1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2018 Intel Corporation
3 */
4
5 #include <stdlib.h>
6 #include <string.h>
7 #include <unistd.h>
8 #include <signal.h>
9 #include <sys/socket.h>
10 #include <linux/netlink.h>
11
12 #include <rte_string_fns.h>
13 #include <rte_log.h>
14 #include <rte_dev.h>
15 #include <rte_interrupts.h>
16 #include <rte_alarm.h>
17 #include <bus_driver.h>
18 #include <rte_spinlock.h>
19 #include <rte_errno.h>
20
21 #include "eal_private.h"
22
23 static struct rte_intr_handle *intr_handle;
24 static rte_rwlock_t monitor_lock = RTE_RWLOCK_INITIALIZER;
25 static uint32_t monitor_refcount;
26 static bool hotplug_handle;
27
28 #define EAL_UEV_MSG_LEN 4096
29 #define EAL_UEV_MSG_ELEM_LEN 128
30
31 /*
32 * spinlock for device hot-unplug failure handling. If it try to access bus or
33 * device, such as handle sigbus on bus or handle memory failure for device
34 * just need to use this lock. It could protect the bus and the device to avoid
35 * race condition.
36 */
37 static rte_spinlock_t failure_handle_lock = RTE_SPINLOCK_INITIALIZER;
38
39 static struct sigaction sigbus_action_old;
40
41 static int sigbus_need_recover;
42
43 static void dev_uev_handler(__rte_unused void *param);
44
45 /* identify the system layer which reports this event. */
46 enum eal_dev_event_subsystem {
47 EAL_DEV_EVENT_SUBSYSTEM_PCI, /* PCI bus device event */
48 EAL_DEV_EVENT_SUBSYSTEM_UIO, /* UIO driver device event */
49 EAL_DEV_EVENT_SUBSYSTEM_VFIO, /* VFIO driver device event */
50 EAL_DEV_EVENT_SUBSYSTEM_MAX
51 };
52
53 static void
sigbus_action_recover(void)54 sigbus_action_recover(void)
55 {
56 if (sigbus_need_recover) {
57 sigaction(SIGBUS, &sigbus_action_old, NULL);
58 sigbus_need_recover = 0;
59 }
60 }
61
sigbus_handler(int signum,siginfo_t * info,void * ctx __rte_unused)62 static void sigbus_handler(int signum, siginfo_t *info,
63 void *ctx __rte_unused)
64 {
65 int ret;
66
67 EAL_LOG(DEBUG, "Thread catch SIGBUS, fault address:%p",
68 info->si_addr);
69
70 rte_spinlock_lock(&failure_handle_lock);
71 ret = rte_bus_sigbus_handler(info->si_addr);
72 rte_spinlock_unlock(&failure_handle_lock);
73 if (ret == -1) {
74 rte_exit(EXIT_FAILURE,
75 "Failed to handle SIGBUS for hot-unplug, "
76 "(rte_errno: %s)!", strerror(rte_errno));
77 } else if (ret == 1) {
78 if (sigbus_action_old.sa_flags == SA_SIGINFO
79 && sigbus_action_old.sa_sigaction) {
80 (*(sigbus_action_old.sa_sigaction))(signum,
81 info, ctx);
82 } else if (sigbus_action_old.sa_flags != SA_SIGINFO
83 && sigbus_action_old.sa_handler) {
84 (*(sigbus_action_old.sa_handler))(signum);
85 } else {
86 rte_exit(EXIT_FAILURE,
87 "Failed to handle generic SIGBUS!");
88 }
89 }
90
91 EAL_LOG(DEBUG, "Success to handle SIGBUS for hot-unplug!");
92 }
93
cmp_dev_name(const struct rte_device * dev,const void * _name)94 static int cmp_dev_name(const struct rte_device *dev,
