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