xref: /linux/kernel/power/main.c (revision b233b28eac0cc37d07c2d007ea08c86c778c5af4)
1 /*
2  * kernel/power/main.c - PM subsystem core functionality.
3  *
4  * Copyright (c) 2003 Patrick Mochel
5  * Copyright (c) 2003 Open Source Development Lab
6  *
7  * This file is released under the GPLv2
8  *
9  */
10 
11 #include <linux/module.h>
12 #include <linux/suspend.h>
13 #include <linux/kobject.h>
14 #include <linux/string.h>
15 #include <linux/delay.h>
16 #include <linux/errno.h>
17 #include <linux/kmod.h>
18 #include <linux/init.h>
19 #include <linux/console.h>
20 #include <linux/cpu.h>
21 #include <linux/resume-trace.h>
22 #include <linux/freezer.h>
23 #include <linux/vmstat.h>
24 #include <linux/syscalls.h>
25 
26 #include "power.h"
27 
28 DEFINE_MUTEX(pm_mutex);
29 
30 unsigned int pm_flags;
31 EXPORT_SYMBOL(pm_flags);
32 
33 #ifdef CONFIG_PM_SLEEP
34 
35 /* Routines for PM-transition notifications */
36 
37 static BLOCKING_NOTIFIER_HEAD(pm_chain_head);
38 
39 int register_pm_notifier(struct notifier_block *nb)
40 {
41 	return blocking_notifier_chain_register(&pm_chain_head, nb);
42 }
43 EXPORT_SYMBOL_GPL(register_pm_notifier);
44 
45 int unregister_pm_notifier(struct notifier_block *nb)
46 {
47 	return blocking_notifier_chain_unregister(&pm_chain_head, nb);
48 }
49 EXPORT_SYMBOL_GPL(unregister_pm_notifier);
50 
51 int pm_notifier_call_chain(unsigned long val)
52 {
53 	return (blocking_notifier_call_chain(&pm_chain_head, val, NULL)
54 			== NOTIFY_BAD) ? -EINVAL : 0;
55 }
56 
57 #ifdef CONFIG_PM_DEBUG
58 int pm_test_level = TEST_NONE;
59 
60 static const char * const pm_tests[__TEST_AFTER_LAST] = {
61 	[TEST_NONE] = "none",
62 	[TEST_CORE] = "core",
63 	[TEST_CPUS] = "processors",
64 	[TEST_PLATFORM] = "platform",
65 	[TEST_DEVICES] = "devices",
66 	[TEST_FREEZER] = "freezer",
67 };
68 
69 static ssize_t pm_test_show(struct kobject *kobj, struct kobj_attribute *attr,
70 				char *buf)
71 {
72 	char *s = buf;
73 	int level;
74 
75 	for (level = TEST_FIRST; level <= TEST_MAX; level++)
76 		if (pm_tests[level]) {
77 			if (level == pm_test_level)
78 				s += sprintf(s, "[%s] ", pm_tests[level]);
79 			else
80 				s += sprintf(s, "%s ", pm_tests[level]);
81 		}
82 
83 	if (s != buf)
84 		/* convert the last space to a newline */
85 		*(s-1) = '\n';
86 
87 	return (s - buf);
88 }
89 
90 static ssize_t pm_test_store(struct kobject *kobj, struct kobj_attribute *attr,
91 				const char *buf, size_t n)
92 {
93 	const char * const *s;
94 	int level;
95 	char *p;
96 	int len;
97 	int error = -EINVAL;
98 
99 	p = memchr(buf, '\n', n);
100 	len = p ? p - buf : n;
101 
102 	mutex_lock(&pm_mutex);
103 
104 	level = TEST_FIRST;
105 	for (s = &pm_tests[level]; level <= TEST_MAX; s++, level++)
106 		if (*s && len == strlen(*s) && !strncmp(buf, *s, len)) {
107 			pm_test_level = level;
108 			error = 0;
109 			break;
110 		}
111 
112 	mutex_unlock(&pm_mutex);
113 
114 	return error ? error : n;
115 }
116 
117 power_attr(pm_test);
118 #endif /* CONFIG_PM_DEBUG */
119 
120 #endif /* CONFIG_PM_SLEEP */
121 
122 #ifdef CONFIG_SUSPEND
123 
124 static int suspend_test(int level)
125 {
126 #ifdef CONFIG_PM_DEBUG
127 	if (pm_test_level == level) {
128 		printk(KERN_INFO "suspend debug: Waiting for 5 seconds.\n");
129 		mdelay(5000);
130 		return 1;
131 	}
132 #endif /* !CONFIG_PM_DEBUG */
133 	return 0;
134 }
135 
136 #ifdef CONFIG_PM_TEST_SUSPEND
137 
138 /*
139  * We test the system suspend code by setting an RTC wakealarm a short
140  * time in the future, then suspending.  Suspending the devices won't
141  * normally take long ... some systems only need a few milliseconds.
