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_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 			set_bit(WDOG_HW_RUNNING, &wdd->status);
272 			wd_data->last_keepalive = started_at;
273 			wd_data->last_hw_keepalive = started_at;
274 			watchdog_update_worker(wdd);
275 			watchdog_hrtimer_pretimeout_start(wdd);
276 		}
277 	}
278 
279 	return err;
280 }
281 
282 /*
283  * watchdog_stop - wrapper to stop the watchdog
284  * @wdd: The watchdog device to stop
285  *
286  * Stop the watchdog if it is still active and unmark it active.
287  * If the 'nowayout' feature was set, the watchdog cannot be stopped.
288  * The caller must hold wd_data->lock.
289  *
290  * Return: 0 on success or a negative errno code for failure.
291  */
292 static int watchdog_stop(struct watchdog_device *wdd)
293 {
294 	int err = 0;
295 
296 	if (!watchdog_active(wdd))
297 		return 0;
298 
299 	if (test_bit(WDOG_NO_WAY_OUT, &wdd->status)) {
300 		pr_info("watchdog%d: nowayout prevents watchdog being stopped!\n",
301 			wdd->id);
302 		return -EBUSY;
303 	}
304 
305 	if (wdd->ops->stop) {
306 		clear_bit(WDOG_HW_RUNNING, &wdd->status);
307 		err = wdd->ops->stop(wdd);
308 		trace_watchdog_stop(wdd, err);
309 	} else {
310 		set_bit(WDOG_HW_RUNNING, &wdd->status);
311 	}
312 
313 	if (err == 0) {
314 		clear_bit(WDOG_ACTIVE, &wdd->status);
315 		watchdog_update_worker(wdd);
316 		watchdog_hrtimer_pretimeout_stop(wdd);
317 	}
318 
319 	return err;
320 }
321 
322 /*
323  * watchdog_get_status - wrapper to get the watchdog status
324  * @wdd: The watchdog device to get the status from
325  *
326  * Get the watchdog's status flags.
327  * The caller must hold wd_data->lock.
328  *
329  * Return: watchdog's status flags.
330  */
331 static unsigned int watchdog_get_status(struct watchdog_device *wdd)
332 {
333 	struct watchdog_core_data *wd_data = wdd->wd_data;
334 	unsigned int status;
335 
336 	if (wdd->ops->status)
337 		status = wdd->ops->status(wdd);
338 	else
339 		status = wdd->bootstatus & (WDIOF_CARDRESET |
340 					    WDIOF_OVERHEAT |
341 					    WDIOF_FANFAULT |
342 					    WDIOF_EXTERN1 |
343 					    WDIOF_EXTERN2 |
344 					    WDIOF_POWERUNDER |
345 					    WDIOF_POWEROVER);
346 
347 	if (test_bit(_WDOG_ALLOW_RELEASE, &wd_data->status))
348 		status |= WDIOF_MAGICCLOSE;
349 
350 	if (test_and_clear_bit(_WDOG_KEEPALIVE, &wd_data->status))
351 		status |= WDIOF_KEEPALIVEPING;
352 
353 	if (IS_ENABLED(CONFIG_WATCHDOG_HRTIMER_PRETIMEOUT))
354 		status |= WDIOF_PRETIMEOUT;
355 
356 	return status;
357 }
358 
359 /*
360  * watchdog_set_timeout - set the watchdog timer timeout
361  * @wdd:	The watchdog device to set the timeout for
362  * @timeout:	Timeout to set in seconds
363  *
364  * The caller must hold wd_data->lock.
365  *
366  * Return: 0 if successful, error otherwise.
367  */
368 static int watchdog_set_timeout(struct watchdog_device *wdd,
369 							unsigned int timeout)
370 {
371 	int err = 0;
372 
373 	if (!(wdd->info->options & WDIOF_SETTIMEOUT))
374 		return -EOPNOTSUPP;
375 
376 	if (watchdog_timeout_invalid(wdd, timeout))
377 		return -EINVAL;
378 
379 	if (wdd->ops->set_timeout) {
380 		err = wdd->ops->set_timeout(wdd, timeout);
381 		trace_watchdog_set_timeout(wdd, timeout, err);
382 	} else {
383 		wdd->timeout = timeout;
384 		/* Disable pretimeout if it doesn't fit the new timeout */
385 		if (wdd->pretimeout >= wdd->timeout)
386 			wdd->pretimeout = 0;
387 	}
388 
389 	watchdog_update_worker(wdd);
390 
391 	return err;
392 }
393 
394 /*
395  * watchdog_set_pretimeout - set the watchdog timer pretimeout
396  * @wdd:	The watchdog device to set the timeout for
397  * @timeout:	pretimeout to set in seconds
398  *
399  * Return: 0 if successful, error otherwise.
