xref: /dpdk/drivers/net/virtio/virtio_user_ethdev.c (revision e11bdd37745229bf26b557305c07d118c3dbaad7)
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 
72 	connectfd = accept(dev->listenfd, NULL, NULL);
73 	if (connectfd < 0)
74 		return -1;
75 
76 	dev->vhostfd = connectfd;
77 	if (dev->ops->send_request(dev, VHOST_USER_GET_FEATURES,
78 				   &dev->device_features) < 0) {
79 		PMD_INIT_LOG(ERR, "get_features failed: %s",
80 			     strerror(errno));
81 		return -1;
82 	}
83 
84 	dev->device_features |= dev->frontend_features;
85 
86 	/* umask vhost-user unsupported features */
87 	dev->device_features &= ~(dev->unsupported_features);
88 
89 	dev->features &= dev->device_features;
90 
91 	/* For packed ring, resetting queues is required in reconnection. */
92 	if (vtpci_packed_queue(hw)) {
93 		PMD_INIT_LOG(NOTICE, "Packets on the fly will be dropped"
94 				" when packed ring reconnecting.");
95 		virtio_user_reset_queues_packed(eth_dev);
96 	}
97 
98 	ret = virtio_user_start_device(dev);
99 	if (ret < 0)
100 		return -1;
101 
102 	if (dev->queue_pairs > 1) {
103 		ret = virtio_user_handle_mq(dev, dev->queue_pairs);
104 		if (ret != 0) {
105 			PMD_INIT_LOG(ERR, "Fails to enable multi-queue pairs!");
106 			return -1;
107 		}
108 	}
109 	if (eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC) {
110 		if (rte_intr_disable(eth_dev->intr_handle) < 0) {
111 			PMD_DRV_LOG(ERR, "interrupt disable failed");
112 			return -1;
113 		}
114 		rte_intr_callback_unregister(eth_dev->intr_handle,
115 					     virtio_interrupt_handler,
116 					     eth_dev);
117 		eth_dev->intr_handle->fd = connectfd;
118 		rte_intr_callback_register(eth_dev->intr_handle,
119 					   virtio_interrupt_handler, eth_dev);
120 
121 		if (rte_intr_enable(eth_dev->intr_handle) < 0) {
122 			PMD_DRV_LOG(ERR, "interrupt enable failed");
123 			return -1;
124 		}
125 	}
126 	PMD_INIT_LOG(NOTICE, "server mode virtio-user reconnection succeeds!");
127 	return 0;
128 }
129 
130 static void
131 virtio_user_delayed_handler(void *param)
132 {
133 	struct virtio_hw *hw = (struct virtio_hw *)param;
134 	struct rte_eth_dev *eth_dev = &rte_eth_devices[hw->port_id];
135 	struct virtio_user_dev *dev = virtio_user_get_dev(hw);
136 
137 	if (rte_intr_disable(eth_dev->intr_handle) < 0) {
138 		PMD_DRV_LOG(ERR, "interrupt disable failed");
139 		return;
140 	}
141 	rte_intr_callback_unregister(eth_dev->intr_handle,
142 				     virtio_interrupt_handler, eth_dev);
143 	if (dev->is_server) {
144 		if (dev->vhostfd >= 0) {
145 			close(dev->vhostfd);
146 			dev->vhostfd = -1;
147 		}
148 		eth_dev->intr_handle->fd = dev->listenfd;
149 		rte_intr_callback_register(eth_dev->intr_handle,
150 					   virtio_interrupt_handler, eth_dev);
151 		if (rte_intr_enable(eth_dev->intr_handle) < 0) {
152 			PMD_DRV_LOG(ERR, "interrupt enable failed");
153 			return;
154 		}
155 	}
156 }
157 
158 static void
