xref: /freebsd/sys/dev/watchdog/watchdog.c (revision 5b63afc09a86318fa8bf09a786a1b084e000cc51)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 2004 Poul-Henning Kamp
5  * Copyright (c) 2013 iXsystems.com,
6  *               author: Alfred Perlstein <alfred@freebsd.org>
7  *
8  * All rights reserved.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer
15  *    in this position and unchanged.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE AUTHORS ``AS IS'' AND ANY EXPRESS OR
21  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
22  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
23  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
24  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
25  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
26  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
27  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
28  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
29  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
30  *
31  */
32 
33 #include "opt_ddb.h"
34 
35 #include <sys/param.h>
36 #include <sys/bus.h>
37 #include <sys/conf.h>
38 #include <sys/eventhandler.h>
39 #include <sys/kdb.h>
40 #include <sys/kernel.h>
41 #include <sys/malloc.h>
42 #include <sys/module.h>
43 #include <sys/mutex.h>
44 #include <sys/sysctl.h>
45 #include <sys/syslog.h>
46 #include <sys/systm.h>
47 #include <sys/uio.h>
48 #include <sys/watchdog.h>
49 #include <machine/bus.h>
50 
51 #include <sys/syscallsubr.h> /* kern_clock_gettime() */
52 
53 #ifdef	COMPAT_FREEBSD14
54 #define WDIOCPATPAT_14	_IOW('W', 42, u_int)	/* pat the watchdog */
55 #define WDIOC_SETTIMEOUT_14   _IOW('W', 43, int)	/* set/reset the timer */
56 #define WDIOC_GETTIMEOUT_14    _IOR('W', 44, int)	/* get total timeout */
57 #define WDIOC_GETTIMELEFT_14   _IOR('W', 45, int)	/* get time left */
58 #define WDIOC_GETPRETIMEOUT_14 _IOR('W', 46, int)	/* get the pre-timeout */
59 #define WDIOC_SETPRETIMEOUT_14 _IOW('W', 47, int)	/* set the pre-timeout */
60 #endif
61 
62 static int wd_set_pretimeout(sbintime_t newtimeout, int disableiftoolong);
63 static void wd_timeout_cb(void *arg);
64 
65 static struct callout wd_pretimeo_handle;
66 static sbintime_t wd_pretimeout;
67 static int wd_pretimeout_act = WD_SOFT_LOG;
68 
69 static struct callout wd_softtimeo_handle;
70 static int wd_softtimer;	/* true = use softtimer instead of hardware
71 				   watchdog */
72 static int wd_softtimeout_act = WD_SOFT_LOG;	/* action for the software timeout */
73 
74 static struct cdev *wd_dev;
75 static volatile sbintime_t wd_last_sbt;	/* last timeout value (sbt) */
76 static sbintime_t wd_last_sbt_sysctl;	/* last timeout value (sbt) */
77 static volatile u_int wd_last_u;    /* last timeout value set by kern_do_pat */
78 static u_int wd_last_u_sysctl;    /* last timeout value set by kern_do_pat */
79 static u_int wd_last_u_sysctl_secs;    /* wd_last_u in seconds */
80 
81 SYSCTL_NODE(_hw, OID_AUTO, watchdog, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
82     "Main watchdog device");
83 SYSCTL_UINT(_hw_watchdog, OID_AUTO, wd_last_u, CTLFLAG_RD,
84     &wd_last_u_sysctl, 0, "Watchdog last update time");
85 SYSCTL_UINT(_hw_watchdog, OID_AUTO, wd_last_u_secs, CTLFLAG_RD,
86     &wd_last_u_sysctl_secs, 0, "Watchdog last update time");
87 SYSCTL_SBINTIME_MSEC(_hw_watchdog, OID_AUTO, wd_last_msecs, CTLFLAG_RD,
