xref: /dpdk/drivers/net/virtio/virtio_user_ethdev.c (revision 6a42e7ef9dd86a554e060fcef0be03aa9ddd394c)
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 	ret = virtio_user_start_device(dev);
98 	if (ret < 0)
99 		return -1;
100 
101 	if (dev->queue_pairs > 1) {
102 		ret = virtio_user_handle_mq(dev, dev->queue_pairs);
103 		if (ret != 0) {
104 			PMD_INIT_LOG(ERR, "Fails to enable multi-queue pairs!");
105 			return -1;
106 		}
107 	}
108 	if (eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC) {
109 		if (rte_intr_disable(eth_dev->intr_handle) < 0) {
110 			PMD_DRV_LOG(ERR, "interrupt disable failed");
111 			return -1;
112 		}
113 		rte_intr_callback_unregister(eth_dev->intr_handle,
114 					     virtio_interrupt_handler,
115 					     eth_dev);
116 		eth_dev->intr_handle->fd = connectfd;
117 		rte_intr_callback_register(eth_dev->intr_handle,
118 					   virtio_interrupt_handler, eth_dev);
119 
120 		if (rte_intr_enable(eth_dev->intr_handle) < 0) {
121 			PMD_DRV_LOG(ERR, "interrupt enable failed");
122 			return -1;
123 		}
124 	}
125 	PMD_INIT_LOG(NOTICE, "server mode virtio-user reconnection succeeds!");
126 	return 0;
127 }
128 
129 static void
130 virtio_user_delayed_handler(void *param)
131 {
132 	struct virtio_hw *hw = (struct virtio_hw *)param;
133 	struct rte_eth_dev *eth_dev = &rte_eth_devices[hw->port_id];
134 	struct virtio_user_dev *dev = virtio_user_get_dev(hw);
135 
136 	if (rte_intr_disable(eth_dev->intr_handle) < 0) {
137 		PMD_DRV_LOG(ERR, "interrupt disable failed");
138 		return;
139 	}
140 	rte_intr_callback_unregister(eth_dev->intr_handle,
141 				     virtio_interrupt_handler, eth_dev);
142 	if (dev->is_server) {
143 		if (dev->vhostfd >= 0) {
144 			close(dev->vhostfd);
145 			dev->vhostfd = -1;
146 		}
147 		eth_dev->intr_handle->fd = dev->listenfd;
148 		rte_intr_callback_register(eth_dev->intr_handle,
149 					   virtio_interrupt_handler, eth_dev);
150 		if (rte_intr_enable(eth_dev->intr_handle) < 0) {
151 			PMD_DRV_LOG(ERR, "interrupt enable failed");
152 			return;
153 		}
154 	}
155 }
156 
157 static void
158 virtio_user_read_dev_config(struct virtio_hw *hw, size_t offset,
159 		     void *dst, int length)
160 {
161 	int i;
162 	struct virtio_user_dev *dev = virtio_user_get_dev(hw);
163 
164 	if (offset == offsetof(struct virtio_net_config, mac) &&
165 	    length == RTE_ETHER_ADDR_LEN) {
166 		for (i = 0; i < RTE_ETHER_ADDR_LEN; ++i)
167 			((uint8_t *)dst)[i] = dev->mac_addr[i];
168 		return;
169 	}
170 
171 	if (offset == offsetof(struct virtio_net_config, status)) {
172 		char buf[128];
173 
174 		if (dev->vhostfd >= 0) {
175 			int r;
176 			int flags;
177 
178 			flags = fcntl(dev->vhostfd, F_GETFL);
179 			if (fcntl(dev->vhostfd, F_SETFL,
180 					flags | O_NONBLOCK) == -1) {
181 				PMD_DRV_LOG(ERR, "error setting O_NONBLOCK flag");
182 				return;
183 			}
184 			r = recv(dev->vhostfd, buf, 128, MSG_PEEK);
185 			if (r == 0 || (r < 0 && errno != EAGAIN)) {
186 				dev->status &= (~VIRTIO_NET_S_LINK_UP);
187 				PMD_DRV_LOG(ERR, "virtio-user port %u is down",
188 					    hw->port_id);
189 
190 				/* This function could be called in the process
191 				 * of interrupt handling, callback cannot be
192 				 * unregistered here, set an alarm to do it.