159 virtio_user_read_dev_config(struct virtio_hw *hw, size_t offset,
160 		     void *dst, int length)
161 {
162 	int i;
163 	struct virtio_user_dev *dev = virtio_user_get_dev(hw);
164 
165 	if (offset == offsetof(struct virtio_net_config, mac) &&
166 	    length == RTE_ETHER_ADDR_LEN) {
167 		for (i = 0; i < RTE_ETHER_ADDR_LEN; ++i)
168 			((uint8_t *)dst)[i] = dev->mac_addr[i];
169 		return;
170 	}
171 
172 	if (offset == offsetof(struct virtio_net_config, status)) {
173 		char buf[128];
174 
175 		if (dev->vhostfd >= 0) {
176 			int r;
177 			int flags;
178 
179 			flags = fcntl(dev->vhostfd, F_GETFL);
180 			if (fcntl(dev->vhostfd, F_SETFL,
181 					flags | O_NONBLOCK) == -1) {
182 				PMD_DRV_LOG(ERR, "error setting O_NONBLOCK flag");
183 				return;
184 			}
185 			r = recv(dev->vhostfd, buf, 128, MSG_PEEK);
186 			if (r == 0 || (r < 0 && errno != EAGAIN)) {
187 				dev->status &= (~VIRTIO_NET_S_LINK_UP);
188 				PMD_DRV_LOG(ERR, "virtio-user port %u is down",
189 					    hw->port_id);
190 
191 				/* This function could be called in the process
192 				 * of interrupt handling, callback cannot be
193 				 * unregistered here, set an alarm to do it.
194 				 */
195 				rte_eal_alarm_set(1,
196 						  virtio_user_delayed_handler,
197 						  (void *)hw);
198 			} else {
199 				dev->status |= VIRTIO_NET_S_LINK_UP;
200 			}
201 			if (fcntl(dev->vhostfd, F_SETFL,
202 					flags & ~O_NONBLOCK) == -1) {
203 				PMD_DRV_LOG(ERR, "error clearing O_NONBLOCK flag");
204 				return;
205 			}
206 		} else if (dev->is_server) {
207 			dev->status &= (~VIRTIO_NET_S_LINK_UP);
208 			if (virtio_user_server_reconnect(dev) >= 0)
209 				dev->status |= VIRTIO_NET_S_LINK_UP;
210 		}
211 
212 		*(uint16_t *)dst = dev->status;
213 	}
214 
215 	if (offset == offsetof(struct virtio_net_config, max_virtqueue_pairs))
216 		*(uint16_t *)dst = dev->max_queue_pairs;
217 }
218 
219 static void
220 virtio_user_write_dev_config(struct virtio_hw *hw, size_t offset,
221 		      const void *src, int length)
222 {
223 	int i;
224 	struct virtio_user_dev *dev = virtio_user_get_dev(hw);
225 
226 	if ((offset == offsetof(struct virtio_net_config, mac)) &&
227 	    (length == RTE_ETHER_ADDR_LEN))
228 		for (i = 0; i < RTE_ETHER_ADDR_LEN; ++i)
229 			dev->mac_addr[i] = ((const uint8_t *)src)[i];
230 	else
231 		PMD_DRV_LOG(ERR, "not supported offset=%zu, len=%d",
232 			    offset, length);
233 }
234 
235 static void
236 virtio_user_reset(struct virtio_hw *hw)
237 {
238 	struct virtio_user_dev *dev = virtio_user_get_dev(hw);
239 
240 	if (dev->status & VIRTIO_CONFIG_STATUS_DRIVER_OK)
241 		virtio_user_stop_device(dev);
242 }
243 
244 static void
245 virtio_user_set_status(struct virtio_hw *hw, uint8_t status)
246 {
247 	struct virtio_user_dev *dev = virtio_user_get_dev(hw);
248 
249 	if (status & VIRTIO_CONFIG_STATUS_DRIVER_OK)
250 		virtio_user_start_device(dev);
251 	else if (status == VIRTIO_CONFIG_STATUS_RESET)
