1 // SPDX-License-Identifier: GPL-2.0+
2 /*
3  *	watchdog_dev.c
4  *
5  *	(c) Copyright 2008-2011 Alan Cox <alan@lxorguk.ukuu.org.uk>,
6  *						All Rights Reserved.
7  *
8  *	(c) Copyright 2008-2011 Wim Van Sebroeck <wim@iguana.be>.
9  *
10  *	(c) Copyright 2021 Hewlett Packard Enterprise Development LP.
11  *
12  *	This source code is part of the generic code that can be used
13  *	by all the watchdog timer drivers.
14  *
15  *	This part of the generic code takes care of the following
16  *	misc device: /dev/watchdog.
17  *
18  *	Based on source code of the following authors:
19  *	  Matt Domsch <Matt_Domsch@dell.com>,
20  *	  Rob Radez <rob@osinvestor.com>,
21  *	  Rusty Lynch <rusty@linux.co.intel.com>
22  *	  Satyam Sharma <satyam@infradead.org>
23  *	  Randy Dunlap <randy.dunlap@oracle.com>
24  *
25  *	Neither Alan Cox, CymruNet Ltd., Wim Van Sebroeck nor Iguana vzw.
26  *	admit liability nor provide warranty for any of this software.
27  *	This material is provided "AS-IS" and at no charge.
28  */
29 
30 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
31 
32 #include <linux/cdev.h>		/* For character device */
33 #include <linux/errno.h>	/* For the -ENODEV/... values */
34 #include <linux/fs.h>		/* For file operations */
35 #include <linux/init.h>		/* For __init/__exit/... */
36 #include <linux/hrtimer.h>	/* For hrtimers */
37 #include <linux/kernel.h>	/* For printk/panic/... */
38 #include <linux/kstrtox.h>	/* For kstrto* */
39 #include <linux/kthread.h>	/* For kthread_work */
40 #include <linux/miscdevice.h>	/* For handling misc devices */
41 #include <linux/module.h>	/* For module stuff/... */
42 #include <linux/mutex.h>	/* For mutexes */
43 #include <linux/slab.h>		/* For memory functions */
44 #include <linux/types.h>	/* For standard types (like size_t) */
45 #include <linux/watchdog.h>	/* For watchdog specific items */
46 #include <linux/uaccess.h>	/* For copy_to_user/put_user/... */
47 
48 #include "watchdog_core.h"
49 #include "watchdog_pretimeout.h"
50 
51 #include <trace/events/watchdog.h>
52 
53 /* the dev_t structure to store the dynamically allocated watchdog devices */
54 static dev_t watchdog_devt;
55 /* Reference to watchdog device behind /dev/watchdog */
56 static struct watchdog_core_data *old_wd_data;
57 
58 static struct kthread_worker *watchdog_kworker;
59 
60 static bool handle_boot_enabled =
61 	IS_ENABLED(CONFIG_WATCHDOG_HANDLE_BOOT_ENABLED);
62 
63 static unsigned open_timeout = CONFIG_WATCHDOG_OPEN_TIMEOUT;
64 
65 static bool watchdog_past_open_deadline(struct watchdog_core_data *data)
66 {
67 	return ktime_after(ktime_get(), data->open_deadline);
68 }
69 
70 static void watchdog_set_open_deadline(struct watchdog_core_data *data)
71 {
72 	data->open_deadline = open_timeout ?
73 		ktime_get() + ktime_set(open_timeout, 0) : KTIME_MAX;
74 }
75 
76 static inline bool watchdog_need_worker(struct watchdog_device *wdd)
77 {
78 	/* All variables in milli-seconds */
79 	unsigned int hm = wdd->max_hw_heartbeat_ms;
80 	unsigned int t = wdd->timeout * 1000;
81 
82 	/*
83 	 * A worker to generate heartbeat requests is needed if all of the
84 	 * following conditions are true.
85 	 * - Userspace activated the watchdog.
86 	 * - The driver provided a value for the maximum hardware timeout, and
87 	 *   thus is aware that the framework supports generating heartbeat
88 	 *   requests.
89 	 * - Userspace requests a longer timeout than the hardware can handle.
90 	 *
91 	 * Alternatively, if userspace has not opened the watchdog
92 	 * device, we take care of feeding the watchdog if it is
93 	 * running.
94 	 */
95 	return (hm && watchdog_active(wdd) && t > hm) ||
96 		(t && !watchdog_active(wdd) && watchdog_hw_running(wdd));
97 }
98 
99 static ktime_t watchdog_next_keepalive(struct watchdog_device *wdd)
100 {
101 	struct watchdog_core_data *wd_data = wdd->wd_data;
102 	unsigned int timeout_ms = wdd->timeout * 1000;
103 	ktime_t keepalive_interval;
104 	ktime_t last_heartbeat, latest_heartbeat;
105 	ktime_t virt_timeout;
106 	unsigned int hw_heartbeat_ms;
107 
108 	if (watchdog_active(wdd))
109 		virt_timeout = ktime_add(wd_data->last_keepalive,
110 					 ms_to_ktime(timeout_ms));
111 	else
112 		virt_timeout = wd_data->open_deadline;
113 
114 	hw_heartbeat_ms = min_not_zero(timeout_ms, wdd->max_hw_heartbeat_ms);
115 	keepalive_interval = ms_to_ktime(hw_heartbeat_ms / 2);
116 
117 	/*
118 	 * To ensure that the watchdog times out wdd->timeout seconds
119 	 * after the most recent ping from userspace, the last
120 	 * worker ping has to come in hw_heartbeat_ms before this timeout.
