xref: /dpdk/drivers/net/failsafe/failsafe.c (revision e2b4cfd6a40d9361928c60834ef676e75d9be9bd)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright 2017 6WIND S.A.
3  * Copyright 2017 Mellanox Technologies, Ltd
4  */
5 
6 #include <stdbool.h>
7 
8 #include <rte_alarm.h>
9 #include <rte_malloc.h>
10 #include <rte_ethdev_driver.h>
11 #include <rte_ethdev_vdev.h>
12 #include <rte_devargs.h>
13 #include <rte_kvargs.h>
14 #include <rte_bus_vdev.h>
15 
16 #include "failsafe_private.h"
17 
18 int failsafe_logtype;
19 
20 const char pmd_failsafe_driver_name[] = FAILSAFE_DRIVER_NAME;
21 static const struct rte_eth_link eth_link = {
22 	.link_speed = ETH_SPEED_NUM_10G,
23 	.link_duplex = ETH_LINK_FULL_DUPLEX,
24 	.link_status = ETH_LINK_UP,
25 	.link_autoneg = ETH_LINK_AUTONEG,
26 };
27 
28 static int
29 fs_sub_device_alloc(struct rte_eth_dev *dev,
30 		const char *params)
31 {
32 	uint8_t nb_subs;
33 	int ret;
34 	int i;
35 	struct sub_device *sdev;
36 	uint8_t sdev_iterator;
37 
38 	ret = failsafe_args_count_subdevice(dev, params);
39 	if (ret)
40 		return ret;
41 	if (PRIV(dev)->subs_tail > FAILSAFE_MAX_ETHPORTS) {
42 		ERROR("Cannot allocate more than %d ports",
43 			FAILSAFE_MAX_ETHPORTS);
44 		return -ENOSPC;
45 	}
46 	nb_subs = PRIV(dev)->subs_tail;
47 	PRIV(dev)->subs = rte_zmalloc(NULL,
48 			sizeof(struct sub_device) * nb_subs,
49 			RTE_CACHE_LINE_SIZE);
50 	if (PRIV(dev)->subs == NULL) {
51 		ERROR("Could not allocate sub_devices");
52 		return -ENOMEM;
53 	}
54 	/* Initiate static sub devices linked list. */
55 	for (i = 1; i < nb_subs; i++)
56 		PRIV(dev)->subs[i - 1].next = PRIV(dev)->subs + i;
57 	PRIV(dev)->subs[i - 1].next = PRIV(dev)->subs;
58 
59 	FOREACH_SUBDEV(sdev, sdev_iterator, dev) {
60 		sdev->sdev_port_id = RTE_MAX_ETHPORTS;
61 	}
62 	return 0;
63 }
64 
65 static void
66 fs_sub_device_free(struct rte_eth_dev *dev)
67 {
68 	rte_free(PRIV(dev)->subs);
69 }
70 
71 static void fs_hotplug_alarm(void *arg);
72 
73 int
74 failsafe_hotplug_alarm_install(struct rte_eth_dev *dev)
75 {
76 	int ret;
77 
78 	if (dev == NULL)
79 		return -EINVAL;
80 	if (PRIV(dev)->pending_alarm)
81 		return 0;
82 	ret = rte_eal_alarm_set(failsafe_hotplug_poll * 1000,
83 				fs_hotplug_alarm,
84 				dev);
85 	if (ret) {
86 		ERROR("Could not set up plug-in event detection");
87 		return ret;
88 	}
89 	PRIV(dev)->pending_alarm = 1;
90 	return 0;
91 }
92 
93 int
94 failsafe_hotplug_alarm_cancel(struct rte_eth_dev *dev)
95 {
96 	int ret = 0;
97 
98 	rte_errno = 0;
99 	rte_eal_alarm_cancel(fs_hotplug_alarm, dev);
100 	if (rte_errno) {
101 		ERROR("rte_eal_alarm_cancel failed (errno: %s)",
102 		      strerror(rte_errno));
103 		ret = -rte_errno;
104 	} else {
105 		PRIV(dev)->pending_alarm = 0;
106 	}
107 	return ret;
108 }
109 
110 static void
111 fs_hotplug_alarm(void *arg)
112 {
113 	struct rte_eth_dev *dev = arg;
114 	struct sub_device *sdev;
115 	int ret;
116 	uint8_t i;
117 
118 	if (!PRIV(dev)->pending_alarm)
119 		return;
120 	PRIV(dev)->pending_alarm = 0;
121 	FOREACH_SUBDEV(sdev, i, dev)
122 		if (sdev->state != PRIV(dev)->state)
123 			break;
124 	/* if we have non-probed device */
125 	if (i != PRIV(dev)->subs_tail) {
126 		if (fs_lock(dev, 1) != 0)
127 			goto reinstall;
128 		ret = failsafe_eth_dev_state_sync(dev);
129 		fs_unlock(dev, 1);
130 		if (ret)
