xref: /dpdk/drivers/net/virtio/virtio_user_ethdev.c (revision dc9e658013d9c2be798f9ff43b8cd2c63c844f17)
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 <sys/socket.h>
10 
11 #include <rte_malloc.h>
12 #include <rte_kvargs.h>
13 #include <rte_ethdev_vdev.h>
14 #include <rte_bus_vdev.h>
15 #include <rte_alarm.h>
16 #include <rte_cycles.h>
17 
18 #include "virtio_ethdev.h"
19 #include "virtio_logs.h"
20 #include "virtio_pci.h"
21 #include "virtqueue.h"
22 #include "virtio_rxtx.h"
23 #include "virtio_user/virtio_user_dev.h"
24 #include "virtio_user/vhost.h"
25 
26 #define virtio_user_get_dev(hw) \
27 	((struct virtio_user_dev *)(hw)->virtio_user_dev)
28 
29 static void
30 virtio_user_reset_queues_packed(struct rte_eth_dev *dev)
31 {
32 	struct virtio_hw *hw = dev->data->dev_private;
33 	struct virtnet_rx *rxvq;
34 	struct virtnet_tx *txvq;
35 	uint16_t i;
36 
37 	/* Add lock to avoid queue contention. */
38 	rte_spinlock_lock(&hw->state_lock);
39 	hw->started = 0;
40 
41 	/*
42 	 * Waitting for datapath to complete before resetting queues.
43 	 * 1 ms should be enough for the ongoing Tx/Rx function to finish.
44 	 */
45 	rte_delay_ms(1);
46 
47 	/* Vring reset for each Tx queue and Rx queue. */
48 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
49 		rxvq = dev->data->rx_queues[i];
50 		virtqueue_rxvq_reset_packed(rxvq->vq);
51 		virtio_dev_rx_queue_setup_finish(dev, i);
52 	}
53 
54 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
55 		txvq = dev->data->tx_queues[i];
56 		virtqueue_txvq_reset_packed(txvq->vq);
57 	}
58 
59 	hw->started = 1;
60 	rte_spinlock_unlock(&hw->state_lock);
61 }
62 
63 
64 static int
65 virtio_user_server_reconnect(struct virtio_user_dev *dev)
66 {
67 	int ret;
68 	int connectfd;
69 	struct rte_eth_dev *eth_dev = &rte_eth_devices[dev->port_id];
70 	struct virtio_hw *hw = eth_dev->data->dev_private;
71 	uint64_t protocol_features;
72 
73 	connectfd = accept(dev->listenfd, NULL, NULL);
74 	if (connectfd < 0)
75 		return -1;
76 
77 	dev->vhostfd = connectfd;
78 	if (dev->ops->send_request(dev, VHOST_USER_GET_FEATURES,
79 				   &dev->device_features) < 0) {
80 		PMD_INIT_LOG(ERR, "get_features failed: %s",
81 			     strerror(errno));
82 		return -1;
83 	}
84 
85 	if (dev->device_features &
86 			(1ULL << VHOST_USER_F_PROTOCOL_FEATURES)) {
87 		if (dev->ops->send_request(dev,
88 					VHOST_USER_GET_PROTOCOL_FEATURES,
89 					&protocol_features))
90 			return -1;
91 
92 		dev->protocol_features &= protocol_features;
93 
94 		if (dev->ops->send_request(dev,
95 					VHOST_USER_SET_PROTOCOL_FEATURES,
96 					&dev->protocol_features))
97 			return -1;
98 
99 		if (!(dev->protocol_features &
100 				(1ULL << VHOST_USER_PROTOCOL_F_MQ)))
101 			dev->unsupported_features |= (1ull << VIRTIO_NET_F_MQ);
102 	}
103 
104 	dev->device_features |= dev->frontend_features;
105 
106 	/* umask vhost-user unsupported features */
107 	dev->device_features &= ~(dev->unsupported_features);
108 
109 	dev->features &= dev->device_features;
110 
111 	/* For packed ring, resetting queues is required in reconnection. */
112 	if (vtpci_packed_queue(hw) &&
113 	   (vtpci_get_status(hw) & VIRTIO_CONFIG_STATUS_DRIVER_OK)) {
114 		PMD_INIT_LOG(NOTICE, "Packets on the fly will be dropped"
