Home | History | Annotate | Line # | Download | only in sysmon
sysmon_wdog.c revision 1.18
      1 /*	$NetBSD: sysmon_wdog.c,v 1.18 2007/03/04 06:02:45 christos Exp $	*/
      2 
      3 /*-
      4  * Copyright (c) 2000 Zembu Labs, Inc.
      5  * All rights reserved.
      6  *
      7  * Author: Jason R. Thorpe <thorpej (at) zembu.com>
      8  *
      9  * Redistribution and use in source and binary forms, with or without
     10  * modification, are permitted provided that the following conditions
     11  * are met:
     12  * 1. Redistributions of source code must retain the above copyright
     13  *    notice, this list of conditions and the following disclaimer.
     14  * 2. Redistributions in binary form must reproduce the above copyright
     15  *    notice, this list of conditions and the following disclaimer in the
     16  *    documentation and/or other materials provided with the distribution.
     17  * 3. All advertising materials mentioning features or use of this software
     18  *    must display the following acknowledgement:
     19  *	This product includes software developed by Zembu Labs, Inc.
     20  * 4. Neither the name of Zembu Labs nor the names of its employees may
     21  *    be used to endorse or promote products derived from this software
     22  *    without specific prior written permission.
     23  *
     24  * THIS SOFTWARE IS PROVIDED BY ZEMBU LABS, INC. ``AS IS'' AND ANY EXPRESS
     25  * OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WAR-
     26  * RANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DIS-
     27  * CLAIMED.  IN NO EVENT SHALL ZEMBU LABS BE LIABLE FOR ANY DIRECT, INDIRECT,
     28  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
     29  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
     30  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
     31  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
     32  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
     33  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
     34  */
     35 
     36 /*
     37  * Watchdog timer framework for sysmon.  Hardware (and software)
     38  * watchdog timers can register themselves here to provide a
     39  * watchdog function, which provides an abstract interface to the
     40  * user.
     41  */
     42 
     43 #include <sys/cdefs.h>
     44 __KERNEL_RCSID(0, "$NetBSD: sysmon_wdog.c,v 1.18 2007/03/04 06:02:45 christos Exp $");
     45 
     46 #include <sys/param.h>
     47 #include <sys/conf.h>
     48 #include <sys/errno.h>
     49 #include <sys/fcntl.h>
     50 #include <sys/lock.h>
     51 #include <sys/callout.h>
     52 #include <sys/kernel.h>
     53 #include <sys/systm.h>
     54 #include <sys/proc.h>
     55 
     56 #include <dev/sysmon/sysmonvar.h>
     57 
     58 LIST_HEAD(, sysmon_wdog) sysmon_wdog_list =
     59     LIST_HEAD_INITIALIZER(&sysmon_wdog_list);
     60 int sysmon_wdog_count;
     61 struct simplelock sysmon_wdog_list_slock = SIMPLELOCK_INITIALIZER;
     62 
     63 struct simplelock sysmon_wdog_slock = SIMPLELOCK_INITIALIZER;
     64 struct sysmon_wdog *sysmon_armed_wdog;
     65 struct callout sysmon_wdog_callout = CALLOUT_INITIALIZER;
     66 void *sysmon_wdog_sdhook;
     67 
     68 #define	SYSMON_WDOG_LOCK(s)						\
     69 do {									\
     70 	s = splsoftclock();						\
     71 	simple_lock(&sysmon_wdog_slock);				\
     72 } while (0)
     73 
     74 #define	SYSMON_WDOG_UNLOCK(s)						\
     75 do {									\
     76 	simple_unlock(&sysmon_wdog_slock);				\
     77 	splx(s);							\
     78 } while (0)
     79 
     80 struct sysmon_wdog *sysmon_wdog_find(const char *);
     81 void	sysmon_wdog_release(struct sysmon_wdog *);
     82 int	sysmon_wdog_setmode(struct sysmon_wdog *, int, u_int);
     83 void	sysmon_wdog_ktickle(void *);
     84 void	sysmon_wdog_shutdown(void *);
     85 
     86 /*
     87  * sysmonopen_wdog:
     88  *
     89  *	Open the system monitor device.