193 				 */
194 				rte_eal_alarm_set(1,
195 						  virtio_user_delayed_handler,
196 						  (void *)hw);
197 			} else {
198 				dev->status |= VIRTIO_NET_S_LINK_UP;
199 			}
200 			if (fcntl(dev->vhostfd, F_SETFL,
201 					flags & ~O_NONBLOCK) == -1) {
202 				PMD_DRV_LOG(ERR, "error clearing O_NONBLOCK flag");
203 				return;
204 			}
205 		} else if (dev->is_server) {
206 			dev->status &= (~VIRTIO_NET_S_LINK_UP);
207 			if (virtio_user_server_reconnect(dev) >= 0)
208 				dev->status |= VIRTIO_NET_S_LINK_UP;
209 		}
210 
211 		*(uint16_t *)dst = dev->status;
212 	}
213 
214 	if (offset == offsetof(struct virtio_net_config, max_virtqueue_pairs))
215 		*(uint16_t *)dst = dev->max_queue_pairs;
216 }
217 
218 static void
219 virtio_user_write_dev_config(struct virtio_hw *hw, size_t offset,
220 		      const void *src, int length)
221 {
222 	int i;
223 	struct virtio_user_dev *dev = virtio_user_get_dev(hw);
224 
225 	if ((offset == offsetof(struct virtio_net_config, mac)) &&
226 	    (length == RTE_ETHER_ADDR_LEN))
227 		for (i = 0; i < RTE_ETHER_ADDR_LEN; ++i)
228 			dev->mac_addr[i] = ((const uint8_t *)src)[i];
229 	else
230 		PMD_DRV_LOG(ERR, "not supported offset=%zu, len=%d",
231 			    offset, length);
232 }
233 
234 static void
235 virtio_user_reset(struct virtio_hw *hw)
236 {
237 	struct virtio_user_dev *dev = virtio_user_get_dev(hw);
238 
239 	if (dev->status & VIRTIO_CONFIG_STATUS_DRIVER_OK)
240 		virtio_user_stop_device(dev);
241 }
242 
243 static void
244 virtio_user_set_status(struct virtio_hw *hw, uint8_t status)
245 {
246 	struct virtio_user_dev *dev = virtio_user_get_dev(hw);
247 
248 	if (status & VIRTIO_CONFIG_STATUS_DRIVER_OK)
249 		virtio_user_start_device(dev);
250 	else if (status == VIRTIO_CONFIG_STATUS_RESET)
251 		virtio_user_reset(hw);
252 	dev->status = status;
253 }
254 
255 static uint8_t
256 virtio_user_get_status(struct virtio_hw *hw)
257 {
258 	struct virtio_user_dev *dev = virtio_user_get_dev(hw);
259 
260 	return dev->status;
261 }
262 
263 static uint64_t
264 virtio_user_get_features(struct virtio_hw *hw)
265 {
266 	struct virtio_user_dev *dev = virtio_user_get_dev(hw);
267 
268 	/* unmask feature bits defined in vhost user protocol */
269 	return dev->device_features & VIRTIO_PMD_SUPPORTED_GUEST_FEATURES;
270 }
271 
272 static void
273 virtio_user_set_features(struct virtio_hw *hw, uint64_t features)
274 {
275 	struct virtio_user_dev *dev = virtio_user_get_dev(hw);
276 
277 	dev->features = features & dev->device_features;
278 }
279 
280 static uint8_t
281 virtio_user_get_isr(struct virtio_hw *hw __rte_unused)
282 {
283 	/* rxq interrupts and config interrupt are separated in virtio-user,
284 	 * here we only report config change.