121 	 */
122 	last_heartbeat = ktime_sub(virt_timeout, ms_to_ktime(hw_heartbeat_ms));
123 	latest_heartbeat = ktime_sub(last_heartbeat, ktime_get());
124 	if (ktime_before(latest_heartbeat, keepalive_interval))
125 		return latest_heartbeat;
126 	return keepalive_interval;
127 }
128 
129 static inline void watchdog_update_worker(struct watchdog_device *wdd)
130 {
131 	struct watchdog_core_data *wd_data = wdd->wd_data;
132 
133 	if (watchdog_need_worker(wdd)) {
134 		ktime_t t = watchdog_next_keepalive(wdd);
135 
136 		if (t > 0)
137 			hrtimer_start(&wd_data->timer, t,
138 				      HRTIMER_MODE_REL_HARD);
139 	} else {
140 		hrtimer_cancel(&wd_data->timer);
141 	}
142 }
143 
144 static int __watchdog_ping(struct watchdog_device *wdd)
145 {
146 	struct watchdog_core_data *wd_data = wdd->wd_data;
147 	ktime_t earliest_keepalive, now;
148 	int err;
149 
150 	earliest_keepalive = ktime_add(wd_data->last_hw_keepalive,
151 				       ms_to_ktime(wdd->min_hw_heartbeat_ms));
152 	now = ktime_get();
153 
154 	if (ktime_after(earliest_keepalive, now)) {
155 		hrtimer_start(&wd_data->timer,
156 			      ktime_sub(earliest_keepalive, now),
157 			      HRTIMER_MODE_REL_HARD);
158 		return 0;
159 	}
160 
161 	wd_data->last_hw_keepalive = now;
162 
163 	if (wdd->ops->ping) {
164 		err = wdd->ops->ping(wdd);  /* ping the watchdog */
165 		trace_watchdog_ping(wdd, err);
166 	} else {
167 		err = wdd->ops->start(wdd); /* restart watchdog */
168 		trace_watchdog_start(wdd, err);
169 	}
170 
171 	if (err == 0)
172 		watchdog_hrtimer_pretimeout_start(wdd);
173 
174 	watchdog_update_worker(wdd);
175 
176 	return err;
177 }
178 
179 /*
180  * watchdog_ping - ping the watchdog
181  * @wdd: The watchdog device to ping
182  *
183  * If the watchdog has no own ping operation then it needs to be
184  * restarted via the start operation. This wrapper function does
185  * exactly that.
186  * We only ping when the watchdog device is running.
187  * The caller must hold wd_data->lock.
188  *
189  * Return: 0 on success, error otherwise.
190  */
191 static int watchdog_ping(struct watchdog_device *wdd)
192 {
193 	struct watchdog_core_data *wd_data = wdd->wd_data;
194 
195 	if (!watchdog_active(wdd) && !watchdog_hw_running(wdd))
196 		return 0;
197 
198 	set_bit(_WDOG_KEEPALIVE, &wd_data->status);
199 
200 	wd_data->last_keepalive = ktime_get();
201 	return __watchdog_ping(wdd);
202 }
203 
204 static bool watchdog_worker_should_ping(struct watchdog_core_data *wd_data)
205 {
206 	struct watchdog_device *wdd = wd_data->wdd;
207 
208 	if (!wdd)
209 		return false;
210 
211 	if (watchdog_active(wdd))
212 		return true;
213 
214 	return watchdog_hw_running(wdd) && !watchdog_past_open_deadline(wd_data);
215 }
216 
217 static void watchdog_ping_work(struct kthread_work *work)
218 {
219 	struct watchdog_core_data *wd_data;
220 
221 	wd_data = container_of(work, struct watchdog_core_data, work);
222 
223 	mutex_lock(&wd_data->lock);
224 	if (watchdog_worker_should_ping(wd_data))
225 		__watchdog_ping(wd_data->wdd);
226 	mutex_unlock(&wd_data->lock);
227 }
228 
229 static enum hrtimer_restart watchdog_timer_expired(struct hrtimer *timer)
230 {
231 	struct watchdog_core_data *wd_data;
232 
233 	wd_data = container_of(timer, struct watchdog_core_data, timer);
234 
235 	kthread_queue_work(watchdog_kworker, &wd_data->work);
236 	return HRTIMER_NORESTART;
237 }
238 
239 /*
240  * watchdog_start - wrapper to start the watchdog
241  * @wdd: The watchdog device to start
242  *
243  * Start the watchdog if it is not active and mark it active.
244  * The caller must hold wd_data->lock.
245  *
246  * Return: 0 on success or a negative errno code for failure.
247  */
248 static int watchdog_start(struct watchdog_device *wdd)
249 {
250 	struct watchdog_core_data *wd_data = wdd->wd_data;
251 	ktime_t started_at;
252 	int err;
253 
254 	if (watchdog_active(wdd))
255 		return 0;
256 
257 	set_bit(_WDOG_KEEPALIVE, &wd_data->status);
258 
259 	started_at = ktime_get();
260 	if (watchdog_hw_running(wdd) && wdd->ops->ping) {
261 		err = __watchdog_ping(wdd);
262 		if (err == 0) {
263 			set_bit(WDOG_ACTIVE, &wdd->status);
264 			watchdog_hrtimer_pretimeout_start(wdd);
265 		}
266 	} else {
267 		err = wdd->ops->start(wdd);
268 		trace_watchdog_start(wdd, err);
269 		if (err == 0) {
270 			set_bit(WDOG_ACTIVE, &wdd->status);
271 			wd_data->last_keepalive = started_at;
272 			wd_data->last_hw_keepalive = started_at;
273 			watchdog_update_worker(wdd);
274 			watchdog_hrtimer_pretimeout_start(wdd);
275 		}
276 	}
277 
278 	return err;
279 }
280 
281 /*
282  * watchdog_stop - wrapper to stop the watchdog
283  * @wdd: The watchdog device to stop
284  *
285  * Stop the watchdog if it is still active and unmark it active.
286  * If the 'nowayout' feature was set, the watchdog cannot be stopped.
287  * The caller must hold wd_data->lock.
288  *
289  * Return: 0 on success or a negative errno code for failure.
290  */
291 static int watchdog_stop(struct watchdog_device *wdd)
292 {
293 	int err = 0;
294 
295 	if (!watchdog_active(wdd))
296 		return 0;
297 
298 	if (test_bit(WDOG_NO_WAY_OUT, &wdd->status)) {
299 		pr_info("watchdog%d: nowayout prevents watchdog being stopped!\n",
300 			wdd->id);
301 		return -EBUSY;
302 	}
303 
304 	if (wdd->ops->stop) {
305 		clear_bit(WDOG_HW_RUNNING, &wdd->status);
306 		err = wdd->ops->stop(wdd);
307 		trace_watchdog_stop(wdd, err);
308 	} else {
309 		set_bit(WDOG_HW_RUNNING, &wdd->status);
310 	}
311 
312 	if (err == 0) {
313 		clear_bit(WDOG_ACTIVE, &wdd->status);
314 		watchdog_update_worker(wdd);
315 		watchdog_hrtimer_pretimeout_stop(wdd);
316 	}
317 
318 	return err;
319 }
320 
321 /*
322  * watchdog_get_status - wrapper to get the watchdog status
323  * @wdd: The watchdog device to get the status from
324  *
325  * Get the watchdog's status flags.