     90  */
     91 int
     92 sysmonopen_wdog(dev_t dev, int flag, int mode,
     93     struct lwp *l)
     94 {
     95 
     96 	simple_lock(&sysmon_wdog_list_slock);
     97 	if (sysmon_wdog_sdhook == NULL) {
     98 		sysmon_wdog_sdhook =
     99 		    shutdownhook_establish(sysmon_wdog_shutdown, NULL);
    100 		if (sysmon_wdog_sdhook == NULL)
    101 			printf("WARNING: unable to register watchdog "
    102 			    "shutdown hook\n");
    103 	}
    104 	simple_unlock(&sysmon_wdog_list_slock);
    105 
    106 	return (0);
    107 }
    108 
    109 /*
    110  * sysmonclose_wdog:
    111  *
    112  *	Close the system monitor device.
    113  */
    114 int
    115 sysmonclose_wdog(dev_t dev, int flag, int mode,
    116     struct lwp *l)
    117 {
    118 	struct sysmon_wdog *smw;
    119 	int s, error = 0;
    120 
    121 	/*
    122 	 * If this is the last close, and there is a watchdog
    123 	 * running in UTICKLE mode, we need to disable it,
    124 	 * otherwise the system will reset in short order.
    125 	 *
    126 	 * XXX Maybe we should just go into KTICKLE mode?
    127 	 */
    128 	SYSMON_WDOG_LOCK(s);
    129 	if ((smw = sysmon_armed_wdog) != NULL) {
    130 		if ((smw->smw_mode & WDOG_MODE_MASK) == WDOG_MODE_UTICKLE) {
    131 			error = sysmon_wdog_setmode(smw,
    132 			    WDOG_MODE_DISARMED, smw->smw_period);
    133 			if (error) {
    134 				printf("WARNING: UNABLE TO DISARM "
    135 				    "WATCHDOG %s ON CLOSE!\n",
    136 				    smw->smw_name);
    137 				/*
    138 				 * ...we will probably reboot soon.
    139 				 */
    140 			}
    141 		}
    142 	}
    143 	SYSMON_WDOG_UNLOCK(s);
    144 
    145 	return (error);
    146 }
    147 
    148 /*
    149  * sysmonioctl_wdog:
    150  *
    151  *	Perform a watchdog control request.
    152  */
    153 int
    154 sysmonioctl_wdog(dev_t dev, u_long cmd, void *data, int flag,
    155     struct lwp *l)
    156 {
    157 	struct sysmon_wdog *smw;
    158 	int s, error = 0;
    159 
    160 	switch (cmd) {
    161 	case WDOGIOC_GMODE:
    162 	    {
    163 		struct wdog_mode *wm = (void *) data;
    164 
    165 		wm->wm_name[sizeof(wm->wm_name) - 1] = '\0';
    166 		smw = sysmon_wdog_find(wm->wm_name);
    167 		if (smw == NULL) {
    168 			error = ESRCH;
    169 			break;
    170 		}
    171 
    172 		wm->wm_mode = smw->smw_mode;
    173 		wm->wm_period = smw->smw_period;
    174 		sysmon_wdog_release(smw);
    175 		break;
    176 	    }
    177 
    178 	case WDOGIOC_SMODE:
    179 	    {
    180 		struct wdog_mode *wm = (void *) data;
    181 
    182 		if ((flag & FWRITE) == 0) {
    183 			error = EPERM;
    184 			break;
    185 		}
    186 
    187 		wm->wm_name[sizeof(wm->wm_name) - 1] = '\0';
    188 		smw = sysmon_wdog_find(wm->wm_name);
    189 		if (smw == NULL) {
    190 			error = ESRCH;
    191 			break;
    192 		}
    193 
    194 		if (wm->wm_mode & ~(WDOG_MODE_MASK|WDOG_FEATURE_MASK))
    195 			error = EINVAL;
    196 		else {
    197 			SYSMON_WDOG_LOCK(s);
    198 			error = sysmon_wdog_setmode(smw, wm->wm_mode,
    199 			    wm->wm_period);
    200 			SYSMON_WDOG_UNLOCK(s);
    201 		}
    202 
    203 		sysmon_wdog_release(smw);
    204 		break;
    205 	    }
    206 
    207 	case WDOGIOC_WHICH:
    208 	    {
    209 		struct wdog_mode *wm = (void *) data;
    210 
    211 		SYSMON_WDOG_LOCK(s);
    212 		if ((smw = sysmon_armed_wdog) != NULL) {
    213 			strcpy(wm->wm_name, smw->smw_name);
    214 			wm->wm_mode = smw->smw_mode;
    215 			wm->wm_period = smw->smw_period;
    216 		} else
    217 			error = ESRCH;