285 	 */
286 	return VIRTIO_PCI_ISR_CONFIG;
287 }
288 
289 static uint16_t
290 virtio_user_set_config_irq(struct virtio_hw *hw __rte_unused,
291 		    uint16_t vec __rte_unused)
292 {
293 	return 0;
294 }
295 
296 static uint16_t
297 virtio_user_set_queue_irq(struct virtio_hw *hw __rte_unused,
298 			  struct virtqueue *vq __rte_unused,
299 			  uint16_t vec)
300 {
301 	/* pretend we have done that */
302 	return vec;
303 }
304 
305 /* This function is to get the queue size, aka, number of descs, of a specified
306  * queue. Different with the VHOST_USER_GET_QUEUE_NUM, which is used to get the
307  * max supported queues.
308  */
309 static uint16_t
310 virtio_user_get_queue_num(struct virtio_hw *hw, uint16_t queue_id __rte_unused)
311 {
312 	struct virtio_user_dev *dev = virtio_user_get_dev(hw);
313 
314 	/* Currently, each queue has same queue size */
315 	return dev->queue_size;
316 }
317 
318 static void
319 virtio_user_setup_queue_packed(struct virtqueue *vq,
320 			       struct virtio_user_dev *dev)
321 {
322 	uint16_t queue_idx = vq->vq_queue_index;
323 	struct vring_packed *vring;
324 	uint64_t desc_addr;
325 	uint64_t avail_addr;
326 	uint64_t used_addr;
327 	uint16_t i;
328 
329 	vring  = &dev->packed_vrings[queue_idx];
330 	desc_addr = (uintptr_t)vq->vq_ring_virt_mem;
331 	avail_addr = desc_addr + vq->vq_nentries *
332 		sizeof(struct vring_packed_desc);
333 	used_addr = RTE_ALIGN_CEIL(avail_addr +
334 			   sizeof(struct vring_packed_desc_event),
335 			   VIRTIO_PCI_VRING_ALIGN);
336 	vring->num = vq->vq_nentries;
337 	vring->desc = (void *)(uintptr_t)desc_addr;
338 	vring->driver = (void *)(uintptr_t)avail_addr;
339 	vring->device = (void *)(uintptr_t)used_addr;
340 	dev->packed_queues[queue_idx].avail_wrap_counter = true;
341 	dev->packed_queues[queue_idx].used_wrap_counter = true;
342 
343 	for (i = 0; i < vring->num; i++)
344 		vring->desc[i].flags = 0;
345 }
346 
347 static void
348 virtio_user_setup_queue_split(struct virtqueue *vq, struct virtio_user_dev *dev)
349 {
350 	uint16_t queue_idx = vq->vq_queue_index;
351 	uint64_t desc_addr, avail_addr, used_addr;
352 
353 	desc_addr = (uintptr_t)vq->vq_ring_virt_mem;
354 	avail_addr = desc_addr + vq->vq_nentries * sizeof(struct vring_desc);
355 	used_addr = RTE_ALIGN_CEIL(avail_addr + offsetof(struct vring_avail,
356 							 ring[vq->vq_nentries]),
357 				   VIRTIO_PCI_VRING_ALIGN);
358 
359 	dev->vrings[queue_idx].num = vq->vq_nentries;
360 	dev->vrings[queue_idx].desc = (void *)(uintptr_t)desc_addr;
361 	dev->vrings[queue_idx].avail = (void *)(uintptr_t)avail_addr;
362 	dev->vrings[queue_idx].used = (void *)(uintptr_t)used_addr;
363 }
364 
365 static int
366 virtio_user_setup_queue(struct virtio_hw *hw, struct virtqueue *vq)
367 {
368 	struct virtio_user_dev *dev = virtio_user_get_dev(hw);
369 
370 	if (vtpci_packed_queue(hw))
371 		virtio_user_setup_queue_packed(vq, dev);
372 	else
373 		virtio_user_setup_queue_split(vq, dev);
374 
375 	return 0;
376 }
377 
378 static void
379 virtio_user_del_queue(struct virtio_hw *hw, struct virtqueue *vq)
380 {
381 	/* For legacy devices, write 0 to VIRTIO_PCI_QUEUE_PFN port, QEMU
382 	 * correspondingly stops the ioeventfds, and reset the status of
383 	 * the device.