326  * The caller must hold wd_data->lock.
327  *
328  * Return: watchdog's status flags.
329  */
330 static unsigned int watchdog_get_status(struct watchdog_device *wdd)
331 {
332 	struct watchdog_core_data *wd_data = wdd->wd_data;
333 	unsigned int status;
334 
335 	if (wdd->ops->status)
336 		status = wdd->ops->status(wdd);
337 	else
338 		status = wdd->bootstatus & (WDIOF_CARDRESET |
339 					    WDIOF_OVERHEAT |
340 					    WDIOF_FANFAULT |
341 					    WDIOF_EXTERN1 |
342 					    WDIOF_EXTERN2 |
343 					    WDIOF_POWERUNDER |
344 					    WDIOF_POWEROVER);
345 
346 	if (test_bit(_WDOG_ALLOW_RELEASE, &wd_data->status))
347 		status |= WDIOF_MAGICCLOSE;
348 
349 	if (test_and_clear_bit(_WDOG_KEEPALIVE, &wd_data->status))
350 		status |= WDIOF_KEEPALIVEPING;
351 
352 	if (IS_ENABLED(CONFIG_WATCHDOG_HRTIMER_PRETIMEOUT))
353 		status |= WDIOF_PRETIMEOUT;
354 
355 	return status;
356 }
357 
358 /*
359  * watchdog_set_timeout - set the watchdog timer timeout
360  * @wdd:	The watchdog device to set the timeout for
361  * @timeout:	Timeout to set in seconds
362  *
363  * The caller must hold wd_data->lock.
364  *
365  * Return: 0 if successful, error otherwise.
366  */
367 static int watchdog_set_timeout(struct watchdog_device *wdd,
368 							unsigned int timeout)
369 {
370 	int err = 0;
371 
372 	if (!(wdd->info->options & WDIOF_SETTIMEOUT))
373 		return -EOPNOTSUPP;
374 
375 	if (watchdog_timeout_invalid(wdd, timeout))
376 		return -EINVAL;
377 
378 	if (wdd->ops->set_timeout) {
379 		err = wdd->ops->set_timeout(wdd, timeout);
380 		trace_watchdog_set_timeout(wdd, timeout, err);
381 	} else {
382 		wdd->timeout = timeout;
383 		/* Disable pretimeout if it doesn't fit the new timeout */
384 		if (wdd->pretimeout >= wdd->timeout)
385 			wdd->pretimeout = 0;
386 	}
387 
388 	watchdog_update_worker(wdd);
389 
390 	return err;
391 }
392 
393 /*
394  * watchdog_set_pretimeout - set the watchdog timer pretimeout
395  * @wdd:	The watchdog device to set the timeout for
396  * @timeout:	pretimeout to set in seconds
397  *
398  * Return: 0 if successful, error otherwise.
399  */
400 static int watchdog_set_pretimeout(struct watchdog_device *wdd,
401 				   unsigned int timeout)
402 {
403 	int err = 0;
404 
405 	if (!watchdog_have_pretimeout(wdd))
406 		return -EOPNOTSUPP;
407 
408 	if (watchdog_pretimeout_invalid(wdd, timeout))
409 		return -EINVAL;
410 
411 	if (wdd->ops->set_pretimeout && (wdd->info->options & WDIOF_PRETIMEOUT))
412 		err = wdd->ops->set_pretimeout(wdd, timeout);
413 	else
414 		wdd->pretimeout = timeout;
415 
416 	return err;
417 }
418 
419 /*
420  * watchdog_get_timeleft - wrapper to get the time left before a reboot
421  * @wdd:	The watchdog device to get the remaining time from
422  * @timeleft:	The time that's left
423  *
424  * Get the time before a watchdog will reboot (if not pinged).
425  * The caller must hold wd_data->lock.
426  *
427  * Return: 0 if successful, error otherwise.
428  */
429 static int watchdog_get_timeleft(struct watchdog_device *wdd,
430 							unsigned int *timeleft)
431 {
432 	*timeleft = 0;
433 
434 	if (!wdd->ops->get_timeleft)
435 		return -EOPNOTSUPP;
436 
437 	*timeleft = wdd->ops->get_timeleft(wdd);
438 
439 	return 0;
440 }
441 
442 #ifdef CONFIG_WATCHDOG_SYSFS
443 static ssize_t nowayout_show(struct device *dev, struct device_attribute *attr,
444 				char *buf)
445 {
446 	struct watchdog_device *wdd = dev_get_drvdata(dev);
447 
448 	return sysfs_emit(buf, "%d\n", !!test_bit(WDOG_NO_WAY_OUT,
449 						  &wdd->status));
450 }
451 
452 static ssize_t nowayout_store(struct device *dev, struct device_attribute *attr,
453 				const char *buf, size_t len)
454 {
455 	struct watchdog_device *wdd = dev_get_drvdata(dev);
456 	unsigned int value;
457 	int ret;
458 
459 	ret = kstrtouint(buf, 0, &value);
460 	if (ret)
461 		return ret;
462 	if (value > 1)
463 		return -EINVAL;
464 	/* nowayout cannot be disabled once set */
465 	if (test_bit(WDOG_NO_WAY_OUT, &wdd->status) && !value)
466 		return -EPERM;
467 	watchdog_set_nowayout(wdd, value);
468 	return len;
469 }
470 static DEVICE_ATTR_RW(nowayout);
471 
472 static ssize_t status_show(struct device *dev, struct device_attribute *attr,
473 				char *buf)
474 {
475 	struct watchdog_device *wdd = dev_get_drvdata(dev);
476 	struct watchdog_core_data *wd_data = wdd->wd_data;
477 	unsigned int status;
478 
479 	mutex_lock(&wd_data->lock);
480 	status = watchdog_get_status(wdd);
481 	mutex_unlock(&wd_data->lock);
482 
483 	return sysfs_emit(buf, "0x%x\n", status);
484 }
485 static DEVICE_ATTR_RO(status);
486 
487 static ssize_t bootstatus_show(struct device *dev,
488 				struct device_attribute *attr, char *buf)
489 {
490 	struct watchdog_device *wdd = dev_get_drvdata(dev);
491 
492 	return sysfs_emit(buf, "%u\n", wdd->bootstatus);
493 }
494 static DEVICE_ATTR_RO(bootstatus);
495 
496 static ssize_t timeleft_show(struct device *dev, struct device_attribute *attr,
497 				char *buf)
498 {
499 	struct watchdog_device *wdd = dev_get_drvdata(dev);
500 	struct watchdog_core_data *wd_data = wdd->wd_data;
501 	ssize_t status;
502 	unsigned int val;
