xref: /dpdk/drivers/net/virtio/virtio_user_ethdev.c (revision f5057be340e44f3edc0fe90fa875eb89a4c49b4f)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright(c) 2010-2016 Intel Corporation
3  */
4 
5 #include <stdint.h>
6 #include <sys/types.h>
7 #include <unistd.h>
8 #include <fcntl.h>
9 #include <linux/major.h>
10 #include <sys/stat.h>
11 #include <sys/sysmacros.h>
12 #include <sys/socket.h>
13 
14 #include <rte_malloc.h>
15 #include <rte_kvargs.h>
16 #include <rte_ethdev_vdev.h>
17 #include <rte_bus_vdev.h>
18 #include <rte_alarm.h>
19 #include <rte_cycles.h>
20 
21 #include "virtio_ethdev.h"
22 #include "virtio_logs.h"
23 #include "virtio_pci.h"
24 #include "virtqueue.h"
25 #include "virtio_rxtx.h"
26 #include "virtio_user/virtio_user_dev.h"
27 #include "virtio_user/vhost.h"
28 
29 #define virtio_user_get_dev(hw) \
30 	((struct virtio_user_dev *)(hw)->virtio_user_dev)
31 
32 static void
33 virtio_user_reset_queues_packed(struct rte_eth_dev *dev)
34 {
35 	struct virtio_hw *hw = dev->data->dev_private;
36 	struct virtnet_rx *rxvq;
37 	struct virtnet_tx *txvq;
38 	uint16_t i;
39 
40 	/* Add lock to avoid queue contention. */
41 	rte_spinlock_lock(&hw->state_lock);
42 	hw->started = 0;
43 
44 	/*
45 	 * Waitting for datapath to complete before resetting queues.
46 	 * 1 ms should be enough for the ongoing Tx/Rx function to finish.
47 	 */
48 	rte_delay_ms(1);
49 
50 	/* Vring reset for each Tx queue and Rx queue. */
51 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
52 		rxvq = dev->data->rx_queues[i];
53 		virtqueue_rxvq_reset_packed(rxvq->vq);
54 		virtio_dev_rx_queue_setup_finish(dev, i);
55 	}
56 
57 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
58 		txvq = dev->data->tx_queues[i];
59 		virtqueue_txvq_reset_packed(txvq->vq);
60 	}
61 
62 	hw->started = 1;
63 	rte_spinlock_unlock(&hw->state_lock);
64 }
65 
66 
67 static int
68 virtio_user_server_reconnect(struct virtio_user_dev *dev)
69 {
70 	int ret;
71 	int connectfd;
72 	struct rte_eth_dev *eth_dev = &rte_eth_devices[dev->port_id];
73 	struct virtio_hw *hw = eth_dev->data->dev_private;
74 	uint64_t protocol_features;
75 
76 	connectfd = accept(dev->listenfd, NULL, NULL);
77 	if (connectfd < 0)
78 		return -1;
79 
80 	dev->vhostfd = connectfd;
81 	if (dev->ops->send_request(dev, VHOST_USER_GET_FEATURES,
82 				   &dev->device_features) < 0) {
83 		PMD_INIT_LOG(ERR, "get_features failed: %s",
84 			     strerror(errno));
85 		return -1;
86 	}
87 
88 	if (dev->device_features &
89 			(1ULL << VHOST_USER_F_PROTOCOL_FEATURES)) {
90 		if (dev->ops->send_request(dev,
91 					VHOST_USER_GET_PROTOCOL_FEATURES,
92 					&protocol_features))
93 			return -1;
94 
95 		dev->protocol_features &= protocol_features;
96 
97 		if (dev->ops->send_request(dev,
98 					VHOST_USER_SET_PROTOCOL_FEATURES,
99 					&dev->protocol_features))
100 			return -1;
101 
102 		if (!(dev->protocol_features &
103 				(1ULL << VHOST_USER_PROTOCOL_F_MQ)))
104 			dev->unsupported_features |= (1ull << VIRTIO_NET_F_MQ);
105 	}
106 
107 	dev->device_features |= dev->frontend_features;
108 
109 	/* umask vhost-user unsupported features */
110 	dev->device_features &= ~(dev->unsupported_features);
111 
112 	dev->features &= dev->device_features;
113 
114 	/* For packed ring, resetting queues is required in reconnection. */
115 	if (vtpci_packed_queue(hw) &&