115 				" when packed ring reconnecting.");
116 		virtio_user_reset_queues_packed(eth_dev);
117 	}
118 
119 	ret = virtio_user_start_device(dev);
120 	if (ret < 0)
121 		return -1;
122 
123 	if (dev->queue_pairs > 1) {
124 		ret = virtio_user_handle_mq(dev, dev->queue_pairs);
125 		if (ret != 0) {
126 			PMD_INIT_LOG(ERR, "Fails to enable multi-queue pairs!");
127 			return -1;
128 		}
129 	}
130 	if (eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC) {
131 		if (rte_intr_disable(eth_dev->intr_handle) < 0) {
132 			PMD_DRV_LOG(ERR, "interrupt disable failed");
133 			return -1;
134 		}
135 		rte_intr_callback_unregister(eth_dev->intr_handle,
136 					     virtio_interrupt_handler,
137 					     eth_dev);
138 		eth_dev->intr_handle->fd = connectfd;
139 		rte_intr_callback_register(eth_dev->intr_handle,
140 					   virtio_interrupt_handler, eth_dev);
141 
142 		if (rte_intr_enable(eth_dev->intr_handle) < 0) {
143 			PMD_DRV_LOG(ERR, "interrupt enable failed");
144 			return -1;
145 		}
146 	}
147 	PMD_INIT_LOG(NOTICE, "server mode virtio-user reconnection succeeds!");
148 	return 0;
149 }
150 
151 static void
152 virtio_user_delayed_handler(void *param)
153 {
154 	struct virtio_hw *hw = (struct virtio_hw *)param;
155 	struct rte_eth_dev *eth_dev = &rte_eth_devices[hw->port_id];
156 	struct virtio_user_dev *dev = virtio_user_get_dev(hw);
157 
158 	if (rte_intr_disable(eth_dev->intr_handle) < 0) {
159 		PMD_DRV_LOG(ERR, "interrupt disable failed");
160 		return;
161 	}
162 	rte_intr_callback_unregister(eth_dev->intr_handle,
163 				     virtio_interrupt_handler, eth_dev);
164 	if (dev->is_server) {
165 		if (dev->vhostfd >= 0) {
166 			close(dev->vhostfd);
167 			dev->vhostfd = -1;
168 		}
169 		eth_dev->intr_handle->fd = dev->listenfd;
170 		rte_intr_callback_register(eth_dev->intr_handle,
171 					   virtio_interrupt_handler, eth_dev);
172 		if (rte_intr_enable(eth_dev->intr_handle) < 0) {
173 			PMD_DRV_LOG(ERR, "interrupt enable failed");
174 			return;
175 		}
176 	}
177 }
178 
179 static void
180 virtio_user_read_dev_config(struct virtio_hw *hw, size_t offset,
181 		     void *dst, int length)
182 {
183 	int i;
184 	struct virtio_user_dev *dev = virtio_user_get_dev(hw);
185 
186 	if (offset == offsetof(struct virtio_net_config, mac) &&
187 	    length == RTE_ETHER_ADDR_LEN) {
188 		for (i = 0; i < RTE_ETHER_ADDR_LEN; ++i)
189 			((uint8_t *)dst)[i] = dev->mac_addr[i];
190 		return;
191 	}
192 
193 	if (offset == offsetof(struct virtio_net_config, status)) {
194 		char buf[128];
195 
196 		if (dev->vhostfd >= 0) {
197 			int r;
198 			int flags;
199 
200 			flags = fcntl(dev->vhostfd, F_GETFL);
201 			if (fcntl(dev->vhostfd, F_SETFL,
202 					flags | O_NONBLOCK) == -1) {
203 				PMD_DRV_LOG(ERR, "error setting O_NONBLOCK flag");
204 				return;
205 			}
206 			r = recv(dev->vhostfd, buf, 128, MSG_PEEK);
207 			if (r == 0 || (r < 0 && errno != EAGAIN)) {
208 				dev->net_status &= (~VIRTIO_NET_S_LINK_UP);
209 				PMD_DRV_LOG(ERR, "virtio-user port %u is down",
210 					    hw->port_id);
211 
212 				/* This function could be called in the process
213 				 * of interrupt handling, callback cannot be
214 				 * unregistered here, set an alarm to do it.
