xref: /dpdk/drivers/net/virtio/virtio_user_ethdev.c (revision 3998e2a07220844d3f3c17f76a781ced3efe0de0)
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 
17 #include "virtio_ethdev.h"
18 #include "virtio_logs.h"
19 #include "virtio_pci.h"
20 #include "virtqueue.h"
21 #include "virtio_rxtx.h"
22 #include "virtio_user/virtio_user_dev.h"
23 
24 #define virtio_user_get_dev(hw) \
25 	((struct virtio_user_dev *)(hw)->virtio_user_dev)
26 
27 static void
28 virtio_user_delayed_handler(void *param)
29 {
30 	struct virtio_hw *hw = (struct virtio_hw *)param;
31 	struct rte_eth_dev *dev = &rte_eth_devices[hw->port_id];
32 
33 	rte_intr_callback_unregister(dev->intr_handle,
34 				     virtio_interrupt_handler,
35 				     dev);
36 }
37 
38 static void
39 virtio_user_read_dev_config(struct virtio_hw *hw, size_t offset,
40 		     void *dst, int length)
41 {
42 	int i;
43 	struct virtio_user_dev *dev = virtio_user_get_dev(hw);
44 
45 	if (offset == offsetof(struct virtio_net_config, mac) &&
46 	    length == ETHER_ADDR_LEN) {
47 		for (i = 0; i < ETHER_ADDR_LEN; ++i)
48 			((uint8_t *)dst)[i] = dev->mac_addr[i];
49 		return;
50 	}
51 
52 	if (offset == offsetof(struct virtio_net_config, status)) {
53 		char buf[128];
54 
55 		if (dev->vhostfd >= 0) {
56 			int r;
57 			int flags;
58 
59 			flags = fcntl(dev->vhostfd, F_GETFL);
60 			if (fcntl(dev->vhostfd, F_SETFL,
61 					flags | O_NONBLOCK) == -1) {
62 				PMD_DRV_LOG(ERR, "error setting O_NONBLOCK flag");
63 				return;
64 			}
65 			r = recv(dev->vhostfd, buf, 128, MSG_PEEK);
66 			if (r == 0 || (r < 0 && errno != EAGAIN)) {
67 				dev->status &= (~VIRTIO_NET_S_LINK_UP);
68 				PMD_DRV_LOG(ERR, "virtio-user port %u is down",
69 					    hw->port_id);
70 				/* Only client mode is available now. Once the
71 				 * connection is broken, it can never be up
72 				 * again. Besides, this function could be called
73 				 * in the process of interrupt handling,
74 				 * callback cannot be unregistered here, set an
75 				 * alarm to do it.
76 				 */
77 				rte_eal_alarm_set(1,
78 						  virtio_user_delayed_handler,
79 						  (void *)hw);
80 			} else {
81 				dev->status |= VIRTIO_NET_S_LINK_UP;
82 			}
83 			fcntl(dev->vhostfd, F_SETFL, flags & (~O_NONBLOCK));
84 		}
85 		*(uint16_t *)dst = dev->status;
86 	}
87 
88 	if (offset == offsetof(struct virtio_net_config, max_virtqueue_pairs))
89 		*(uint16_t *)dst = dev->max_queue_pairs;
90 }
91 
92 static void
93 virtio_user_write_dev_config(struct virtio_hw *hw, size_t offset,
94 		      const void *src, int length)
95 {
96 	int i;
97 	struct virtio_user_dev *dev = virtio_user_get_dev(hw);
98 
99 	if ((offset == offsetof(struct virtio_net_config, mac)) &&
100 	    (length == ETHER_ADDR_LEN))
101 		for (i = 0; i < ETHER_ADDR_LEN; ++i)
102 			dev->mac_addr[i] = ((const uint8_t *)src)[i];
103 	else
104 		PMD_DRV_LOG(ERR, "not supported offset=%zu, len=%d",
105 			    offset, length);
106 }
107 
108 static void
109 virtio_user_reset(struct virtio_hw *hw)
110 {
111 	struct virtio_user_dev *dev = virtio_user_get_dev(hw);
112 
113 	if (dev->status & VIRTIO_CONFIG_STATUS_DRIVER_OK)
114 		virtio_user_stop_device(dev);
115 }
116 
117 static void
118 virtio_user_set_status(struct virtio_hw *hw, uint8_t status)
119 {
120 	struct virtio_user_dev *dev = virtio_user_get_dev(hw);
121 