95 const void *_name)
96 {
97 const char *name = _name;
98
99 return strcmp(dev->name, name);
100 }
101
102 static int
dev_uev_socket_fd_create(void)103 dev_uev_socket_fd_create(void)
104 {
105 struct sockaddr_nl addr;
106 int ret, fd;
107
108 fd = socket(PF_NETLINK, SOCK_RAW | SOCK_CLOEXEC | SOCK_NONBLOCK,
109 NETLINK_KOBJECT_UEVENT);
110 if (fd < 0) {
111 EAL_LOG(ERR, "create uevent fd failed.");
112 return -1;
113 }
114
115 memset(&addr, 0, sizeof(addr));
116 addr.nl_family = AF_NETLINK;
117 addr.nl_pid = 0;
118 addr.nl_groups = 0xffffffff;
119
120 ret = bind(fd, (struct sockaddr *) &addr, sizeof(addr));
121 if (ret < 0) {
122 EAL_LOG(ERR, "Failed to bind uevent socket.");
123 goto err;
124 }
125
126 if (rte_intr_fd_set(intr_handle, fd))
127 goto err;
128
129 return 0;
130 err:
131 close(fd);
132 fd = -1;
133 return ret;
134 }
135
136 struct rte_dev_event {
137 enum rte_dev_event_type type; /**< device event type */
138 int subsystem; /**< subsystem id */
139 char *devname; /**< device name */
140 };
141
142 static int
dev_uev_parse(const char * buf,struct rte_dev_event * event,int length)143 dev_uev_parse(const char *buf, struct rte_dev_event *event, int length)
144 {
145 char action[EAL_UEV_MSG_ELEM_LEN];
146 char subsystem[EAL_UEV_MSG_ELEM_LEN];
147 char pci_slot_name[EAL_UEV_MSG_ELEM_LEN];
148 int i = 0;
149
150 memset(action, 0, EAL_UEV_MSG_ELEM_LEN);
151 memset(subsystem, 0, EAL_UEV_MSG_ELEM_LEN);
152 memset(pci_slot_name, 0, EAL_UEV_MSG_ELEM_LEN);
153
154 while (i < length) {
155 for (; i < length; i++) {
156 if (*buf)
157 break;
158 buf++;
159 }
160 if (i >= length)
161 break;
162
163 /**
164 * check device uevent from kernel side, no need to check
165 * uevent from udev.
166 */
167 if (!strncmp(buf, "libudev", 7)) {
168 buf += 7;
169 i += 7;
170 return -1;
171 }
172 if (!strncmp(buf, "ACTION=", 7)) {
173 buf += 7;
174 i += 7;
175 strlcpy(action, buf, sizeof(action));
176 } else if (!strncmp(buf, "SUBSYSTEM=", 10)) {
177 buf += 10;
178 i += 10;
179 strlcpy(subsystem, buf, sizeof(subsystem));
180 } else if (!strncmp(buf, "PCI_SLOT_NAME=", 14)) {
181 buf += 14;
182 i += 14;
183 strlcpy(pci_slot_name, buf, sizeof(subsystem));
184 event->devname = strdup(pci_slot_name);
185 if (event->devname == NULL)
186 return -1;
187 }
188 for (; i < length; i++) {
189 if (*buf == '\0')
190 break;
191 buf++;
192 }
193 }
194
195 /* parse the subsystem layer */
196 if (!strncmp(subsystem, "uio", 3))
197 event->subsystem = EAL_DEV_EVENT_SUBSYSTEM_UIO;
198 else if (!strncmp(subsystem, "pci", 3))
199 event->subsystem = EAL_DEV_EVENT_SUBSYSTEM_PCI;
200 else if (!strncmp(subsystem, "vfio", 4))
201 event->subsystem = EAL_DEV_EVENT_SUBSYSTEM_VFIO;
202 else
203 goto err;
204
205 /* parse the action type */
206 if (!strncmp(action, "add", 3))
207 event->type = RTE_DEV_EVENT_ADD;
208 else if (!strncmp(action, "remove", 6))