400  */
401 static int watchdog_set_pretimeout(struct watchdog_device *wdd,
402 				   unsigned int timeout)
403 {
404 	int err = 0;
405 
406 	if (!watchdog_have_pretimeout(wdd))
407 		return -EOPNOTSUPP;
408 
409 	if (watchdog_pretimeout_invalid(wdd, timeout))
410 		return -EINVAL;
411 
412 	if (wdd->ops->set_pretimeout && (wdd->info->options & WDIOF_PRETIMEOUT))
413 		err = wdd->ops->set_pretimeout(wdd, timeout);
414 	else
415 		wdd->pretimeout = timeout;
416 
417 	return err;
418 }
419 
420 /*
421  * watchdog_get_timeleft - wrapper to get the time left before a reboot
422  * @wdd:	The watchdog device to get the remaining time from
423  * @timeleft:	The time that's left
424  *
425  * Get the time before a watchdog will reboot (if not pinged).
426  * The caller must hold wd_data->lock.
427  *
428  * Return: 0 if successful, error otherwise.
429  */
430 static int watchdog_get_timeleft(struct watchdog_device *wdd,
431 							unsigned int *timeleft)
432 {
433 	*timeleft = 0;
434 
435 	if (!wdd->ops->get_timeleft)
436 		return -EOPNOTSUPP;
437 
438 	*timeleft = wdd->ops->get_timeleft(wdd);
439 
440 	return 0;
441 }
442 
443 #ifdef CONFIG_WATCHDOG_SYSFS
444 static ssize_t nowayout_show(struct device *dev, struct device_attribute *attr,
445 				char *buf)
446 {
447 	struct watchdog_device *wdd = dev_get_drvdata(dev);
448 
449 	return sysfs_emit(buf, "%d\n", !!test_bit(WDOG_NO_WAY_OUT,
450 						  &wdd->status));
451 }
452 
453 static ssize_t nowayout_store(struct device *dev, struct device_attribute *attr,
454 				const char *buf, size_t len)
455 {
456 	struct watchdog_device *wdd = dev_get_drvdata(dev);
457 	unsigned int value;
458 	int ret;
459 
460 	ret = kstrtouint(buf, 0, &value);
461 	if (ret)
462 		return ret;
463 	if (value > 1)
464 		return -EINVAL;
465 	/* nowayout cannot be disabled once set */
466 	if (test_bit(WDOG_NO_WAY_OUT, &wdd->status) && !value)
467 		return -EPERM;
468 	watchdog_set_nowayout(wdd, value);
469 	return len;
470 }
471 static DEVICE_ATTR_RW(nowayout);
472 
473 static ssize_t status_show(struct device *dev, struct device_attribute *attr,
474 				char *buf)
475 {
476 	struct watchdog_device *wdd = dev_get_drvdata(dev);
477 	struct watchdog_core_data *wd_data = wdd->wd_data;
478 	unsigned int status;
479 
480 	mutex_lock(&wd_data->lock);
481 	status = watchdog_get_status(wdd);
482 	mutex_unlock(&wd_data->lock);
483 
484 	return sysfs_emit(buf, "0x%x\n", status);
485 }
486 static DEVICE_ATTR_RO(status);
487 
488 static ssize_t bootstatus_show(struct device *dev,
489 				struct device_attribute *attr, char *buf)
490 {
491 	struct watchdog_device *wdd = dev_get_drvdata(dev);
492 
493 	return sysfs_emit(buf, "%u\n", wdd->bootstatus);
494 }
495 static DEVICE_ATTR_RO(bootstatus);
496 
497 static ssize_t timeleft_show(struct device *dev, struct device_attribute *attr,
498 				char *buf)
499 {
500 	struct watchdog_device *wdd = dev_get_drvdata(dev);
501 	struct watchdog_core_data *wd_data = wdd->wd_data;
502 	ssize_t status;
503 	unsigned int val;
504 
505 	mutex_lock(&wd_data->lock);
506 	status = watchdog_get_timeleft(wdd, &val);
507 	mutex_unlock(&wd_data->lock);
508 	if (!status)
509 		status = sysfs_emit(buf, "%u\n", val);
510 
511 	return status;
512 }
513 static DEVICE_ATTR_RO(timeleft);
514 
515 static ssize_t timeout_show(struct device *dev, struct device_attribute *attr,
516 				char *buf)
517 {
518 	struct watchdog_device *wdd = dev_get_drvdata(dev);