88     &wd_last_sbt_sysctl, "Watchdog last update time (milliseconds)");
89 
90 static int wd_lastpat_valid = 0;
91 static time_t wd_lastpat = 0;	/* when the watchdog was last patted */
92 
93 /* Hook for external software watchdog to register for use if needed */
94 void (*wdog_software_attach)(void);
95 
96 /* Legacy interface to watchdog. */
97 int
wdog_kern_pat(u_int utim)98 wdog_kern_pat(u_int utim)
99 {
100 	sbintime_t sbt;
101 
102 	if ((utim & WD_LASTVAL) != 0 && (utim & WD_INTERVAL) > 0)
103 		return (EINVAL);
104 
105 	if ((utim & WD_LASTVAL) != 0) {
106 		return (wdog_control(WD_CTRL_RESET));
107 	}
108 
109 	utim &= WD_INTERVAL;
110 	if (utim == WD_TO_NEVER)
111 		sbt = 0;
112 	else
113 		sbt = nstosbt(1 << utim);
114 
115 	return (wdog_kern_pat_sbt(sbt));
116 }
117 
118 int
wdog_control(int ctrl)119 wdog_control(int ctrl)
120 {
121 	/* Disable takes precedence */
122 	if (ctrl == WD_CTRL_DISABLE) {
123 		wdog_kern_pat(0);
124 	}
125 
126 	if ((ctrl & WD_CTRL_RESET) != 0) {
127 		wdog_kern_pat_sbt(wd_last_sbt);
128 	} else if ((ctrl & WD_CTRL_ENABLE) != 0) {
129 		wdog_kern_pat_sbt(wd_last_sbt);
130 	}
131 
132 	return (0);
133 }
134 
135 int
wdog_kern_pat_sbt(sbintime_t sbt)136 wdog_kern_pat_sbt(sbintime_t sbt)
137 {
138 	sbintime_t error_sbt = 0;
139 	int pow2ns = 0;
140 	int error = 0;
141 	static bool first = true;
142 
143 	/* legacy uses power-of-2-nanoseconds time. */
144 	if (sbt != 0) {
145 		pow2ns = flsl(sbttons(sbt));
146 	}
147 	if (wd_last_sbt != sbt) {
148 		wd_last_u = pow2ns;
149 		wd_last_u_sysctl = wd_last_u;
150 		wd_last_u_sysctl_secs = sbt / SBT_1S;
151 
152 		wd_last_sbt = sbt;
153 	}
154 
155 	if (sbt != 0)
156 		error = EOPNOTSUPP;
157 
158 	if (wd_softtimer) {
159 		if (sbt == 0) {
160 			callout_stop(&wd_softtimeo_handle);
161 		} else {
162 			(void) callout_reset_sbt(&wd_softtimeo_handle,
163 			    sbt, 0, wd_timeout_cb, "soft", 0);
164 		}
165 		error = 0;
166 	} else {
167 		EVENTHANDLER_INVOKE(watchdog_sbt_list, sbt, &error_sbt, &error);
168 		EVENTHANDLER_INVOKE(watchdog_list, pow2ns, &error);
169 	}
170 	/*
171 	 * If no hardware watchdog responded, we have not tried to
172 	 * attach an external software watchdog, and one is available,
173 	 * attach it now and retry.
174 	 */
175 	if (error == EOPNOTSUPP && first && wdog_software_attach != NULL) {
176 		(*wdog_software_attach)();
177 		EVENTHANDLER_INVOKE(watchdog_sbt_list, sbt, &error_sbt, &error);
178 		EVENTHANDLER_INVOKE(watchdog_list, pow2ns, &error);
179 	}
180 	first = false;
181 
182 	/* TODO: Print a (rate limited?) warning if error_sbt is too far away */
183 	wd_set_pretimeout(wd_pretimeout, true);
184 	if (!error) {
185 		struct timespec ts;
186 
187 		error = kern_clock_gettime(curthread /* XXX */,
188 		    CLOCK_MONOTONIC_FAST, &ts);
189 		if (!error) {
190 			wd_lastpat = ts.tv_sec;
191 			wd_lastpat_valid = 1;
192 		}
193 	}
194 
195 	return (error);
196 }
197 
198 static int
wd_valid_act(int act)199 wd_valid_act(int act)
200 {
201 
202 	if ((act & ~(WD_SOFT_MASK)) != 0)
203 		return false;
204 	return true;
205 }
206 
207 #ifdef	COMPAT_FREEBSD14
208 static int
wd_ioctl_patpat(caddr_t data)209 wd_ioctl_patpat(caddr_t data)
210 {
211 	u_int u;
212 
213 	u = *(u_int *)data;
214 	if (u & ~(WD_ACTIVE | WD_PASSIVE | WD_LASTVAL | WD_INTERVAL))
215 		return (EINVAL);
216 	if ((u & (WD_ACTIVE | WD_PASSIVE)) == (WD_ACTIVE | WD_PASSIVE))
217 		return (EINVAL);