116 	   (vtpci_get_status(hw) & VIRTIO_CONFIG_STATUS_DRIVER_OK)) {
117 		PMD_INIT_LOG(NOTICE, "Packets on the fly will be dropped"
118 				" when packed ring reconnecting.");
119 		virtio_user_reset_queues_packed(eth_dev);
120 	}
121 
122 	ret = virtio_user_start_device(dev);
123 	if (ret < 0)
124 		return -1;
125 
126 	if (dev->queue_pairs > 1) {
127 		ret = virtio_user_handle_mq(dev, dev->queue_pairs);
128 		if (ret != 0) {
129 			PMD_INIT_LOG(ERR, "Fails to enable multi-queue pairs!");
130 			return -1;
131 		}
132 	}
133 	if (eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC) {
134 		if (rte_intr_disable(eth_dev->intr_handle) < 0) {
135 			PMD_DRV_LOG(ERR, "interrupt disable failed");
136 			return -1;
137 		}
138 		rte_intr_callback_unregister(eth_dev->intr_handle,
139 					     virtio_interrupt_handler,
140 					     eth_dev);
141 		eth_dev->intr_handle->fd = connectfd;
142 		rte_intr_callback_register(eth_dev->intr_handle,
143 					   virtio_interrupt_handler, eth_dev);
144 
145 		if (rte_intr_enable(eth_dev->intr_handle) < 0) {
146 			PMD_DRV_LOG(ERR, "interrupt enable failed");
147 			return -1;
148 		}
149 	}
150 	PMD_INIT_LOG(NOTICE, "server mode virtio-user reconnection succeeds!");
151 	return 0;
152 }
153 
154 static void
155 virtio_user_delayed_handler(void *param)
156 {
157 	struct virtio_hw *hw = (struct virtio_hw *)param;
158 	struct rte_eth_dev *eth_dev = &rte_eth_devices[hw->port_id];
159 	struct virtio_user_dev *dev = virtio_user_get_dev(hw);
160 
161 	if (rte_intr_disable(eth_dev->intr_handle) < 0) {
162 		PMD_DRV_LOG(ERR, "interrupt disable failed");
163 		return;
164 	}
165 	rte_intr_callback_unregister(eth_dev->intr_handle,
166 				     virtio_interrupt_handler, eth_dev);
167 	if (dev->is_server) {
168 		if (dev->vhostfd >= 0) {
169 			close(dev->vhostfd);
170 			dev->vhostfd = -1;
171 		}
172 		eth_dev->intr_handle->fd = dev->listenfd;
173 		rte_intr_callback_register(eth_dev->intr_handle,
174 					   virtio_interrupt_handler, eth_dev);
175 		if (rte_intr_enable(eth_dev->intr_handle) < 0) {
176 			PMD_DRV_LOG(ERR, "interrupt enable failed");
177 			return;
178 		}
179 	}
180 }
181 
182 static void
183 virtio_user_read_dev_config(struct virtio_hw *hw, size_t offset,
184 		     void *dst, int length)
185 {
186 	int i;
187 	struct virtio_user_dev *dev = virtio_user_get_dev(hw);
188 
189 	if (offset == offsetof(struct virtio_net_config, mac) &&
190 	    length == RTE_ETHER_ADDR_LEN) {
191 		for (i = 0; i < RTE_ETHER_ADDR_LEN; ++i)
192 			((uint8_t *)dst)[i] = dev->mac_addr[i];
193 		return;
194 	}
195 
196 	if (offset == offsetof(struct virtio_net_config, status)) {
197 		char buf[128];
198 
199 		if (dev->vhostfd >= 0) {
200 			int r;
201 			int flags;
202 
203 			flags = fcntl(dev->vhostfd, F_GETFL);
204 			if (fcntl(dev->vhostfd, F_SETFL,
205 					flags | O_NONBLOCK) == -1) {
206 				PMD_DRV_LOG(ERR, "error setting O_NONBLOCK flag");
207 				return;
208 			}
209 			r = recv(dev->vhostfd, buf, 128, MSG_PEEK);
210 			if (r == 0 || (r < 0 && errno != EAGAIN)) {
211 				dev->net_status &= (~VIRTIO_NET_S_LINK_UP);
212 				PMD_DRV_LOG(ERR, "virtio-user port %u is down",
213 					    hw->port_id);
214 
215 				/* This function could be called in the process
216 				 * of interrupt handling, callback cannot be
217 				 * unregistered here, set an alarm to do it.