    218 		SYSMON_WDOG_UNLOCK(s);
    219 		break;
    220 	    }
    221 
    222 	case WDOGIOC_TICKLE:
    223 		if ((flag & FWRITE) == 0) {
    224 			error = EPERM;
    225 			break;
    226 		}
    227 
    228 		SYSMON_WDOG_LOCK(s);
    229 		if ((smw = sysmon_armed_wdog) != NULL) {
    230 			error = (*smw->smw_tickle)(smw);
    231 			if (error == 0)
    232 				smw->smw_tickler = l->l_proc->p_pid;
    233 		} else
    234 			error = ESRCH;
    235 		SYSMON_WDOG_UNLOCK(s);
    236 		break;
    237 
    238 	case WDOGIOC_GTICKLER:
    239 		if ((smw = sysmon_armed_wdog) != NULL)
    240 			*(pid_t *)data = smw->smw_tickler;
    241 		else
    242 			error = ESRCH;
    243 		break;
    244 
    245 	case WDOGIOC_GWDOGS:
    246 	    {
    247 		struct wdog_conf *wc = (void *) data;
    248 		char *cp;
    249 		int i;
    250 
    251 		simple_lock(&sysmon_wdog_list_slock);
    252 		if (wc->wc_names == NULL)
    253 			wc->wc_count = sysmon_wdog_count;
    254 		else {
    255 			for (i = 0, cp = wc->wc_names,
    256 			       smw = LIST_FIRST(&sysmon_wdog_list);
    257 			     i < sysmon_wdog_count && smw != NULL && error == 0;
    258 			     i++, cp += WDOG_NAMESIZE,
    259 			       smw = LIST_NEXT(smw, smw_list))
    260 				error = copyout(smw->smw_name, cp,
    261 				    strlen(smw->smw_name) + 1);
    262 			wc->wc_count = i;
    263 		}
    264 		simple_unlock(&sysmon_wdog_list_slock);
    265 		break;
    266 	    }
    267 
    268 	default:
    269 		error = ENOTTY;
    270 	}
    271 
    272 	return (error);
    273 }
    274 
    275 /*
    276  * sysmon_wdog_register:
    277  *
    278  *	Register a watchdog device.
    279  */
    280 int
    281 sysmon_wdog_register(struct sysmon_wdog *smw)
    282 {
    283 	struct sysmon_wdog *lsmw;
    284 	int error = 0;
    285 
    286 	simple_lock(&sysmon_wdog_list_slock);
    287 
    288 	for (lsmw = LIST_FIRST(&sysmon_wdog_list); lsmw != NULL;
    289 	     lsmw = LIST_NEXT(lsmw, smw_list)) {
    290 		if (strcmp(lsmw->smw_name, smw->smw_name) == 0) {
    291 			error = EEXIST;
    292 			goto out;
    293 		}
    294 	}
    295 
    296 	smw->smw_mode = WDOG_MODE_DISARMED;
    297 	smw->smw_tickler = (pid_t) -1;
    298 	smw->smw_refcnt = 0;
    299 	sysmon_wdog_count++;
    300 	LIST_INSERT_HEAD(&sysmon_wdog_list, smw, smw_list);
    301 
    302  out:
    303 	simple_unlock(&sysmon_wdog_list_slock);
    304 	return (error);
    305 }
    306 
    307 /*
    308  * sysmon_wdog_unregister:
    309  *
    310  *	Unregister a watchdog device.
    311  */
    312 void
    313 sysmon_wdog_unregister(struct sysmon_wdog *smw)
    314 {
    315 
    316 	simple_lock(&sysmon_wdog_list_slock);
    317 	sysmon_wdog_count--;
    318 	LIST_REMOVE(smw, smw_list);
    319 	simple_unlock(&sysmon_wdog_list_slock);
    320 }
    321 
    322 /*
    323  * sysmon_wdog_find:
    324  *
    325  *	Find a watchdog device.  We increase the reference
    326  *	count on a match.
    327  */
    328 struct sysmon_wdog *
    329 sysmon_wdog_find(const char *name)
    330 {
    331 	struct sysmon_wdog *smw;
    332 
    333 	simple_lock(&sysmon_wdog_list_slock);
    334 
    335 	for (smw = LIST_FIRST(&sysmon_wdog_list); smw != NULL;
    336 	     smw = LIST_NEXT(smw, smw_list)) {
    337 		if (strcmp(smw->smw_name, name) == 0)
    338 			break;
    339 	}
    340 
    341 	if (smw != NULL)
    342 		smw->smw_refcnt++;
    343 
    344 	simple_unlock(&sysmon_wdog_list_slock);
    345 	return (smw);
    346 }
    347 
    348 /*
    349  * sysmon_wdog_release:
    350  *
    351  *	Release a watchdog device.