215 				 */
216 				rte_eal_alarm_set(1,
217 						  virtio_user_delayed_handler,
218 						  (void *)hw);
219 			} else {
220 				dev->net_status |= VIRTIO_NET_S_LINK_UP;
221 			}
222 			if (fcntl(dev->vhostfd, F_SETFL,
223 					flags & ~O_NONBLOCK) == -1) {
224 				PMD_DRV_LOG(ERR, "error clearing O_NONBLOCK flag");
225 				return;
226 			}
227 		} else if (dev->is_server) {
228 			dev->net_status &= (~VIRTIO_NET_S_LINK_UP);
229 			if (virtio_user_server_reconnect(dev) >= 0)
230 				dev->net_status |= VIRTIO_NET_S_LINK_UP;
231 		}
232 
233 		*(uint16_t *)dst = dev->net_status;
234 	}
235 
236 	if (offset == offsetof(struct virtio_net_config, max_virtqueue_pairs))
237 		*(uint16_t *)dst = dev->max_queue_pairs;
238 }
239 
240 static void
241 virtio_user_write_dev_config(struct virtio_hw *hw, size_t offset,
242 		      const void *src, int length)
243 {
244 	int i;
245 	struct virtio_user_dev *dev = virtio_user_get_dev(hw);
246 
247 	if ((offset == offsetof(struct virtio_net_config, mac)) &&
248 	    (length == RTE_ETHER_ADDR_LEN))
249 		for (i = 0; i < RTE_ETHER_ADDR_LEN; ++i)
250 			dev->mac_addr[i] = ((const uint8_t *)src)[i];
251 	else
252 		PMD_DRV_LOG(ERR, "not supported offset=%zu, len=%d",
253 			    offset, length);
254 }
255 
256 static void
257 virtio_user_reset(struct virtio_hw *hw)
258 {
259 	struct virtio_user_dev *dev = virtio_user_get_dev(hw);
260 
261 	if (dev->status & VIRTIO_CONFIG_STATUS_DRIVER_OK)
262 		virtio_user_stop_device(dev);
263 }
264 
265 static void
266 virtio_user_set_status(struct virtio_hw *hw, uint8_t status)
267 {
268 	struct virtio_user_dev *dev = virtio_user_get_dev(hw);
269 
270 	if (status & VIRTIO_CONFIG_STATUS_DRIVER_OK)
271 		virtio_user_start_device(dev);
272 	else if (status == VIRTIO_CONFIG_STATUS_RESET)
273 		virtio_user_reset(hw);
274 	dev->status = status;
275 	virtio_user_send_status_update(dev, status);
276 }
277 
278 static uint8_t
279 virtio_user_get_status(struct virtio_hw *hw)
280 {
281 	struct virtio_user_dev *dev = virtio_user_get_dev(hw);
282 
283 	virtio_user_update_status(dev);
284 
285 	return dev->status;
286 }
287 
288 static uint64_t
289 virtio_user_get_features(struct virtio_hw *hw)
290 {
291 	struct virtio_user_dev *dev = virtio_user_get_dev(hw);
292 
293 	/* unmask feature bits defined in vhost user protocol */
294 	return dev->device_features & VIRTIO_PMD_SUPPORTED_GUEST_FEATURES;
295 }
296 
297 static void
298 virtio_user_set_features(struct virtio_hw *hw, uint64_t features)
299 {
300 	struct virtio_user_dev *dev = virtio_user_get_dev(hw);
301 
302 	dev->features = features & dev->device_features;
303 }
304 
305 static uint8_t
306 virtio_user_get_isr(struct virtio_hw *hw __rte_unused)
307 {
308 	/* rxq interrupts and config interrupt are separated in virtio-user,
309 	 * here we only report config change.
310 	 */
311 	return VIRTIO_PCI_ISR_CONFIG;
312 }
313 
314 static uint16_t
315 virtio_user_set_config_irq(struct virtio_hw *hw __rte_unused,
316 		    uint16_t vec __rte_unused)
317 {
318 	return 0;
319 }
320 
321 static uint16_t
322 virtio_user_set_queue_irq(struct virtio_hw *hw __rte_unused,
323 			  struct virtqueue *vq __rte_unused,
324 			  uint16_t vec)
325 {
326 	/* pretend we have done that */
327 	return vec;
328 }
329 
330 /* This function is to get the queue size, aka, number of descs, of a specified
331  * queue. Different with the VHOST_USER_GET_QUEUE_NUM, which is used to get the
332  * max supported queues.
