xref: /dpdk/drivers/net/virtio/virtio_user/virtio_user_dev.c (revision bbbe38a6d59ccdda25917712701e629d0b10af6f)
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 int
264 virtio_user_dev_set_mac(struct virtio_user_dev *dev)
265 {
266 	int ret = 0;
267 
268 	if (!(dev->device_features & (1ULL << VIRTIO_NET_F_MAC)))
269 		return -ENOTSUP;
270 
271 	if (!dev->ops->set_config)
272 		return -ENOTSUP;
273 
274 	ret = dev->ops->set_config(dev, dev->mac_addr,
275 			offsetof(struct virtio_net_config, mac),
276 			RTE_ETHER_ADDR_LEN);
277 	if (ret)
278 		PMD_DRV_LOG(ERR, "(%s) Failed to set MAC address in device", dev->path);
279 
280 	return ret;
281 }
282 
283 int
284 virtio_user_dev_get_mac(struct virtio_user_dev *dev)
285 {
286 	int ret = 0;
287 
288 	if (!(dev->device_features & (1ULL << VIRTIO_NET_F_MAC)))
289 		return -ENOTSUP;
290 
291 	if (!dev->ops->get_config)
292 		return -ENOTSUP;
293 
294 	ret = dev->ops->get_config(dev, dev->mac_addr,
295 			offsetof(struct virtio_net_config, mac),
296 			RTE_ETHER_ADDR_LEN);
297 	if (ret)
298 		PMD_DRV_LOG(ERR, "(%s) Failed to get MAC address from device", dev->path);
299 
300 	return ret;
301 }
302 
303 static void
304 virtio_user_dev_init_mac(struct virtio_user_dev *dev, const char *mac)
305 {
306 	struct rte_ether_addr cmdline_mac;
307 	char buf[RTE_ETHER_ADDR_FMT_SIZE];
308 	int ret;
309 
310 	if (mac && rte_ether_unformat_addr(mac, &cmdline_mac) == 0) {
311 		/*
312 		 * MAC address was passed from command-line, try to store
313 		 * it in the device if it supports it. Otherwise try to use
314 		 * the device one.
315 		 */
316 		memcpy(dev->mac_addr, &cmdline_mac, RTE_ETHER_ADDR_LEN);
317 		dev->mac_specified = 1;
318 
319 		/* Setting MAC may fail, continue to get the device one in this case */
320 		virtio_user_dev_set_mac(dev);
321 		ret = virtio_user_dev_get_mac(dev);
322 		if (ret == -ENOTSUP)
323 			goto out;
324 
325 		if (memcmp(&cmdline_mac, dev->mac_addr, RTE_ETHER_ADDR_LEN))
326 			PMD_DRV_LOG(INFO, "(%s) Device MAC update failed", dev->path);
327 	} else {
328 		ret = virtio_user_dev_get_mac(dev);
329 		if (ret) {
330 			PMD_DRV_LOG(ERR, "(%s) No valid MAC in devargs or device, use random",
331 					dev->path);
332 			return;
333 		}
334 
335 		dev->mac_specified = 1;
336 	}
337 out:
338 	rte_ether_format_addr(buf, RTE_ETHER_ADDR_FMT_SIZE,
339 			(struct rte_ether_addr *)dev->mac_addr);
340 	PMD_DRV_LOG(INFO, "(%s) MAC %s specified", dev->path, buf);
341 }
342 
343 static int
344 virtio_user_dev_init_notify(struct virtio_user_dev *dev)
345 {
346 	uint32_t i, j;
347 	int callfd;
348 	int kickfd;
349 
350 	for (i = 0; i < dev->max_queue_pairs * 2; i++) {
351 		/* May use invalid flag, but some backend uses kickfd and
352 		 * callfd as criteria to judge if dev is alive. so finally we
353 		 * use real event_fd.