142  *
143  * The time it takes is system-specific though, so when we test this
144  * during system bootup we allow a LOT of time.
145  */
146 #define TEST_SUSPEND_SECONDS	5
147 
148 static unsigned long suspend_test_start_time;
149 
150 static void suspend_test_start(void)
151 {
152 	/* FIXME Use better timebase than "jiffies", ideally a clocksource.
153 	 * What we want is a hardware counter that will work correctly even
154 	 * during the irqs-are-off stages of the suspend/resume cycle...
155 	 */
156 	suspend_test_start_time = jiffies;
157 }
158 
159 static void suspend_test_finish(const char *label)
160 {
161 	long nj = jiffies - suspend_test_start_time;
162 	unsigned msec;
163 
164 	msec = jiffies_to_msecs(abs(nj));
165 	pr_info("PM: %s took %d.%03d seconds\n", label,
166 			msec / 1000, msec % 1000);
167 
168 	/* Warning on suspend means the RTC alarm period needs to be
169 	 * larger -- the system was sooo slooowwww to suspend that the
170 	 * alarm (should have) fired before the system went to sleep!
171 	 *
172 	 * Warning on either suspend or resume also means the system
173 	 * has some performance issues.  The stack dump of a WARN_ON
174 	 * is more likely to get the right attention than a printk...
175 	 */
176 	WARN(msec > (TEST_SUSPEND_SECONDS * 1000), "Component: %s\n", label);
177 }
178 
179 #else
180 
181 static void suspend_test_start(void)
182 {
183 }
184 
185 static void suspend_test_finish(const char *label)
186 {
187 }
188 
189 #endif
190 
191 /* This is just an arbitrary number */
192 #define FREE_PAGE_NUMBER (100)
193 
194 static struct platform_suspend_ops *suspend_ops;
195 
196 /**
197  *	suspend_set_ops - Set the global suspend method table.
198  *	@ops:	Pointer to ops structure.
199  */
200 
201 void suspend_set_ops(struct platform_suspend_ops *ops)
202 {
203 	mutex_lock(&pm_mutex);
204 	suspend_ops = ops;
205 	mutex_unlock(&pm_mutex);
206 }
207 
208 /**
209  * suspend_valid_only_mem - generic memory-only valid callback
210  *
211  * Platform drivers that implement mem suspend only and only need
212  * to check for that in their .valid callback can use this instead
213  * of rolling their own .valid callback.
214  */
215 int suspend_valid_only_mem(suspend_state_t state)
216 {
217 	return state == PM_SUSPEND_MEM;
218 }
219 
220 /**
221  *	suspend_prepare - Do prep work before entering low-power state.
222  *
223  *	This is common code that is called for each state that we're entering.
224  *	Run suspend notifiers, allocate a console and stop all processes.
225  */
226 static int suspend_prepare(void)
227 {
228 	int error;
229 	unsigned int free_pages;
230 
231 	if (!suspend_ops || !suspend_ops->enter)
232 		return -EPERM;
233 
234 	pm_prepare_console();
235 
236 	error = pm_notifier_call_chain(PM_SUSPEND_PREPARE);
237 	if (error)
238 		goto Finish;
239 
240 	error = usermodehelper_disable();
241 	if (error)
242 		goto Finish;
243 
244 	if (suspend_freeze_processes()) {
245 		error = -EAGAIN;
246 		goto Thaw;
247 	}
248 
249 	free_pages = global_page_state(NR_FREE_PAGES);
250 	if (free_pages < FREE_PAGE_NUMBER) {
251 		pr_debug("PM: free some memory\n");
252 		shrink_all_memory(FREE_PAGE_NUMBER - free_pages);
253 		if (nr_free_pages() < FREE_PAGE_NUMBER) {
254 			error = -ENOMEM;
255 			printk(KERN_ERR "PM: No enough memory\n");
256 		}
257 	}
258 	if (!error)
259 		return 0;
260 
261  Thaw:
262 	suspend_thaw_processes();
263 	usermodehelper_enable();
264  Finish:
265 	pm_notifier_call_chain(PM_POST_SUSPEND);
266 	pm_restore_console();
267 	return error;
268 }
269 
270 /* default implementation */
271 void __attribute__ ((weak)) arch_suspend_disable_irqs(void)
272 {
273 	local_irq_disable();
274 }
275 
276 /* default implementation */
277 void __attribute__ ((weak)) arch_suspend_enable_irqs(void)
278 {
279 	local_irq_enable();
280 }
281 
282 /**
283  *	suspend_enter - enter the desired system sleep state.