122 	if (status & VIRTIO_CONFIG_STATUS_DRIVER_OK)
123 		virtio_user_start_device(dev);
124 	else if (status == VIRTIO_CONFIG_STATUS_RESET)
125 		virtio_user_reset(hw);
126 	dev->status = status;
127 }
128 
129 static uint8_t
130 virtio_user_get_status(struct virtio_hw *hw)
131 {
132 	struct virtio_user_dev *dev = virtio_user_get_dev(hw);
133 
134 	return dev->status;
135 }
136 
137 static uint64_t
138 virtio_user_get_features(struct virtio_hw *hw)
139 {
140 	struct virtio_user_dev *dev = virtio_user_get_dev(hw);
141 
142 	/* unmask feature bits defined in vhost user protocol */
143 	return dev->device_features & VIRTIO_PMD_SUPPORTED_GUEST_FEATURES;
144 }
145 
146 static void
147 virtio_user_set_features(struct virtio_hw *hw, uint64_t features)
148 {
149 	struct virtio_user_dev *dev = virtio_user_get_dev(hw);
150 
151 	dev->features = features & dev->device_features;
152 }
153 
154 static uint8_t
155 virtio_user_get_isr(struct virtio_hw *hw __rte_unused)
156 {
157 	/* rxq interrupts and config interrupt are separated in virtio-user,
158 	 * here we only report config change.
159 	 */
160 	return VIRTIO_PCI_ISR_CONFIG;
161 }
162 
163 static uint16_t
164 virtio_user_set_config_irq(struct virtio_hw *hw __rte_unused,
165 		    uint16_t vec __rte_unused)
166 {
167 	return 0;
168 }
169 
170 static uint16_t
171 virtio_user_set_queue_irq(struct virtio_hw *hw __rte_unused,
172 			  struct virtqueue *vq __rte_unused,
173 			  uint16_t vec)
174 {
175 	/* pretend we have done that */
176 	return vec;
177 }
178 
179 /* This function is to get the queue size, aka, number of descs, of a specified
180  * queue. Different with the VHOST_USER_GET_QUEUE_NUM, which is used to get the
181  * max supported queues.
182  */
183 static uint16_t
184 virtio_user_get_queue_num(struct virtio_hw *hw, uint16_t queue_id __rte_unused)
185 {
186 	struct virtio_user_dev *dev = virtio_user_get_dev(hw);
187 
188 	/* Currently, each queue has same queue size */
189 	return dev->queue_size;
190 }
191 
192 static int
193 virtio_user_setup_queue(struct virtio_hw *hw, struct virtqueue *vq)
194 {
195 	struct virtio_user_dev *dev = virtio_user_get_dev(hw);
196 	uint16_t queue_idx = vq->vq_queue_index;
197 	uint64_t desc_addr, avail_addr, used_addr;
198 
199 	desc_addr = (uintptr_t)vq->vq_ring_virt_mem;
200 	avail_addr = desc_addr + vq->vq_nentries * sizeof(struct vring_desc);
201 	used_addr = RTE_ALIGN_CEIL(avail_addr + offsetof(struct vring_avail,
202 							 ring[vq->vq_nentries]),
203 				   VIRTIO_PCI_VRING_ALIGN);
204 
205 	dev->vrings[queue_idx].num = vq->vq_nentries;
206 	dev->vrings[queue_idx].desc = (void *)(uintptr_t)desc_addr;
207 	dev->vrings[queue_idx].avail = (void *)(uintptr_t)avail_addr;
208 	dev->vrings[queue_idx].used = (void *)(uintptr_t)used_addr;
209 
210 	return 0;
211 }
212 
213 static void
214 virtio_user_del_queue(struct virtio_hw *hw, struct virtqueue *vq)
215 {
216 	/* For legacy devices, write 0 to VIRTIO_PCI_QUEUE_PFN port, QEMU
217 	 * correspondingly stops the ioeventfds, and reset the status of
218 	 * the device.
219 	 * For modern devices, set queue desc, avail, used in PCI bar to 0,
220 	 * not see any more behavior in QEMU.
221 	 *
222 	 * Here we just care about what information to deliver to vhost-user
223 	 * or vhost-kernel. So we just close ioeventfd for now.