519 
520 	return sysfs_emit(buf, "%u\n", wdd->timeout);
521 }
522 static DEVICE_ATTR_RO(timeout);
523 
524 static ssize_t min_timeout_show(struct device *dev,
525 				struct device_attribute *attr, char *buf)
526 {
527 	struct watchdog_device *wdd = dev_get_drvdata(dev);
528 
529 	return sysfs_emit(buf, "%u\n", wdd->min_timeout);
530 }
531 static DEVICE_ATTR_RO(min_timeout);
532 
533 static ssize_t max_timeout_show(struct device *dev,
534 				struct device_attribute *attr, char *buf)
535 {
536 	struct watchdog_device *wdd = dev_get_drvdata(dev);
537 
538 	return sysfs_emit(buf, "%u\n", wdd->max_timeout);
539 }
540 static DEVICE_ATTR_RO(max_timeout);
541 
542 static ssize_t pretimeout_show(struct device *dev,
543 			       struct device_attribute *attr, char *buf)
544 {
545 	struct watchdog_device *wdd = dev_get_drvdata(dev);
546 
547 	return sysfs_emit(buf, "%u\n", wdd->pretimeout);
548 }
549 static DEVICE_ATTR_RO(pretimeout);
550 
551 static ssize_t options_show(struct device *dev, struct device_attribute *attr,
552 			    char *buf)
553 {
554 	struct watchdog_device *wdd = dev_get_drvdata(dev);
555 
556 	return sysfs_emit(buf, "0x%x\n", wdd->info->options);
557 }
558 static DEVICE_ATTR_RO(options);
559 
560 static ssize_t fw_version_show(struct device *dev, struct device_attribute *attr,
561 			       char *buf)
562 {
563 	struct watchdog_device *wdd = dev_get_drvdata(dev);
564 
565 	return sysfs_emit(buf, "%d\n", wdd->info->firmware_version);
566 }
567 static DEVICE_ATTR_RO(fw_version);
568 
569 static ssize_t identity_show(struct device *dev, struct device_attribute *attr,
570 				char *buf)
571 {
572 	struct watchdog_device *wdd = dev_get_drvdata(dev);
573 
574 	return sysfs_emit(buf, "%s\n", wdd->info->identity);
575 }
576 static DEVICE_ATTR_RO(identity);
577 
578 static ssize_t state_show(struct device *dev, struct device_attribute *attr,
579 				char *buf)
580 {
581 	struct watchdog_device *wdd = dev_get_drvdata(dev);
582 
583 	if (watchdog_active(wdd))
584 		return sysfs_emit(buf, "active\n");
585 
586 	return sysfs_emit(buf, "inactive\n");
587 }
588 static DEVICE_ATTR_RO(state);
589 
590 static ssize_t pretimeout_available_governors_show(struct device *dev,
591 				   struct device_attribute *attr, char *buf)
592 {
593 	return watchdog_pretimeout_available_governors_get(buf);
594 }
595 static DEVICE_ATTR_RO(pretimeout_available_governors);
596 
597 static ssize_t pretimeout_governor_show(struct device *dev,
598 					struct device_attribute *attr,
599 					char *buf)
600 {
601 	struct watchdog_device *wdd = dev_get_drvdata(dev);
602 
603 	return watchdog_pretimeout_governor_get(wdd, buf);
604 }
605 
606 static ssize_t pretimeout_governor_store(struct device *dev,
607 					 struct device_attribute *attr,
608 					 const char *buf, size_t count)
609 {
610 	struct watchdog_device *wdd = dev_get_drvdata(dev);
611 	int ret = watchdog_pretimeout_governor_set(wdd, buf);
612 
613 	if (!ret)
614 		ret = count;
615 
616 	return ret;
617 }
618 static DEVICE_ATTR_RW(pretimeout_governor);
619 
620 static umode_t wdt_is_visible(struct kobject *kobj, struct attribute *attr,
621 				int n)
622 {
623 	struct device *dev = kobj_to_dev(kobj);
624 	struct watchdog_device *wdd = dev_get_drvdata(dev);
625 	umode_t mode = attr->mode;
626 
627 	if (attr == &dev_attr_timeleft.attr && !wdd->ops->get_timeleft)
628 		mode = 0;