503 
504 	mutex_lock(&wd_data->lock);
505 	status = watchdog_get_timeleft(wdd, &val);
506 	mutex_unlock(&wd_data->lock);
507 	if (!status)
508 		status = sysfs_emit(buf, "%u\n", val);
509 
510 	return status;
511 }
512 static DEVICE_ATTR_RO(timeleft);
513 
514 static ssize_t timeout_show(struct device *dev, struct device_attribute *attr,
515 				char *buf)
516 {
517 	struct watchdog_device *wdd = dev_get_drvdata(dev);
518 
519 	return sysfs_emit(buf, "%u\n", wdd->timeout);
520 }
521 static DEVICE_ATTR_RO(timeout);
522 
523 static ssize_t min_timeout_show(struct device *dev,
524 				struct device_attribute *attr, char *buf)
525 {
526 	struct watchdog_device *wdd = dev_get_drvdata(dev);
527 
528 	return sysfs_emit(buf, "%u\n", wdd->min_timeout);
529 }
530 static DEVICE_ATTR_RO(min_timeout);
531 
532 static ssize_t max_timeout_show(struct device *dev,
533 				struct device_attribute *attr, char *buf)
534 {
535 	struct watchdog_device *wdd = dev_get_drvdata(dev);
536 
537 	return sysfs_emit(buf, "%u\n", wdd->max_timeout);
538 }
539 static DEVICE_ATTR_RO(max_timeout);
540 
541 static ssize_t pretimeout_show(struct device *dev,
542 			       struct device_attribute *attr, char *buf)
543 {
544 	struct watchdog_device *wdd = dev_get_drvdata(dev);
545 
546 	return sysfs_emit(buf, "%u\n", wdd->pretimeout);
547 }
548 static DEVICE_ATTR_RO(pretimeout);
549 
550 static ssize_t options_show(struct device *dev, struct device_attribute *attr,
551 			    char *buf)
552 {
553 	struct watchdog_device *wdd = dev_get_drvdata(dev);
554 
555 	return sysfs_emit(buf, "0x%x\n", wdd->info->options);
556 }
557 static DEVICE_ATTR_RO(options);
558 
559 static ssize_t fw_version_show(struct device *dev, struct device_attribute *attr,
560 			       char *buf)
561 {
562 	struct watchdog_device *wdd = dev_get_drvdata(dev);
563 
564 	return sysfs_emit(buf, "%d\n", wdd->info->firmware_version);
565 }
566 static DEVICE_ATTR_RO(fw_version);
567 
568 static ssize_t identity_show(struct device *dev, struct device_attribute *attr,
569 				char *buf)
570 {
571 	struct watchdog_device *wdd = dev_get_drvdata(dev);
572 
573 	return sysfs_emit(buf, "%s\n", wdd->info->identity);
574 }
575 static DEVICE_ATTR_RO(identity);
576 
577 static ssize_t state_show(struct device *dev, struct device_attribute *attr,
578 				char *buf)
579 {
580 	struct watchdog_device *wdd = dev_get_drvdata(dev);
581 
582 	if (watchdog_active(wdd))
583 		return sysfs_emit(buf, "active\n");
584 
585 	return sysfs_emit(buf, "inactive\n");
586 }
587 static DEVICE_ATTR_RO(state);
588 
589 static ssize_t pretimeout_available_governors_show(struct device *dev,
590 				   struct device_attribute *attr, char *buf)
591 {
592 	return watchdog_pretimeout_available_governors_get(buf);
593 }
594 static DEVICE_ATTR_RO(pretimeout_available_governors);
595 
596 static ssize_t pretimeout_governor_show(struct device *dev,
597 					struct device_attribute *attr,
598 					char *buf)
599 {
600 	struct watchdog_device *wdd = dev_get_drvdata(dev);
601 
602 	return watchdog_pretimeout_governor_get(wdd, buf);
603 }
604 
605 static ssize_t pretimeout_governor_store(struct device *dev,
606 					 struct device_attribute *attr,
607 					 const char *buf, size_t count)
608 {
609 	struct watchdog_device *wdd = dev_get_drvdata(dev);
610 	int ret = watchdog_pretimeout_governor_set(wdd, buf);
611 
612 	if (!ret)
613 		ret = count;
614 
615 	return ret;
616 }
617 static DEVICE_ATTR_RW(pretimeout_governor);
618 
619 static umode_t wdt_is_visible(struct kobject *kobj, struct attribute *attr,
620 				int n)
621 {
622 	struct device *dev = kobj_to_dev(kobj);
623 	struct watchdog_device *wdd = dev_get_drvdata(dev);
624 	umode_t mode = attr->mode;
625 
626 	if (attr == &dev_attr_timeleft.attr && !wdd->ops->get_timeleft)
627 		mode = 0;
628 	else if (attr == &dev_attr_pretimeout.attr && !watchdog_have_pretimeout(wdd))
629 		mode = 0;
630 	else if ((attr == &dev_attr_pretimeout_governor.attr ||
631 		  attr == &dev_attr_pretimeout_available_governors.attr) &&
632 		 (!watchdog_have_pretimeout(wdd) || !IS_ENABLED(CONFIG_WATCHDOG_PRETIMEOUT_GOV)))
633 		mode = 0;
634 
635 	return mode;
636 }
637 static struct attribute *wdt_attrs[] = {
638 	&dev_attr_state.attr,
639 	&dev_attr_options.attr,
640 	&dev_attr_fw_version.attr,
641 	&dev_attr_identity.attr,
642 	&dev_attr_timeout.attr,
643 	&dev_attr_min_timeout.attr,
644 	&dev_attr_max_timeout.attr,
645 	&dev_attr_pretimeout.attr,
646 	&dev_attr_timeleft.attr,
647 	&dev_attr_bootstatus.attr,
648 	&dev_attr_status.attr,
649 	&dev_attr_nowayout.attr,
650 	&dev_attr_pretimeout_governor.attr,
651 	&dev_attr_pretimeout_available_governors.attr,
652 	NULL,
653 };
654 
655 static const struct attribute_group wdt_group = {
656 	.attrs = wdt_attrs,
657 	.is_visible = wdt_is_visible,
658 };
659 __ATTRIBUTE_GROUPS(wdt);
660 #else
661 #define wdt_groups	NULL
662 #endif
663 
664 /*
665  * watchdog_ioctl_op - call the watchdog drivers ioctl op if defined
666  * @wdd: The watchdog device to do the ioctl on
667  * @cmd: Watchdog command
668  * @arg: Argument pointer
669  *
670  * The caller must hold wd_data->lock.