354 		 */
355 		callfd = eventfd(0, EFD_CLOEXEC | EFD_NONBLOCK);
356 		if (callfd < 0) {
357 			PMD_DRV_LOG(ERR, "(%s) callfd error, %s", dev->path, strerror(errno));
358 			goto err;
359 		}
360 		kickfd = eventfd(0, EFD_CLOEXEC | EFD_NONBLOCK);
361 		if (kickfd < 0) {
362 			close(callfd);
363 			PMD_DRV_LOG(ERR, "(%s) kickfd error, %s", dev->path, strerror(errno));
364 			goto err;
365 		}
366 		dev->callfds[i] = callfd;
367 		dev->kickfds[i] = kickfd;
368 	}
369 
370 	return 0;
371 err:
372 	for (j = 0; j < i; j++) {
373 		if (dev->kickfds[j] >= 0) {
374 			close(dev->kickfds[j]);
375 			dev->kickfds[j] = -1;
376 		}
377 		if (dev->callfds[j] >= 0) {
378 			close(dev->callfds[j]);
379 			dev->callfds[j] = -1;
380 		}
381 	}
382 
383 	return -1;
384 }
385 
386 static void
387 virtio_user_dev_uninit_notify(struct virtio_user_dev *dev)
388 {
389 	uint32_t i;
390 
391 	for (i = 0; i < dev->max_queue_pairs * 2; ++i) {
392 		if (dev->kickfds[i] >= 0) {
393 			close(dev->kickfds[i]);
394 			dev->kickfds[i] = -1;
395 		}
396 		if (dev->callfds[i] >= 0) {
397 			close(dev->callfds[i]);
398 			dev->callfds[i] = -1;
399 		}
400 	}
401 }
402 
403 static int
404 virtio_user_fill_intr_handle(struct virtio_user_dev *dev)
405 {
406 	uint32_t i;
407 	struct rte_eth_dev *eth_dev = &rte_eth_devices[dev->hw.port_id];
408 
409 	if (!eth_dev->intr_handle) {
410 		eth_dev->intr_handle = malloc(sizeof(*eth_dev->intr_handle));
411 		if (!eth_dev->intr_handle) {
412 			PMD_DRV_LOG(ERR, "(%s) failed to allocate intr_handle", dev->path);
413 			return -1;
414 		}
415 		memset(eth_dev->intr_handle, 0, sizeof(*eth_dev->intr_handle));
416 	}
417 
418 	for (i = 0; i < dev->max_queue_pairs; ++i)
419 		eth_dev->intr_handle->efds[i] = dev->callfds[i];
420 	eth_dev->intr_handle->nb_efd = dev->max_queue_pairs;
421 	eth_dev->intr_handle->max_intr = dev->max_queue_pairs + 1;
422 	eth_dev->intr_handle->type = RTE_INTR_HANDLE_VDEV;
423 	/* For virtio vdev, no need to read counter for clean */
424 	eth_dev->intr_handle->efd_counter_size = 0;
425 	eth_dev->intr_handle->fd = dev->ops->get_intr_fd(dev);
426 
427 	return 0;
428 }
429 
430 static void
431 virtio_user_mem_event_cb(enum rte_mem_event type __rte_unused,
432 			 const void *addr,
433 			 size_t len __rte_unused,
434 			 void *arg)
435 {
436 	struct virtio_user_dev *dev = arg;
437 	struct rte_memseg_list *msl;
438 	uint16_t i;
439 	int ret = 0;
440 
441 	/* ignore externally allocated memory */
442 	msl = rte_mem_virt2memseg_list(addr);
443 	if (msl->external)
444 		return;
445 
446 	pthread_mutex_lock(&dev->mutex);
447 
448 	if (dev->started == false)
449 		goto exit;
450 
451 	/* Step 1: pause the active queues */
452 	for (i = 0; i < dev->queue_pairs; i++) {
453 		ret = dev->ops->enable_qp(dev, i, 0);
454 		if (ret < 0)
455 			goto exit;
456 	}
457 
458 	/* Step 2: update memory regions */
459 	ret = dev->ops->set_memory_table(dev);
460 	if (ret < 0)
461 		goto exit;
462 
463 	/* Step 3: resume the active queues */
464 	for (i = 0; i < dev->queue_pairs; i++) {
465 		ret = dev->ops->enable_qp(dev, i, 1);
466 		if (ret < 0)
467 			goto exit;
468 	}
469 
470 exit:
471 	pthread_mutex_unlock(&dev->mutex);
472 
473 	if (ret < 0)
474 		PMD_DRV_LOG(ERR, "(%s) Failed to update memory table\n", dev->path);
475 }
476 
477 static int
478 virtio_user_dev_setup(struct virtio_user_dev *dev)