131 			ERROR("Unable to synchronize sub_device state");
132 	}
133 	failsafe_dev_remove(dev);
134 reinstall:
135 	ret = failsafe_hotplug_alarm_install(dev);
136 	if (ret)
137 		ERROR("Unable to set up next alarm");
138 }
139 
140 static int
141 fs_mutex_init(struct fs_priv *priv)
142 {
143 	int ret;
144 	pthread_mutexattr_t attr;
145 
146 	ret = pthread_mutexattr_init(&attr);
147 	if (ret) {
148 		ERROR("Cannot initiate mutex attributes - %s", strerror(ret));
149 		return ret;
150 	}
151 	/* Allow mutex relocks for the thread holding the mutex. */
152 	ret = pthread_mutexattr_settype(&attr, PTHREAD_MUTEX_RECURSIVE);
153 	if (ret) {
154 		ERROR("Cannot set mutex type - %s", strerror(ret));
155 		return ret;
156 	}
157 	ret = pthread_mutex_init(&priv->hotplug_mutex, &attr);
158 	if (ret) {
159 		ERROR("Cannot initiate mutex - %s", strerror(ret));
160 		return ret;
161 	}
162 	return 0;
163 }
164 
165 static int
166 fs_eth_dev_create(struct rte_vdev_device *vdev)
167 {
168 	struct rte_eth_dev *dev;
169 	struct rte_ether_addr *mac;
170 	struct fs_priv *priv;
171 	struct sub_device *sdev;
172 	const char *params;
173 	unsigned int socket_id;
174 	uint8_t i;
175 	int ret;
176 
177 	dev = NULL;
178 	priv = NULL;
179 	socket_id = rte_socket_id();
180 	INFO("Creating fail-safe device on NUMA socket %u", socket_id);
181 	params = rte_vdev_device_args(vdev);
182 	if (params == NULL) {
183 		ERROR("This PMD requires sub-devices, none provided");
184 		return -1;
185 	}
186 	dev = rte_eth_vdev_allocate(vdev, sizeof(*priv));
187 	if (dev == NULL) {
188 		ERROR("Unable to allocate rte_eth_dev");
189 		return -1;
190 	}
191 	priv = PRIV(dev);
192 	priv->data = dev->data;
193 	priv->rxp = FS_RX_PROXY_INIT;
194 	dev->dev_ops = &failsafe_ops;
195 	dev->data->mac_addrs = &PRIV(dev)->mac_addrs[0];
196 	dev->data->dev_link = eth_link;
197 	PRIV(dev)->nb_mac_addr = 1;
198 	TAILQ_INIT(&PRIV(dev)->flow_list);
199 	dev->rx_pkt_burst = (eth_rx_burst_t)&failsafe_rx_burst;
200 	dev->tx_pkt_burst = (eth_tx_burst_t)&failsafe_tx_burst;
201 	ret = fs_sub_device_alloc(dev, params);
202 	if (ret) {
203 		ERROR("Could not allocate sub_devices");
204 		goto free_dev;
205 	}
206 	ret = failsafe_args_parse(dev, params);
207 	if (ret)
208 		goto free_subs;
209 	ret = rte_eth_dev_owner_new(&priv->my_owner.id);
210 	if (ret) {
211 		ERROR("Failed to get unique owner identifier");
212 		goto free_args;
213 	}
214 	snprintf(priv->my_owner.name, sizeof(priv->my_owner.name),
215 		 FAILSAFE_OWNER_NAME);
216 	DEBUG("Failsafe port %u owner info: %s_%016"PRIX64, dev->data->port_id,
217 	      priv->my_owner.name, priv->my_owner.id);
218 	ret = rte_eth_dev_callback_register(RTE_ETH_ALL, RTE_ETH_EVENT_NEW,
219 					    failsafe_eth_new_event_callback,
220 					    dev);
221 	if (ret) {
222 		ERROR("Failed to register NEW callback");
223 		goto free_args;
224 	}
225 	ret = failsafe_eal_init(dev);
226 	if (ret)
227 		goto unregister_new_callback;
228 	ret = fs_mutex_init(priv);
229 	if (ret)
230 		goto unregister_new_callback;
231 	ret = failsafe_hotplug_alarm_install(dev);
232 	if (ret) {
233 		ERROR("Could not set up plug-in event detection");
234 		goto unregister_new_callback;
235 	}
236 	mac = &dev->data->mac_addrs[0];
237 	if (failsafe_mac_from_arg) {
238 		/*
239 		 * If MAC address was provided as a parameter,
240 		 * apply to all probed slaves.