284  *	@state:		state to enter
285  *
286  *	This function should be called after devices have been suspended.
287  */
288 static int suspend_enter(suspend_state_t state)
289 {
290 	int error = 0;
291 
292 	device_pm_lock();
293 	arch_suspend_disable_irqs();
294 	BUG_ON(!irqs_disabled());
295 
296 	if ((error = device_power_down(PMSG_SUSPEND))) {
297 		printk(KERN_ERR "PM: Some devices failed to power down\n");
298 		goto Done;
299 	}
300 
301 	if (!suspend_test(TEST_CORE))
302 		error = suspend_ops->enter(state);
303 
304 	device_power_up(PMSG_RESUME);
305  Done:
306 	arch_suspend_enable_irqs();
307 	BUG_ON(irqs_disabled());
308 	device_pm_unlock();
309 	return error;
310 }
311 
312 /**
313  *	suspend_devices_and_enter - suspend devices and enter the desired system
314  *				    sleep state.
315  *	@state:		  state to enter
316  */
317 int suspend_devices_and_enter(suspend_state_t state)
318 {
319 	int error;
320 
321 	if (!suspend_ops)
322 		return -ENOSYS;
323 
324 	if (suspend_ops->begin) {
325 		error = suspend_ops->begin(state);
326 		if (error)
327 			goto Close;
328 	}
329 	suspend_console();
330 	suspend_test_start();
331 	error = device_suspend(PMSG_SUSPEND);
332 	if (error) {
333 		printk(KERN_ERR "PM: Some devices failed to suspend\n");
334 		goto Recover_platform;
335 	}
336 	suspend_test_finish("suspend devices");
337 	if (suspend_test(TEST_DEVICES))
338 		goto Recover_platform;
339 
340 	if (suspend_ops->prepare) {
341 		error = suspend_ops->prepare();
342 		if (error)
343 			goto Resume_devices;
344 	}
345 
346 	if (suspend_test(TEST_PLATFORM))
347 		goto Finish;
348 
349 	error = disable_nonboot_cpus();
350 	if (!error && !suspend_test(TEST_CPUS))
351 		suspend_enter(state);
352 
353 	enable_nonboot_cpus();
354  Finish:
355 	if (suspend_ops->finish)
356 		suspend_ops->finish();
357  Resume_devices:
358 	suspend_test_start();
359 	device_resume(PMSG_RESUME);
360 	suspend_test_finish("resume devices");
361 	resume_console();
362  Close:
363 	if (suspend_ops->end)
364 		suspend_ops->end();
365 	return error;
366 
367  Recover_platform:
368 	if (suspend_ops->recover)
369 		suspend_ops->recover();
370 	goto Resume_devices;
371 }
372 
373 /**
374  *	suspend_finish - Do final work before exiting suspend sequence.
375  *
376  *	Call platform code to clean up, restart processes, and free the
377  *	console that we've allocated. This is not called for suspend-to-disk.
378  */
379 static void suspend_finish(void)
380 {
381 	suspend_thaw_processes();
382 	usermodehelper_enable();
383 	pm_notifier_call_chain(PM_POST_SUSPEND);
384 	pm_restore_console();
385 }
386 
387 
388 
389 
390 static const char * const pm_states[PM_SUSPEND_MAX] = {
391 	[PM_SUSPEND_STANDBY]	= "standby",
392 	[PM_SUSPEND_MEM]	= "mem",
393 };
394 
395 static inline int valid_state(suspend_state_t state)
396 {
397 	/* All states need lowlevel support and need to be valid
398 	 * to the lowlevel implementation, no valid callback
399 	 * implies that none are valid. */
400 	if (!suspend_ops || !suspend_ops->valid || !suspend_ops->valid(state))
401 		return 0;
402 	return 1;
403 }
404 
405 
406 /**
407  *	enter_state - Do common work of entering low-power state.
