintptr_t)vring->desc[idx_status].addr = status;
933 
934 	return n_descs;
935 }
936 
937 static inline int
938 desc_is_avail(struct vring_packed_desc *desc, bool wrap_counter)
939 {
940 	uint16_t flags = __atomic_load_n(&desc->flags, __ATOMIC_ACQUIRE);
941 
942 	return wrap_counter == !!(flags & VRING_PACKED_DESC_F_AVAIL) &&
943 		wrap_counter != !!(flags & VRING_PACKED_DESC_F_USED);
944 }
945 
946 static uint32_t
947 virtio_user_handle_ctrl_msg_packed(struct virtio_user_dev *dev,
948 				   struct vring_packed *vring,
949 				   uint16_t idx_hdr)
950 {
951 	struct virtio_net_ctrl_hdr *hdr;
952 	virtio_net_ctrl_ack status = ~0;
953 	uint16_t idx_data, idx_status;
954 	/* initialize to one, header is first */
955 	uint32_t n_descs = 1;
956 	int dlen[CVQ_MAX_DATA_DESCS], nb_dlen = 0;
957 
958 	/* locate desc for header, data, and status */
959 	idx_data = idx_hdr + 1;
960 	if (idx_data >= dev->queue_size)
961 		idx_data -= dev->queue_size;
962 
963 	n_descs++;
964 
965 	idx_status = idx_data;
966 	while (vring->desc[idx_status].flags & VRING_DESC_F_NEXT) {
967 		dlen[nb_dlen++] = vring->desc[idx_status].len;
968 		idx_status++;
969 		if (idx_status >= dev->queue_size)
970 			idx_status -= dev->queue_size;
971 		n_descs++;
972 	}
973 
974 	hdr = (void *)(uintptr_t)vring->desc[idx_hdr].addr;
975 	if (hdr->class == VIRTIO_NET_CTRL_MQ &&
976 	    hdr->cmd == VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET) {
977 		uint16_t queues;
978 
979 		queues = *(uint16_t *)(uintptr_t)
980 				vring->desc[idx_data].addr;
981 		status = virtio_user_handle_mq(dev, queues);
982 	} else if (hdr->class == VIRTIO_NET_CTRL_MQ && hdr->cmd == VIRTIO_NET_CTRL_MQ_RSS_CONFIG) {
983 		struct virtio_net_ctrl_rss *rss;
984 
985 		rss = (struct virtio_net_ctrl_rss *)(uintptr_t)vring->desc[idx_data].addr;
986 		status = virtio_user_handle_mq(dev, rss->max_tx_vq);
987 	} else if (hdr->class == VIRTIO_NET_CTRL_RX  ||
988 		   hdr->class == VIRTIO_NET_CTRL_MAC ||
989 		   hdr->class == VIRTIO_NET_CTRL_VLAN) {
990 		status = 0;
991 	}
992 
993 	if (!status && dev->scvq)
994 		status = virtio_send_command(&dev->scvq->cq,
995 				(struct virtio_pmd_ctrl *)hdr, dlen, nb_dlen);
996 
997 	/* Update status */
998 	*(virtio_net_ctrl_ack *)(uintptr_t)
999 		vring->desc[idx_status].addr = status;
1000 
1001 	/* Update used descriptor */
1002 	vring->desc[idx_hdr].id = vring->desc[idx_status].id;
1003 	vring->desc[idx_hdr].len = sizeof(status);
1004 
1005 	return n_descs;
1006 }
1007 
1008 static void
1009 virtio_user_handle_cq_packed(struct virtio_user_dev *dev, uint16_t queue_idx)
1010 {
1011 	struct virtio_user_queue *vq = &dev->packed_queues[queue_idx];
1012 	struct vring_packed *vring = &dev->vrings.packed[queue_idx];
1013 	uint16_t n_descs, flags;
1014 
1015 	/* Perform a load-acquire barrier in desc_is_avail to
1016 	 * enforce the ordering between desc flags and desc
1017 	 * content.