252 		virtio_user_reset(hw);
253 	dev->status = status;
254 }
255 
256 static uint8_t
257 virtio_user_get_status(struct virtio_hw *hw)
258 {
259 	struct virtio_user_dev *dev = virtio_user_get_dev(hw);
260 
261 	return dev->status;
262 }
263 
264 static uint64_t
265 virtio_user_get_features(struct virtio_hw *hw)
266 {
267 	struct virtio_user_dev *dev = virtio_user_get_dev(hw);
268 
269 	/* unmask feature bits defined in vhost user protocol */
270 	return dev->device_features & VIRTIO_PMD_SUPPORTED_GUEST_FEATURES;
271 }
272 
273 static void
274 virtio_user_set_features(struct virtio_hw *hw, uint64_t features)
275 {
276 	struct virtio_user_dev *dev = virtio_user_get_dev(hw);
277 
278 	dev->features = features & dev->device_features;
279 }
280 
281 static uint8_t
282 virtio_user_get_isr(struct virtio_hw *hw __rte_unused)
283 {
284 	/* rxq interrupts and config interrupt are separated in virtio-user,
285 	 * here we only report config change.
286 	 */
287 	return VIRTIO_PCI_ISR_CONFIG;
288 }
289 
290 static uint16_t
291 virtio_user_set_config_irq(struct virtio_hw *hw __rte_unused,
292 		    uint16_t vec __rte_unused)
293 {
294 	return 0;
295 }
296 
297 static uint16_t
298 virtio_user_set_queue_irq(struct virtio_hw *hw __rte_unused,
299 			  struct virtqueue *vq __rte_unused,
300 			  uint16_t vec)
301 {
302 	/* pretend we have done that */
303 	return vec;
304 }
305 
306 /* This function is to get the queue size, aka, number of descs, of a specified
307  * queue. Different with the VHOST_USER_GET_QUEUE_NUM, which is used to get the
308  * max supported queues.
309  */
310 static uint16_t
311 virtio_user_get_queue_num(struct virtio_hw *hw, uint16_t queue_id __rte_unused)
312 {
313 	struct virtio_user_dev *dev = virtio_user_get_dev(hw);
314 
315 	/* Currently, each queue has same queue size */
316 	return dev->queue_size;
317 }
318 
319 static void
320 virtio_user_setup_queue_packed(struct virtqueue *vq,
321 			       struct virtio_user_dev *dev)
322 {
323 	uint16_t queue_idx = vq->vq_queue_index;
324 	struct vring_packed *vring;
325 	uint64_t desc_addr;
326 	uint64_t avail_addr;
327 	uint64_t used_addr;
328 	uint16_t i;
329 
330 	vring  = &dev->packed_vrings[queue_idx];
331 	desc_addr = (uintptr_t)vq->vq_ring_virt_mem;
332 	avail_addr = desc_addr + vq->vq_nentries *
333 		sizeof(struct vring_packed_desc);
334 	used_addr = RTE_ALIGN_CEIL(avail_addr +
335 			   sizeof(struct vring_packed_desc_event),
336 			   VIRTIO_PCI_VRING_ALIGN);
337 	vring->num = vq->vq_nentries;
338 	vring->desc = (void *)(uintptr_t)desc_addr;
339 	vring->driver = (void *)(uintptr_t)avail_addr;
340 	vring->device = (void *)(uintptr_t)used_addr;
341 	dev->packed_queues[queue_idx].avail_wrap_counter = true;
342 	dev->packed_queues[queue_idx].used_wrap_counter = true;
343 
344 	for (i = 0; i < vring->num; i++)
345 		vring->desc[i].flags = 0;
346 }
347 
348 static void