224 	 */
225 	struct virtio_user_dev *dev = virtio_user_get_dev(hw);
226 
227 	close(dev->callfds[vq->vq_queue_index]);
228 	close(dev->kickfds[vq->vq_queue_index]);
229 }
230 
231 static void
232 virtio_user_notify_queue(struct virtio_hw *hw, struct virtqueue *vq)
233 {
234 	uint64_t buf = 1;
235 	struct virtio_user_dev *dev = virtio_user_get_dev(hw);
236 
237 	if (hw->cvq && (hw->cvq->vq == vq)) {
238 		virtio_user_handle_cq(dev, vq->vq_queue_index);
239 		return;
240 	}
241 
242 	if (write(dev->kickfds[vq->vq_queue_index], &buf, sizeof(buf)) < 0)
243 		PMD_DRV_LOG(ERR, "failed to kick backend: %s",
244 			    strerror(errno));
245 }
246 
247 const struct virtio_pci_ops virtio_user_ops = {
248 	.read_dev_cfg	= virtio_user_read_dev_config,
249 	.write_dev_cfg	= virtio_user_write_dev_config,
250 	.reset		= virtio_user_reset,
251 	.get_status	= virtio_user_get_status,
252 	.set_status	= virtio_user_set_status,
253 	.get_features	= virtio_user_get_features,
254 	.set_features	= virtio_user_set_features,
255 	.get_isr	= virtio_user_get_isr,
256 	.set_config_irq	= virtio_user_set_config_irq,
257 	.set_queue_irq	= virtio_user_set_queue_irq,
258 	.get_queue_num	= virtio_user_get_queue_num,
259 	.setup_queue	= virtio_user_setup_queue,
260 	.del_queue	= virtio_user_del_queue,
261 	.notify_queue	= virtio_user_notify_queue,
262 };
263 
264 static const char *valid_args[] = {
265 #define VIRTIO_USER_ARG_QUEUES_NUM     "queues"
266 	VIRTIO_USER_ARG_QUEUES_NUM,
267 #define VIRTIO_USER_ARG_CQ_NUM         "cq"
268 	VIRTIO_USER_ARG_CQ_NUM,
269 #define VIRTIO_USER_ARG_MAC            "mac"
270 	VIRTIO_USER_ARG_MAC,
271 #define VIRTIO_USER_ARG_PATH           "path"
272 	VIRTIO_USER_ARG_PATH,
273 #define VIRTIO_USER_ARG_QUEUE_SIZE     "queue_size"
274 	VIRTIO_USER_ARG_QUEUE_SIZE,
275 #define VIRTIO_USER_ARG_INTERFACE_NAME "iface"
276 	VIRTIO_USER_ARG_INTERFACE_NAME,
277 	NULL
278 };
279 
280 #define VIRTIO_USER_DEF_CQ_EN	0
281 #define VIRTIO_USER_DEF_Q_NUM	1
282 #define VIRTIO_USER_DEF_Q_SZ	256
283 
284 static int
285 get_string_arg(const char *key __rte_unused,
286 	       const char *value, void *extra_args)
287 {
288 	if (!value || !extra_args)
289 		return -EINVAL;
290 
291 	*(char **)extra_args = strdup(value);
292 
293 	if (!*(char **)extra_args)
294 		return -ENOMEM;
295 
296 	return 0;
297 }
298 
299 static int
300 get_integer_arg(const char *key __rte_unused,
301 		const char *value, void *extra_args)
302 {
303 	if (!value || !extra_args)
304 		return -EINVAL;
305 
306 	*(uint64_t *)extra_args = strtoull(value, NULL, 0);
307 
308 	return 0;
309 }
310 
311 static struct rte_vdev_driver virtio_user_driver;
312 
313 static struct rte_eth_dev *
314 virtio_user_eth_dev_alloc(struct rte_vdev_device *vdev)
315 {
316 	struct rte_eth_dev *eth_dev;
317 	struct rte_eth_dev_data *data;
318 	struct virtio_hw *hw;
319 	struct virtio_user_dev *dev;
320 
321 	eth_dev = rte_eth_vdev_allocate(vdev, sizeof(*hw));
322 	if (!eth_dev) {
323 		PMD_INIT_LOG(ERR, "cannot alloc rte_eth_dev");
324 		return NULL;
325 	}
326 
327 	data = eth_dev->data;
328 	hw = eth_dev->data->dev_private;
329 
330 	dev = rte_zmalloc(NULL, sizeof(*dev), 0);
331 	if (!dev) {
332 		PMD_INIT_LOG(ERR, "malloc virtio_user_dev failed");
333 		rte_eth_dev_release_port(eth_dev);
334 		rte_free(hw);
335 		return NULL;
336 	}
337 
338 	hw->port_id = data->port_id;
339 	dev->port_id = data->port_id;
340 	virtio_hw_internal[hw->port_id].vtpci_ops = &virtio_user_ops;
341 	/*
342 	 * MSIX is required to enable LSC (see virtio_init_device).