333  */
334 static uint16_t
335 virtio_user_get_queue_num(struct virtio_hw *hw, uint16_t queue_id __rte_unused)
336 {
337 	struct virtio_user_dev *dev = virtio_user_get_dev(hw);
338 
339 	/* Currently, each queue has same queue size */
340 	return dev->queue_size;
341 }
342 
343 static void
344 virtio_user_setup_queue_packed(struct virtqueue *vq,
345 			       struct virtio_user_dev *dev)
346 {
347 	uint16_t queue_idx = vq->vq_queue_index;
348 	struct vring_packed *vring;
349 	uint64_t desc_addr;
350 	uint64_t avail_addr;
351 	uint64_t used_addr;
352 	uint16_t i;
353 
354 	vring  = &dev->packed_vrings[queue_idx];
355 	desc_addr = (uintptr_t)vq->vq_ring_virt_mem;
356 	avail_addr = desc_addr + vq->vq_nentries *
357 		sizeof(struct vring_packed_desc);
358 	used_addr = RTE_ALIGN_CEIL(avail_addr +
359 			   sizeof(struct vring_packed_desc_event),
360 			   VIRTIO_PCI_VRING_ALIGN);
361 	vring->num = vq->vq_nentries;
362 	vring->desc = (void *)(uintptr_t)desc_addr;
363 	vring->driver = (void *)(uintptr_t)avail_addr;
364 	vring->device = (void *)(uintptr_t)used_addr;
365 	dev->packed_queues[queue_idx].avail_wrap_counter = true;
366 	dev->packed_queues[queue_idx].used_wrap_counter = true;
367 
368 	for (i = 0; i < vring->num; i++)
369 		vring->desc[i].flags = 0;
370 }
371 
372 static void
373 virtio_user_setup_queue_split(struct virtqueue *vq, struct virtio_user_dev *dev)
374 {
375 	uint16_t queue_idx = vq->vq_queue_index;
376 	uint64_t desc_addr, avail_addr, used_addr;
377 
378 	desc_addr = (uintptr_t)vq->vq_ring_virt_mem;
379 	avail_addr = desc_addr + vq->vq_nentries * sizeof(struct vring_desc);
380 	used_addr = RTE_ALIGN_CEIL(avail_addr + offsetof(struct vring_avail,
381 							 ring[vq->vq_nentries]),
382 				   VIRTIO_PCI_VRING_ALIGN);
383 
384 	dev->vrings[queue_idx].num = vq->vq_nentries;
385 	dev->vrings[queue_idx].desc = (void *)(uintptr_t)desc_addr;
386 	dev->vrings[queue_idx].avail = (void *)(uintptr_t)avail_addr;
387 	dev->vrings[queue_idx].used = (void *)(uintptr_t)used_addr;
388 }
389 
390 static int
391 virtio_user_setup_queue(struct virtio_hw *hw, struct virtqueue *vq)
392 {
393 	struct virtio_user_dev *dev = virtio_user_get_dev(hw);
394 
395 	if (vtpci_packed_queue(hw))
396 		virtio_user_setup_queue_packed(vq, dev);
397 	else
398 		virtio_user_setup_queue_split(vq, dev);
399 
400 	return 0;
401 }
402 
403 static void
404 virtio_user_del_queue(struct virtio_hw *hw, struct virtqueue *vq)
405 {
406 	/* For legacy devices, write 0 to VIRTIO_PCI_QUEUE_PFN port, QEMU
407 	 * correspondingly stops the ioeventfds, and reset the status of
408 	 * the device.
409 	 * For modern devices, set queue desc, avail, used in PCI bar to 0,
410 	 * not see any more behavior in QEMU.
411 	 *
412 	 * Here we just care about what information to deliver to vhost-user
413 	 * or vhost-kernel. So we just close ioeventfd for now.