479 {
480 	if (dev->is_server) {
481 		if (dev->backend_type != VIRTIO_USER_BACKEND_VHOST_USER) {
482 			PMD_DRV_LOG(ERR, "Server mode only supports vhost-user!");
483 			return -1;
484 		}
485 	}
486 
487 	switch (dev->backend_type) {
488 	case VIRTIO_USER_BACKEND_VHOST_USER:
489 		dev->ops = &virtio_ops_user;
490 		break;
491 	case VIRTIO_USER_BACKEND_VHOST_KERNEL:
492 		dev->ops = &virtio_ops_kernel;
493 		break;
494 	case VIRTIO_USER_BACKEND_VHOST_VDPA:
495 		dev->ops = &virtio_ops_vdpa;
496 		break;
497 	default:
498 		PMD_DRV_LOG(ERR, "(%s) Unknown backend type", dev->path);
499 		return -1;
500 	}
501 
502 	if (dev->ops->setup(dev) < 0) {
503 		PMD_INIT_LOG(ERR, "(%s) Failed to setup backend\n", dev->path);
504 		return -1;
505 	}
506 
507 	if (virtio_user_dev_init_notify(dev) < 0) {
508 		PMD_INIT_LOG(ERR, "(%s) Failed to init notifiers\n", dev->path);
509 		goto destroy;
510 	}
511 
512 	if (virtio_user_fill_intr_handle(dev) < 0) {
513 		PMD_INIT_LOG(ERR, "(%s) Failed to init interrupt handler\n", dev->path);
514 		goto uninit;
515 	}
516 
517 	return 0;
518 
519 uninit:
520 	virtio_user_dev_uninit_notify(dev);
521 destroy:
522 	dev->ops->destroy(dev);
523 
524 	return -1;
525 }
526 
527 /* Use below macro to filter features from vhost backend */
528 #define VIRTIO_USER_SUPPORTED_FEATURES			\
529 	(1ULL << VIRTIO_NET_F_MAC		|	\
530 	 1ULL << VIRTIO_NET_F_STATUS		|	\
531 	 1ULL << VIRTIO_NET_F_MQ		|	\
532 	 1ULL << VIRTIO_NET_F_CTRL_MAC_ADDR	|	\
533 	 1ULL << VIRTIO_NET_F_CTRL_VQ		|	\
534 	 1ULL << VIRTIO_NET_F_CTRL_RX		|	\
535 	 1ULL << VIRTIO_NET_F_CTRL_VLAN		|	\
536 	 1ULL << VIRTIO_NET_F_CSUM		|	\
537 	 1ULL << VIRTIO_NET_F_HOST_TSO4		|	\
538 	 1ULL << VIRTIO_NET_F_HOST_TSO6		|	\
539 	 1ULL << VIRTIO_NET_F_MRG_RXBUF		|	\
540 	 1ULL << VIRTIO_RING_F_INDIRECT_DESC	|	\
541 	 1ULL << VIRTIO_NET_F_GUEST_CSUM	|	\
542 	 1ULL << VIRTIO_NET_F_GUEST_TSO4	|	\
543 	 1ULL << VIRTIO_NET_F_GUEST_TSO6	|	\
544 	 1ULL << VIRTIO_F_IN_ORDER		|	\
545 	 1ULL << VIRTIO_F_VERSION_1		|	\
546 	 1ULL << VIRTIO_F_RING_PACKED)
547 
548 int
549 virtio_user_dev_init(struct virtio_user_dev *dev, char *path, int queues,
550 		     int cq, int queue_size, const char *mac, char **ifname,
551 		     int server, int mrg_rxbuf, int in_order, int packed_vq,
552 		     enum virtio_user_backend_type backend_type)
553 {
554 	uint64_t backend_features;
555 	int i;
556 
557 	pthread_mutex_init(&dev->mutex, NULL);
558 	strlcpy(dev->path, path, PATH_MAX);
559 
560 	for (i = 0; i < VIRTIO_MAX_VIRTQUEUES; i++) {
561 		dev->kickfds[i] = -1;
562 		dev->callfds[i] = -1;
563 	}
564 
565 	dev->started = 0;
566 	dev->max_queue_pairs = queues;
567 	dev->queue_pairs = 1; /* mq disabled by default */
568 	dev->queue_size = queue_size;
569 	dev->is_server = server;
570 	dev->mac_specified = 0;
571 	dev->frontend_features = 0;
572 	dev->unsupported_features = 0;
573 	dev->backend_type = backend_type;
574 
575 	if (*ifname) {
576 		dev->ifname = *ifname;
577 		*ifname = NULL;
578 	}
579 
580 	if (virtio_user_dev_setup(dev) < 0) {
581 		PMD_INIT_LOG(ERR, "(%s) backend set up fails", dev->path);
582 		return -1;
583 	}
584 
585 	if (dev->ops->set_owner(dev) < 0) {
586 		PMD_INIT_LOG(ERR, "(%s) Failed to set backend owner", dev->path);
587 		return -1;
588 	}
589 
590 	if (dev->ops->get_backend_features(&backend_features) < 0) {