209 event->type = RTE_DEV_EVENT_REMOVE;
210 else
211 goto err;
212 return 0;
213 err:
214 free(event->devname);
215 return -1;
216 }
217
218 static void
dev_delayed_unregister(void * param)219 dev_delayed_unregister(void *param)
220 {
221 rte_intr_callback_unregister(intr_handle, dev_uev_handler, param);
222 if (rte_intr_fd_get(intr_handle) >= 0) {
223 close(rte_intr_fd_get(intr_handle));
224 rte_intr_fd_set(intr_handle, -1);
225 }
226 }
227
228 static void
dev_uev_handler(__rte_unused void * param)229 dev_uev_handler(__rte_unused void *param)
230 {
231 struct rte_dev_event uevent;
232 int ret;
233 char buf[EAL_UEV_MSG_LEN + 1];
234 struct rte_bus *bus;
235 struct rte_device *dev;
236 const char *busname = "";
237
238 memset(&uevent, 0, sizeof(struct rte_dev_event));
239 memset(buf, 0, EAL_UEV_MSG_LEN + 1);
240
241 if (rte_intr_fd_get(intr_handle) < 0)
242 return;
243
244 ret = recv(rte_intr_fd_get(intr_handle), buf, EAL_UEV_MSG_LEN,
245 MSG_DONTWAIT);
246 if (ret < 0 && errno == EAGAIN)
247 return;
248 else if (ret <= 0) {
249 /* connection is closed or broken, can not up again. */
250 EAL_LOG(ERR, "uevent socket connection is broken.");
251 rte_eal_alarm_set(1, dev_delayed_unregister, NULL);
252 return;
253 }
254
255 ret = dev_uev_parse(buf, &uevent, EAL_UEV_MSG_LEN);
256 if (ret < 0) {
257 EAL_LOG(DEBUG, "Ignoring uevent '%s'", buf);
258 return;
259 }
260
261 EAL_LOG(DEBUG, "receive uevent(name:%s, type:%d, subsystem:%d)",
262 uevent.devname, uevent.type, uevent.subsystem);
263
264 switch (uevent.subsystem) {
265 case EAL_DEV_EVENT_SUBSYSTEM_PCI:
266 case EAL_DEV_EVENT_SUBSYSTEM_UIO:
267 busname = "pci";
268 break;
269 default:
270 break;
271 }
272
273 if (uevent.devname) {
274 if (uevent.type == RTE_DEV_EVENT_REMOVE && hotplug_handle) {
275 rte_spinlock_lock(&failure_handle_lock);
276 bus = rte_bus_find_by_name(busname);
277 if (bus == NULL) {
278 EAL_LOG(ERR, "Cannot find bus (%s)",
279 busname);
280 goto failure_handle_err;
281 }
282
283 dev = bus->find_device(NULL, cmp_dev_name,
284 uevent.devname);
285 if (dev == NULL) {
286 EAL_LOG(ERR, "Cannot find device (%s) on "
287 "bus (%s)", uevent.devname, busname);
288 goto failure_handle_err;
289 }
290
291 ret = bus->hot_unplug_handler(dev);
292 if (ret) {
293 EAL_LOG(ERR, "Can not handle hot-unplug "
294 "for device (%s)", dev->name);
295 }
296 rte_spinlock_unlock(&failure_handle_lock);
297 }
298 rte_dev_event_callback_process(uevent.devname, uevent.type);
299 free(uevent.devname);
300 }
301
302 return;
303
304 failure_handle_err:
305 rte_spinlock_unlock(&failure_handle_lock);
306 free(uevent.devname);
307 }
308
309 int
rte_dev_event_monitor_start(void)310 rte_dev_event_monitor_start(void)
311 {
312 int ret = 0;
313
314 rte_rwlock_write_lock(&monitor_lock);
315
316 if (monitor_refcount) {
317 monitor_refcount++;
318 goto exit;
319 }
320
