8af31d71b3df1e3ca672cd45279452eab44660af
[dpdk.git] / drivers / net / failsafe / failsafe.c
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         dev->dev_ops = &failsafe_ops;
194         dev->data->mac_addrs = &PRIV(dev)->mac_addrs[0];
195         dev->data->dev_link = eth_link;
196         PRIV(dev)->nb_mac_addr = 1;
197         TAILQ_INIT(&PRIV(dev)->flow_list);
198         dev->rx_pkt_burst = (eth_rx_burst_t)&failsafe_rx_burst;
199         dev->tx_pkt_burst = (eth_tx_burst_t)&failsafe_tx_burst;
200         ret = fs_sub_device_alloc(dev, params);
201         if (ret) {
202                 ERROR("Could not allocate sub_devices");
203                 goto free_dev;
204         }
205         ret = failsafe_args_parse(dev, params);
206         if (ret)
207                 goto free_subs;
208         ret = rte_eth_dev_owner_new(&priv->my_owner.id);
209         if (ret) {
210                 ERROR("Failed to get unique owner identifier");
211                 goto free_args;
212         }
213         snprintf(priv->my_owner.name, sizeof(priv->my_owner.name),
214                  FAILSAFE_OWNER_NAME);
215         DEBUG("Failsafe port %u owner info: %s_%016"PRIX64, dev->data->port_id,
216               priv->my_owner.name, priv->my_owner.id);
217         ret = rte_eth_dev_callback_register(RTE_ETH_ALL, RTE_ETH_EVENT_NEW,
218                                             failsafe_eth_new_event_callback,
219                                             dev);
220         if (ret) {
221                 ERROR("Failed to register NEW callback");
222                 goto free_args;
223         }
224         ret = failsafe_eal_init(dev);
225         if (ret)
226                 goto unregister_new_callback;
227         ret = fs_mutex_init(priv);
228         if (ret)
229                 goto unregister_new_callback;
230         ret = failsafe_hotplug_alarm_install(dev);
231         if (ret) {
232                 ERROR("Could not set up plug-in event detection");
233                 goto unregister_new_callback;
234         }
235         mac = &dev->data->mac_addrs[0];
236         if (failsafe_mac_from_arg) {
237                 /*
238                  * If MAC address was provided as a parameter,
239                  * apply to all probed slaves.
240                  */
241                 FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_PROBED) {
242                         ret = rte_eth_dev_default_mac_addr_set(PORT_ID(sdev),
243                                                                mac);
244                         if (ret) {
245                                 ERROR("Failed to set default MAC address");
246                                 goto cancel_alarm;
247                         }
248                 }
249         } else {
250                 /*
251                  * Use the ether_addr from first probed
252                  * device, either preferred or fallback.
253                  */
254                 FOREACH_SUBDEV(sdev, i, dev)
255                         if (sdev->state >= DEV_PROBED) {
256                                 rte_ether_addr_copy(
257                                         &ETH(sdev)->data->mac_addrs[0], mac);
258                                 break;
259                         }
260                 /*
261                  * If no device has been probed and no ether_addr
262                  * has been provided on the command line, use a random
263                  * valid one.
264                  * It will be applied during future slave state syncs to
265                  * probed slaves.
266                  */
267                 if (i == priv->subs_tail)
268                         rte_eth_random_addr(&mac->addr_bytes[0]);
269         }
270         INFO("MAC address is %02x:%02x:%02x:%02x:%02x:%02x",
271                 mac->addr_bytes[0], mac->addr_bytes[1],
272                 mac->addr_bytes[2], mac->addr_bytes[3],
273                 mac->addr_bytes[4], mac->addr_bytes[5]);
274         dev->data->dev_flags |= RTE_ETH_DEV_INTR_LSC;
275         PRIV(dev)->intr_handle = (struct rte_intr_handle){
276                 .fd = -1,
277                 .type = RTE_INTR_HANDLE_EXT,
278         };
279         rte_eth_dev_probing_finish(dev);
280         return 0;
281 cancel_alarm:
282         failsafe_hotplug_alarm_cancel(dev);
283 unregister_new_callback:
284         rte_eth_dev_callback_unregister(RTE_ETH_ALL, RTE_ETH_EVENT_NEW,
285                                         failsafe_eth_new_event_callback, dev);
286 free_args:
287         failsafe_args_free(dev);
288 free_subs:
289         fs_sub_device_free(dev);
290 free_dev:
291         /* mac_addrs must not be freed alone because part of dev_private */
292         dev->data->mac_addrs = NULL;
293         rte_eth_dev_release_port(dev);
294         return -1;