384 	 * For modern devices, set queue desc, avail, used in PCI bar to 0,
385 	 * not see any more behavior in QEMU.
386 	 *
387 	 * Here we just care about what information to deliver to vhost-user
388 	 * or vhost-kernel. So we just close ioeventfd for now.
389 	 */
390 	struct virtio_user_dev *dev = virtio_user_get_dev(hw);
391 
392 	close(dev->callfds[vq->vq_queue_index]);
393 	close(dev->kickfds[vq->vq_queue_index]);
394 }
395 
396 static void
397 virtio_user_notify_queue(struct virtio_hw *hw, struct virtqueue *vq)
398 {
399 	uint64_t buf = 1;
400 	struct virtio_user_dev *dev = virtio_user_get_dev(hw);
401 
402 	if (hw->cvq && (hw->cvq->vq == vq)) {
403 		if (vtpci_packed_queue(vq->hw))
404 			virtio_user_handle_cq_packed(dev, vq->vq_queue_index);
405 		else
406 			virtio_user_handle_cq(dev, vq->vq_queue_index);
407 		return;
408 	}
409 
410 	if (write(dev->kickfds[vq->vq_queue_index], &buf, sizeof(buf)) < 0)
411 		PMD_DRV_LOG(ERR, "failed to kick backend: %s",
412 			    strerror(errno));
413 }
414 
415 const struct virtio_pci_ops virtio_user_ops = {
416 	.read_dev_cfg	= virtio_user_read_dev_config,
417 	.write_dev_cfg	= virtio_user_write_dev_config,
418 	.get_status	= virtio_user_get_status,
419 	.set_status	= virtio_user_set_status,
420 	.get_features	= virtio_user_get_features,
421 	.set_features	= virtio_user_set_features,
422 	.get_isr	= virtio_user_get_isr,
423 	.set_config_irq	= virtio_user_set_config_irq,
424 	.set_queue_irq	= virtio_user_set_queue_irq,
425 	.get_queue_num	= virtio_user_get_queue_num,
426 	.setup_queue	= virtio_user_setup_queue,
427 	.del_queue	= virtio_user_del_queue,
428 	.notify_queue	= virtio_user_notify_queue,
429 };
430 
431 static const char *valid_args[] = {
432 #define VIRTIO_USER_ARG_QUEUES_NUM     "queues"
433 	VIRTIO_USER_ARG_QUEUES_NUM,
434 #define VIRTIO_USER_ARG_CQ_NUM         "cq"
435 	VIRTIO_USER_ARG_CQ_NUM,
436 #define VIRTIO_USER_ARG_MAC            "mac"
437 	VIRTIO_USER_ARG_MAC,
438 #define VIRTIO_USER_ARG_PATH           "path"
439 	VIRTIO_USER_ARG_PATH,
440 #define VIRTIO_USER_ARG_QUEUE_SIZE     "queue_size"
441 	VIRTIO_USER_ARG_QUEUE_SIZE,
442 #define VIRTIO_USER_ARG_INTERFACE_NAME "iface"
443 	VIRTIO_USER_ARG_INTERFACE_NAME,
444 #define VIRTIO_USER_ARG_SERVER_MODE    "server"
445 	VIRTIO_USER_ARG_SERVER_MODE,
446 #define VIRTIO_USER_ARG_MRG_RXBUF      "mrg_rxbuf"
447 	VIRTIO_USER_ARG_MRG_RXBUF,
448 #define VIRTIO_USER_ARG_IN_ORDER       "in_order"
449 	VIRTIO_USER_ARG_IN_ORDER,
450 #define VIRTIO_USER_ARG_PACKED_VQ      "packed_vq"
451 	VIRTIO_USER_ARG_PACKED_VQ,
452 	NULL
453 };
454 
455 #define VIRTIO_USER_DEF_CQ_EN	0
456 #define VIRTIO_USER_DEF_Q_NUM	1
457 #define VIRTIO_USER_DEF_Q_SZ	256
458 #define VIRTIO_USER_DEF_SERVER_MODE	0
459 
460 static int
461 get_string_arg(const char *key __rte_unused,
462 	       const char *value, void *extra_args)
463 {
464 	if (!value || !extra_args)
465 		return -EINVAL;
466 
467 	*(char **)extra_args = strdup(value);
468 
469 	if (!*(char **)extra_args)
470 		return -ENOMEM;
471 
472 	return 0;
473 }
474 
475 static int