343 	 * Here just pretend that we support msix.
344 	 */
345 	hw->use_msix = 1;
346 	hw->modern   = 0;
347 	hw->use_simple_rx = 0;
348 	hw->use_simple_tx = 0;
349 	hw->virtio_user_dev = dev;
350 	return eth_dev;
351 }
352 
353 static void
354 virtio_user_eth_dev_free(struct rte_eth_dev *eth_dev)
355 {
356 	struct rte_eth_dev_data *data = eth_dev->data;
357 	struct virtio_hw *hw = data->dev_private;
358 
359 	rte_free(hw->virtio_user_dev);
360 	rte_free(hw);
361 	rte_eth_dev_release_port(eth_dev);
362 }
363 
364 /* Dev initialization routine. Invoked once for each virtio vdev at
365  * EAL init time, see rte_bus_probe().
366  * Returns 0 on success.
367  */
368 static int
369 virtio_user_pmd_probe(struct rte_vdev_device *dev)
370 {
371 	struct rte_kvargs *kvlist = NULL;
372 	struct rte_eth_dev *eth_dev;
373 	struct virtio_hw *hw;
374 	uint64_t queues = VIRTIO_USER_DEF_Q_NUM;
375 	uint64_t cq = VIRTIO_USER_DEF_CQ_EN;
376 	uint64_t queue_size = VIRTIO_USER_DEF_Q_SZ;
377 	char *path = NULL;
378 	char *ifname = NULL;
379 	char *mac_addr = NULL;
380 	int ret = -1;
381 
382 	kvlist = rte_kvargs_parse(rte_vdev_device_args(dev), valid_args);
383 	if (!kvlist) {
384 		PMD_INIT_LOG(ERR, "error when parsing param");
385 		goto end;
386 	}
387 
388 	if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_PATH) == 1) {
389 		if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_PATH,
390 				       &get_string_arg, &path) < 0) {
391 			PMD_INIT_LOG(ERR, "error to parse %s",
392 				     VIRTIO_USER_ARG_PATH);
393 			goto end;
394 		}
395 	} else {
396 		PMD_INIT_LOG(ERR, "arg %s is mandatory for virtio_user",
397 			  VIRTIO_USER_ARG_QUEUE_SIZE);
398 		goto end;
399 	}
400 
401 	if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_INTERFACE_NAME) == 1) {
402 		if (is_vhost_user_by_type(path)) {
403 			PMD_INIT_LOG(ERR,
404 				"arg %s applies only to vhost-kernel backend",
405 				VIRTIO_USER_ARG_INTERFACE_NAME);
406 			goto end;
407 		}
408 
409 		if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_INTERFACE_NAME,
410 				       &get_string_arg, &ifname) < 0) {
411 			PMD_INIT_LOG(ERR, "error to parse %s",
412 				     VIRTIO_USER_ARG_INTERFACE_NAME);
413 			goto end;
414 		}
415 	}
416 
417 	if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_MAC) == 1) {
418 		if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_MAC,
419 				       &get_string_arg, &mac_addr) < 0) {
420 			PMD_INIT_LOG(ERR, "error to parse %s",
421 				     VIRTIO_USER_ARG_MAC);
422 			goto end;
423 		}
424 	}
425 
426 	if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_QUEUE_SIZE) == 1) {
427 		if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_QUEUE_SIZE,
428 				       &get_integer_arg, &queue_size) < 0) {
429 			PMD_INIT_LOG(ERR, "error to parse %s",
430 				     VIRTIO_USER_ARG_QUEUE_SIZE);
431 			goto end;
432 		}
433 	}
434 
435 	if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_QUEUES_NUM) == 1) {
436 		if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_QUEUES_NUM,
437 				       &get_integer_arg, &queues) < 0) {
438 			PMD_INIT_LOG(ERR, "error to parse %s",