671  *
672  * Return: 0 if successful, error otherwise.
673  */
674 static int watchdog_ioctl_op(struct watchdog_device *wdd, unsigned int cmd,
675 							unsigned long arg)
676 {
677 	if (!wdd->ops->ioctl)
678 		return -ENOIOCTLCMD;
679 
680 	return wdd->ops->ioctl(wdd, cmd, arg);
681 }
682 
683 /*
684  * watchdog_write - writes to the watchdog
685  * @file:	File from VFS
686  * @data:	User address of data
687  * @len:	Length of data
688  * @ppos:	Pointer to the file offset
689  *
690  * A write to a watchdog device is defined as a keepalive ping.
691  * Writing the magic 'V' sequence allows the next close to turn
692  * off the watchdog (if 'nowayout' is not set).
693  *
694  * Return: @len if successful, error otherwise.
695  */
696 static ssize_t watchdog_write(struct file *file, const char __user *data,
697 						size_t len, loff_t *ppos)
698 {
699 	struct watchdog_core_data *wd_data = file->private_data;
700 	struct watchdog_device *wdd;
701 	int err;
702 	size_t i;
703 	char c;
704 
705 	if (len == 0)
706 		return 0;
707 
708 	/*
709 	 * Note: just in case someone wrote the magic character
710 	 * five months ago...
711 	 */
712 	clear_bit(_WDOG_ALLOW_RELEASE, &wd_data->status);
713 
714 	/* scan to see whether or not we got the magic character */
715 	for (i = 0; i != len; i++) {
716 		if (get_user(c, data + i))
717 			return -EFAULT;
718 		if (c == 'V')
719 			set_bit(_WDOG_ALLOW_RELEASE, &wd_data->status);
720 	}
721 
722 	/* someone wrote to us, so we send the watchdog a keepalive ping */
723 
724 	err = -ENODEV;
725 	mutex_lock(&wd_data->lock);
726 	wdd = wd_data->wdd;
727 	if (wdd)
728 		err = watchdog_ping(wdd);
729 	mutex_unlock(&wd_data->lock);
730 
731 	if (err < 0)
732 		return err;
733 
734 	return len;
735 }
736 
737 /*
738  * watchdog_ioctl - handle the different ioctl's for the watchdog device
739  * @file:	File handle to the device
740  * @cmd:	Watchdog command
741  * @arg:	Argument pointer
742  *
743  * The watchdog API defines a common set of functions for all watchdogs
744  * according to their available features.
745  *
746  * Return: 0 if successful, error otherwise.
747  */
748 
749 static long watchdog_ioctl(struct file *file, unsigned int cmd,
750 							unsigned long arg)
751 {
752 	struct watchdog_core_data *wd_data = file->private_data;
753 	void __user *argp = (void __user *)arg;
754 	struct watchdog_device *wdd;
755 	int __user *p = argp;
756 	unsigned int val;
757 	int err;
758 
759 	mutex_lock(&wd_data->lock);
760 
761 	wdd = wd_data->wdd;
762 	if (!wdd) {
763 		err = -ENODEV;
764 		goto out_ioctl;
765 	}
766 
767 	err = watchdog_ioctl_op(wdd, cmd, arg);
768 	if (err != -ENOIOCTLCMD)
769 		goto out_ioctl;
770 
771 	switch (cmd) {
772 	case WDIOC_GETSUPPORT:
773 		err = copy_to_user(argp, wdd->info,
774 			sizeof(struct watchdog_info)) ? -EFAULT : 0;
775 		break;
776 	case WDIOC_GETSTATUS:
777 		val = watchdog_get_status(wdd);
778 		err = put_user(val, p);
779 		break;
780 	case WDIOC_GETBOOTSTATUS:
781 		err = put_user(wdd->bootstatus, p);
782 		break;
783 	case WDIOC_SETOPTIONS:
784 		if (get_user(val, p)) {
785 			err = -EFAULT;
786 			break;
787 		}
788 		if (val & WDIOS_DISABLECARD) {
789 			err = watchdog_stop(wdd);
790 			if (err < 0)
791 				break;
792 		}
793 		if (val & WDIOS_ENABLECARD)
794 			err = watchdog_start(wdd);
795 		break;
796 	case WDIOC_KEEPALIVE:
797 		if (!(wdd->info->options & WDIOF_KEEPALIVEPING)) {
798 			err = -EOPNOTSUPP;
799 			break;
800 		}
801 		err = watchdog_ping(wdd);
802 		break;
803 	case WDIOC_SETTIMEOUT:
804 		if (get_user(val, p)) {
805 			err = -EFAULT;
806 			break;
807 		}
808 		err = watchdog_set_timeout(wdd, val);
809 		if (err < 0)
810 			break;
811 		/* If the watchdog is active then we send a keepalive ping
812 		 * to make sure that the watchdog keep's running (and if
813 		 * possible that it takes the new timeout) */
814 		err = watchdog_ping(wdd);
815 		if (err < 0)
816 			break;
817 		fallthrough;
818 	case WDIOC_GETTIMEOUT:
819 		/* timeout == 0 means that we don't know the timeout */
820 		if (wdd->timeout == 0) {
821 			err = -EOPNOTSUPP;
822 			break;
823 		}
824 		err = put_user(wdd->timeout, p);
825 		break;
826 	case WDIOC_GETTIMELEFT:
827 		err = watchdog_get_timeleft(wdd, &val);
828 		if (err < 0)
829 			break;
830 		err = put_user(val, p);
831 		break;
832 	case WDIOC_SETPRETIMEOUT:
833 		if (get_user(val, p)) {
834 			err = -EFAULT;
835 			break;
836 		}
837 		err = watchdog_set_pretimeout(wdd, val);
838 		break;
839 	case WDIOC_GETPRETIMEOUT:
840 		err = put_user(wdd->pretimeout, p);
841 		break;
842 	default:
843 		err = -ENOTTY;
844 		break;
845 	}
846 
847 out_ioctl:
848 	mutex_unlock(&wd_data->lock);
849 	return err;
850 }
851 
852 /*
853  * watchdog_open - open the /dev/watchdog* devices
854  * @inode:	Inode of device
855  * @file:	File handle to device
856  *
857  * When the /dev/watchdog* device gets opened, we start the watchdog.