591 		PMD_INIT_LOG(ERR, "(%s) Failed to get backend features", dev->path);
592 		return -1;
593 	}
594 
595 	dev->unsupported_features = ~(VIRTIO_USER_SUPPORTED_FEATURES | backend_features);
596 
597 	if (dev->ops->get_features(dev, &dev->device_features) < 0) {
598 		PMD_INIT_LOG(ERR, "(%s) Failed to get device features", dev->path);
599 		return -1;
600 	}
601 
602 	virtio_user_dev_init_mac(dev, mac);
603 
604 	if (!mrg_rxbuf)
605 		dev->unsupported_features |= (1ull << VIRTIO_NET_F_MRG_RXBUF);
606 
607 	if (!in_order)
608 		dev->unsupported_features |= (1ull << VIRTIO_F_IN_ORDER);
609 
610 	if (!packed_vq)
611 		dev->unsupported_features |= (1ull << VIRTIO_F_RING_PACKED);
612 
613 	if (dev->mac_specified)
614 		dev->frontend_features |= (1ull << VIRTIO_NET_F_MAC);
615 	else
616 		dev->unsupported_features |= (1ull << VIRTIO_NET_F_MAC);
617 
618 	if (cq) {
619 		/* device does not really need to know anything about CQ,
620 		 * so if necessary, we just claim to support CQ
621 		 */
622 		dev->frontend_features |= (1ull << VIRTIO_NET_F_CTRL_VQ);
623 	} else {
624 		dev->unsupported_features |= (1ull << VIRTIO_NET_F_CTRL_VQ);
625 		/* Also disable features that depend on VIRTIO_NET_F_CTRL_VQ */
626 		dev->unsupported_features |= (1ull << VIRTIO_NET_F_CTRL_RX);
627 		dev->unsupported_features |= (1ull << VIRTIO_NET_F_CTRL_VLAN);
628 		dev->unsupported_features |=
629 			(1ull << VIRTIO_NET_F_GUEST_ANNOUNCE);
630 		dev->unsupported_features |= (1ull << VIRTIO_NET_F_MQ);
631 		dev->unsupported_features |=
632 			(1ull << VIRTIO_NET_F_CTRL_MAC_ADDR);
633 	}
634 
635 	/* The backend will not report this feature, we add it explicitly */
636 	if (dev->backend_type == VIRTIO_USER_BACKEND_VHOST_USER)
637 		dev->frontend_features |= (1ull << VIRTIO_NET_F_STATUS);
638 
639 	dev->frontend_features &= ~dev->unsupported_features;
640 	dev->device_features &= ~dev->unsupported_features;
641 
642 	if (rte_mem_event_callback_register(VIRTIO_USER_MEM_EVENT_CLB_NAME,
643 				virtio_user_mem_event_cb, dev)) {
644 		if (rte_errno != ENOTSUP) {
645 			PMD_INIT_LOG(ERR, "(%s) Failed to register mem event callback\n",
646 					dev->path);
647 			return -1;
648 		}
649 	}
650 
651 	return 0;
652 }
653 
654 void
655 virtio_user_dev_uninit(struct virtio_user_dev *dev)
656 {
657 	virtio_user_stop_device(dev);
658 
659 	rte_mem_event_callback_unregister(VIRTIO_USER_MEM_EVENT_CLB_NAME, dev);
660 
661 	virtio_user_dev_uninit_notify(dev);
662 
663 	free(dev->ifname);
664 
665 	if (dev->is_server)
666 		unlink(dev->path);
667 
668 	dev->ops->destroy(dev);
669 }
670 
671 uint8_t
672 virtio_user_handle_mq(struct virtio_user_dev *dev, uint16_t q_pairs)
673 {
674 	uint16_t i;
675 	uint8_t ret = 0;
676 
677 	if (q_pairs > dev->max_queue_pairs) {
678 		PMD_INIT_LOG(ERR, "(%s) multi-q config %u, but only %u supported",
679 			     dev->path, q_pairs, dev->max_queue_pairs);
680 		return -1;
681 	}
682 
683 	for (i = 0; i < q_pairs; ++i)
684 		ret |= dev->ops->enable_qp(dev, i, 1);
685 	for (i = q_pairs; i < dev->max_queue_pairs; ++i)
686 		ret |= dev->ops->enable_qp(dev, i, 0);
687 
688 	dev->queue_pairs = q_pairs;
689 
690 	return ret;
691 }
692 
693 static uint32_t
694 virtio_user_handle_ctrl_msg(struct virtio_user_dev *dev, struct vring *vring,
695 			    uint16_t idx_hdr)
696 {
697 	struct virtio_net_ctrl_hdr *hdr;