295 }
296
297 static int
298 fs_rte_eth_free(const char *name)
299 {
300         struct rte_eth_dev *dev;
301         int ret;
302
303         dev = rte_eth_dev_allocated(name);
304         if (dev == NULL)
305                 return -ENODEV;
306         rte_eth_dev_callback_unregister(RTE_ETH_ALL, RTE_ETH_EVENT_NEW,
307                                         failsafe_eth_new_event_callback, dev);
308         ret = failsafe_eal_uninit(dev);
309         if (ret)
310                 ERROR("Error while uninitializing sub-EAL");
311         failsafe_args_free(dev);
312         fs_sub_device_free(dev);
313         ret = pthread_mutex_destroy(&PRIV(dev)->hotplug_mutex);
314         if (ret)
315                 ERROR("Error while destroying hotplug mutex");
316         rte_free(PRIV(dev)->mcast_addrs);
317         /* mac_addrs must not be freed alone because part of dev_private */
318         dev->data->mac_addrs = NULL;
319         rte_eth_dev_release_port(dev);
320         return ret;
321 }
322
323 static bool
324 devargs_already_listed(struct rte_devargs *devargs)
325 {
326         struct rte_devargs *list_da;
327
328         RTE_EAL_DEVARGS_FOREACH(devargs->bus->name, list_da) {
329                 if (strcmp(list_da->name, devargs->name) == 0)
330                         /* devargs already in the list */
331                         return true;
332         }
333         return false;
334 }
335
336 static int
337 rte_pmd_failsafe_probe(struct rte_vdev_device *vdev)
338 {
339         const char *name;
340         struct rte_eth_dev *eth_dev;
341         struct sub_device  *sdev;
342         struct rte_devargs devargs;
343         uint8_t i;
344         int ret;
345
346         name = rte_vdev_device_name(vdev);
347         INFO("Initializing " FAILSAFE_DRIVER_NAME " for %s",
348                         name);
349
350         if (rte_eal_process_type() == RTE_PROC_SECONDARY &&
351             strlen(rte_vdev_device_args(vdev)) == 0) {
352                 eth_dev = rte_eth_dev_attach_secondary(name);
353                 if (!eth_dev) {
354                         ERROR("Failed to probe %s", name);
355                         return -1;
356                 }
357                 eth_dev->dev_ops = &failsafe_ops;
358                 eth_dev->device = &vdev->device;
359                 eth_dev->rx_pkt_burst = (eth_rx_burst_t)&failsafe_rx_burst;
360                 eth_dev->tx_pkt_burst = (eth_tx_burst_t)&failsafe_tx_burst;
361                 /*
362                  * Failsafe will attempt to probe all of its sub-devices.
363                  * Any failure in sub-devices is not a fatal error.
364                  * A sub-device can be plugged later.
365                  */
366                 FOREACH_SUBDEV(sdev, i, eth_dev) {
367                         /* skip empty devargs */
368                         if (sdev->devargs.name[0] == '\0')
369                                 continue;
370
371                         /* rebuild devargs to be able to get the bus name. */
372                         ret = rte_devargs_parse(&devargs,
373                                                 sdev->devargs.name);
374                         if (ret != 0) {
375                                 ERROR("Failed to parse devargs %s",
376                                         devargs.name);
377                                 continue;
378                         }
379                         if (!devargs_already_listed(&devargs)) {
380                                 ret = rte_dev_probe(devargs.name);
381                                 if (ret < 0) {
382                                         ERROR("Failed to probe devargs %s",
383                                               devargs.name);
384                                         continue;
385                                 }
386                         }
387                 }
388                 rte_eth_dev_probing_finish(eth_dev);
389                 return 0;
390         }
391
392         return fs_eth_dev_create(vdev);
393 }
394
395 static int
396 rte_pmd_failsafe_remove(struct rte_vdev_device *vdev)
397 {
398         const char *name;
399
400         name = rte_vdev_device_name(vdev);
401         INFO("Uninitializing " FAILSAFE_DRIVER_NAME " for %s", name);
402         return fs_rte_eth_free(name);
403 }
404
405 static struct rte_vdev_driver failsafe_drv = {
406         .probe = rte_pmd_failsafe_probe,
407         .remove = rte_pmd_failsafe_remove,
408 };
409
410 RTE_PMD_REGISTER_VDEV(net_failsafe, failsafe_drv);
411 RTE_PMD_REGISTER_PARAM_STRING(net_failsafe, PMD_FAILSAFE_PARAM_STRING);
412
413 RTE_INIT(failsafe_init_log)
414 {
415         failsafe_logtype = rte_log_register("pmd.net.failsafe");
416         if (failsafe_logtype >= 0)
417                 rte_log_set_level(failsafe_logtype, RTE_LOG_NOTICE);
418 }