629 	else if (attr == &dev_attr_pretimeout.attr && !watchdog_have_pretimeout(wdd))
630 		mode = 0;
631 	else if ((attr == &dev_attr_pretimeout_governor.attr ||
632 		  attr == &dev_attr_pretimeout_available_governors.attr) &&
633 		 (!watchdog_have_pretimeout(wdd) || !IS_ENABLED(CONFIG_WATCHDOG_PRETIMEOUT_GOV)))
634 		mode = 0;
635 
636 	return mode;
637 }
638 static struct attribute *wdt_attrs[] = {
639 	&dev_attr_state.attr,
640 	&dev_attr_options.attr,
641 	&dev_attr_fw_version.attr,
642 	&dev_attr_identity.attr,
643 	&dev_attr_timeout.attr,
644 	&dev_attr_min_timeout.attr,
645 	&dev_attr_max_timeout.attr,
646 	&dev_attr_pretimeout.attr,
647 	&dev_attr_timeleft.attr,
648 	&dev_attr_bootstatus.attr,
649 	&dev_attr_status.attr,
650 	&dev_attr_nowayout.attr,
651 	&dev_attr_pretimeout_governor.attr,
652 	&dev_attr_pretimeout_available_governors.attr,
653 	NULL,
654 };
655 
656 static const struct attribute_group wdt_group = {
657 	.attrs = wdt_attrs,
658 	.is_visible = wdt_is_visible,
659 };
660 __ATTRIBUTE_GROUPS(wdt);
661 #else
662 #define wdt_groups	NULL
663 #endif
664 
665 /*
666  * watchdog_ioctl_op - call the watchdog drivers ioctl op if defined
667  * @wdd: The watchdog device to do the ioctl on
668  * @cmd: Watchdog command
669  * @arg: Argument pointer
670  *
671  * The caller must hold wd_data->lock.
672  *
673  * Return: 0 if successful, error otherwise.
674  */
675 static int watchdog_ioctl_op(struct watchdog_device *wdd, unsigned int cmd,
676 							unsigned long arg)
677 {
678 	if (!wdd->ops->ioctl)
679 		return -ENOIOCTLCMD;
680 
681 	return wdd->ops->ioctl(wdd, cmd, arg);
682 }
683 
684 /*
685  * watchdog_write - writes to the watchdog
686  * @file:	File from VFS
687  * @data:	User address of data
688  * @len:	Length of data
689  * @ppos:	Pointer to the file offset
690  *
691  * A write to a watchdog device is defined as a keepalive ping.
692  * Writing the magic 'V' sequence allows the next close to turn
693  * off the watchdog (if 'nowayout' is not set).
694  *
695  * Return: @len if successful, error otherwise.
696  */
697 static ssize_t watchdog_write(struct file *file, const char __user *data,
698 						size_t len, loff_t *ppos)
699 {
700 	struct watchdog_core_data *wd_data = file->private_data;
701 	struct watchdog_device *wdd;
702 	int err;
703 	size_t i;
704 	char c;
705 
706 	if (len == 0)
707 		return 0;
708 
709 	/*
710 	 * Note: just in case someone wrote the magic character
711 	 * five months ago...
712 	 */
713 	clear_bit(_WDOG_ALLOW_RELEASE, &wd_data->status);
714 
715 	/* scan to see whether or not we got the magic character */
716 	for (i = 0; i != len; i++) {
717 		if (get_user(c, data + i))
718 			return -EFAULT;
719 		if (c == 'V')
720 			set_bit(_WDOG_ALLOW_RELEASE, &wd_data->status);
721 	}
722 
723 	/* someone wrote to us, so we send the watchdog a keepalive ping */
724 
725 	err = -ENODEV;
726 	mutex_lock(&wd_data->lock);
727 	wdd = wd_data->wdd;
728 	if (wdd)
729 		err = watchdog_ping(wdd);
730 	mutex_unlock(&wd_data->lock);
731 
732 	if (err < 0)
733 		return err;
734 
735 	return len;
736 }
737 
738 /*
739  * watchdog_ioctl - handle the different ioctl's for the watchdog device
740  * @file:	File handle to the device
741  * @cmd:	Watchdog command
742  * @arg:	Argument pointer
743  *
744  * The watchdog API defines a common set of functions for all watchdogs
745  * according to their available features.
746  *
747  * Return: 0 if successful, error otherwise.