698 	virtio_net_ctrl_ack status = ~0;
699 	uint16_t i, idx_data, idx_status;
700 	uint32_t n_descs = 0;
701 
702 	/* locate desc for header, data, and status */
703 	idx_data = vring->desc[idx_hdr].next;
704 	n_descs++;
705 
706 	i = idx_data;
707 	while (vring->desc[i].flags == VRING_DESC_F_NEXT) {
708 		i = vring->desc[i].next;
709 		n_descs++;
710 	}
711 
712 	/* locate desc for status */
713 	idx_status = i;
714 	n_descs++;
715 
716 	hdr = (void *)(uintptr_t)vring->desc[idx_hdr].addr;
717 	if (hdr->class == VIRTIO_NET_CTRL_MQ &&
718 	    hdr->cmd == VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET) {
719 		uint16_t queues;
720 
721 		queues = *(uint16_t *)(uintptr_t)vring->desc[idx_data].addr;
722 		status = virtio_user_handle_mq(dev, queues);
723 	} else if (hdr->class == VIRTIO_NET_CTRL_RX  ||
724 		   hdr->class == VIRTIO_NET_CTRL_MAC ||
725 		   hdr->class == VIRTIO_NET_CTRL_VLAN) {
726 		status = 0;
727 	}
728 
729 	/* Update status */
730 	*(virtio_net_ctrl_ack *)(uintptr_t)vring->desc[idx_status].addr = status;
731 
732 	return n_descs;
733 }
734 
735 static inline int
736 desc_is_avail(struct vring_packed_desc *desc, bool wrap_counter)
737 {
738 	uint16_t flags = __atomic_load_n(&desc->flags, __ATOMIC_ACQUIRE);
739 
740 	return wrap_counter == !!(flags & VRING_PACKED_DESC_F_AVAIL) &&
741 		wrap_counter != !!(flags & VRING_PACKED_DESC_F_USED);
742 }
743 
744 static uint32_t
745 virtio_user_handle_ctrl_msg_packed(struct virtio_user_dev *dev,
746 				   struct vring_packed *vring,
747 				   uint16_t idx_hdr)
748 {
749 	struct virtio_net_ctrl_hdr *hdr;
750 	virtio_net_ctrl_ack status = ~0;
751 	uint16_t idx_data, idx_status;
752 	/* initialize to one, header is first */
753 	uint32_t n_descs = 1;
754 
755 	/* locate desc for header, data, and status */
756 	idx_data = idx_hdr + 1;
757 	if (idx_data >= dev->queue_size)
758 		idx_data -= dev->queue_size;
759 
760 	n_descs++;
761 
762 	idx_status = idx_data;
763 	while (vring->desc[idx_status].flags & VRING_DESC_F_NEXT) {
764 		idx_status++;
765 		if (idx_status >= dev->queue_size)
766 			idx_status -= dev->queue_size;
767 		n_descs++;
768 	}
769 
770 	hdr = (void *)(uintptr_t)vring->desc[idx_hdr].addr;
771 	if (hdr->class == VIRTIO_NET_CTRL_MQ &&
772 	    hdr->cmd == VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET) {
773 		uint16_t queues;
774 
775 		queues = *(uint16_t *)(uintptr_t)
776 				vring->desc[idx_data].addr;
777 		status = virtio_user_handle_mq(dev, queues);
778 	} else if (hdr->class == VIRTIO_NET_CTRL_RX  ||
779 		   hdr->class == VIRTIO_NET_CTRL_MAC ||
780 		   hdr->class == VIRTIO_NET_CTRL_VLAN) {
781 		status = 0;
782 	}
783 
784 	/* Update status */
785 	*(virtio_net_ctrl_ack *)(uintptr_t)
786 		vring->desc[idx_status].addr = status;
787 
788 	/* Update used descriptor */
789 	vring->desc[idx_hdr].id = vring->desc[idx_status].id;
790 	vring->desc[idx_hdr].len = sizeof(status);
791 
792 	return n_descs;
793 }
794 
795 void
796 virtio_user_handle_cq_packed(struct virtio_user_dev *dev, uint16_t queue_idx)
797 {
798 	struct virtio_user_queue *vq = &dev->packed_queues[queue_idx];
799 	struct vring_packed *vring = &dev->packed_vrings[queue_idx];
800 	uint16_t n_descs, flags;
801 
802 	/* Perform a load-acquire barrier in desc_is_avail to
803 	 * enforce the ordering between desc flags and desc
804 	 * content.