1018 	 */
1019 	while (desc_is_avail(&vring->desc[vq->used_idx],
1020 			     vq->used_wrap_counter)) {
1021 
1022 		n_descs = virtio_user_handle_ctrl_msg_packed(dev, vring,
1023 				vq->used_idx);
1024 
1025 		flags = VRING_DESC_F_WRITE;
1026 		if (vq->used_wrap_counter)
1027 			flags |= VRING_PACKED_DESC_F_AVAIL_USED;
1028 
1029 		__atomic_store_n(&vring->desc[vq->used_idx].flags, flags,
1030 				 __ATOMIC_RELEASE);
1031 
1032 		vq->used_idx += n_descs;
1033 		if (vq->used_idx >= dev->queue_size) {
1034 			vq->used_idx -= dev->queue_size;
1035 			vq->used_wrap_counter ^= 1;
1036 		}
1037 	}
1038 }
1039 
1040 static void
1041 virtio_user_handle_cq_split(struct virtio_user_dev *dev, uint16_t queue_idx)
1042 {
1043 	uint16_t avail_idx, desc_idx;
1044 	struct vring_used_elem *uep;
1045 	uint32_t n_descs;
1046 	struct vring *vring = &dev->vrings.split[queue_idx];
1047 
1048 	/* Consume avail ring, using used ring idx as first one */
1049 	while (__atomic_load_n(&vring->used->idx, __ATOMIC_RELAXED)
1050 	       != vring->avail->idx) {
1051 		avail_idx = __atomic_load_n(&vring->used->idx, __ATOMIC_RELAXED)
1052 			    & (vring->num - 1);
1053 		desc_idx = vring->avail->ring[avail_idx];
1054 
1055 		n_descs = virtio_user_handle_ctrl_msg_split(dev, vring, desc_idx);
1056 
1057 		/* Update used ring */
1058 		uep = &vring->used->ring[avail_idx];
1059 		uep->id = desc_idx;
1060 		uep->len = n_descs;
1061 
1062 		__atomic_fetch_add(&vring->used->idx, 1, __ATOMIC_RELAXED);
1063 	}
1064 }
1065 
1066 void
1067 virtio_user_handle_cq(struct virtio_user_dev *dev, uint16_t queue_idx)
1068 {
1069 	if (virtio_with_packed_queue(&dev->hw))
1070 		virtio_user_handle_cq_packed(dev, queue_idx);
1071 	else
1072 		virtio_user_handle_cq_split(dev, queue_idx);
1073 }
1074 
1075 static void
1076 virtio_user_control_queue_notify(struct virtqueue *vq, void *cookie)
1077 {
1078 	struct virtio_user_dev *dev = cookie;
1079 	uint64_t notify_data = 1;
1080 
1081 	if (!dev->notify_area) {
1082 		if (write(dev->kickfds[vq->vq_queue_index], &notify_data, sizeof(notify_data)) < 0)
1083 			PMD_DRV_LOG(ERR, "failed to kick backend: %s",
1084 				    strerror(errno));
1085 		return;
1086 	} else if (!virtio_with_feature(&dev->hw, VIRTIO_F_NOTIFICATION_DATA)) {
1087 		rte_write16(vq->vq_queue_index, vq->notify_addr);
1088 		return;
1089 	}
1090 
1091 	if (virtio_with_packed_queue(&dev->hw)) {
1092 		/* Bit[0:15]: vq queue index
1093 		 * Bit[16:30]: avail index
1094 		 * Bit[31]: avail wrap counter
1095 		 */
1096 		notify_data = ((uint32_t)(!!(vq->vq_packed.cached_flags &
1097 				VRING_PACKED_DESC_F_AVAIL)) << 31) |
1098 				((uint32_t)vq->vq_avail_idx << 16) |
1099 				vq->vq_queue_index;
1100 	} else {
1101 		/* Bit[0:15]: vq queue index
1102 		 * Bit[16:31]: avail index
1103 		 */
1104 		notify_data = ((uint32_t)vq->vq_avail_idx << 16) |
1105 				vq->vq_queue_index;
1106 	}
1107 	rte_write32(notify_data, vq->notify_addr);
1108 }
1109 
1110 int
1111 virtio_user_dev_create_shadow_cvq(struct virtio_user_dev *dev, struct virtqueue *vq)