476 get_integer_arg(const char *key __rte_unused,
477 		const char *value, void *extra_args)
478 {
479 	if (!value || !extra_args)
480 		return -EINVAL;
481 
482 	*(uint64_t *)extra_args = strtoull(value, NULL, 0);
483 
484 	return 0;
485 }
486 
487 static struct rte_eth_dev *
488 virtio_user_eth_dev_alloc(struct rte_vdev_device *vdev)
489 {
490 	struct rte_eth_dev *eth_dev;
491 	struct rte_eth_dev_data *data;
492 	struct virtio_hw *hw;
493 	struct virtio_user_dev *dev;
494 
495 	eth_dev = rte_eth_vdev_allocate(vdev, sizeof(*hw));
496 	if (!eth_dev) {
497 		PMD_INIT_LOG(ERR, "cannot alloc rte_eth_dev");
498 		return NULL;
499 	}
500 
501 	data = eth_dev->data;
502 	hw = eth_dev->data->dev_private;
503 
504 	dev = rte_zmalloc(NULL, sizeof(*dev), 0);
505 	if (!dev) {
506 		PMD_INIT_LOG(ERR, "malloc virtio_user_dev failed");
507 		rte_eth_dev_release_port(eth_dev);
508 		return NULL;
509 	}
510 
511 	hw->port_id = data->port_id;
512 	dev->port_id = data->port_id;
513 	virtio_hw_internal[hw->port_id].vtpci_ops = &virtio_user_ops;
514 	/*
515 	 * MSIX is required to enable LSC (see virtio_init_device).
516 	 * Here just pretend that we support msix.
517 	 */
518 	hw->use_msix = 1;
519 	hw->modern   = 0;
520 	hw->use_simple_rx = 0;
521 	hw->use_inorder_rx = 0;
522 	hw->use_inorder_tx = 0;
523 	hw->virtio_user_dev = dev;
524 	return eth_dev;
525 }
526 
527 static void
528 virtio_user_eth_dev_free(struct rte_eth_dev *eth_dev)
529 {
530 	struct rte_eth_dev_data *data = eth_dev->data;
531 	struct virtio_hw *hw = data->dev_private;
532 
533 	rte_free(hw->virtio_user_dev);
534 	rte_eth_dev_release_port(eth_dev);
535 }
536 
537 /* Dev initialization routine. Invoked once for each virtio vdev at
538  * EAL init time, see rte_bus_probe().
539  * Returns 0 on success.
540  */
541 static int
542 virtio_user_pmd_probe(struct rte_vdev_device *dev)
543 {
544 	struct rte_kvargs *kvlist = NULL;
545 	struct rte_eth_dev *eth_dev;
546 	struct virtio_hw *hw;
547 	uint64_t queues = VIRTIO_USER_DEF_Q_NUM;
548 	uint64_t cq = VIRTIO_USER_DEF_CQ_EN;
549 	uint64_t queue_size = VIRTIO_USER_DEF_Q_SZ;
550 	uint64_t server_mode = VIRTIO_USER_DEF_SERVER_MODE;
551 	uint64_t mrg_rxbuf = 1;
552 	uint64_t in_order = 1;
553 	uint64_t packed_vq = 0;
554 	char *path = NULL;
555 	char *ifname = NULL;
556 	char *mac_addr = NULL;
557 	int ret = -1;
558 
559 	if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
560 		const char *name = rte_vdev_device_name(dev);
561 		eth_dev = rte_eth_dev_attach_secondary(name);
562 		if (!eth_dev) {
563 			RTE_LOG(ERR, PMD, "Failed to probe %s\n", name);
564 			return -1;
565 		}
566 
567 		if (eth_virtio_dev_init(eth_dev) < 0) {
568 			PMD_INIT_LOG(ERR, "eth_virtio_dev_init fails");
569 			rte_eth_dev_release_port(eth_dev);
570 			return -1;
571 		}
572 
573 		eth_dev->dev_ops = &virtio_user_secondary_eth_dev_ops;
574 		eth_dev->device = &dev->device;
575 		rte_eth_dev_probing_finish(eth_dev);
576 		return 0;
577 	}
578 
579 	kvlist = rte_kvargs_parse(rte_vdev_device_args(dev), valid_args);
580 	if (!kvlist) {
581 		PMD_INIT_LOG(ERR, "error when parsing param");