349 virtio_user_setup_queue_split(struct virtqueue *vq, struct virtio_user_dev *dev)
350 {
351 	uint16_t queue_idx = vq->vq_queue_index;
352 	uint64_t desc_addr, avail_addr, used_addr;
353 
354 	desc_addr = (uintptr_t)vq->vq_ring_virt_mem;
355 	avail_addr = desc_addr + vq->vq_nentries * sizeof(struct vring_desc);
356 	used_addr = RTE_ALIGN_CEIL(avail_addr + offsetof(struct vring_avail,
357 							 ring[vq->vq_nentries]),
358 				   VIRTIO_PCI_VRING_ALIGN);
359 
360 	dev->vrings[queue_idx].num = vq->vq_nentries;
361 	dev->vrings[queue_idx].desc = (void *)(uintptr_t)desc_addr;
362 	dev->vrings[queue_idx].avail = (void *)(uintptr_t)avail_addr;
363 	dev->vrings[queue_idx].used = (void *)(uintptr_t)used_addr;
364 }
365 
366 static int
367 virtio_user_setup_queue(struct virtio_hw *hw, struct virtqueue *vq)
368 {
369 	struct virtio_user_dev *dev = virtio_user_get_dev(hw);
370 
371 	if (vtpci_packed_queue(hw))
372 		virtio_user_setup_queue_packed(vq, dev);
373 	else
374 		virtio_user_setup_queue_split(vq, dev);
375 
376 	return 0;
377 }
378 
379 static void
380 virtio_user_del_queue(struct virtio_hw *hw, struct virtqueue *vq)
381 {
382 	/* For legacy devices, write 0 to VIRTIO_PCI_QUEUE_PFN port, QEMU
383 	 * correspondingly stops the ioeventfds, and reset the status of
384 	 * the device.
385 	 * For modern devices, set queue desc, avail, used in PCI bar to 0,
386 	 * not see any more behavior in QEMU.
387 	 *
388 	 * Here we just care about what information to deliver to vhost-user
389 	 * or vhost-kernel. So we just close ioeventfd for now.
390 	 */
391 	struct virtio_user_dev *dev = virtio_user_get_dev(hw);
392 
393 	close(dev->callfds[vq->vq_queue_index]);
394 	close(dev->kickfds[vq->vq_queue_index]);
395 }
396 
397 static void
398 virtio_user_notify_queue(struct virtio_hw *hw, struct virtqueue *vq)
399 {
400 	uint64_t buf = 1;
401 	struct virtio_user_dev *dev = virtio_user_get_dev(hw);
402 
403 	if (hw->cvq && (hw->cvq->vq == vq)) {
404 		if (vtpci_packed_queue(vq->hw))
405 			virtio_user_handle_cq_packed(dev, vq->vq_queue_index);
406 		else
407 			virtio_user_handle_cq(dev, vq->vq_queue_index);
408 		return;
409 	}
410 
411 	if (write(dev->kickfds[vq->vq_queue_index], &buf, sizeof(buf)) < 0)
412 		PMD_DRV_LOG(ERR, "failed to kick backend: %s",
413 			    strerror(errno));
414 }
415 
416 const struct virtio_pci_ops virtio_user_ops = {
417 	.read_dev_cfg	= virtio_user_read_dev_config,
418 	.write_dev_cfg	= virtio_user_write_dev_config,
419 	.get_status	= virtio_user_get_status,
420 	.set_status	= virtio_user_set_status,
421 	.get_features	= virtio_user_get_features,
422 	.set_features	= virtio_user_set_features,
423 	.get_isr	= virtio_user_get_isr,
424 	.set_config_irq	= virtio_user_set_config_irq,
425 	.set_queue_irq	= virtio_user_set_queue_irq,
426 	.get_queue_num	= virtio_user_get_queue_num,
427 	.setup_queue	= virtio_user_setup_queue,
428 	.del_queue	= virtio_user_del_queue,