1112 {
1113 	char name[VIRTQUEUE_MAX_NAME_SZ];
1114 	struct virtqueue *scvq;
1115 
1116 	snprintf(name, sizeof(name), "port%d_shadow_cvq", vq->hw->port_id);
1117 	scvq = virtqueue_alloc(&dev->hw, vq->vq_queue_index, vq->vq_nentries,
1118 			VTNET_CQ, SOCKET_ID_ANY, name);
1119 	if (!scvq) {
1120 		PMD_INIT_LOG(ERR, "(%s) Failed to alloc shadow control vq\n", dev->path);
1121 		return -ENOMEM;
1122 	}
1123 
1124 	scvq->cq.notify_queue = &virtio_user_control_queue_notify;
1125 	scvq->cq.notify_cookie = dev;
1126 	scvq->notify_addr = vq->notify_addr;
1127 	dev->scvq = scvq;
1128 
1129 	return 0;
1130 }
1131 
1132 void
1133 virtio_user_dev_destroy_shadow_cvq(struct virtio_user_dev *dev)
1134 {
1135 	if (!dev->scvq)
1136 		return;
1137 
1138 	virtqueue_free(dev->scvq);
1139 	dev->scvq = NULL;
1140 }
1141 
1142 int
1143 virtio_user_dev_set_status(struct virtio_user_dev *dev, uint8_t status)
1144 {
1145 	int ret;
1146 
1147 	pthread_mutex_lock(&dev->mutex);
1148 	dev->status = status;
1149 	ret = dev->ops->set_status(dev, status);
1150 	if (ret && ret != -ENOTSUP)
1151 		PMD_INIT_LOG(ERR, "(%s) Failed to set backend status", dev->path);
1152 
1153 	pthread_mutex_unlock(&dev->mutex);
1154 	return ret;
1155 }
1156 
1157 int
1158 virtio_user_dev_update_status(struct virtio_user_dev *dev)
1159 {
1160 	int ret;
1161 	uint8_t status;
1162 
1163 	pthread_mutex_lock(&dev->mutex);
1164 
1165 	ret = dev->ops->get_status(dev, &status);
1166 	if (!ret) {
1167 		dev->status = status;
1168 		PMD_INIT_LOG(DEBUG, "Updated Device Status(0x%08x):\n"
1169 			"\t-RESET: %u\n"
1170 			"\t-ACKNOWLEDGE: %u\n"
1171 			"\t-DRIVER: %u\n"
1172 			"\t-DRIVER_OK: %u\n"
1173 			"\t-FEATURES_OK: %u\n"
1174 			"\t-DEVICE_NEED_RESET: %u\n"
1175 			"\t-FAILED: %u",
1176 			dev->status,
1177 			(dev->status == VIRTIO_CONFIG_STATUS_RESET),
1178 			!!(dev->status & VIRTIO_CONFIG_STATUS_ACK),
1179 			!!(dev->status & VIRTIO_CONFIG_STATUS_DRIVER),
1180 			!!(dev->status & VIRTIO_CONFIG_STATUS_DRIVER_OK),
1181 			!!(dev->status & VIRTIO_CONFIG_STATUS_FEATURES_OK),
1182 			!!(dev->status & VIRTIO_CONFIG_STATUS_DEV_NEED_RESET),
1183 			!!(dev->status & VIRTIO_CONFIG_STATUS_FAILED));
1184 	} else if (ret != -ENOTSUP) {
1185 		PMD_INIT_LOG(ERR, "(%s) Failed to get backend status", dev->path);
1186 	}
1187 
1188 	pthread_mutex_unlock(&dev->mutex);
1189 	return ret;
1190 }
1191 
1192 int
1193 virtio_user_dev_update_link_state(struct virtio_user_dev *dev)
1194 {
1195 	if (dev->ops->update_link_state)
1196 		return dev->ops->update_link_state(dev);
1197 
1198 	return 0;
1199 }
1200 
1201 static void
1202 virtio_user_dev_reset_queues_packed(struct rte_eth_dev *eth_dev)
1203 {
1204 	struct virtio_user_dev *dev = eth_dev->data->dev_private;
1205 	struct virtio_hw *hw = &dev->hw;
1206 	struct virtnet_rx *rxvq;
1207 	struct virtnet_tx *txvq;
1208 	uint16_t i;
1209 
1210 	/* Add lock to avoid queue contention. */
1211 	rte_spinlock_lock(&hw->state_lock);
1212 	hw->started = 0;
1213 
1214 	/*
1215 	 * Waiting for datapath to complete before resetting queues.