321 intr_handle = rte_intr_instance_alloc(RTE_INTR_INSTANCE_F_PRIVATE);
322 if (intr_handle == NULL) {
323 EAL_LOG(ERR, "Fail to allocate intr_handle");
324 goto exit;
325 }
326
327 ret = rte_intr_type_set(intr_handle, RTE_INTR_HANDLE_DEV_EVENT);
328 if (ret)
329 goto exit;
330
331 ret = rte_intr_fd_set(intr_handle, -1);
332 if (ret)
333 goto exit;
334
335 ret = dev_uev_socket_fd_create();
336 if (ret) {
337 EAL_LOG(ERR, "error create device event fd.");
338 goto exit;
339 }
340
341 ret = rte_intr_callback_register(intr_handle, dev_uev_handler, NULL);
342
343 if (ret) {
344 close(rte_intr_fd_get(intr_handle));
345 goto exit;
346 }
347
348 monitor_refcount++;
349
350 exit:
351 if (ret) {
352 rte_intr_instance_free(intr_handle);
353 intr_handle = NULL;
354 }
355 rte_rwlock_write_unlock(&monitor_lock);
356 return ret;
357 }
358
359 int
rte_dev_event_monitor_stop(void)360 rte_dev_event_monitor_stop(void)
361 {
362 int ret = 0;
363
364 rte_rwlock_write_lock(&monitor_lock);
365
366 if (!monitor_refcount) {
367 EAL_LOG(ERR, "device event monitor already stopped");
368 goto exit;
369 }
370
371 if (monitor_refcount > 1) {
372 monitor_refcount--;
373 goto exit;
374 }
375
376 ret = rte_intr_callback_unregister(intr_handle, dev_uev_handler,
377 (void *)-1);
378 if (ret < 0) {
379 EAL_LOG(ERR, "fail to unregister uevent callback.");
380 goto exit;
381 }
382
383 close(rte_intr_fd_get(intr_handle));
384 rte_intr_instance_free(intr_handle);
385 intr_handle = NULL;
386 ret = 0;
387
388 monitor_refcount--;
389
390 exit:
391 rte_rwlock_write_unlock(&monitor_lock);
392
393 return ret;
394 }
395
396 static int
dev_sigbus_handler_register(void)397 dev_sigbus_handler_register(void)
398 {
399 sigset_t mask;
400 struct sigaction action;
401
402 rte_errno = 0;
403
404 if (sigbus_need_recover)
405 return 0;
406
407 sigemptyset(&mask);
408 sigaddset(&mask, SIGBUS);
409 action.sa_flags = SA_SIGINFO;
410 action.sa_mask = mask;
411 action.sa_sigaction = sigbus_handler;
412 sigbus_need_recover = !sigaction(SIGBUS, &action, &sigbus_action_old);
413
414 return rte_errno;
415 }
416
417 static int
dev_sigbus_handler_unregister(void)418 dev_sigbus_handler_unregister(void)
419 {
420 rte_errno = 0;
421
422 sigbus_action_recover();
423
424 return rte_errno;
425 }
426
427 int
rte_dev_hotplug_handle_enable(void)428 rte_dev_hotplug_handle_enable(void)
429 {
430 int ret = 0;
431
432 ret = dev_sigbus_handler_register();
433 if (ret < 0)
434 EAL_LOG(ERR,
435 "fail to register sigbus handler for devices.");
436
437 hotplug_handle = true;
438
439 return ret;
440 }
441
442 int
rte_dev_hotplug_handle_disable(void)443 rte_dev_hotplug_handle_disable(void)
444 {
445 int ret = 0;
446
447 ret = dev_sigbus_handler_unregister();
448 if (ret < 0)
449 EAL_LOG(ERR,
450 "fail to unregister sigbus handler for devices.");
451
452 hotplug_handle = false;
453
454 return ret;
455 }
456