858  * Watch out: the /dev/watchdog device is single open, so we make sure
859  * it can only be opened once.
860  *
861  * Return: 0 if successful, error otherwise.
862  */
863 static int watchdog_open(struct inode *inode, struct file *file)
864 {
865 	struct watchdog_core_data *wd_data;
866 	struct watchdog_device *wdd;
867 	bool hw_running;
868 	int err;
869 
870 	/* Get the corresponding watchdog device */
871 	if (imajor(inode) == MISC_MAJOR)
872 		wd_data = old_wd_data;
873 	else
874 		wd_data = container_of(inode->i_cdev, struct watchdog_core_data,
875 				       cdev);
876 
877 	/* the watchdog is single open! */
878 	if (test_and_set_bit(_WDOG_DEV_OPEN, &wd_data->status))
879 		return -EBUSY;
880 
881 	wdd = wd_data->wdd;
882 
883 	/*
884 	 * If the /dev/watchdog device is open, we don't want the module
885 	 * to be unloaded.
886 	 */
887 	hw_running = watchdog_hw_running(wdd);
888 	if (!hw_running && !try_module_get(wdd->ops->owner)) {
889 		err = -EBUSY;
890 		goto out_clear;
891 	}
892 
893 	err = watchdog_start(wdd);
894 	if (err < 0)
895 		goto out_mod;
896 
897 	file->private_data = wd_data;
898 
899 	if (!hw_running)
900 		get_device(&wd_data->dev);
901 
902 	/*
903 	 * open_timeout only applies for the first open from
904 	 * userspace. Set open_deadline to infinity so that the kernel
905 	 * will take care of an always-running hardware watchdog in
906 	 * case the device gets magic-closed or WDIOS_DISABLECARD is
907 	 * applied.
908 	 */
909 	wd_data->open_deadline = KTIME_MAX;
910 
911 	/* dev/watchdog is a virtual (and thus non-seekable) filesystem */
912 	return stream_open(inode, file);
913 
914 out_mod:
915 	module_put(wd_data->wdd->ops->owner);
916 out_clear:
917 	clear_bit(_WDOG_DEV_OPEN, &wd_data->status);
918 	return err;
919 }
920 
921 static void watchdog_core_data_release(struct device *dev)
922 {
923 	struct watchdog_core_data *wd_data;
924 
925 	wd_data = container_of(dev, struct watchdog_core_data, dev);
926 
927 	kfree(wd_data);
928 }
929 
930 /*
931  * watchdog_release - release the watchdog device
932  * @inode:	Inode of device
933  * @file:	File handle to device
934  *
935  * This is the code for when /dev/watchdog gets closed. We will only
936  * stop the watchdog when we have received the magic char (and nowayout
937  * was not set), else the watchdog will keep running.
938  *
939  * Always returns 0.
940  */
941 static int watchdog_release(struct inode *inode, struct file *file)
942 {
943 	struct watchdog_core_data *wd_data = file->private_data;
944 	struct watchdog_device *wdd;
945 	int err = -EBUSY;
946 	bool running;
947 
948 	mutex_lock(&wd_data->lock);
949 
950 	wdd = wd_data->wdd;
951 	if (!wdd)
952 		goto done;
953 
954 	/*
955 	 * We only stop the watchdog if we received the magic character
956 	 * or if WDIOF_MAGICCLOSE is not set. If nowayout was set then
957 	 * watchdog_stop will fail.
958 	 */
959 	if (!watchdog_active(wdd))
960 		err = 0;
961 	else if (test_and_clear_bit(_WDOG_ALLOW_RELEASE, &wd_data->status) ||
962 		 !(wdd->info->options & WDIOF_MAGICCLOSE))
963 		err = watchdog_stop(wdd);
964 
965 	/* If the watchdog was not stopped, send a keepalive ping */
966 	if (err < 0) {
967 		pr_crit("watchdog%d: watchdog did not stop!\n", wdd->id);
968 		watchdog_ping(wdd);
969 	}
970 
971 	watchdog_update_worker(wdd);
972 
973 	/* make sure that /dev/watchdog can be re-opened */
974 	clear_bit(_WDOG_DEV_OPEN, &wd_data->status);
975 
976 done:
977 	running = wdd && watchdog_hw_running(wdd);
978 	mutex_unlock(&wd_data->lock);
979 	/*
980 	 * Allow the owner module to be unloaded again unless the watchdog
981 	 * is still running. If the watchdog is still running, it can not
982 	 * be stopped, and its driver must not be unloaded.
983 	 */
984 	if (!running) {
985 		module_put(wd_data->cdev.owner);
986 		put_device(&wd_data->dev);
987 	}
988 	return 0;
989 }
990 
991 static const struct file_operations watchdog_fops = {
992 	.owner		= THIS_MODULE,
993 	.write		= watchdog_write,
994 	.unlocked_ioctl	= watchdog_ioctl,
995 	.compat_ioctl	= compat_ptr_ioctl,
996 	.open		= watchdog_open,
997 	.release	= watchdog_release,
998 };
999 
1000 static struct miscdevice watchdog_miscdev = {
1001 	.minor		= WATCHDOG_MINOR,
1002 	.name		= "watchdog",
1003 	.fops		= &watchdog_fops,
1004 };
1005 
1006 static struct class watchdog_class = {
1007 	.name =		"watchdog",
1008 	.owner =	THIS_MODULE,
1009 	.dev_groups =	wdt_groups,
1010 };
1011 
1012 /*
1013  * watchdog_cdev_register - register watchdog character device
1014  * @wdd: Watchdog device
1015  *
1016  * Register a watchdog character device including handling the legacy
1017  * /dev/watchdog node. /dev/watchdog is actually a miscdevice and
1018  * thus we set it up like that.