218 	if ((u & (WD_ACTIVE | WD_PASSIVE)) == 0 && ((u & WD_INTERVAL) > 0 ||
219 	    (u & WD_LASTVAL) != 0))
220 		return (EINVAL);
221 	if (u & WD_PASSIVE)
222 		return (ENOSYS);	/* XXX Not implemented yet */
223 	u &= ~(WD_ACTIVE | WD_PASSIVE);
224 
225 	return (wdog_kern_pat(u));
226 }
227 #endif
228 
229 static int
wd_get_time_left(struct thread * td,time_t * remainp)230 wd_get_time_left(struct thread *td, time_t *remainp)
231 {
232 	struct timespec ts;
233 	int error;
234 
235 	error = kern_clock_gettime(td, CLOCK_MONOTONIC_FAST, &ts);
236 	if (error)
237 		return (error);
238 	if (!wd_lastpat_valid)
239 		return (ENOENT);
240 	*remainp = ts.tv_sec - wd_lastpat;
241 	return (0);
242 }
243 
244 static void
wd_timeout_cb(void * arg)245 wd_timeout_cb(void *arg)
246 {
247 	const char *type = arg;
248 
249 #ifdef DDB
250 	if ((wd_pretimeout_act & WD_SOFT_DDB)) {
251 		char kdb_why[80];
252 		snprintf(kdb_why, sizeof(kdb_why), "watchdog %s-timeout", type);
253 		kdb_backtrace();
254 		kdb_enter(KDB_WHY_WATCHDOG, kdb_why);
255 	}
256 #endif
257 	if ((wd_pretimeout_act & WD_SOFT_LOG))
258 		log(LOG_EMERG, "watchdog %s-timeout, WD_SOFT_LOG\n", type);
259 	if ((wd_pretimeout_act & WD_SOFT_PRINTF))
260 		printf("watchdog %s-timeout, WD_SOFT_PRINTF\n", type);
261 	if ((wd_pretimeout_act & WD_SOFT_PANIC))
262 		panic("watchdog %s-timeout, WD_SOFT_PANIC set", type);
263 }
264 
265 /*
266  * Called to manage timeouts.
267  * newtimeout needs to be in the range of 0 to actual watchdog timeout.
268  * if 0, we disable the pre-timeout.
269  * otherwise we set the pre-timeout provided it's not greater than the
270  * current actual watchdog timeout.
271  */
272 static int
wd_set_pretimeout(sbintime_t newtimeout,int disableiftoolong)273 wd_set_pretimeout(sbintime_t newtimeout, int disableiftoolong)
274 {
275 	sbintime_t utime;
276 	sbintime_t timeout_left;
277 
278 	utime = wdog_kern_last_timeout_sbt();
279 	/* do not permit a pre-timeout >= than the timeout. */
280 	if (newtimeout >= utime) {
281 		/*
282 		 * If 'disableiftoolong' then just fall through
283 		 * so as to disable the pre-watchdog
284 		 */
285 		if (disableiftoolong)
286 			newtimeout = 0;
287 		else
288 			return EINVAL;
289 	}
290 
291 	/* disable the pre-timeout */
292 	if (newtimeout == 0) {
293 		wd_pretimeout = 0;
294 		callout_stop(&wd_pretimeo_handle);
295 		return 0;
296 	}
297 
298 	timeout_left = utime - newtimeout;
299 #if 0
300 	printf("wd_set_pretimeout: "
301 	    "newtimeout: %d, "
302 	    "utime: %d -> utime_ticks: %d, "
303 	    "hz*newtimeout: %d, "
304 	    "timeout_ticks: %d -> sec: %d\n",
305 	    newtimeout,
306 	    utime, pow2ns_to_ticks(utime),
307 	    hz*newtimeout,
308 	    timeout_ticks, timeout_ticks / hz);
309 #endif
310 
311 	/* We determined the value is sane, so reset the callout */
312 	(void) callout_reset_sbt(&wd_pretimeo_handle,
313 	    timeout_left, 0, wd_timeout_cb, "pre", 0);
314 	wd_pretimeout = newtimeout;
315 	return 0;
316 }
317 
318 static int
wd_ioctl(struct cdev * dev __unused,u_long cmd,caddr_t data,int flags __unused,struct thread * td)319 wd_ioctl(struct cdev *dev __unused, u_long cmd, caddr_t data,
320     int flags __unused, struct thread *td)
321 {
322 	sbintime_t sb;
323 	u_int u;
324 	time_t timeleft;
325 	int error;
326 
327 	error = 0;
328 
329 	switch (cmd) {
330 	case WDIOC_SETSOFT:
331 		u = *(int *)data;
332 		/* do nothing? */