408  *	@state:		pm_state structure for state we're entering.
409  *
410  *	Make sure we're the only ones trying to enter a sleep state. Fail
411  *	if someone has beat us to it, since we don't want anything weird to
412  *	happen when we wake up.
413  *	Then, do the setup for suspend, enter the state, and cleaup (after
414  *	we've woken up).
415  */
416 static int enter_state(suspend_state_t state)
417 {
418 	int error;
419 
420 	if (!valid_state(state))
421 		return -ENODEV;
422 
423 	if (!mutex_trylock(&pm_mutex))
424 		return -EBUSY;
425 
426 	printk(KERN_INFO "PM: Syncing filesystems ... ");
427 	sys_sync();
428 	printk("done.\n");
429 
430 	pr_debug("PM: Preparing system for %s sleep\n", pm_states[state]);
431 	error = suspend_prepare();
432 	if (error)
433 		goto Unlock;
434 
435 	if (suspend_test(TEST_FREEZER))
436 		goto Finish;
437 
438 	pr_debug("PM: Entering %s sleep\n", pm_states[state]);
439 	error = suspend_devices_and_enter(state);
440 
441  Finish:
442 	pr_debug("PM: Finishing wakeup.\n");
443 	suspend_finish();
444  Unlock:
445 	mutex_unlock(&pm_mutex);
446 	return error;
447 }
448 
449 
450 /**
451  *	pm_suspend - Externally visible function for suspending system.
452  *	@state:		Enumerated value of state to enter.
453  *
454  *	Determine whether or not value is within range, get state
455  *	structure, and enter (above).
456  */
457 
458 int pm_suspend(suspend_state_t state)
459 {
460 	if (state > PM_SUSPEND_ON && state <= PM_SUSPEND_MAX)
461 		return enter_state(state);
462 	return -EINVAL;
463 }
464 
465 EXPORT_SYMBOL(pm_suspend);
466 
467 #endif /* CONFIG_SUSPEND */
468 
469 struct kobject *power_kobj;
470 
471 /**
472  *	state - control system power state.
473  *
474  *	show() returns what states are supported, which is hard-coded to
475  *	'standby' (Power-On Suspend), 'mem' (Suspend-to-RAM), and
476  *	'disk' (Suspend-to-Disk).
477  *
478  *	store() accepts one of those strings, translates it into the
479  *	proper enumerated value, and initiates a suspend transition.
480  */
481 
482 static ssize_t state_show(struct kobject *kobj, struct kobj_attribute *attr,
483 			  char *buf)
484 {
485 	char *s = buf;
486 #ifdef CONFIG_SUSPEND
487 	int i;
488 
489 	for (i = 0; i < PM_SUSPEND_MAX; i++) {
490 		if (pm_states[i] && valid_state(i))
491 			s += sprintf(s,"%s ", pm_states[i]);
492 	}
493 #endif
494 #ifdef CONFIG_HIBERNATION
495 	s += sprintf(s, "%s\n", "disk");
496 #else
497 	if (s != buf)
498 		/* convert the last space to a newline */
499 		*(s-1) = '\n';
500 #endif
501 	return (s - buf);
502 }
503 
504 static ssize_t state_store(struct kobject *kobj, struct kobj_attribute *attr,
505 			   const char *buf, size_t n)
506 {
507 #ifdef CONFIG_SUSPEND
508 	suspend_state_t state = PM_SUSPEND_STANDBY;
509 	const char * const *s;
510 #endif
511 	char *p;
512 	int len;
513 	int error = -EINVAL;
514 
515 	p = memchr(buf, '\n', n);
516 	len = p ? p - buf : n;
517 
518 	/* First, check if we are requested to hibernate */
519 	if (len == 4 && !strncmp(buf, "disk", len)) {
520 		error = hibernate();
521   goto Exit;
522 	}
523 
524 #ifdef CONFIG_SUSPEND
525 	for (s = &pm_states[state]; state < PM_SUSPEND_MAX; s++, state++) {
526 		if (*s && len == strlen(*s) && !strncmp(buf, *s, len))
527 			break;
528 	}
529 	if (state < PM_SUSPEND_MAX && *s)