241 		 */
242 		FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_PROBED) {
243 			ret = rte_eth_dev_default_mac_addr_set(PORT_ID(sdev),
244 							       mac);
245 			if (ret) {
246 				ERROR("Failed to set default MAC address");
247 				goto cancel_alarm;
248 			}
249 		}
250 	} else {
251 		/*
252 		 * Use the ether_addr from first probed
253 		 * device, either preferred or fallback.
254 		 */
255 		FOREACH_SUBDEV(sdev, i, dev)
256 			if (sdev->state >= DEV_PROBED) {
257 				rte_ether_addr_copy(
258 					&ETH(sdev)->data->mac_addrs[0], mac);
259 				break;
260 			}
261 		/*
262 		 * If no device has been probed and no ether_addr
263 		 * has been provided on the command line, use a random
264 		 * valid one.
265 		 * It will be applied during future slave state syncs to
266 		 * probed slaves.
267 		 */
268 		if (i == priv->subs_tail)
269 			rte_eth_random_addr(&mac->addr_bytes[0]);
270 	}
271 	INFO("MAC address is %02x:%02x:%02x:%02x:%02x:%02x",
272 		mac->addr_bytes[0], mac->addr_bytes[1],
273 		mac->addr_bytes[2], mac->addr_bytes[3],
274 		mac->addr_bytes[4], mac->addr_bytes[5]);
275 	dev->data->dev_flags |= RTE_ETH_DEV_INTR_LSC;
276 	PRIV(dev)->intr_handle = (struct rte_intr_handle){
277 		.fd = -1,
278 		.type = RTE_INTR_HANDLE_EXT,
279 	};
280 	rte_eth_dev_probing_finish(dev);
281 	return 0;
282 cancel_alarm:
283 	failsafe_hotplug_alarm_cancel(dev);
284 unregister_new_callback:
285 	rte_eth_dev_callback_unregister(RTE_ETH_ALL, RTE_ETH_EVENT_NEW,
286 					failsafe_eth_new_event_callback, dev);
287 free_args:
288 	failsafe_args_free(dev);
289 free_subs:
290 	fs_sub_device_free(dev);
291 free_dev:
292 	/* mac_addrs must not be freed alone because part of dev_private */
293 	dev->data->mac_addrs = NULL;
294 	rte_eth_dev_release_port(dev);
295 	return -1;
296 }
297 
298 static int
299 fs_rte_eth_free(const char *name)
300 {
301 	struct rte_eth_dev *dev;
302 	int ret;
303 
304 	dev = rte_eth_dev_allocated(name);
305 	if (dev == NULL)
306 		return -ENODEV;
307 	rte_eth_dev_callback_unregister(RTE_ETH_ALL, RTE_ETH_EVENT_NEW,
308 					failsafe_eth_new_event_callback, dev);
309 	ret = failsafe_eal_uninit(dev);
310 	if (ret)
311 		ERROR("Error while uninitializing sub-EAL");
312 	failsafe_args_free(dev);
313 	fs_sub_device_free(dev);
314 	ret = pthread_mutex_destroy(&PRIV(dev)->hotplug_mutex);
315 	if (ret)
316 		ERROR("Error while destroying hotplug mutex");
317 	rte_free(PRIV(dev)->mcast_addrs);
318 	/* mac_addrs must not be freed alone because part of dev_private */
319 	dev->data->mac_addrs = NULL;
320 	rte_eth_dev_release_port(dev);
321 	return ret;
322 }
323 
324 static bool
325 devargs_already_listed(struct rte_devargs *devargs)
326 {