429 	.notify_queue	= virtio_user_notify_queue,
430 };
431 
432 static const char *valid_args[] = {
433 #define VIRTIO_USER_ARG_QUEUES_NUM     "queues"
434 	VIRTIO_USER_ARG_QUEUES_NUM,
435 #define VIRTIO_USER_ARG_CQ_NUM         "cq"
436 	VIRTIO_USER_ARG_CQ_NUM,
437 #define VIRTIO_USER_ARG_MAC            "mac"
438 	VIRTIO_USER_ARG_MAC,
439 #define VIRTIO_USER_ARG_PATH           "path"
440 	VIRTIO_USER_ARG_PATH,
441 #define VIRTIO_USER_ARG_QUEUE_SIZE     "queue_size"
442 	VIRTIO_USER_ARG_QUEUE_SIZE,
443 #define VIRTIO_USER_ARG_INTERFACE_NAME "iface"
444 	VIRTIO_USER_ARG_INTERFACE_NAME,
445 #define VIRTIO_USER_ARG_SERVER_MODE    "server"
446 	VIRTIO_USER_ARG_SERVER_MODE,
447 #define VIRTIO_USER_ARG_MRG_RXBUF      "mrg_rxbuf"
448 	VIRTIO_USER_ARG_MRG_RXBUF,
449 #define VIRTIO_USER_ARG_IN_ORDER       "in_order"
450 	VIRTIO_USER_ARG_IN_ORDER,
451 #define VIRTIO_USER_ARG_PACKED_VQ      "packed_vq"
452 	VIRTIO_USER_ARG_PACKED_VQ,
453 #define VIRTIO_USER_ARG_SPEED          "speed"
454 	VIRTIO_USER_ARG_SPEED,
455 #define VIRTIO_USER_ARG_VECTORIZED     "vectorized"
456 	VIRTIO_USER_ARG_VECTORIZED,
457 	NULL
458 };
459 
460 #define VIRTIO_USER_DEF_CQ_EN	0
461 #define VIRTIO_USER_DEF_Q_NUM	1
462 #define VIRTIO_USER_DEF_Q_SZ	256
463 #define VIRTIO_USER_DEF_SERVER_MODE	0
464 
465 static int
466 get_string_arg(const char *key __rte_unused,
467 	       const char *value, void *extra_args)
468 {
469 	if (!value || !extra_args)
470 		return -EINVAL;
471 
472 	*(char **)extra_args = strdup(value);
473 
474 	if (!*(char **)extra_args)
475 		return -ENOMEM;
476 
477 	return 0;
478 }
479 
480 static int
481 get_integer_arg(const char *key __rte_unused,
482 		const char *value, void *extra_args)
483 {
484 	uint64_t integer = 0;
485 	if (!value || !extra_args)
486 		return -EINVAL;
487 	errno = 0;
488 	integer = strtoull(value, NULL, 0);
489 	/* extra_args keeps default value, it should be replaced
490 	 * only in case of successful parsing of the 'value' arg
491 	 */
492 	if (errno == 0)
493 		*(uint64_t *)extra_args = integer;
494 	return -errno;
495 }
496 
497 static struct rte_eth_dev *
498 virtio_user_eth_dev_alloc(struct rte_vdev_device *vdev)
499 {
500 	struct rte_eth_dev *eth_dev;
501 	struct rte_eth_dev_data *data;
502 	struct virtio_hw *hw;
503 	struct virtio_user_dev *dev;
504 
505 	eth_dev = rte_eth_vdev_allocate(vdev, sizeof(*hw));
506 	if (!eth_dev) {
507 		PMD_INIT_LOG(ERR, "cannot alloc rte_eth_dev");
508 		return NULL;
509 	}
510 
511 	data = eth_dev->data;
512 	hw = eth_dev->data->dev_private;
513 
514 	dev = rte_zmalloc(NULL, sizeof(*dev), 0);
515 	if (!dev) {
516 		PMD_INIT_LOG(ERR, "malloc virtio_user_dev failed");
517 		rte_eth_dev_release_port(eth_dev);
518 		return NULL;
519 	}
520 
521 	hw->port_id = data->port_id;
522 	dev->port_id = data->port_id;
523 	virtio_hw_internal[hw->port_id].vtpci_ops = &virtio_user_ops;
524 	/*
525 	 * MSIX is required to enable LSC (see virtio_init_device).