1019  *
1020  * Return: 0 if successful, error otherwise.
1021  */
1022 static int watchdog_cdev_register(struct watchdog_device *wdd)
1023 {
1024 	struct watchdog_core_data *wd_data;
1025 	int err;
1026 
1027 	wd_data = kzalloc(sizeof(struct watchdog_core_data), GFP_KERNEL);
1028 	if (!wd_data)
1029 		return -ENOMEM;
1030 	mutex_init(&wd_data->lock);
1031 
1032 	wd_data->wdd = wdd;
1033 	wdd->wd_data = wd_data;
1034 
1035 	if (IS_ERR_OR_NULL(watchdog_kworker)) {
1036 		kfree(wd_data);
1037 		return -ENODEV;
1038 	}
1039 
1040 	device_initialize(&wd_data->dev);
1041 	wd_data->dev.devt = MKDEV(MAJOR(watchdog_devt), wdd->id);
1042 	wd_data->dev.class = &watchdog_class;
1043 	wd_data->dev.parent = wdd->parent;
1044 	wd_data->dev.groups = wdd->groups;
1045 	wd_data->dev.release = watchdog_core_data_release;
1046 	dev_set_drvdata(&wd_data->dev, wdd);
1047 	err = dev_set_name(&wd_data->dev, "watchdog%d", wdd->id);
1048 	if (err) {
1049 		put_device(&wd_data->dev);
1050 		return err;
1051 	}
1052 
1053 	kthread_init_work(&wd_data->work, watchdog_ping_work);
1054 	hrtimer_init(&wd_data->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_HARD);
1055 	wd_data->timer.function = watchdog_timer_expired;
1056 	watchdog_hrtimer_pretimeout_init(wdd);
1057 
1058 	if (wdd->id == 0) {
1059 		old_wd_data = wd_data;
1060 		watchdog_miscdev.parent = wdd->parent;
1061 		err = misc_register(&watchdog_miscdev);
1062 		if (err != 0) {
1063 			pr_err("%s: cannot register miscdev on minor=%d (err=%d).\n",
1064 				wdd->info->identity, WATCHDOG_MINOR, err);
1065 			if (err == -EBUSY)
1066 				pr_err("%s: a legacy watchdog module is probably present.\n",
1067 					wdd->info->identity);
1068 			old_wd_data = NULL;
1069 			put_device(&wd_data->dev);
1070 			return err;
1071 		}
1072 	}
1073 
1074 	/* Fill in the data structures */
1075 	cdev_init(&wd_data->cdev, &watchdog_fops);
1076 
1077 	/* Add the device */
1078 	err = cdev_device_add(&wd_data->cdev, &wd_data->dev);
1079 	if (err) {
1080 		pr_err("watchdog%d unable to add device %d:%d\n",
1081 			wdd->id,  MAJOR(watchdog_devt), wdd->id);
1082 		if (wdd->id == 0) {
1083 			misc_deregister(&watchdog_miscdev);
1084 			old_wd_data = NULL;
1085 		}
1086 		put_device(&wd_data->dev);
1087 		return err;
1088 	}
1089 
1090 	wd_data->cdev.owner = wdd->ops->owner;
1091 
1092 	/* Record time of most recent heartbeat as 'just before now'. */
1093 	wd_data->last_hw_keepalive = ktime_sub(ktime_get(), 1);
1094 	watchdog_set_open_deadline(wd_data);
1095 
1096 	/*
1097 	 * If the watchdog is running, prevent its driver from being unloaded,
1098 	 * and schedule an immediate ping.
1099 	 */
1100 	if (watchdog_hw_running(wdd)) {
1101 		__module_get(wdd->ops->owner);
1102 		get_device(&wd_data->dev);
1103 		if (handle_boot_enabled)
1104 			hrtimer_start(&wd_data->timer, 0,
1105 				      HRTIMER_MODE_REL_HARD);
1106 		else
1107 			pr_info("watchdog%d running and kernel based pre-userspace handler disabled\n",
1108 				wdd->id);
1109 	}
1110 
1111 	return 0;
1112 }
1113 
1114 /*
1115  * watchdog_cdev_unregister - unregister watchdog character device
1116  * @wdd: Watchdog device
1117  *
1118  * Unregister watchdog character device and if needed the legacy
1119  * /dev/watchdog device.
1120  */
1121 static void watchdog_cdev_unregister(struct watchdog_device *wdd)
1122 {
1123 	struct watchdog_core_data *wd_data = wdd->wd_data;
1124 
1125 	cdev_device_del(&wd_data->cdev, &wd_data->dev);
1126 	if (wdd->id == 0) {
1127 		misc_deregister(&watchdog_miscdev);
1128 		old_wd_data = NULL;
1129 	}
1130 
1131 	if (watchdog_active(wdd) &&
1132 	    test_bit(WDOG_STOP_ON_UNREGISTER, &wdd->status)) {
1133 		watchdog_stop(wdd);
1134 	}
1135 
1136 	watchdog_hrtimer_pretimeout_stop(wdd);
1137 
1138 	mutex_lock(&wd_data->lock);
1139 	wd_data->wdd = NULL;
1140 	wdd->wd_data = NULL;
1141 	mutex_unlock(&wd_data->lock);
1142 
1143 	hrtimer_cancel(&wd_data->timer);
1144 	kthread_cancel_work_sync(&wd_data->work);
1145 
1146 	put_device(&wd_data->dev);
1147 }
1148 
1149 /**
1150  * watchdog_dev_register - register a watchdog device
1151  * @wdd: Watchdog device
1152  *
1153  * Register a watchdog device including handling the legacy
1154  * /dev/watchdog node. /dev/watchdog is actually a miscdevice and
1155  * thus we set it up like that.
1156  *
1157  * Return: 0 if successful, error otherwise.