218 				 */
219 				rte_eal_alarm_set(1,
220 						  virtio_user_delayed_handler,
221 						  (void *)hw);
222 			} else {
223 				dev->net_status |= VIRTIO_NET_S_LINK_UP;
224 			}
225 			if (fcntl(dev->vhostfd, F_SETFL,
226 					flags & ~O_NONBLOCK) == -1) {
227 				PMD_DRV_LOG(ERR, "error clearing O_NONBLOCK flag");
228 				return;
229 			}
230 		} else if (dev->is_server) {
231 			dev->net_status &= (~VIRTIO_NET_S_LINK_UP);
232 			if (virtio_user_server_reconnect(dev) >= 0)
233 				dev->net_status |= VIRTIO_NET_S_LINK_UP;
234 		}
235 
236 		*(uint16_t *)dst = dev->net_status;
237 	}
238 
239 	if (offset == offsetof(struct virtio_net_config, max_virtqueue_pairs))
240 		*(uint16_t *)dst = dev->max_queue_pairs;
241 }
242 
243 static void
244 virtio_user_write_dev_config(struct virtio_hw *hw, size_t offset,
245 		      const void *src, int length)
246 {
247 	int i;
248 	struct virtio_user_dev *dev = virtio_user_get_dev(hw);
249 
250 	if ((offset == offsetof(struct virtio_net_config, mac)) &&
251 	    (length == RTE_ETHER_ADDR_LEN))
252 		for (i = 0; i < RTE_ETHER_ADDR_LEN; ++i)
253 			dev->mac_addr[i] = ((const uint8_t *)src)[i];
254 	else
255 		PMD_DRV_LOG(ERR, "not supported offset=%zu, len=%d",
256 			    offset, length);
257 }
258 
259 static void
260 virtio_user_reset(struct virtio_hw *hw)
261 {
262 	struct virtio_user_dev *dev = virtio_user_get_dev(hw);
263 
264 	if (dev->status & VIRTIO_CONFIG_STATUS_DRIVER_OK)
265 		virtio_user_stop_device(dev);
266 }
267 
268 static void
269 virtio_user_set_status(struct virtio_hw *hw, uint8_t status)
270 {
271 	struct virtio_user_dev *dev = virtio_user_get_dev(hw);
272 	uint8_t old_status = dev->status;
273 
274 	if (status & VIRTIO_CONFIG_STATUS_FEATURES_OK &&
275 			~old_status & VIRTIO_CONFIG_STATUS_FEATURES_OK)
276 		virtio_user_dev_set_features(dev);
277 	if (status & VIRTIO_CONFIG_STATUS_DRIVER_OK)
278 		virtio_user_start_device(dev);
279 	else if (status == VIRTIO_CONFIG_STATUS_RESET)
280 		virtio_user_reset(hw);
281 	dev->status = status;
282 	virtio_user_send_status_update(dev, status);
283 }
284 
285 static uint8_t
286 virtio_user_get_status(struct virtio_hw *hw)
287 {
288 	struct virtio_user_dev *dev = virtio_user_get_dev(hw);
289 
290 	virtio_user_update_status(dev);
291 
292 	return dev->status;
293 }
294 
295 static uint64_t
296 virtio_user_get_features(struct virtio_hw *hw)
297 {
298 	struct virtio_user_dev *dev = virtio_user_get_dev(hw);
299 
300 	/* unmask feature bits defined in vhost user protocol */
301 	return dev->device_features & VIRTIO_PMD_SUPPORTED_GUEST_FEATURES;
302 }
303 
304 static void
305 virtio_user_set_features(struct virtio_hw *hw, uint64_t features)
306 {
307 	struct virtio_user_dev *dev = virtio_user_get_dev(hw);
308 
309 	dev->features = features & dev->device_features;
310 }
311 
312 static uint8_t
313 virtio_user_get_isr(struct virtio_hw *hw __rte_unused)
314 {
315 	/* rxq interrupts and config interrupt are separated in virtio-user,
316 	 * here we only report config change.
317 	 */
318 	return VIRTIO_PCI_ISR_CONFIG;
319 }
320 
321 static uint16_t
322 virtio_user_set_config_irq(struct virtio_hw *hw __rte_unused,
323 		    uint16_t vec __rte_unused)
324 {
325 	return 0;
326 }
327 
328 static uint16_t
329 virtio_user_set_queue_irq(struct virtio_hw *hw __rte_unused,
330 			  struct virtqueue *vq __rte_unused,
331 			  uint16_t vec)
332 {
333 	/* pretend we have done that */
334 	return vec;
335 }
336 
337 /* This function is to get the queue size, aka, number of descs, of a specified
338  * queue. Different with the VHOST_USER_GET_QUEUE_NUM, which is used to get the
339  * max supported queues.