    352  */
    353 void
    354 sysmon_wdog_release(struct sysmon_wdog *smw)
    355 {
    356 
    357 	simple_lock(&sysmon_wdog_list_slock);
    358 	KASSERT(smw->smw_refcnt != 0);
    359 	smw->smw_refcnt--;
    360 	simple_unlock(&sysmon_wdog_list_slock);
    361 }
    362 
    363 /*
    364  * sysmon_wdog_setmode:
    365  *
    366  *	Set the mode of a watchdog device.
    367  */
    368 int
    369 sysmon_wdog_setmode(struct sysmon_wdog *smw, int mode, u_int period)
    370 {
    371 	u_int operiod = smw->smw_period;
    372 	int omode = smw->smw_mode;
    373 	int error = 0;
    374 
    375 	smw->smw_period = period;
    376 	smw->smw_mode = mode;
    377 
    378 	switch (mode & WDOG_MODE_MASK) {
    379 	case WDOG_MODE_DISARMED:
    380 		if (smw != sysmon_armed_wdog) {
    381 			error = EINVAL;
    382 			goto out;
    383 		}
    384 		break;
    385 
    386 	case WDOG_MODE_KTICKLE:
    387 	case WDOG_MODE_UTICKLE:
    388 	case WDOG_MODE_ETICKLE:
    389 		if (sysmon_armed_wdog != NULL) {
    390 			error = EBUSY;
    391 			goto out;
    392 		}
    393 		break;
    394 
    395 	default:
    396 		error = EINVAL;
    397 		goto out;
    398 	}
    399 
    400 	error = (*smw->smw_setmode)(smw);
    401 
    402  out:
    403 	if (error) {
    404 		smw->smw_period = operiod;
    405 		smw->smw_mode = omode;
    406 	} else {
    407 		if ((mode & WDOG_MODE_MASK) == WDOG_MODE_DISARMED) {
    408 			sysmon_armed_wdog = NULL;
    409 			smw->smw_tickler = (pid_t) -1;
    410 			smw->smw_refcnt--;
    411 			if ((omode & WDOG_MODE_MASK) == WDOG_MODE_KTICKLE)
    412 				callout_stop(&sysmon_wdog_callout);
    413 		} else {
    414 			sysmon_armed_wdog = smw;
    415 			smw->smw_refcnt++;
    416 			if ((mode & WDOG_MODE_MASK) == WDOG_MODE_KTICKLE) {
    417 				callout_reset(&sysmon_wdog_callout,
    418 				    WDOG_PERIOD_TO_TICKS(smw->smw_period) / 2,
    419 				    sysmon_wdog_ktickle, NULL);
    420 			}
    421 		}
    422 	}
    423 	return (error);
    424 }
    425 
    426 /*
    427  * sysmon_wdog_ktickle:
    428  *
    429  *	Kernel watchdog tickle routine.
    430  */
    431 void
    432 sysmon_wdog_ktickle(void *arg)
    433 {
    434 	struct sysmon_wdog *smw;
    435 	int s;
    436 
    437 	SYSMON_WDOG_LOCK(s);
    438 	if ((smw = sysmon_armed_wdog) != NULL) {
    439 		if ((*smw->smw_tickle)(smw) != 0) {
    440 			printf("WARNING: KERNEL TICKLE OF WATCHDOG %s "
    441 			    "FAILED!\n", smw->smw_name);
    442 			/*
    443 			 * ...we will probably reboot soon.
    444 			 */
    445 		}
    446 		callout_reset(&sysmon_wdog_callout,
    447 		    WDOG_PERIOD_TO_TICKS(smw->smw_period) / 2,
    448 		    sysmon_wdog_ktickle, NULL);
    449 	}
    450 	SYSMON_WDOG_UNLOCK(s);
    451 }
    452 
    453 /*
    454  * sysmon_wdog_shutdown:
    455  *
    456  *	Perform shutdown-time operations.
    457  */
    458 void
    459 sysmon_wdog_shutdown(void *arg)
    460 {
    461 	struct sysmon_wdog *smw;
    462 
    463 	/*
    464 	 * XXX Locking here?  I don't think it's necessary.
    465 	 */
    466 
    467 	if ((smw = sysmon_armed_wdog) != NULL) {
    468 		if (sysmon_wdog_setmode(smw, WDOG_MODE_DISARMED,
    469 		    smw->smw_period))
    470 			printf("WARNING: FAILED TO SHUTDOWN WATCHDOG %s!\n",
    471 			    smw->smw_name);
    472 	}
    473 }
    474