805 	 */
806 	while (desc_is_avail(&vring->desc[vq->used_idx],
807 			     vq->used_wrap_counter)) {
808 
809 		n_descs = virtio_user_handle_ctrl_msg_packed(dev, vring,
810 				vq->used_idx);
811 
812 		flags = VRING_DESC_F_WRITE;
813 		if (vq->used_wrap_counter)
814 			flags |= VRING_PACKED_DESC_F_AVAIL_USED;
815 
816 		__atomic_store_n(&vring->desc[vq->used_idx].flags, flags,
817 				 __ATOMIC_RELEASE);
818 
819 		vq->used_idx += n_descs;
820 		if (vq->used_idx >= dev->queue_size) {
821 			vq->used_idx -= dev->queue_size;
822 			vq->used_wrap_counter ^= 1;
823 		}
824 	}
825 }
826 
827 void
828 virtio_user_handle_cq(struct virtio_user_dev *dev, uint16_t queue_idx)
829 {
830 	uint16_t avail_idx, desc_idx;
831 	struct vring_used_elem *uep;
832 	uint32_t n_descs;
833 	struct vring *vring = &dev->vrings[queue_idx];
834 
835 	/* Consume avail ring, using used ring idx as first one */
836 	while (__atomic_load_n(&vring->used->idx, __ATOMIC_RELAXED)
837 	       != vring->avail->idx) {
838 		avail_idx = __atomic_load_n(&vring->used->idx, __ATOMIC_RELAXED)
839 			    & (vring->num - 1);
840 		desc_idx = vring->avail->ring[avail_idx];
841 
842 		n_descs = virtio_user_handle_ctrl_msg(dev, vring, desc_idx);
843 
844 		/* Update used ring */
845 		uep = &vring->used->ring[avail_idx];
846 		uep->id = desc_idx;
847 		uep->len = n_descs;
848 
849 		__atomic_add_fetch(&vring->used->idx, 1, __ATOMIC_RELAXED);
850 	}
851 }
852 
853 int
854 virtio_user_dev_set_status(struct virtio_user_dev *dev, uint8_t status)
855 {
856 	int ret;
857 
858 	pthread_mutex_lock(&dev->mutex);
859 	dev->status = status;
860 	ret = dev->ops->set_status(dev, status);
861 	if (ret && ret != -ENOTSUP)
862 		PMD_INIT_LOG(ERR, "(%s) Failed to set backend status\n", dev->path);
863 
864 	pthread_mutex_unlock(&dev->mutex);
865 	return ret;
866 }
867 
868 int
869 virtio_user_dev_update_status(struct virtio_user_dev *dev)
870 {
871 	int ret;
872 	uint8_t status;
873 
874 	pthread_mutex_lock(&dev->mutex);
875 
876 	ret = dev->ops->get_status(dev, &status);
877 	if (!ret) {
878 		dev->status = status;
879 		PMD_INIT_LOG(DEBUG, "Updated Device Status(0x%08x):\n"
880 			"\t-RESET: %u\n"
881 			"\t-ACKNOWLEDGE: %u\n"
882 			"\t-DRIVER: %u\n"
883 			"\t-DRIVER_OK: %u\n"
884 			"\t-FEATURES_OK: %u\n"
885 			"\t-DEVICE_NEED_RESET: %u\n"
886 			"\t-FAILED: %u\n",
887 			dev->status,
888 			(dev->status == VIRTIO_CONFIG_STATUS_RESET),
889 			!!(dev->status & VIRTIO_CONFIG_STATUS_ACK),
890 			!!(dev->status & VIRTIO_CONFIG_STATUS_DRIVER),
891 			!!(dev->status & VIRTIO_CONFIG_STATUS_DRIVER_OK),
892 			!!(dev->status & VIRTIO_CONFIG_STATUS_FEATURES_OK),
893 			!!(dev->status & VIRTIO_CONFIG_STATUS_DEV_NEED_RESET),
894 			!!(dev->status & VIRTIO_CONFIG_STATUS_FAILED));
895 	} else if (ret != -ENOTSUP) {
896 		PMD_INIT_LOG(ERR, "(%s) Failed to get backend status\n", dev->path);
897 	}
898 
899 	pthread_mutex_unlock(&dev->mutex);
900 	return ret;
901 }
902 
903 int