340  */
341 static uint16_t
342 virtio_user_get_queue_num(struct virtio_hw *hw, uint16_t queue_id __rte_unused)
343 {
344 	struct virtio_user_dev *dev = virtio_user_get_dev(hw);
345 
346 	/* Currently, each queue has same queue size */
347 	return dev->queue_size;
348 }
349 
350 static void
351 virtio_user_setup_queue_packed(struct virtqueue *vq,
352 			       struct virtio_user_dev *dev)
353 {
354 	uint16_t queue_idx = vq->vq_queue_index;
355 	struct vring_packed *vring;
356 	uint64_t desc_addr;
357 	uint64_t avail_addr;
358 	uint64_t used_addr;
359 	uint16_t i;
360 
361 	vring  = &dev->packed_vrings[queue_idx];
362 	desc_addr = (uintptr_t)vq->vq_ring_virt_mem;
363 	avail_addr = desc_addr + vq->vq_nentries *
364 		sizeof(struct vring_packed_desc);
365 	used_addr = RTE_ALIGN_CEIL(avail_addr +
366 			   sizeof(struct vring_packed_desc_event),
367 			   VIRTIO_PCI_VRING_ALIGN);
368 	vring->num = vq->vq_nentries;
369 	vring->desc = (void *)(uintptr_t)desc_addr;
370 	vring->driver = (void *)(uintptr_t)avail_addr;
371 	vring->device = (void *)(uintptr_t)used_addr;
372 	dev->packed_queues[queue_idx].avail_wrap_counter = true;
373 	dev->packed_queues[queue_idx].used_wrap_counter = true;
374 
375 	for (i = 0; i < vring->num; i++)
376 		vring->desc[i].flags = 0;
377 }
378 
379 static void
380 virtio_user_setup_queue_split(struct virtqueue *vq, struct virtio_user_dev *dev)
381 {
382 	uint16_t queue_idx = vq->vq_queue_index;
383 	uint64_t desc_addr, avail_addr, used_addr;
384 
385 	desc_addr = (uintptr_t)vq->vq_ring_virt_mem;
386 	avail_addr = desc_addr + vq->vq_nentries * sizeof(struct vring_desc);
387 	used_addr = RTE_ALIGN_CEIL(avail_addr + offsetof(struct vring_avail,
388 							 ring[vq->vq_nentries]),
389 				   VIRTIO_PCI_VRING_ALIGN);
390 
391 	dev->vrings[queue_idx].num = vq->vq_nentries;
392 	dev->vrings[queue_idx].desc = (void *)(uintptr_t)desc_addr;
393 	dev->vrings[queue_idx].avail = (void *)(uintptr_t)avail_addr;
394 	dev->vrings[queue_idx].used = (void *)(uintptr_t)used_addr;
395 }
396 
397 static int
398 virtio_user_setup_queue(struct virtio_hw *hw, struct virtqueue *vq)
399 {
400 	struct virtio_user_dev *dev = virtio_user_get_dev(hw);
401 
402 	if (vtpci_packed_queue(hw))
403 		virtio_user_setup_queue_packed(vq, dev);
404 	else
405 		virtio_user_setup_queue_split(vq, dev);
406 
407 	return 0;
408 }
409 
410 static void
411 virtio_user_del_queue(struct virtio_hw *hw, struct virtqueue *vq)
412 {
413 	/* For legacy devices, write 0 to VIRTIO_PCI_QUEUE_PFN port, QEMU
414 	 * correspondingly stops the ioeventfds, and reset the status of
415 	 * the device.
416 	 * For modern devices, set queue desc, avail, used in PCI bar to 0,
417 	 * not see any more behavior in QEMU.
418 	 *
419 	 * Here we just care about what information to deliver to vhost-user
420 	 * or vhost-kernel. So we just close ioeventfd for now.
421 	 */
422 	struct virtio_user_dev *dev = virtio_user_get_dev(hw);
423 
424 	close(dev->callfds[vq->vq_queue_index]);
425 	close(dev->kickfds[vq->vq_queue_index]);
426 }
427 
428 static void
429 virtio_user_notify_queue(struct virtio_hw *hw, struct virtqueue *vq)
430 {
431 	uint64_t buf = 1;
432 	struct virtio_user_dev *dev = virtio_user_get_dev(hw);
433 
434 	if (hw->cvq && (hw->cvq->vq == vq)) {
435 		if (vtpci_packed_queue(vq->hw))
436 			virtio_user_handle_cq_packed(dev, vq->vq_queue_index);
437 		else
438 			virtio_user_handle_cq(dev, vq->vq_queue_index);
439 		return;
440 	}
441 
442 	if (write(dev->kickfds[vq->vq_queue_index], &buf, sizeof(buf)) < 0)
443 		PMD_DRV_LOG(ERR, "failed to kick backend: %s",
444 			    strerror(errno));
445 }
446 
447 const struct virtio_pci_ops virtio_user_ops = {
448 	.read_dev_cfg	= virtio_user_read_dev_config,
449 	.write_dev_cfg	= virtio_user_write_dev_config,
450 	.get_status	= virtio_user_get_status,
451 	.set_status	= virtio_user_set_status,
452 	.get_features	= virtio_user_get_features,
453 	.set_features	= virtio_user_set_features,
454 	.get_isr	= virtio_user_get_isr,
455 	.set_config_irq	= virtio_user_set_config_irq,
456 	.set_queue_irq	= virtio_user_set_queue_irq,
457 	.get_queue_num	= virtio_user_get_queue_num,
458 	.setup_queue	= virtio_user_setup_queue,
459 	.del_queue	= virtio_user_del_queue,
460 	.notify_queue	= virtio_user_notify_queue,