748  */
749 
750 static long watchdog_ioctl(struct file *file, unsigned int cmd,
751 							unsigned long arg)
752 {
753 	struct watchdog_core_data *wd_data = file->private_data;
754 	void __user *argp = (void __user *)arg;
755 	struct watchdog_device *wdd;
756 	int __user *p = argp;
757 	unsigned int val;
758 	int err;
759 
760 	mutex_lock(&wd_data->lock);
761 
762 	wdd = wd_data->wdd;
763 	if (!wdd) {
764 		err = -ENODEV;
765 		goto out_ioctl;
766 	}
767 
768 	err = watchdog_ioctl_op(wdd, cmd, arg);
769 	if (err != -ENOIOCTLCMD)
770 		goto out_ioctl;
771 
772 	switch (cmd) {
773 	case WDIOC_GETSUPPORT:
774 		err = copy_to_user(argp, wdd->info,
775 			sizeof(struct watchdog_info)) ? -EFAULT : 0;
776 		break;
777 	case WDIOC_GETSTATUS:
778 		val = watchdog_get_status(wdd);
779 		err = put_user(val, p);
780 		break;
781 	case WDIOC_GETBOOTSTATUS:
782 		err = put_user(wdd->bootstatus, p);
783 		break;
784 	case WDIOC_SETOPTIONS:
785 		if (get_user(val, p)) {
786 			err = -EFAULT;
787 			break;
788 		}
789 		if (val & WDIOS_DISABLECARD) {
790 			err = watchdog_stop(wdd);
791 			if (err < 0)
792 				break;
793 		}
794 		if (val & WDIOS_ENABLECARD)
795 			err = watchdog_start(wdd);
796 		break;
797 	case WDIOC_KEEPALIVE:
798 		if (!(wdd->info->options & WDIOF_KEEPALIVEPING)) {
799 			err = -EOPNOTSUPP;
800 			break;
801 		}
802 		err = watchdog_ping(wdd);
803 		break;
804 	case WDIOC_SETTIMEOUT:
805 		if (get_user(val, p)) {
806 			err = -EFAULT;
807 			break;
808 		}
809 		err = watchdog_set_timeout(wdd, val);
810 		if (err < 0)
811 			break;
812 		/* If the watchdog is active then we send a keepalive ping
813 		 * to make sure that the watchdog keep's running (and if
814 		 * possible that it takes the new timeout) */
815 		err = watchdog_ping(wdd);
816 		if (err < 0)
817 			break;
818 		fallthrough;
819 	case WDIOC_GETTIMEOUT:
820 		/* timeout == 0 means that we don't know the timeout */
821 		if (wdd->timeout == 0) {
822 			err = -EOPNOTSUPP;
823 			break;
824 		}
825 		err = put_user(wdd->timeout, p);
826 		break;
827 	case WDIOC_GETTIMELEFT:
828 		err = watchdog_get_timeleft(wdd, &val);
829 		if (err < 0)
830 			break;
831 		err = put_user(val, p);
832 		break;
833 	case WDIOC_SETPRETIMEOUT:
834 		if (get_user(val, p)) {
835 			err = -EFAULT;
836 			break;
837 		}
838 		err = watchdog_set_pretimeout(wdd, val);
839 		break;
840 	case WDIOC_GETPRETIMEOUT:
841 		err = put_user(wdd->pretimeout, p);
842 		break;
843 	default:
844 		err = -ENOTTY;
845 		break;
846 	}
847 
848 out_ioctl:
849 	mutex_unlock(&wd_data->lock);
850 	return err;
851 }
852 
853 /*
854  * watchdog_open - open the /dev/watchdog* devices
855  * @inode:	Inode of device
856  * @file:	File handle to device
857  *
858  * When the /dev/watchdog* device gets opened, we start the watchdog.
859  * Watch out: the /dev/watchdog device is single open, so we make sure
860  * it can only be opened once.
861  *
862  * Return: 0 if successful, error otherwise.
863  */
864 static int watchdog_open(struct inode *inode, struct file *file)
865 {
866 	struct watchdog_core_data *wd_data;
867 	struct watchdog_device *wdd;
868 	bool hw_running;
869 	int err;
870 
871 	/* Get the corresponding watchdog device */
872 	if (imajor(inode) == MISC_MAJOR)
873 		wd_data = old_wd_data;
874 	else
875 		wd_data = container_of(inode->i_cdev, struct watchdog_core_data,
876 				       cdev);
877 
878 	/* the watchdog is single open! */
879 	if (test_and_set_bit(_WDOG_DEV_OPEN, &wd_data->status))
880 		return -EBUSY;
881 
882 	wdd = wd_data->wdd;
883 
884 	/*
885 	 * If the /dev/watchdog device is open, we don't want the module
886 	 * to be unloaded.