904 virtio_user_dev_update_link_state(struct virtio_user_dev *dev)
905 {
906 	if (dev->ops->update_link_state)
907 		return dev->ops->update_link_state(dev);
908 
909 	return 0;
910 }
911 
912 static void
913 virtio_user_dev_reset_queues_packed(struct rte_eth_dev *eth_dev)
914 {
915 	struct virtio_user_dev *dev = eth_dev->data->dev_private;
916 	struct virtio_hw *hw = &dev->hw;
917 	struct virtnet_rx *rxvq;
918 	struct virtnet_tx *txvq;
919 	uint16_t i;
920 
921 	/* Add lock to avoid queue contention. */
922 	rte_spinlock_lock(&hw->state_lock);
923 	hw->started = 0;
924 
925 	/*
926 	 * Waiting for datapath to complete before resetting queues.
927 	 * 1 ms should be enough for the ongoing Tx/Rx function to finish.
928 	 */
929 	rte_delay_ms(1);
930 
931 	/* Vring reset for each Tx queue and Rx queue. */
932 	for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
933 		rxvq = eth_dev->data->rx_queues[i];
934 		virtqueue_rxvq_reset_packed(virtnet_rxq_to_vq(rxvq));
935 		virtio_dev_rx_queue_setup_finish(eth_dev, i);
936 	}
937 
938 	for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
939 		txvq = eth_dev->data->tx_queues[i];
940 		virtqueue_txvq_reset_packed(virtnet_txq_to_vq(txvq));
941 	}
942 
943 	hw->started = 1;
944 	rte_spinlock_unlock(&hw->state_lock);
945 }
946 
947 void
948 virtio_user_dev_delayed_disconnect_handler(void *param)
949 {
950 	struct virtio_user_dev *dev = param;
951 	struct rte_eth_dev *eth_dev = &rte_eth_devices[dev->hw.port_id];
952 
953 	if (rte_intr_disable(eth_dev->intr_handle) < 0) {
954 		PMD_DRV_LOG(ERR, "interrupt disable failed");
955 		return;
956 	}
957 	PMD_DRV_LOG(DEBUG, "Unregistering intr fd: %d",
958 		    eth_dev->intr_handle->fd);
959 	if (rte_intr_callback_unregister(eth_dev->intr_handle,
960 					 virtio_interrupt_handler,
961 					 eth_dev) != 1)
962 		PMD_DRV_LOG(ERR, "interrupt unregister failed");
963 
964 	if (dev->is_server) {
965 		if (dev->ops->server_disconnect)
966 			dev->ops->server_disconnect(dev);
967 
968 		eth_dev->intr_handle->fd = dev->ops->get_intr_fd(dev);
969 
970 		PMD_DRV_LOG(DEBUG, "Registering intr fd: %d",
971 			    eth_dev->intr_handle->fd);
972 
973 		if (rte_intr_callback_register(eth_dev->intr_handle,
974 					       virtio_interrupt_handler,
975 					       eth_dev))
976 			PMD_DRV_LOG(ERR, "interrupt register failed");
977 
978 		if (rte_intr_enable(eth_dev->intr_handle) < 0) {
979 			PMD_DRV_LOG(ERR, "interrupt enable failed");
980 			return;
981 		}
982 	}
983 }
984 
985 static void
986 virtio_user_dev_delayed_intr_reconfig_handler(void *param)
987 {
988 	struct virtio_user_dev *dev = param;
989 	struct rte_eth_dev *eth_dev = &rte_eth_devices[dev->hw.port_id];
990 
991 	PMD_DRV_LOG(DEBUG, "Unregistering intr fd: %d",
992 		    eth_dev->intr_handle->fd);
993 
994 	if (rte_intr_callback_unregister(eth_dev->intr_handle,
995 					 virtio_interrupt_handler,
996 					 eth_dev) != 1)
997 		PMD_DRV_LOG(ERR, "interrupt unregister failed");
998 