1216 	 * 1 ms should be enough for the ongoing Tx/Rx function to finish.
1217 	 */
1218 	rte_delay_ms(1);
1219 
1220 	/* Vring reset for each Tx queue and Rx queue. */
1221 	for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
1222 		rxvq = eth_dev->data->rx_queues[i];
1223 		virtqueue_rxvq_reset_packed(virtnet_rxq_to_vq(rxvq));
1224 		virtio_dev_rx_queue_setup_finish(eth_dev, i);
1225 	}
1226 
1227 	for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
1228 		txvq = eth_dev->data->tx_queues[i];
1229 		virtqueue_txvq_reset_packed(virtnet_txq_to_vq(txvq));
1230 	}
1231 
1232 	hw->started = 1;
1233 	rte_spinlock_unlock(&hw->state_lock);
1234 }
1235 
1236 void
1237 virtio_user_dev_delayed_disconnect_handler(void *param)
1238 {
1239 	struct virtio_user_dev *dev = param;
1240 	struct rte_eth_dev *eth_dev = &rte_eth_devices[dev->hw.port_id];
1241 
1242 	if (rte_intr_disable(eth_dev->intr_handle) < 0) {
1243 		PMD_DRV_LOG(ERR, "interrupt disable failed");
1244 		return;
1245 	}
1246 	PMD_DRV_LOG(DEBUG, "Unregistering intr fd: %d",
1247 		    rte_intr_fd_get(eth_dev->intr_handle));
1248 	if (rte_intr_callback_unregister(eth_dev->intr_handle,
1249 					 virtio_interrupt_handler,
1250 					 eth_dev) != 1)
1251 		PMD_DRV_LOG(ERR, "interrupt unregister failed");
1252 
1253 	if (dev->is_server) {
1254 		if (dev->ops->server_disconnect)
1255 			dev->ops->server_disconnect(dev);
1256 
1257 		rte_intr_fd_set(eth_dev->intr_handle,
1258 			dev->ops->get_intr_fd(dev));
1259 
1260 		PMD_DRV_LOG(DEBUG, "Registering intr fd: %d",
1261 			    rte_intr_fd_get(eth_dev->intr_handle));
1262 
1263 		if (rte_intr_callback_register(eth_dev->intr_handle,
1264 					       virtio_interrupt_handler,
1265 					       eth_dev))
1266 			PMD_DRV_LOG(ERR, "interrupt register failed");
1267 
1268 		if (rte_intr_enable(eth_dev->intr_handle) < 0) {
1269 			PMD_DRV_LOG(ERR, "interrupt enable failed");
1270 			return;
1271 		}
1272 	}
1273 }
1274 
1275 static void
1276 virtio_user_dev_delayed_intr_reconfig_handler(void *param)
1277 {
1278 	struct virtio_user_dev *dev = param;
1279 	struct rte_eth_dev *eth_dev = &rte_eth_devices[dev->hw.port_id];
1280 
1281 	PMD_DRV_LOG(DEBUG, "Unregistering intr fd: %d",
1282 		    rte_intr_fd_get(eth_dev->intr_handle));
1283 
1284 	if (rte_intr_callback_unregister(eth_dev->intr_handle,
1285 					 virtio_interrupt_handler,
1286 					 eth_dev) != 1)
1287 		PMD_DRV_LOG(ERR, "interrupt unregister failed");
1288 
1289 	rte_intr_fd_set(eth_dev->intr_handle, dev->ops->get_intr_fd(dev));
1290 
1291 	PMD_DRV_LOG(DEBUG, "Registering intr fd: %d",
1292 		    rte_intr_fd_get(eth_dev->intr_handle));
1293 
1294 	if (rte_intr_callback_register(eth_dev->intr_handle,
1295 				       virtio_interrupt_handler, eth_dev))