414 	 */
415 	struct virtio_user_dev *dev = virtio_user_get_dev(hw);
416 
417 	close(dev->callfds[vq->vq_queue_index]);
418 	close(dev->kickfds[vq->vq_queue_index]);
419 }
420 
421 static void
422 virtio_user_notify_queue(struct virtio_hw *hw, struct virtqueue *vq)
423 {
424 	uint64_t buf = 1;
425 	struct virtio_user_dev *dev = virtio_user_get_dev(hw);
426 
427 	if (hw->cvq && (hw->cvq->vq == vq)) {
428 		if (vtpci_packed_queue(vq->hw))
429 			virtio_user_handle_cq_packed(dev, vq->vq_queue_index);
430 		else
431 			virtio_user_handle_cq(dev, vq->vq_queue_index);
432 		return;
433 	}
434 
435 	if (write(dev->kickfds[vq->vq_queue_index], &buf, sizeof(buf)) < 0)
436 		PMD_DRV_LOG(ERR, "failed to kick backend: %s",
437 			    strerror(errno));
438 }
439 
440 const struct virtio_pci_ops virtio_user_ops = {
441 	.read_dev_cfg	= virtio_user_read_dev_config,
442 	.write_dev_cfg	= virtio_user_write_dev_config,
443 	.get_status	= virtio_user_get_status,
444 	.set_status	= virtio_user_set_status,
445 	.get_features	= virtio_user_get_features,
446 	.set_features	= virtio_user_set_features,
447 	.get_isr	= virtio_user_get_isr,
448 	.set_config_irq	= virtio_user_set_config_irq,
449 	.set_queue_irq	= virtio_user_set_queue_irq,
450 	.get_queue_num	= virtio_user_get_queue_num,
451 	.setup_queue	= virtio_user_setup_queue,
452 	.del_queue	= virtio_user_del_queue,
453 	.notify_queue	= virtio_user_notify_queue,
454 };
455 
456 static const char *valid_args[] = {
457 #define VIRTIO_USER_ARG_QUEUES_NUM     "queues"
458 	VIRTIO_USER_ARG_QUEUES_NUM,
459 #define VIRTIO_USER_ARG_CQ_NUM         "cq"
460 	VIRTIO_USER_ARG_CQ_NUM,
461 #define VIRTIO_USER_ARG_MAC            "mac"
462 	VIRTIO_USER_ARG_MAC,
463 #define VIRTIO_USER_ARG_PATH           "path"
464 	VIRTIO_USER_ARG_PATH,
465 #define VIRTIO_USER_ARG_QUEUE_SIZE     "queue_size"
466 	VIRTIO_USER_ARG_QUEUE_SIZE,
467 #define VIRTIO_USER_ARG_INTERFACE_NAME "iface"
468 	VIRTIO_USER_ARG_INTERFACE_NAME,
469 #define VIRTIO_USER_ARG_SERVER_MODE    "server"
470 	VIRTIO_USER_ARG_SERVER_MODE,
471 #define VIRTIO_USER_ARG_MRG_RXBUF      "mrg_rxbuf"
472 	VIRTIO_USER_ARG_MRG_RXBUF,
473 #define VIRTIO_USER_ARG_IN_ORDER       "in_order"
474 	VIRTIO_USER_ARG_IN_ORDER,
475 #define VIRTIO_USER_ARG_PACKED_VQ      "packed_vq"
476 	VIRTIO_USER_ARG_PACKED_VQ,
477 #define VIRTIO_USER_ARG_SPEED          "speed"
478 	VIRTIO_USER_ARG_SPEED,
479 #define VIRTIO_USER_ARG_VECTORIZED     "vectorized"
480 	VIRTIO_USER_ARG_VECTORIZED,
481 	NULL
482 };
483 
484 #define VIRTIO_USER_DEF_CQ_EN	0
485 #define VIRTIO_USER_DEF_Q_NUM	1
486 #define VIRTIO_USER_DEF_Q_SZ	256
487 #define VIRTIO_USER_DEF_SERVER_MODE	0
488 
489 static int
490 get_string_arg(const char *key __rte_unused,
491 	       const char *value, void *extra_args)
492 {
493 	if (!value || !extra_args)
494 		return -EINVAL;
495 
496 	*(char **)extra_args = strdup(value);
497 
498 	if (!*(char **)extra_args)
499 		return -ENOMEM;
500 
501 	return 0;
502 }
503 
504 static int
505 get_integer_arg(const char *key __rte_unused,
506 		const char *value, void *extra_args)
507 {
508 	uint64_t integer = 0;
509 	if (!value || !extra_args)
510 		return -EINVAL;
511 	errno = 0;
512 	integer = strtoull(value, NULL, 0);
513 	/* extra_args keeps default value, it should be replaced