582 		goto end;
583 	}
584 
585 	if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_PATH) == 1) {
586 		if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_PATH,
587 				       &get_string_arg, &path) < 0) {
588 			PMD_INIT_LOG(ERR, "error to parse %s",
589 				     VIRTIO_USER_ARG_PATH);
590 			goto end;
591 		}
592 	} else {
593 		PMD_INIT_LOG(ERR, "arg %s is mandatory for virtio_user",
594 			     VIRTIO_USER_ARG_PATH);
595 		goto end;
596 	}
597 
598 	if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_INTERFACE_NAME) == 1) {
599 		if (is_vhost_user_by_type(path)) {
600 			PMD_INIT_LOG(ERR,
601 				"arg %s applies only to vhost-kernel backend",
602 				VIRTIO_USER_ARG_INTERFACE_NAME);
603 			goto end;
604 		}
605 
606 		if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_INTERFACE_NAME,
607 				       &get_string_arg, &ifname) < 0) {
608 			PMD_INIT_LOG(ERR, "error to parse %s",
609 				     VIRTIO_USER_ARG_INTERFACE_NAME);
610 			goto end;
611 		}
612 	}
613 
614 	if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_MAC) == 1) {
615 		if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_MAC,
616 				       &get_string_arg, &mac_addr) < 0) {
617 			PMD_INIT_LOG(ERR, "error to parse %s",
618 				     VIRTIO_USER_ARG_MAC);
619 			goto end;
620 		}
621 	}
622 
623 	if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_QUEUE_SIZE) == 1) {
624 		if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_QUEUE_SIZE,
625 				       &get_integer_arg, &queue_size) < 0) {
626 			PMD_INIT_LOG(ERR, "error to parse %s",
627 				     VIRTIO_USER_ARG_QUEUE_SIZE);
628 			goto end;
629 		}
630 	}
631 
632 	if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_QUEUES_NUM) == 1) {
633 		if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_QUEUES_NUM,
634 				       &get_integer_arg, &queues) < 0) {
635 			PMD_INIT_LOG(ERR, "error to parse %s",
636 				     VIRTIO_USER_ARG_QUEUES_NUM);
637 			goto end;
638 		}
639 	}
640 
641 	if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_SERVER_MODE) == 1) {
642 		if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_SERVER_MODE,
643 				       &get_integer_arg, &server_mode) < 0) {
644 			PMD_INIT_LOG(ERR, "error to parse %s",
645 				     VIRTIO_USER_ARG_SERVER_MODE);
646 			goto end;
647 		}
648 	}
649 
650 	if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_CQ_NUM) == 1) {
651 		if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_CQ_NUM,
652 				       &get_integer_arg, &cq) < 0) {
653 			PMD_INIT_LOG(ERR, "error to parse %s",
654 				     VIRTIO_USER_ARG_CQ_NUM);
655 			goto end;
656 		}
657 	} else if (queues > 1) {
658 		cq = 1;
659 	}
660 
661 	if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_PACKED_VQ) == 1) {
662 		if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_PACKED_VQ,
663 				       &get_integer_arg, &packed_vq) < 0) {
664 			PMD_INIT_LOG(ERR, "error to parse %s",
665 				     VIRTIO_USER_ARG_PACKED_VQ);
666 			goto end;
667 		}
668 	}
669 
670 	if (queues > 1 && cq == 0) {
671 		PMD_INIT_LOG(ERR, "multi-q requires ctrl-q");