526 	 * Here just pretend that we support msix.
527 	 */
528 	hw->use_msix = 1;
529 	hw->modern   = 0;
530 	hw->use_vec_rx = 0;
531 	hw->use_vec_tx = 0;
532 	hw->use_inorder_rx = 0;
533 	hw->use_inorder_tx = 0;
534 	hw->virtio_user_dev = dev;
535 	return eth_dev;
536 }
537 
538 static void
539 virtio_user_eth_dev_free(struct rte_eth_dev *eth_dev)
540 {
541 	struct rte_eth_dev_data *data = eth_dev->data;
542 	struct virtio_hw *hw = data->dev_private;
543 
544 	rte_free(hw->virtio_user_dev);
545 	rte_eth_dev_release_port(eth_dev);
546 }
547 
548 /* Dev initialization routine. Invoked once for each virtio vdev at
549  * EAL init time, see rte_bus_probe().
550  * Returns 0 on success.
551  */
552 static int
553 virtio_user_pmd_probe(struct rte_vdev_device *dev)
554 {
555 	struct rte_kvargs *kvlist = NULL;
556 	struct rte_eth_dev *eth_dev;
557 	struct virtio_hw *hw;
558 	uint64_t queues = VIRTIO_USER_DEF_Q_NUM;
559 	uint64_t cq = VIRTIO_USER_DEF_CQ_EN;
560 	uint64_t queue_size = VIRTIO_USER_DEF_Q_SZ;
561 	uint64_t server_mode = VIRTIO_USER_DEF_SERVER_MODE;
562 	uint64_t mrg_rxbuf = 1;
563 	uint64_t in_order = 1;
564 	uint64_t packed_vq = 0;
565 	uint64_t vectorized = 0;
566 	char *path = NULL;
567 	char *ifname = NULL;
568 	char *mac_addr = NULL;
569 	int ret = -1;
570 
571 	if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
572 		const char *name = rte_vdev_device_name(dev);
573 		eth_dev = rte_eth_dev_attach_secondary(name);
574 		if (!eth_dev) {
575 			PMD_INIT_LOG(ERR, "Failed to probe %s", name);
576 			return -1;
577 		}
578 
579 		if (eth_virtio_dev_init(eth_dev) < 0) {
580 			PMD_INIT_LOG(ERR, "eth_virtio_dev_init fails");
581 			rte_eth_dev_release_port(eth_dev);
582 			return -1;
583 		}
584 
585 		eth_dev->dev_ops = &virtio_user_secondary_eth_dev_ops;
586 		eth_dev->device = &dev->device;
587 		rte_eth_dev_probing_finish(eth_dev);
588 		return 0;
589 	}
590 
591 	kvlist = rte_kvargs_parse(rte_vdev_device_args(dev), valid_args);
592 	if (!kvlist) {
593 		PMD_INIT_LOG(ERR, "error when parsing param");
594 		goto end;
595 	}
596 
597 	if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_PATH) == 1) {
598 		if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_PATH,
599 				       &get_string_arg, &path) < 0) {
600 			PMD_INIT_LOG(ERR, "error to parse %s",
601 				     VIRTIO_USER_ARG_PATH);
602 			goto end;
603 		}
604 	} else {
605 		PMD_INIT_LOG(ERR, "arg %s is mandatory for virtio_user",
606 			     VIRTIO_USER_ARG_PATH);
607 		goto end;
608 	}
609 
610 	if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_INTERFACE_NAME) == 1) {
611 		if (is_vhost_user_by_type(path)) {
612 			PMD_INIT_LOG(ERR,
613 				"arg %s applies only to vhost-kernel backend",
614 				VIRTIO_USER_ARG_INTERFACE_NAME);
615 			goto end;
616 		}
617 
618 		if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_INTERFACE_NAME,
619 				       &get_string_arg, &ifname) < 0) {
620 			PMD_INIT_LOG(ERR, "error to parse %s",