514 	 * only in case of successful parsing of the 'value' arg
515 	 */
516 	if (errno == 0)
517 		*(uint64_t *)extra_args = integer;
518 	return -errno;
519 }
520 
521 static struct rte_eth_dev *
522 virtio_user_eth_dev_alloc(struct rte_vdev_device *vdev)
523 {
524 	struct rte_eth_dev *eth_dev;
525 	struct rte_eth_dev_data *data;
526 	struct virtio_hw *hw;
527 	struct virtio_user_dev *dev;
528 
529 	eth_dev = rte_eth_vdev_allocate(vdev, sizeof(*hw));
530 	if (!eth_dev) {
531 		PMD_INIT_LOG(ERR, "cannot alloc rte_eth_dev");
532 		return NULL;
533 	}
534 
535 	data = eth_dev->data;
536 	hw = eth_dev->data->dev_private;
537 
538 	dev = rte_zmalloc(NULL, sizeof(*dev), 0);
539 	if (!dev) {
540 		PMD_INIT_LOG(ERR, "malloc virtio_user_dev failed");
541 		rte_eth_dev_release_port(eth_dev);
542 		return NULL;
543 	}
544 
545 	hw->port_id = data->port_id;
546 	dev->port_id = data->port_id;
547 	virtio_hw_internal[hw->port_id].vtpci_ops = &virtio_user_ops;
548 	/*
549 	 * MSIX is required to enable LSC (see virtio_init_device).
550 	 * Here just pretend that we support msix.
551 	 */
552 	hw->use_msix = 1;
553 	hw->modern   = 0;
554 	hw->use_vec_rx = 0;
555 	hw->use_vec_tx = 0;
556 	hw->use_inorder_rx = 0;
557 	hw->use_inorder_tx = 0;
558 	hw->virtio_user_dev = dev;
559 	return eth_dev;
560 }
561 
562 static void
563 virtio_user_eth_dev_free(struct rte_eth_dev *eth_dev)
564 {
565 	struct rte_eth_dev_data *data = eth_dev->data;
566 	struct virtio_hw *hw = data->dev_private;
567 
568 	rte_free(hw->virtio_user_dev);
569 	rte_eth_dev_release_port(eth_dev);
570 }
571 
572 /* Dev initialization routine. Invoked once for each virtio vdev at
573  * EAL init time, see rte_bus_probe().
574  * Returns 0 on success.
575  */
576 static int
577 virtio_user_pmd_probe(struct rte_vdev_device *dev)
578 {
579 	struct rte_kvargs *kvlist = NULL;
580 	struct rte_eth_dev *eth_dev;
581 	struct virtio_hw *hw;
582 	uint64_t queues = VIRTIO_USER_DEF_Q_NUM;
583 	uint64_t cq = VIRTIO_USER_DEF_CQ_EN;
584 	uint64_t queue_size = VIRTIO_USER_DEF_Q_SZ;
585 	uint64_t server_mode = VIRTIO_USER_DEF_SERVER_MODE;
586 	uint64_t mrg_rxbuf = 1;
587 	uint64_t in_order = 1;
588 	uint64_t packed_vq = 0;
589 	uint64_t vectorized = 0;
590 	char *path = NULL;
591 	char *ifname = NULL;
592 	char *mac_addr = NULL;
593 	int ret = -1;
594 
595 	if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
596 		const char *name = rte_vdev_device_name(dev);
597 		eth_dev = rte_eth_dev_attach_secondary(name);
598 		if (!eth_dev) {
599 			PMD_INIT_LOG(ERR, "Failed to probe %s", name);
600 			return -1;
601 		}
602 
603 		if (eth_virtio_dev_init(eth_dev) < 0) {
604 			PMD_INIT_LOG(ERR, "eth_virtio_dev_init fails");
605 			rte_eth_dev_release_port(eth_dev);
606 			return -1;
607 		}
608 
609 		eth_dev->dev_ops = &virtio_user_secondary_eth_dev_ops;
610 		eth_dev->device = &dev->device;
611 		rte_eth_dev_probing_finish(eth_dev);
612 		return 0;
613 	}
614 
615 	kvlist = rte_kvargs_parse(rte_vdev_device_args(dev), valid_args);
616 	if (!kvlist) {
617 		PMD_INIT_LOG(ERR, "error when parsing param");
618 		goto end;
619 	}
620 
621 	if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_PATH) == 1) {
622 		if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_PATH,
623 				       &get_string_arg, &path) < 0) {
624 			PMD_INIT_LOG(ERR, "error to parse %s",