1296 		PMD_DRV_LOG(ERR, "interrupt register failed");
1297 
1298 	if (rte_intr_enable(eth_dev->intr_handle) < 0)
1299 		PMD_DRV_LOG(ERR, "interrupt enable failed");
1300 }
1301 
1302 int
1303 virtio_user_dev_server_reconnect(struct virtio_user_dev *dev)
1304 {
1305 	int ret, old_status;
1306 	struct rte_eth_dev *eth_dev = &rte_eth_devices[dev->hw.port_id];
1307 	struct virtio_hw *hw = &dev->hw;
1308 
1309 	if (!dev->ops->server_reconnect) {
1310 		PMD_DRV_LOG(ERR, "(%s) Missing server reconnect callback", dev->path);
1311 		return -1;
1312 	}
1313 
1314 	if (dev->ops->server_reconnect(dev)) {
1315 		PMD_DRV_LOG(ERR, "(%s) Reconnect callback call failed", dev->path);
1316 		return -1;
1317 	}
1318 
1319 	old_status = dev->status;
1320 
1321 	virtio_reset(hw);
1322 
1323 	virtio_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1324 
1325 	virtio_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1326 
1327 	if (dev->ops->get_features(dev, &dev->device_features) < 0) {
1328 		PMD_INIT_LOG(ERR, "get_features failed: %s",
1329 			     strerror(errno));
1330 		return -1;
1331 	}
1332 
1333 	/* unmask vhost-user unsupported features */
1334 	dev->device_features &= ~(dev->unsupported_features);
1335 
1336 	dev->features &= (dev->device_features | dev->frontend_features);
1337 
1338 	/* For packed ring, resetting queues is required in reconnection. */
1339 	if (virtio_with_packed_queue(hw) &&
1340 	   (old_status & VIRTIO_CONFIG_STATUS_DRIVER_OK)) {
1341 		PMD_INIT_LOG(NOTICE, "Packets on the fly will be dropped"
1342 				" when packed ring reconnecting.");
1343 		virtio_user_dev_reset_queues_packed(eth_dev);
1344 	}
1345 
1346 	virtio_set_status(hw, VIRTIO_CONFIG_STATUS_FEATURES_OK);
1347 
1348 	/* Start the device */
1349 	virtio_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER_OK);
1350 	if (!dev->started)
1351 		return -1;
1352 
1353 	if (dev->queue_pairs > 1) {
1354 		ret = virtio_user_handle_mq(dev, dev->queue_pairs);
1355 		if (ret != 0) {
1356 			PMD_INIT_LOG(ERR, "Fails to enable multi-queue pairs!");
1357 			return -1;
1358 		}
1359 	}
1360 	if (eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC) {
1361 		if (rte_intr_disable(eth_dev->intr_handle) < 0) {
1362 			PMD_DRV_LOG(ERR, "interrupt disable failed");
1363 			return -1;
1364 		}
1365 		/*
1366 		 * This function can be called from the interrupt handler, so
1367 		 * we can't unregister interrupt handler here.  Setting
1368 		 * alarm to do that later.
1369 		 */
1370 		rte_eal_alarm_set(1,
1371 			virtio_user_dev_delayed_intr_reconfig_handler,
1372 			(void *)dev);
1373 	}
1374 	PMD_INIT_LOG(NOTICE, "server mode virtio-user reconnection succeeds!");
1375 	return 0;
1376 }
1377