672 		goto end;
673 	}
674 
675 	if (queues > VIRTIO_MAX_VIRTQUEUE_PAIRS) {
676 		PMD_INIT_LOG(ERR, "arg %s %" PRIu64 " exceeds the limit %u",
677 			VIRTIO_USER_ARG_QUEUES_NUM, queues,
678 			VIRTIO_MAX_VIRTQUEUE_PAIRS);
679 		goto end;
680 	}
681 
682 	if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_MRG_RXBUF) == 1) {
683 		if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_MRG_RXBUF,
684 				       &get_integer_arg, &mrg_rxbuf) < 0) {
685 			PMD_INIT_LOG(ERR, "error to parse %s",
686 				     VIRTIO_USER_ARG_MRG_RXBUF);
687 			goto end;
688 		}
689 	}
690 
691 	if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_IN_ORDER) == 1) {
692 		if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_IN_ORDER,
693 				       &get_integer_arg, &in_order) < 0) {
694 			PMD_INIT_LOG(ERR, "error to parse %s",
695 				     VIRTIO_USER_ARG_IN_ORDER);
696 			goto end;
697 		}
698 	}
699 
700 	eth_dev = virtio_user_eth_dev_alloc(dev);
701 	if (!eth_dev) {
702 		PMD_INIT_LOG(ERR, "virtio_user fails to alloc device");
703 		goto end;
704 	}
705 
706 	hw = eth_dev->data->dev_private;
707 	if (virtio_user_dev_init(hw->virtio_user_dev, path, queues, cq,
708 			 queue_size, mac_addr, &ifname, server_mode,
709 			 mrg_rxbuf, in_order, packed_vq) < 0) {
710 		PMD_INIT_LOG(ERR, "virtio_user_dev_init fails");
711 		virtio_user_eth_dev_free(eth_dev);
712 		goto end;
713 	}
714 
715 	/* previously called by rte_pci_probe() for physical dev */
716 	if (eth_virtio_dev_init(eth_dev) < 0) {
717 		PMD_INIT_LOG(ERR, "eth_virtio_dev_init fails");
718 		virtio_user_eth_dev_free(eth_dev);
719 		goto end;
720 	}
721 
722 	rte_eth_dev_probing_finish(eth_dev);
723 	ret = 0;
724 
725 end:
726 	if (kvlist)
727 		rte_kvargs_free(kvlist);
728 	if (path)
729 		free(path);
730 	if (mac_addr)
731 		free(mac_addr);
732 	if (ifname)
733 		free(ifname);
734 	return ret;
735 }
736 
737 static int
738 virtio_user_pmd_remove(struct rte_vdev_device *vdev)
739 {
740 	const char *name;
741 	struct rte_eth_dev *eth_dev;
742 
743 	if (!vdev)
744 		return -EINVAL;
745 
746 	name = rte_vdev_device_name(vdev);
747 	PMD_DRV_LOG(INFO, "Un-Initializing %s", name);
748 	eth_dev = rte_eth_dev_allocated(name);
749 	/* Port has already been released by close. */
750 	if (!eth_dev)
751 		return 0;
752 
753 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
754 		return rte_eth_dev_release_port(eth_dev);
755 
756 	/* make sure the device is stopped, queues freed */
757 	rte_eth_dev_close(eth_dev->data->port_id);
758 
759 	return 0;
760 }
761 
762 static struct rte_vdev_driver virtio_user_driver = {
763 	.probe = virtio_user_pmd_probe,
764 	.remove = virtio_user_pmd_remove,
765 };
766 
767 RTE_PMD_REGISTER_VDEV(net_virtio_user, virtio_user_driver);
768 RTE_PMD_REGISTER_ALIAS(net_virtio_user, virtio_user);
769 RTE_PMD_REGISTER_PARAM_STRING(net_virtio_user,
770 	"path=<path> "
771 	"mac=<mac addr> "
772 	"cq=<int> "
773 	"queue_size=<int> "
774 	"queues=<int> "
775 	"iface=<string> "
776 	"server=<0|1> "
777 	"mrg_rxbuf=<0|1> "
778 	"in_order=<0|1> "
779 	"packed_vq=<0|1>");
780