461 };
462 
463 static const char *valid_args[] = {
464 #define VIRTIO_USER_ARG_QUEUES_NUM     "queues"
465 	VIRTIO_USER_ARG_QUEUES_NUM,
466 #define VIRTIO_USER_ARG_CQ_NUM         "cq"
467 	VIRTIO_USER_ARG_CQ_NUM,
468 #define VIRTIO_USER_ARG_MAC            "mac"
469 	VIRTIO_USER_ARG_MAC,
470 #define VIRTIO_USER_ARG_PATH           "path"
471 	VIRTIO_USER_ARG_PATH,
472 #define VIRTIO_USER_ARG_QUEUE_SIZE     "queue_size"
473 	VIRTIO_USER_ARG_QUEUE_SIZE,
474 #define VIRTIO_USER_ARG_INTERFACE_NAME "iface"
475 	VIRTIO_USER_ARG_INTERFACE_NAME,
476 #define VIRTIO_USER_ARG_SERVER_MODE    "server"
477 	VIRTIO_USER_ARG_SERVER_MODE,
478 #define VIRTIO_USER_ARG_MRG_RXBUF      "mrg_rxbuf"
479 	VIRTIO_USER_ARG_MRG_RXBUF,
480 #define VIRTIO_USER_ARG_IN_ORDER       "in_order"
481 	VIRTIO_USER_ARG_IN_ORDER,
482 #define VIRTIO_USER_ARG_PACKED_VQ      "packed_vq"
483 	VIRTIO_USER_ARG_PACKED_VQ,
484 #define VIRTIO_USER_ARG_SPEED          "speed"
485 	VIRTIO_USER_ARG_SPEED,
486 #define VIRTIO_USER_ARG_VECTORIZED     "vectorized"
487 	VIRTIO_USER_ARG_VECTORIZED,
488 	NULL
489 };
490 
491 #define VIRTIO_USER_DEF_CQ_EN	0
492 #define VIRTIO_USER_DEF_Q_NUM	1
493 #define VIRTIO_USER_DEF_Q_SZ	256
494 #define VIRTIO_USER_DEF_SERVER_MODE	0
495 
496 static int
497 get_string_arg(const char *key __rte_unused,
498 	       const char *value, void *extra_args)
499 {
500 	if (!value || !extra_args)
501 		return -EINVAL;
502 
503 	*(char **)extra_args = strdup(value);
504 
505 	if (!*(char **)extra_args)
506 		return -ENOMEM;
507 
508 	return 0;
509 }
510 
511 static int
512 get_integer_arg(const char *key __rte_unused,
513 		const char *value, void *extra_args)
514 {
515 	uint64_t integer = 0;
516 	if (!value || !extra_args)
517 		return -EINVAL;
518 	errno = 0;
519 	integer = strtoull(value, NULL, 0);
520 	/* extra_args keeps default value, it should be replaced
521 	 * only in case of successful parsing of the 'value' arg
522 	 */
523 	if (errno == 0)
524 		*(uint64_t *)extra_args = integer;
525 	return -errno;
526 }
527 
528 static uint32_t
529 vdpa_dynamic_major_num(void)
530 {
531 	FILE *fp;
532 	char *line = NULL;
533 	size_t size;
534 	char name[11];
535 	bool found = false;
536 	uint32_t num;
537 
538 	fp = fopen("/proc/devices", "r");
539 	if (fp == NULL) {
540 		PMD_INIT_LOG(ERR, "Cannot open /proc/devices: %s",
541 			     strerror(errno));
542 		return UNNAMED_MAJOR;
543 	}
544 
545 	while (getline(&line, &size, fp) > 0) {
546 		char *stripped = line + strspn(line, " ");
547 		if ((sscanf(stripped, "%u %10s", &num, name) == 2) &&
548 		    (strncmp(name, "vhost-vdpa", 10) == 0)) {
549 			found = true;
550 			break;
551 		}
552 	}
553 	fclose(fp);
554 	return found ? num : UNNAMED_MAJOR;
555 }
556 
557 static enum virtio_user_backend_type
558 virtio_user_backend_type(const char *path)
559 {
560 	struct stat sb;
561 
562 	if (stat(path, &sb) == -1) {
563 		PMD_INIT_LOG(ERR, "Stat fails: %s (%s)\n", path,
564 			     strerror(errno));
565 		return VIRTIO_USER_BACKEND_UNKNOWN;
566 	}
567 
568 	if (S_ISSOCK(sb.st_mode)) {
569 		return VIRTIO_USER_BACKEND_VHOST_USER;
570 	} else if (S_ISCHR(sb.st_mode)) {
571 		if (major(sb.st_rdev) == MISC_MAJOR)
572 			return VIRTIO_USER_BACKEND_VHOST_KERNEL;
573 		if (major(sb.st_rdev) == vdpa_dynamic_major_num())
574 			return VIRTIO_USER_BACKEND_VHOST_VDPA;
575 	}
576 	return VIRTIO_USER_BACKEND_UNKNOWN;
577 }
578 
579 static struct rte_eth_dev *
580 virtio_user_eth_dev_alloc(struct rte_vdev_device *vdev)
581 {
582 	struct rte_eth_dev *eth_dev;
583 	struct rte_eth_dev_data *data;
584 	struct virtio_hw *hw;
585 	struct virtio_user_dev *dev;
586 
587 	eth_dev = rte_eth_vdev_allocate(vdev, sizeof(*hw));
588 	if (!eth_dev) {
589 		PMD_INIT_LOG(ERR, "cannot alloc rte_eth_dev");
590 		return NULL;
591 	}
592 
593 	data = eth_dev->data;
594 	hw = eth_dev->data->dev_private;
595 
596 	dev = rte_zmalloc(NULL, sizeof(*dev), 0);
597 	if (!dev) {
598 		PMD_INIT_LOG(ERR, "malloc virtio_user_dev failed");
599 		rte_eth_dev_release_port(eth_dev);
600 		return NULL;
601 	}
602 
603 	hw->port_id = data->port_id;
604 	dev->port_id = data->port_id;
605 	virtio_hw_internal[hw->port_id].vtpci_ops = &virtio_user_ops;
606 	/*
607 	 * MSIX is required to enable LSC (see virtio_init_device).