625 				     VIRTIO_USER_ARG_PATH);
626 			goto end;
627 		}
628 	} else {
629 		PMD_INIT_LOG(ERR, "arg %s is mandatory for virtio_user",
630 			     VIRTIO_USER_ARG_PATH);
631 		goto end;
632 	}
633 
634 	if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_INTERFACE_NAME) == 1) {
635 		if (is_vhost_user_by_type(path)) {
636 			PMD_INIT_LOG(ERR,
637 				"arg %s applies only to vhost-kernel backend",
638 				VIRTIO_USER_ARG_INTERFACE_NAME);
639 			goto end;
640 		}
641 
642 		if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_INTERFACE_NAME,
643 				       &get_string_arg, &ifname) < 0) {
644 			PMD_INIT_LOG(ERR, "error to parse %s",
645 				     VIRTIO_USER_ARG_INTERFACE_NAME);
646 			goto end;
647 		}
648 	}
649 
650 	if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_MAC) == 1) {
651 		if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_MAC,
652 				       &get_string_arg, &mac_addr) < 0) {
653 			PMD_INIT_LOG(ERR, "error to parse %s",
654 				     VIRTIO_USER_ARG_MAC);
655 			goto end;
656 		}
657 	}
658 
659 	if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_QUEUE_SIZE) == 1) {
660 		if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_QUEUE_SIZE,
661 				       &get_integer_arg, &queue_size) < 0) {
662 			PMD_INIT_LOG(ERR, "error to parse %s",
663 				     VIRTIO_USER_ARG_QUEUE_SIZE);
664 			goto end;
665 		}
666 	}
667 
668 	if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_QUEUES_NUM) == 1) {
669 		if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_QUEUES_NUM,
670 				       &get_integer_arg, &queues) < 0) {
671 			PMD_INIT_LOG(ERR, "error to parse %s",
672 				     VIRTIO_USER_ARG_QUEUES_NUM);
673 			goto end;
674 		}
675 	}
676 
677 	if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_SERVER_MODE) == 1) {
678 		if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_SERVER_MODE,
679 				       &get_integer_arg, &server_mode) < 0) {
680 			PMD_INIT_LOG(ERR, "error to parse %s",
681 				     VIRTIO_USER_ARG_SERVER_MODE);
682 			goto end;
683 		}
684 	}
685 
686 	if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_CQ_NUM) == 1) {
687 		if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_CQ_NUM,
688 				       &get_integer_arg, &cq) < 0) {
689 			PMD_INIT_LOG(ERR, "error to parse %s",
690 				     VIRTIO_USER_ARG_CQ_NUM);
691 			goto end;
692 		}
693 	} else if (queues > 1) {
694 		cq = 1;
695 	}
696 
697 	if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_PACKED_VQ) == 1) {
698 		if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_PACKED_VQ,
699 				       &get_integer_arg, &packed_vq) < 0) {
700 			PMD_INIT_LOG(ERR, "error to parse %s",
701 				     VIRTIO_USER_ARG_PACKED_VQ);
702 			goto end;
703 		}
704 	}
705 
706 	if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_VECTORIZED) == 1) {
707 		if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_VECTORIZED,
708 				       &get_integer_arg, &vectorized) < 0) {
709 			PMD_INIT_LOG(ERR, "error to parse %s",
710 				     VIRTIO_USER_ARG_VECTORIZED);
711 			goto end;
712 		}
713 	}
714 
715 	if (queues > 1 && cq == 0) {
716 		PMD_INIT_LOG(ERR, "multi-q requires ctrl-q");
717 		goto end;
718 	}
719 
720 	if (queues > VIRTIO_MAX_VIRTQUEUE_PAIRS) {
721 		PMD_INIT_LOG(ERR, "arg %s %" PRIu64 " exceeds the limit %u",