999 	eth_dev->intr_handle->fd = dev->ops->get_intr_fd(dev);
1000 
1001 	PMD_DRV_LOG(DEBUG, "Registering intr fd: %d", eth_dev->intr_handle->fd);
1002 
1003 	if (rte_intr_callback_register(eth_dev->intr_handle,
1004 				       virtio_interrupt_handler, eth_dev))
1005 		PMD_DRV_LOG(ERR, "interrupt register failed");
1006 
1007 	if (rte_intr_enable(eth_dev->intr_handle) < 0)
1008 		PMD_DRV_LOG(ERR, "interrupt enable failed");
1009 }
1010 
1011 int
1012 virtio_user_dev_server_reconnect(struct virtio_user_dev *dev)
1013 {
1014 	int ret, old_status;
1015 	struct rte_eth_dev *eth_dev = &rte_eth_devices[dev->hw.port_id];
1016 	struct virtio_hw *hw = &dev->hw;
1017 
1018 	if (!dev->ops->server_reconnect) {
1019 		PMD_DRV_LOG(ERR, "(%s) Missing server reconnect callback", dev->path);
1020 		return -1;
1021 	}
1022 
1023 	if (dev->ops->server_reconnect(dev)) {
1024 		PMD_DRV_LOG(ERR, "(%s) Reconnect callback call failed", dev->path);
1025 		return -1;
1026 	}
1027 
1028 	old_status = dev->status;
1029 
1030 	virtio_reset(hw);
1031 
1032 	virtio_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1033 
1034 	virtio_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1035 
1036 	if (dev->ops->get_features(dev, &dev->device_features) < 0) {
1037 		PMD_INIT_LOG(ERR, "get_features failed: %s",
1038 			     strerror(errno));
1039 		return -1;
1040 	}
1041 
1042 	/* unmask vhost-user unsupported features */
1043 	dev->device_features &= ~(dev->unsupported_features);
1044 
1045 	dev->features &= (dev->device_features | dev->frontend_features);
1046 
1047 	/* For packed ring, resetting queues is required in reconnection. */
1048 	if (virtio_with_packed_queue(hw) &&
1049 	   (old_status & VIRTIO_CONFIG_STATUS_DRIVER_OK)) {
1050 		PMD_INIT_LOG(NOTICE, "Packets on the fly will be dropped"
1051 				" when packed ring reconnecting.");
1052 		virtio_user_dev_reset_queues_packed(eth_dev);
1053 	}
1054 
1055 	virtio_set_status(hw, VIRTIO_CONFIG_STATUS_FEATURES_OK);
1056 
1057 	/* Start the device */
1058 	virtio_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER_OK);
1059 	if (!dev->started)
1060 		return -1;
1061 
1062 	if (dev->queue_pairs > 1) {
1063 		ret = virtio_user_handle_mq(dev, dev->queue_pairs);
1064 		if (ret != 0) {
1065 			PMD_INIT_LOG(ERR, "Fails to enable multi-queue pairs!");
1066 			return -1;
1067 		}
1068 	}
1069 	if (eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC) {
1070 		if (rte_intr_disable(eth_dev->intr_handle) < 0) {
1071 			PMD_DRV_LOG(ERR, "interrupt disable failed");
1072 			return -1;
1073 		}
1074 		/*
1075 		 * This function can be called from the interrupt handler, so
1076 		 * we can't unregister interrupt handler here.  Setting
1077 		 * alarm to do that later.
1078 		 */
1079 		rte_eal_alarm_set(1,
1080 			virtio_user_dev_delayed_intr_reconfig_handler,
1081 			(void *)dev);
1082 	}
1083 	PMD_INIT_LOG(NOTICE, "server mode virtio-user reconnection succeeds!");
1084 	return 0;
1085 }
1086