608 	 * Here just pretend that we support msix.
609 	 */
610 	hw->use_msix = 1;
611 	hw->modern   = 0;
612 	hw->use_vec_rx = 0;
613 	hw->use_vec_tx = 0;
614 	hw->use_inorder_rx = 0;
615 	hw->use_inorder_tx = 0;
616 	hw->virtio_user_dev = dev;
617 	return eth_dev;
618 }
619 
620 static void
621 virtio_user_eth_dev_free(struct rte_eth_dev *eth_dev)
622 {
623 	struct rte_eth_dev_data *data = eth_dev->data;
624 	struct virtio_hw *hw = data->dev_private;
625 
626 	rte_free(hw->virtio_user_dev);
627 	rte_eth_dev_release_port(eth_dev);
628 }
629 
630 /* Dev initialization routine. Invoked once for each virtio vdev at
631  * EAL init time, see rte_bus_probe().
632  * Returns 0 on success.
633  */
634 static int
635 virtio_user_pmd_probe(struct rte_vdev_device *dev)
636 {
637 	struct rte_kvargs *kvlist = NULL;
638 	struct rte_eth_dev *eth_dev;
639 	struct virtio_hw *hw;
640 	enum virtio_user_backend_type backend_type = VIRTIO_USER_BACKEND_UNKNOWN;
641 	uint64_t queues = VIRTIO_USER_DEF_Q_NUM;
642 	uint64_t cq = VIRTIO_USER_DEF_CQ_EN;
643 	uint64_t queue_size = VIRTIO_USER_DEF_Q_SZ;
644 	uint64_t server_mode = VIRTIO_USER_DEF_SERVER_MODE;
645 	uint64_t mrg_rxbuf = 1;
646 	uint64_t in_order = 1;
647 	uint64_t packed_vq = 0;
648 	uint64_t vectorized = 0;
649 	char *path = NULL;
650 	char *ifname = NULL;
651 	char *mac_addr = NULL;
652 	int ret = -1;
653 
654 	if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
655 		const char *name = rte_vdev_device_name(dev);
656 		eth_dev = rte_eth_dev_attach_secondary(name);
657 		if (!eth_dev) {
658 			PMD_INIT_LOG(ERR, "Failed to probe %s", name);
659 			return -1;
660 		}
661 
662 		if (eth_virtio_dev_init(eth_dev) < 0) {
663 			PMD_INIT_LOG(ERR, "eth_virtio_dev_init fails");
664 			rte_eth_dev_release_port(eth_dev);
665 			return -1;
666 		}
667 
668 		eth_dev->dev_ops = &virtio_user_secondary_eth_dev_ops;
669 		eth_dev->device = &dev->device;
670 		rte_eth_dev_probing_finish(eth_dev);
671 		return 0;
672 	}
673 
674 	kvlist = rte_kvargs_parse(rte_vdev_device_args(dev), valid_args);
675 	if (!kvlist) {
676 		PMD_INIT_LOG(ERR, "error when parsing param");
677 		goto end;
678 	}
679 
680 	if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_PATH) == 1) {
681 		if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_PATH,
682 				       &get_string_arg, &path) < 0) {
683 			PMD_INIT_LOG(ERR, "error to parse %s",
684 				     VIRTIO_USER_ARG_PATH);
685 			goto end;
686 		}
687 	} else {
688 		PMD_INIT_LOG(ERR, "arg %s is mandatory for virtio_user",
689 			     VIRTIO_USER_ARG_PATH);
690 		goto end;
691 	}
692 
693 	backend_type = virtio_user_backend_type(path);
694 	if (backend_type == VIRTIO_USER_BACKEND_UNKNOWN) {
695 		PMD_INIT_LOG(ERR,
696 			     "unable to determine backend type for path %s",
697 			path);
698 		goto end;
699 	}
700 
701 
702 	if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_INTERFACE_NAME) == 1) {
703 		if (backend_type != VIRTIO_USER_BACKEND_VHOST_KERNEL) {
704 			PMD_INIT_LOG(ERR,
705 				"arg %s applies only to vhost-kernel backend",
706 				VIRTIO_USER_ARG_INTERFACE_NAME);
707 			goto end;
708 		}
709 
710 		if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_INTERFACE_NAME,
711 				       &get_string_arg, &ifname) < 0) {
712 			PMD_INIT_LOG(ERR, "error to parse %s",
713 				     VIRTIO_USER_ARG_INTERFACE_NAME);
714 			goto end;
715 		}
716 	}
717 
718 	if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_MAC) == 1) {
719 		if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_MAC,
720 				       &get_string_arg, &mac_addr) < 0) {