327 	struct rte_devargs *list_da;
328 
329 	RTE_EAL_DEVARGS_FOREACH(devargs->bus->name, list_da) {
330 		if (strcmp(list_da->name, devargs->name) == 0)
331 			/* devargs already in the list */
332 			return true;
333 	}
334 	return false;
335 }
336 
337 static int
338 rte_pmd_failsafe_probe(struct rte_vdev_device *vdev)
339 {
340 	const char *name;
341 	struct rte_eth_dev *eth_dev;
342 	struct sub_device  *sdev;
343 	struct rte_devargs devargs;
344 	uint8_t i;
345 	int ret;
346 
347 	name = rte_vdev_device_name(vdev);
348 	INFO("Initializing " FAILSAFE_DRIVER_NAME " for %s",
349 			name);
350 
351 	if (rte_eal_process_type() == RTE_PROC_SECONDARY &&
352 	    strlen(rte_vdev_device_args(vdev)) == 0) {
353 		eth_dev = rte_eth_dev_attach_secondary(name);
354 		if (!eth_dev) {
355 			ERROR("Failed to probe %s", name);
356 			return -1;
357 		}
358 		eth_dev->dev_ops = &failsafe_ops;
359 		eth_dev->device = &vdev->device;
360 		eth_dev->rx_pkt_burst = (eth_rx_burst_t)&failsafe_rx_burst;
361 		eth_dev->tx_pkt_burst = (eth_tx_burst_t)&failsafe_tx_burst;
362 		/*
363 		 * Failsafe will attempt to probe all of its sub-devices.
364 		 * Any failure in sub-devices is not a fatal error.
365 		 * A sub-device can be plugged later.
366 		 */
367 		FOREACH_SUBDEV(sdev, i, eth_dev) {
368 			/* skip empty devargs */
369 			if (sdev->devargs.name[0] == '\0')
370 				continue;
371 
372 			/* rebuild devargs to be able to get the bus name. */
373 			ret = rte_devargs_parse(&devargs,
374 						sdev->devargs.name);
375 			if (ret != 0) {
376 				ERROR("Failed to parse devargs %s",
377 					devargs.name);
378 				continue;
379 			}
380 			if (!devargs_already_listed(&devargs)) {
381 				ret = rte_dev_probe(devargs.name);
382 				if (ret < 0) {
383 					ERROR("Failed to probe devargs %s",
384 					      devargs.name);
385 					continue;
386 				}
387 			}
388 		}
389 		rte_eth_dev_probing_finish(eth_dev);
390 		return 0;
391 	}
392 
393 	return fs_eth_dev_create(vdev);
394 }
395 
396 static int
397 rte_pmd_failsafe_remove(struct rte_vdev_device *vdev)
398 {
399 	const char *name;
400 
401 	name = rte_vdev_device_name(vdev);
402 	INFO("Uninitializing " FAILSAFE_DRIVER_NAME " for %s", name);
403 	return fs_rte_eth_free(name);
404 }
405 
406 static struct rte_vdev_driver failsafe_drv = {
407 	.probe = rte_pmd_failsafe_probe,
408 	.remove = rte_pmd_failsafe_remove,
409 };
410 
411 RTE_PMD_REGISTER_VDEV(net_failsafe, failsafe_drv);
412 RTE_PMD_REGISTER_PARAM_STRING(net_failsafe, PMD_FAILSAFE_PARAM_STRING);
413 
414 RTE_INIT(failsafe_init_log)
415 {
416 	failsafe_logtype = rte_log_register("pmd.net.failsafe");
417 	if (failsafe_logtype >= 0)
418 		rte_log_set_level(failsafe_logtype, RTE_LOG_NOTICE);
419 }
420