1158  */
1159 int watchdog_dev_register(struct watchdog_device *wdd)
1160 {
1161 	int ret;
1162 
1163 	ret = watchdog_cdev_register(wdd);
1164 	if (ret)
1165 		return ret;
1166 
1167 	ret = watchdog_register_pretimeout(wdd);
1168 	if (ret)
1169 		watchdog_cdev_unregister(wdd);
1170 
1171 	return ret;
1172 }
1173 
1174 /**
1175  * watchdog_dev_unregister - unregister a watchdog device
1176  * @wdd: watchdog device
1177  *
1178  * Unregister watchdog device and if needed the legacy
1179  * /dev/watchdog device.
1180  */
1181 void watchdog_dev_unregister(struct watchdog_device *wdd)
1182 {
1183 	watchdog_unregister_pretimeout(wdd);
1184 	watchdog_cdev_unregister(wdd);
1185 }
1186 
1187 /**
1188  * watchdog_set_last_hw_keepalive - set last HW keepalive time for watchdog
1189  * @wdd:		Watchdog device
1190  * @last_ping_ms:	Time since last HW heartbeat
1191  *
1192  * Adjusts the last known HW keepalive time for a watchdog timer.
1193  * This is needed if the watchdog is already running when the probe
1194  * function is called, and it can't be pinged immediately. This
1195  * function must be called immediately after watchdog registration,
1196  * and min_hw_heartbeat_ms must be set for this to be useful.
1197  *
1198  * Return: 0 if successful, error otherwise.
1199  */
1200 int watchdog_set_last_hw_keepalive(struct watchdog_device *wdd,
1201 				   unsigned int last_ping_ms)
1202 {
1203 	struct watchdog_core_data *wd_data;
1204 	ktime_t now;
1205 
1206 	if (!wdd)
1207 		return -EINVAL;
1208 
1209 	wd_data = wdd->wd_data;
1210 
1211 	now = ktime_get();
1212 
1213 	wd_data->last_hw_keepalive = ktime_sub(now, ms_to_ktime(last_ping_ms));
1214 
1215 	if (watchdog_hw_running(wdd) && handle_boot_enabled)
1216 		return __watchdog_ping(wdd);
1217 
1218 	return 0;
1219 }
1220 EXPORT_SYMBOL_GPL(watchdog_set_last_hw_keepalive);
1221 
1222 /**
1223  * watchdog_dev_init - init dev part of watchdog core
1224  *
1225  * Allocate a range of chardev nodes to use for watchdog devices.
1226  *
1227  * Return: 0 if successful, error otherwise.
1228  */
1229 int __init watchdog_dev_init(void)
1230 {
1231 	int err;
1232 
1233 	watchdog_kworker = kthread_create_worker(0, "watchdogd");
1234 	if (IS_ERR(watchdog_kworker)) {
1235 		pr_err("Failed to create watchdog kworker\n");
1236 		return PTR_ERR(watchdog_kworker);
1237 	}
1238 	sched_set_fifo(watchdog_kworker->task);
1239 
1240 	err = class_register(&watchdog_class);
1241 	if (err < 0) {
1242 		pr_err("couldn't register class\n");
1243 		goto err_register;
1244 	}
1245 
1246 	err = alloc_chrdev_region(&watchdog_devt, 0, MAX_DOGS, "watchdog");
1247 	if (err < 0) {
1248 		pr_err("watchdog: unable to allocate char dev region\n");
1249 		goto err_alloc;
1250 	}
1251 
1252 	return 0;
1253 
1254 err_alloc:
1255 	class_unregister(&watchdog_class);
1256 err_register:
1257 	kthread_destroy_worker(watchdog_kworker);
1258 	return err;
1259 }
1260 
1261 /**
1262  * watchdog_dev_exit - exit dev part of watchdog core
1263  *
1264  * Release the range of chardev nodes used for watchdog devices.
1265  */
1266 void __exit watchdog_dev_exit(void)
1267 {
1268 	unregister_chrdev_region(watchdog_devt, MAX_DOGS);
1269 	class_unregister(&watchdog_class);
1270 	kthread_destroy_worker(watchdog_kworker);
1271 }
1272 
1273 int watchdog_dev_suspend(struct watchdog_device *wdd)
1274 {
1275 	struct watchdog_core_data *wd_data = wdd->wd_data;
1276 	int ret = 0;
1277 
1278 	if (!wdd->wd_data)
1279 		return -ENODEV;
1280 
1281 	/* ping for the last time before suspend */
1282 	mutex_lock(&wd_data->lock);
1283 	if (watchdog_worker_should_ping(wd_data))
1284 		ret = __watchdog_ping(wd_data->wdd);
1285 	mutex_unlock(&wd_data->lock);
1286 
1287 	if (ret)
1288 		return ret;
1289 
1290 	/*
1291 	 * make sure that watchdog worker will not kick in when the wdog is
1292 	 * suspended
1293 	 */
1294 	hrtimer_cancel(&wd_data->timer);
1295 	kthread_cancel_work_sync(&wd_data->work);
1296 
1297 	return 0;
1298 }
1299 
1300 int watchdog_dev_resume(struct watchdog_device *wdd)
1301 {
1302 	struct watchdog_core_data *wd_data = wdd->wd_data;
1303 	int ret = 0;
1304 
1305 	if (!wdd->wd_data)
1306 		return -ENODEV;
1307 
1308 	/*
1309 	 * __watchdog_ping will also retrigger hrtimer and therefore restore the
1310 	 * ping worker if needed.
1311 	 */
1312 	mutex_lock(&wd_data->lock);
1313 	if (watchdog_worker_should_ping(wd_data))
1314 		ret = __watchdog_ping(wd_data->wdd);
1315 	mutex_unlock(&wd_data->lock);
1316 
1317 	return ret;
1318 }
1319 
1320 module_param(handle_boot_enabled, bool, 0444);
1321 MODULE_PARM_DESC(handle_boot_enabled,
1322 	"Watchdog core auto-updates boot enabled watchdogs before userspace takes over (default="
1323 	__MODULE_STRING(IS_ENABLED(CONFIG_WATCHDOG_HANDLE_BOOT_ENABLED)) ")");
1324 
1325 module_param(open_timeout, uint, 0644);
1326 MODULE_PARM_DESC(open_timeout,
1327 	"Maximum time (in seconds, 0 means infinity) for userspace to take over a running watchdog (default="
1328 	__MODULE_STRING(CONFIG_WATCHDOG_OPEN_TIMEOUT) ")");
1329