439 				     VIRTIO_USER_ARG_QUEUES_NUM);
440 			goto end;
441 		}
442 	}
443 
444 	if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_CQ_NUM) == 1) {
445 		if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_CQ_NUM,
446 				       &get_integer_arg, &cq) < 0) {
447 			PMD_INIT_LOG(ERR, "error to parse %s",
448 				     VIRTIO_USER_ARG_CQ_NUM);
449 			goto end;
450 		}
451 	} else if (queues > 1) {
452 		cq = 1;
453 	}
454 
455 	if (queues > 1 && cq == 0) {
456 		PMD_INIT_LOG(ERR, "multi-q requires ctrl-q");
457 		goto end;
458 	}
459 
460 	if (queues > VIRTIO_MAX_VIRTQUEUE_PAIRS) {
461 		PMD_INIT_LOG(ERR, "arg %s %" PRIu64 " exceeds the limit %u",
462 			VIRTIO_USER_ARG_QUEUES_NUM, queues,
463 			VIRTIO_MAX_VIRTQUEUE_PAIRS);
464 		goto end;
465 	}
466 
467 	if (rte_eal_process_type() == RTE_PROC_PRIMARY) {
468 		eth_dev = virtio_user_eth_dev_alloc(dev);
469 		if (!eth_dev) {
470 			PMD_INIT_LOG(ERR, "virtio_user fails to alloc device");
471 			goto end;
472 		}
473 
474 		hw = eth_dev->data->dev_private;
475 		if (virtio_user_dev_init(hw->virtio_user_dev, path, queues, cq,
476 				 queue_size, mac_addr, &ifname) < 0) {
477 			PMD_INIT_LOG(ERR, "virtio_user_dev_init fails");
478 			virtio_user_eth_dev_free(eth_dev);
479 			goto end;
480 		}
481 	} else {
482 		eth_dev = rte_eth_dev_attach_secondary(rte_vdev_device_name(dev));
483 		if (!eth_dev)
484 			goto end;
485 	}
486 
487 	/* previously called by rte_pci_probe() for physical dev */
488 	if (eth_virtio_dev_init(eth_dev) < 0) {
489 		PMD_INIT_LOG(ERR, "eth_virtio_dev_init fails");
490 		virtio_user_eth_dev_free(eth_dev);
491 		goto end;
492 	}
493 	ret = 0;
494 
495 end:
496 	if (kvlist)
497 		rte_kvargs_free(kvlist);
498 	if (path)
499 		free(path);
500 	if (mac_addr)
501 		free(mac_addr);
502 	if (ifname)
503 		free(ifname);
504 	return ret;
505 }
506 
507 /** Called by rte_eth_dev_detach() */
508 static int
509 virtio_user_pmd_remove(struct rte_vdev_device *vdev)
510 {
511 	const char *name;
512 	struct rte_eth_dev *eth_dev;
513 	struct virtio_hw *hw;
514 	struct virtio_user_dev *dev;
515 
516 	if (!vdev)
517 		return -EINVAL;
518 
519 	name = rte_vdev_device_name(vdev);
520 	PMD_DRV_LOG(INFO, "Un-Initializing %s", name);
521 	eth_dev = rte_eth_dev_allocated(name);
522 	if (!eth_dev)
523 		return -ENODEV;
524 
525 	/* make sure the device is stopped, queues freed */
526 	rte_eth_dev_close(eth_dev->data->port_id);
527 
528 	hw = eth_dev->data->dev_private;
529 	dev = hw->virtio_user_dev;
530 	virtio_user_dev_uninit(dev);
531 
532 	rte_free(eth_dev->data->dev_private);
533 	rte_eth_dev_release_port(eth_dev);
534 
535 	return 0;
536 }
537 
538 static struct rte_vdev_driver virtio_user_driver = {
539 	.probe = virtio_user_pmd_probe,
540 	.remove = virtio_user_pmd_remove,
541 };
542 
543 RTE_PMD_REGISTER_VDEV(net_virtio_user, virtio_user_driver);
544 RTE_PMD_REGISTER_ALIAS(net_virtio_user, virtio_user);
545 RTE_PMD_REGISTER_PARAM_STRING(net_virtio_user,
546 	"path=<path> "
547 	"mac=<mac addr> "
548 	"cq=<int> "
549 	"queue_size=<int> "
550 	"queues=<int> "
551 	"iface=<string>");
552