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