721 			PMD_INIT_LOG(ERR, "error to parse %s",
722 				     VIRTIO_USER_ARG_MAC);
723 			goto end;
724 		}
725 	}
726 
727 	if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_QUEUE_SIZE) == 1) {
728 		if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_QUEUE_SIZE,
729 				       &get_integer_arg, &queue_size) < 0) {
730 			PMD_INIT_LOG(ERR, "error to parse %s",
731 				     VIRTIO_USER_ARG_QUEUE_SIZE);
732 			goto end;
733 		}
734 	}
735 
736 	if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_QUEUES_NUM) == 1) {
737 		if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_QUEUES_NUM,
738 				       &get_integer_arg, &queues) < 0) {
739 			PMD_INIT_LOG(ERR, "error to parse %s",
740 				     VIRTIO_USER_ARG_QUEUES_NUM);
741 			goto end;
742 		}
743 	}
744 
745 	if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_SERVER_MODE) == 1) {
746 		if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_SERVER_MODE,
747 				       &get_integer_arg, &server_mode) < 0) {
748 			PMD_INIT_LOG(ERR, "error to parse %s",
749 				     VIRTIO_USER_ARG_SERVER_MODE);
750 			goto end;
751 		}
752 	}
753 
754 	if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_CQ_NUM) == 1) {
755 		if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_CQ_NUM,
756 				       &get_integer_arg, &cq) < 0) {
757 			PMD_INIT_LOG(ERR, "error to parse %s",
758 				     VIRTIO_USER_ARG_CQ_NUM);
759 			goto end;
760 		}
761 	} else if (queues > 1) {
762 		cq = 1;
763 	}
764 
765 	if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_PACKED_VQ) == 1) {
766 		if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_PACKED_VQ,
767 				       &get_integer_arg, &packed_vq) < 0) {
768 			PMD_INIT_LOG(ERR, "error to parse %s",
769 				     VIRTIO_USER_ARG_PACKED_VQ);
770 			goto end;
771 		}
772 	}
773 
774 	if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_VECTORIZED) == 1) {
775 		if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_VECTORIZED,
776 				       &get_integer_arg, &vectorized) < 0) {
777 			PMD_INIT_LOG(ERR, "error to parse %s",
778 				     VIRTIO_USER_ARG_VECTORIZED);
779 			goto end;
780 		}
781 	}
782 
783 	if (queues > 1 && cq == 0) {
784 		PMD_INIT_LOG(ERR, "multi-q requires ctrl-q");
785 		goto end;
786 	}
787 
788 	if (queues > VIRTIO_MAX_VIRTQUEUE_PAIRS) {
789 		PMD_INIT_LOG(ERR, "arg %s %" PRIu64 " exceeds the limit %u",
790 			VIRTIO_USER_ARG_QUEUES_NUM, queues,
791 			VIRTIO_MAX_VIRTQUEUE_PAIRS);
792 		goto end;
793 	}
794 
795 	if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_MRG_RXBUF) == 1) {
796 		if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_MRG_RXBUF,
797 				       &get_integer_arg, &mrg_rxbuf) < 0) {
798 			PMD_INIT_LOG(ERR, "error to parse %s",
799 				     VIRTIO_USER_ARG_MRG_RXBUF);
800 			goto end;
801 		}
802 	}
803 
804 	if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_IN_ORDER) == 1) {
805 		if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_IN_ORDER,
806 				       &get_integer_arg, &in_order) < 0) {
807 			PMD_INIT_LOG(ERR, "error to parse %s",
808 				     VIRTIO_USER_ARG_IN_ORDER);
809 			goto end;
810 		}
811 	}
812 
813 	eth_dev = virtio_user_eth_dev_alloc(dev);
814 	if (!eth_dev) {
815 		PMD_INIT_LOG(ERR, "virtio_user fails to alloc device");
816 		goto end;
817 	}
818 
819 	hw = eth_dev->data->dev_private;
820 	if (virtio_user_dev_init(hw->virtio_user_dev, path, queues, cq,
821 			 queue_size, mac_addr, &ifname, server_mode,
822 			 mrg_rxbuf, in_order, packed_vq, backend_type) < 0) {
823 		PMD_INIT_LOG(ERR, "virtio_user_dev_init fails");