621 				     VIRTIO_USER_ARG_INTERFACE_NAME);
622 			goto end;
623 		}
624 	}
625 
626 	if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_MAC) == 1) {
627 		if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_MAC,
628 				       &get_string_arg, &mac_addr) < 0) {
629 			PMD_INIT_LOG(ERR, "error to parse %s",
630 				     VIRTIO_USER_ARG_MAC);
631 			goto end;
632 		}
633 	}
634 
635 	if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_QUEUE_SIZE) == 1) {
636 		if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_QUEUE_SIZE,
637 				       &get_integer_arg, &queue_size) < 0) {
638 			PMD_INIT_LOG(ERR, "error to parse %s",
639 				     VIRTIO_USER_ARG_QUEUE_SIZE);
640 			goto end;
641 		}
642 	}
643 
644 	if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_QUEUES_NUM) == 1) {
645 		if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_QUEUES_NUM,
646 				       &get_integer_arg, &queues) < 0) {
647 			PMD_INIT_LOG(ERR, "error to parse %s",
648 				     VIRTIO_USER_ARG_QUEUES_NUM);
649 			goto end;
650 		}
651 	}
652 
653 	if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_SERVER_MODE) == 1) {
654 		if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_SERVER_MODE,
655 				       &get_integer_arg, &server_mode) < 0) {
656 			PMD_INIT_LOG(ERR, "error to parse %s",
657 				     VIRTIO_USER_ARG_SERVER_MODE);
658 			goto end;
659 		}
660 	}
661 
662 	if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_CQ_NUM) == 1) {
663 		if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_CQ_NUM,
664 				       &get_integer_arg, &cq) < 0) {
665 			PMD_INIT_LOG(ERR, "error to parse %s",
666 				     VIRTIO_USER_ARG_CQ_NUM);
667 			goto end;
668 		}
669 	} else if (queues > 1) {
670 		cq = 1;
671 	}
672 
673 	if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_PACKED_VQ) == 1) {
674 		if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_PACKED_VQ,
675 				       &get_integer_arg, &packed_vq) < 0) {
676 			PMD_INIT_LOG(ERR, "error to parse %s",
677 				     VIRTIO_USER_ARG_PACKED_VQ);
678 			goto end;
679 		}
680 	}
681 
682 	if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_VECTORIZED) == 1) {
683 		if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_VECTORIZED,
684 				       &get_integer_arg, &vectorized) < 0) {
685 			PMD_INIT_LOG(ERR, "error to parse %s",
686 				     VIRTIO_USER_ARG_VECTORIZED);
687 			goto end;
688 		}
689 	}
690 
691 	if (queues > 1 && cq == 0) {
692 		PMD_INIT_LOG(ERR, "multi-q requires ctrl-q");
693 		goto end;
694 	}
695 
696 	if (queues > VIRTIO_MAX_VIRTQUEUE_PAIRS) {
697 		PMD_INIT_LOG(ERR, "arg %s %" PRIu64 " exceeds the limit %u",
698 			VIRTIO_USER_ARG_QUEUES_NUM, queues,
699 			VIRTIO_MAX_VIRTQUEUE_PAIRS);
700 		goto end;
701 	}
702 
703 	if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_MRG_RXBUF) == 1) {
704 		if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_MRG_RXBUF,
705 				       &get_integer_arg, &mrg_rxbuf) < 0) {