333 		if (u == wd_softtimer)
334 			break;
335 		/* If there is a pending timeout disallow this ioctl */
336 		if (wd_last_u != 0) {
337 			error = EINVAL;
338 			break;
339 		}
340 		wd_softtimer = u;
341 		break;
342 	case WDIOC_SETSOFTTIMEOUTACT:
343 		u = *(int *)data;
344 		if (wd_valid_act(u)) {
345 			wd_softtimeout_act = u;
346 		} else {
347 			error = EINVAL;
348 		}
349 		break;
350 	case WDIOC_SETPRETIMEOUTACT:
351 		u = *(int *)data;
352 		if (wd_valid_act(u)) {
353 			wd_pretimeout_act = u;
354 		} else {
355 			error = EINVAL;
356 		}
357 		break;
358 #ifdef	COMPAT_FREEBSD14
359 	case WDIOC_GETPRETIMEOUT_14:
360 		*(int *)data = (int)(wd_pretimeout / SBT_1S);
361 		break;
362 	case WDIOC_SETPRETIMEOUT_14:
363 		error = wd_set_pretimeout(*(int *)data * SBT_1S, false);
364 		break;
365 	case WDIOC_GETTIMELEFT_14:
366 		error = wd_get_time_left(td, &timeleft);
367 		if (error)
368 			break;
369 		*(int *)data = (int)timeleft;
370 		break;
371 	case WDIOC_SETTIMEOUT_14:
372 		u = *(u_int *)data;
373 		error = wdog_kern_pat_sbt(mstosbt(u * 1000ULL));
374 		break;
375 	case WDIOC_GETTIMEOUT_14:
376 		u = wdog_kern_last_timeout();
377 		*(u_int *)data = u;
378 		break;
379 	case WDIOCPATPAT_14:
380 		error = wd_ioctl_patpat(data);
381 		break;
382 #endif
383 
384 	/* New API */
385 	case WDIOC_CONTROL:
386 		wdog_control(*(int *)data);
387 		break;
388 	case WDIOC_SETTIMEOUT:
389 		sb = *(sbintime_t *)data;
390 		error = wdog_kern_pat_sbt(sb);
391 		break;
392 	case WDIOC_GETTIMEOUT:
393 		*(sbintime_t *)data = wdog_kern_last_timeout_sbt();
394 		break;
395 	case WDIOC_GETTIMELEFT:
396 		error = wd_get_time_left(td, &timeleft);
397 		if (error)
398 			break;
399 		*(sbintime_t *)data = (sbintime_t)timeleft * SBT_1S;
400 		break;
401 	case WDIOC_GETPRETIMEOUT:
402 		*(sbintime_t *)data = wd_pretimeout;
403 		break;
404 	case WDIOC_SETPRETIMEOUT:
405 		error = wd_set_pretimeout(*(sbintime_t *)data, false);
406 		break;
407 	default:
408 		error = ENOIOCTL;
409 		break;
410 	}
411 	return (error);
412 }
413 
414 /*
415  * Return the last timeout set, this is NOT the seconds from NOW until timeout,
416  * rather it is the amount of seconds passed to WDIOCPATPAT/WDIOC_SETTIMEOUT.
417  */
418 u_int
wdog_kern_last_timeout(void)419 wdog_kern_last_timeout(void)
420 {
421 
422 	return (wd_last_u);
423 }
424 
425 sbintime_t
wdog_kern_last_timeout_sbt(void)426 wdog_kern_last_timeout_sbt(void)
427 {
428 	return (wd_last_sbt);
429 }
430 
431 static struct cdevsw wd_cdevsw = {
432 	.d_version =	D_VERSION,
433 	.d_ioctl =	wd_ioctl,
434 	.d_name =	"watchdog",
435 };
436 
437 static int
watchdog_modevent(module_t mod __unused,int type,void * data __unused)438 watchdog_modevent(module_t mod __unused, int type, void *data __unused)
439 {
440 	switch(type) {
441 	case MOD_LOAD:
442 		callout_init(&wd_pretimeo_handle, 1);
443 		callout_init(&wd_softtimeo_handle, 1);
444 		wd_dev = make_dev(&wd_cdevsw, 0,
445 		    UID_ROOT, GID_WHEEL, 0600, _PATH_WATCHDOG);
446 		return 0;
447 	case MOD_UNLOAD:
448 		callout_stop(&wd_pretimeo_handle);
449 		callout_stop(&wd_softtimeo_handle);
450 		callout_drain(&wd_pretimeo_handle);
451 		callout_drain(&wd_softtimeo_handle);
452 		destroy_dev(wd_dev);
453 		return 0;
454 	case MOD_SHUTDOWN:
455 		return 0;
456 	default:
457 		return EOPNOTSUPP;
458 	}
459 }
460 
461 DEV_MODULE(watchdog, watchdog_modevent, NULL);
462