824 		virtio_user_eth_dev_free(eth_dev);
825 		goto end;
826 	}
827 
828 	/* previously called by pci probing for physical dev */
829 	if (eth_virtio_dev_init(eth_dev) < 0) {
830 		PMD_INIT_LOG(ERR, "eth_virtio_dev_init fails");
831 		virtio_user_eth_dev_free(eth_dev);
832 		goto end;
833 	}
834 
835 	if (vectorized) {
836 		if (packed_vq) {
837 #if defined(CC_AVX512_SUPPORT)
838 			hw->use_vec_rx = 1;
839 			hw->use_vec_tx = 1;
840 #else
841 			PMD_INIT_LOG(INFO,
842 				"building environment do not support packed ring vectorized");
843 #endif
844 		} else {
845 			hw->use_vec_rx = 1;
846 		}
847 	}
848 
849 	rte_eth_dev_probing_finish(eth_dev);
850 	ret = 0;
851 
852 end:
853 	if (kvlist)
854 		rte_kvargs_free(kvlist);
855 	if (path)
856 		free(path);
857 	if (mac_addr)
858 		free(mac_addr);
859 	if (ifname)
860 		free(ifname);
861 	return ret;
862 }
863 
864 static int
865 virtio_user_pmd_remove(struct rte_vdev_device *vdev)
866 {
867 	const char *name;
868 	struct rte_eth_dev *eth_dev;
869 
870 	if (!vdev)
871 		return -EINVAL;
872 
873 	name = rte_vdev_device_name(vdev);
874 	PMD_DRV_LOG(INFO, "Un-Initializing %s", name);
875 	eth_dev = rte_eth_dev_allocated(name);
876 	/* Port has already been released by close. */
877 	if (!eth_dev)
878 		return 0;
879 
880 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
881 		return rte_eth_dev_release_port(eth_dev);
882 
883 	/* make sure the device is stopped, queues freed */
884 	rte_eth_dev_close(eth_dev->data->port_id);
885 
886 	return 0;
887 }
888 
889 static int virtio_user_pmd_dma_map(struct rte_vdev_device *vdev, void *addr,
890 		uint64_t iova, size_t len)
891 {
892 	const char *name;
893 	struct rte_eth_dev *eth_dev;
894 	struct virtio_user_dev *dev;
895 	struct virtio_hw *hw;
896 
897 	if (!vdev)
898 		return -EINVAL;
899 
900 	name = rte_vdev_device_name(vdev);
901 	eth_dev = rte_eth_dev_allocated(name);
902 	/* Port has already been released by close. */
903 	if (!eth_dev)
904 		return 0;
905 
906 	hw = (struct virtio_hw *)eth_dev->data->dev_private;
907 	dev = hw->virtio_user_dev;
908 
909 	if (dev->ops->dma_map)
910 		return dev->ops->dma_map(dev, addr, iova, len);
911 
912 	return 0;
913 }
914 
915 static int virtio_user_pmd_dma_unmap(struct rte_vdev_device *vdev, void *addr,
916 		uint64_t iova, size_t len)
917 {
918 	const char *name;
919 	struct rte_eth_dev *eth_dev;
920 	struct virtio_user_dev *dev;
921 	struct virtio_hw *hw;
922 
923 	if (!vdev)
924 		return -EINVAL;
925 
926 	name = rte_vdev_device_name(vdev);
927 	eth_dev = rte_eth_dev_allocated(name);
928 	/* Port has already been released by close. */
929 	if (!eth_dev)
930 		return 0;
931 
932 	hw = (struct virtio_hw *)eth_dev->data->dev_private;
933 	dev = hw->virtio_user_dev;
934 
935 	if (dev->ops->dma_unmap)
936 		return dev->ops->dma_unmap(dev, addr, iova, len);
937 
938 	return 0;
939 }
940 
941 static struct rte_vdev_driver virtio_user_driver = {
942 	.probe = virtio_user_pmd_probe,
943 	.remove = virtio_user_pmd_remove,
944 	.dma_map = virtio_user_pmd_dma_map,
945 	.dma_unmap = virtio_user_pmd_dma_unmap,
946 };
947 
948 RTE_PMD_REGISTER_VDEV(net_virtio_user, virtio_user_driver);
949 RTE_PMD_REGISTER_ALIAS(net_virtio_user, virtio_user);
950 RTE_PMD_REGISTER_PARAM_STRING(net_virtio_user,
951 	"path=<path> "
952 	"mac=<mac addr> "
953 	"cq=<int> "
954 	"queue_size=<int> "
955 	"queues=<int> "
956 	"iface=<string> "
957 	"server=<0|1> "
958 	"mrg_rxbuf=<0|1> "
959 	"in_order=<0|1> "
960 	"packed_vq=<0|1> "
961 	"speed=<int> "
962 	"vectorized=<0|1>");
963