706 			PMD_INIT_LOG(ERR, "error to parse %s",
707 				     VIRTIO_USER_ARG_MRG_RXBUF);
708 			goto end;
709 		}
710 	}
711 
712 	if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_IN_ORDER) == 1) {
713 		if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_IN_ORDER,
714 				       &get_integer_arg, &in_order) < 0) {
715 			PMD_INIT_LOG(ERR, "error to parse %s",
716 				     VIRTIO_USER_ARG_IN_ORDER);
717 			goto end;
718 		}
719 	}
720 
721 	eth_dev = virtio_user_eth_dev_alloc(dev);
722 	if (!eth_dev) {
723 		PMD_INIT_LOG(ERR, "virtio_user fails to alloc device");
724 		goto end;
725 	}
726 
727 	hw = eth_dev->data->dev_private;
728 	if (virtio_user_dev_init(hw->virtio_user_dev, path, queues, cq,
729 			 queue_size, mac_addr, &ifname, server_mode,
730 			 mrg_rxbuf, in_order, packed_vq) < 0) {
731 		PMD_INIT_LOG(ERR, "virtio_user_dev_init fails");
732 		virtio_user_eth_dev_free(eth_dev);
733 		goto end;
734 	}
735 
736 	/* previously called by pci probing for physical dev */
737 	if (eth_virtio_dev_init(eth_dev) < 0) {
738 		PMD_INIT_LOG(ERR, "eth_virtio_dev_init fails");
739 		virtio_user_eth_dev_free(eth_dev);
740 		goto end;
741 	}
742 
743 	if (vectorized) {
744 		if (packed_vq) {
745 #if defined(CC_AVX512_SUPPORT)
746 			hw->use_vec_rx = 1;
747 			hw->use_vec_tx = 1;
748 #else
749 			PMD_INIT_LOG(INFO,
750 				"building environment do not support packed ring vectorized");
751 #endif
752 		} else {
753 			hw->use_vec_rx = 1;
754 		}
755 	}
756 
757 	rte_eth_dev_probing_finish(eth_dev);
758 	ret = 0;
759 
760 end:
761 	if (kvlist)
762 		rte_kvargs_free(kvlist);
763 	if (path)
764 		free(path);
765 	if (mac_addr)
766 		free(mac_addr);
767 	if (ifname)
768 		free(ifname);
769 	return ret;
770 }
771 
772 static int
773 virtio_user_pmd_remove(struct rte_vdev_device *vdev)
774 {
775 	const char *name;
776 	struct rte_eth_dev *eth_dev;
777 
778 	if (!vdev)
779 		return -EINVAL;
780 
781 	name = rte_vdev_device_name(vdev);
782 	PMD_DRV_LOG(INFO, "Un-Initializing %s", name);
783 	eth_dev = rte_eth_dev_allocated(name);
784 	/* Port has already been released by close. */
785 	if (!eth_dev)
786 		return 0;
787 
788 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
789 		return rte_eth_dev_release_port(eth_dev);
790 
791 	/* make sure the device is stopped, queues freed */
792 	rte_eth_dev_close(eth_dev->data->port_id);
793 
794 	return 0;
795 }
796 
797 static struct rte_vdev_driver virtio_user_driver = {
798 	.probe = virtio_user_pmd_probe,
799 	.remove = virtio_user_pmd_remove,
800 };
801 
802 RTE_PMD_REGISTER_VDEV(net_virtio_user, virtio_user_driver);
803 RTE_PMD_REGISTER_ALIAS(net_virtio_user, virtio_user);
804 RTE_PMD_REGISTER_PARAM_STRING(net_virtio_user,
805 	"path=<path> "
806 	"mac=<mac addr> "
807 	"cq=<int> "
808 	"queue_size=<int> "
809 	"queues=<int> "
810 	"iface=<string> "
811 	"server=<0|1> "
812 	"mrg_rxbuf=<0|1> "
813 	"in_order=<0|1> "
814 	"packed_vq=<0|1> "
815 	"speed=<int> "
816 	"vectorized=<0|1>");
817