887 	 */
888 	hw_running = watchdog_hw_running(wdd);
889 	if (!hw_running && !try_module_get(wdd->ops->owner)) {
890 		err = -EBUSY;
891 		goto out_clear;
892 	}
893 
894 	err = watchdog_start(wdd);
895 	if (err < 0)
896 		goto out_mod;
897 
898 	file->private_data = wd_data;
899 
900 	if (!hw_running)
901 		get_device(&wd_data->dev);
902 
903 	/*
904 	 * open_timeout only applies for the first open from
905 	 * userspace. Set open_deadline to infinity so that the kernel
906 	 * will take care of an always-running hardware watchdog in
907 	 * case the device gets magic-closed or WDIOS_DISABLECARD is
908 	 * applied.
909 	 */
910 	wd_data->open_deadline = KTIME_MAX;
911 
912 	/* dev/watchdog is a virtual (and thus non-seekable) filesystem */
913 	return stream_open(inode, file);
914 
915 out_mod:
916 	module_put(wd_data->wdd->ops->owner);
917 out_clear:
918 	clear_bit(_WDOG_DEV_OPEN, &wd_data->status);
919 	return err;
920 }
921 
922 static void watchdog_core_data_release(struct device *dev)
923 {
924 	struct watchdog_core_data *wd_data;
925 
926 	wd_data = container_of(dev, struct watchdog_core_data, dev);
927 
928 	kfree(wd_data);
929 }
930 
931 /*
932  * watchdog_release - release the watchdog device
933  * @inode:	Inode of device
934  * @file:	File handle to device
935  *
936  * This is the code for when /dev/watchdog gets closed. We will only
937  * stop the watchdog when we have received the magic char (and nowayout
938  * was not set), else the watchdog will keep running.
939  *
940  * Always returns 0.
941  */
942 static int watchdog_release(struct inode *inode, struct file *file)
943 {
944 	struct watchdog_core_data *wd_data = file->private_data;
945 	struct watchdog_device *wdd;
946 	int err = -EBUSY;
947 	bool running;
948 
949 	mutex_lock(&wd_data->lock);
950 
951 	wdd = wd_data->wdd;
952 	if (!wdd)
953 		goto done;
954 
955 	/*
956 	 * We only stop the watchdog if we received the magic character
957 	 * or if WDIOF_MAGICCLOSE is not set. If nowayout was set then
958 	 * watchdog_stop will fail.
959 	 */
960 	if (!watchdog_active(wdd))
961 		err = 0;
962 	else if (test_and_clear_bit(_WDOG_ALLOW_RELEASE, &wd_data->status) ||
963 		 !(wdd->info->options & WDIOF_MAGICCLOSE))
964 		err = watchdog_stop(wdd);
965 
966 	/* If the watchdog was not stopped, send a keepalive ping */
967 	if (err < 0) {
968 		pr_crit("watchdog%d: watchdog did not stop!\n", wdd->id);
969 		watchdog_ping(wdd);
970 	}
971 
972 	watchdog_update_worker(wdd);
973 
974 	/* make sure that /dev/watchdog can be re-opened */
975 	clear_bit(_WDOG_DEV_OPEN, &wd_data->status);
976 
977 done:
978 	running = wdd && watchdog_hw_running(wdd);
979 	mutex_unlock(&wd_data->lock);
980 	/*
981 	 * Allow the owner module to be unloaded again unless the watchdog
982 	 * is still running. If the watchdog is still running, it can not
983 	 * be stopped, and its driver must not be unloaded.
984 	 */
985 	if (!running) {
986 		module_put(wd_data->cdev.owner);
987 		put_device(&wd_data->dev);
988 	}
989 	return 0;
990 }
991 
992 static const struct file_operations watchdog_fops = {
993 	.owner		= THIS_MODULE,
994 	.write		= watchdog_write,
995 	.unlocked_ioctl	= watchdog_ioctl,
996 	.compat_ioctl	= compat_ptr_ioctl,
997 	.open		= watchdog_open,
998 	.release	= watchdog_release,
999 };
1000 
1001 static struct miscdevice watchdog_miscdev = {
1002 	.minor		= WATCHDOG_MINOR,
1003 	.name		= "watchdog",
1004 	.fops		= &watchdog_fops,
1005 };
1006 
1007 static struct class watchdog_class = {
1008 	.name =		"watchdog",
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