722 			VIRTIO_USER_ARG_QUEUES_NUM, queues,
723 			VIRTIO_MAX_VIRTQUEUE_PAIRS);
724 		goto end;
725 	}
726 
727 	if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_MRG_RXBUF) == 1) {
728 		if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_MRG_RXBUF,
729 				       &get_integer_arg, &mrg_rxbuf) < 0) {
730 			PMD_INIT_LOG(ERR, "error to parse %s",
731 				     VIRTIO_USER_ARG_MRG_RXBUF);
732 			goto end;
733 		}
734 	}
735 
736 	if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_IN_ORDER) == 1) {
737 		if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_IN_ORDER,
738 				       &get_integer_arg, &in_order) < 0) {
739 			PMD_INIT_LOG(ERR, "error to parse %s",
740 				     VIRTIO_USER_ARG_IN_ORDER);
741 			goto end;
742 		}
743 	}
744 
745 	eth_dev = virtio_user_eth_dev_alloc(dev);
746 	if (!eth_dev) {
747 		PMD_INIT_LOG(ERR, "virtio_user fails to alloc device");
748 		goto end;
749 	}
750 
751 	hw = eth_dev->data->dev_private;
752 	if (virtio_user_dev_init(hw->virtio_user_dev, path, queues, cq,
753 			 queue_size, mac_addr, &ifname, server_mode,
754 			 mrg_rxbuf, in_order, packed_vq) < 0) {
755 		PMD_INIT_LOG(ERR, "virtio_user_dev_init fails");
756 		virtio_user_eth_dev_free(eth_dev);
757 		goto end;
758 	}
759 
760 	/* previously called by pci probing for physical dev */
761 	if (eth_virtio_dev_init(eth_dev) < 0) {
762 		PMD_INIT_LOG(ERR, "eth_virtio_dev_init fails");
763 		virtio_user_eth_dev_free(eth_dev);
764 		goto end;
765 	}
766 
767 	if (vectorized) {
768 		if (packed_vq) {
769 #if defined(CC_AVX512_SUPPORT)
770 			hw->use_vec_rx = 1;
771 			hw->use_vec_tx = 1;
772 #else
773 			PMD_INIT_LOG(INFO,
774 				"building environment do not support packed ring vectorized");
775 #endif
776 		} else {
777 			hw->use_vec_rx = 1;
778 		}
779 	}
780 
781 	rte_eth_dev_probing_finish(eth_dev);
782 	ret = 0;
783 
784 end:
785 	if (kvlist)
786 		rte_kvargs_free(kvlist);
787 	if (path)
788 		free(path);
789 	if (mac_addr)
790 		free(mac_addr);
791 	if (ifname)
792 		free(ifname);
793 	return ret;
794 }
795 
796 static int
797 virtio_user_pmd_remove(struct rte_vdev_device *vdev)
798 {
799 	const char *name;
800 	struct rte_eth_dev *eth_dev;
801 
802 	if (!vdev)
803 		return -EINVAL;
804 
805 	name = rte_vdev_device_name(vdev);
806 	PMD_DRV_LOG(INFO, "Un-Initializing %s", name);
807 	eth_dev = rte_eth_dev_allocated(name);
808 	/* Port has already been released by close. */
809 	if (!eth_dev)
810 		return 0;
811 
812 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
813 		return rte_eth_dev_release_port(eth_dev);
814 
815 	/* make sure the device is stopped, queues freed */
816 	rte_eth_dev_close(eth_dev->data->port_id);
817 
818 	return 0;
819 }
820 
821 static struct rte_vdev_driver virtio_user_driver = {
822 	.probe = virtio_user_pmd_probe,
823 	.remove = virtio_user_pmd_remove,
824 };
825 
826 RTE_PMD_REGISTER_VDEV(net_virtio_user, virtio_user_driver);
827 RTE_PMD_REGISTER_ALIAS(net_virtio_user, virtio_user);
828 RTE_PMD_REGISTER_PARAM_STRING(net_virtio_user,
829 	"path=<path> "
830 	"mac=<mac addr> "
831 	"cq=<int> "
832 	"queue_size=<int> "
833 	"queues=<int> "
834 	"iface=<string> "
835 	"server=<0|1> "
836 	"mrg_rxbuf=<0|1> "
837 	"in_order=<0|1> "
838 	"packed_vq=<0|1> "
839 	"speed=<int> "
840 	"vectorized=<0|1>");
841