530 		error = enter_state(state);
531 #endif
532 
533  Exit:
534 	return error ? error : n;
535 }
536 
537 power_attr(state);
538 
539 #ifdef CONFIG_PM_TRACE
540 int pm_trace_enabled;
541 
542 static ssize_t pm_trace_show(struct kobject *kobj, struct kobj_attribute *attr,
543 			     char *buf)
544 {
545 	return sprintf(buf, "%d\n", pm_trace_enabled);
546 }
547 
548 static ssize_t
549 pm_trace_store(struct kobject *kobj, struct kobj_attribute *attr,
550 	       const char *buf, size_t n)
551 {
552 	int val;
553 
554 	if (sscanf(buf, "%d", &val) == 1) {
555 		pm_trace_enabled = !!val;
556 		return n;
557 	}
558 	return -EINVAL;
559 }
560 
561 power_attr(pm_trace);
562 #endif /* CONFIG_PM_TRACE */
563 
564 static struct attribute * g[] = {
565 	&state_attr.attr,
566 #ifdef CONFIG_PM_TRACE
567 	&pm_trace_attr.attr,
568 #endif
569 #if defined(CONFIG_PM_SLEEP) && defined(CONFIG_PM_DEBUG)
570 	&pm_test_attr.attr,
571 #endif
572 	NULL,
573 };
574 
575 static struct attribute_group attr_group = {
576 	.attrs = g,
577 };
578 
579 
580 static int __init pm_init(void)
581 {
582 	power_kobj = kobject_create_and_add("power", NULL);
583 	if (!power_kobj)
584 		return -ENOMEM;
585 	return sysfs_create_group(power_kobj, &attr_group);
586 }
587 
588 core_initcall(pm_init);
589 
590 
591 #ifdef CONFIG_PM_TEST_SUSPEND
592 
593 #include <linux/rtc.h>
594 
595 /*
596  * To test system suspend, we need a hands-off mechanism to resume the
597  * system.  RTCs wake alarms are a common self-contained mechanism.
598  */
599 
600 static void __init test_wakealarm(struct rtc_device *rtc, suspend_state_t state)
601 {
602 	static char err_readtime[] __initdata =
603 		KERN_ERR "PM: can't read %s time, err %d\n";
604 	static char err_wakealarm [] __initdata =
605 		KERN_ERR "PM: can't set %s wakealarm, err %d\n";
606 	static char err_suspend[] __initdata =
607 		KERN_ERR "PM: suspend test failed, error %d\n";
608 	static char info_test[] __initdata =
609 		KERN_INFO "PM: test RTC wakeup from '%s' suspend\n";
610 
611 	unsigned long		now;
612 	struct rtc_wkalrm	alm;
613 	int			status;
614 
615 	/* this may fail if the RTC hasn't been initialized */
616 	status = rtc_read_time(rtc, &alm.time);
617 	if (status < 0) {
618 		printk(err_readtime, dev_name(&rtc->dev), status);
619 		return;
620 	}
621 	rtc_tm_to_time(&alm.time, &now);
622 
623 	memset(&alm, 0, sizeof alm);
624 	rtc_time_to_tm(now + TEST_SUSPEND_SECONDS, &alm.time);
625 	alm.enabled = true;
626 
627 	status = rtc_set_alarm(rtc, &alm);
628 	if (status < 0) {
629 		printk(err_wakealarm, dev_name(&rtc->dev), status);
630 		return;
631 	}
632 
633 	if (state == PM_SUSPEND_MEM) {
634 		printk(info_test, pm_states[state]);
635 		status = pm_suspend(state);
636 		if (status == -ENODEV)
637 			state = PM_SUSPEND_STANDBY;
638 	}
639 	if (state == PM_SUSPEND_STANDBY) {
640 		printk(info_test, pm_states[state]);
641 		status = pm_suspend(state);
642 	}
643 	if (status < 0)
644 		printk(err_suspend, status);
645 
646 	/* Some platforms can't detect that the alarm triggered the
647 	 * wakeup, or (accordingly) disable it after it afterwards.
648 	 * It's supposed to give oneshot behavior; cope.
649 	 */
650 	alm.enabled = false;
651 	rtc_set_alarm(rtc, &alm);
652 }
653 
654 static int __init has_wakealarm(struct device *dev, void *name_ptr)
655 {
656 	struct rtc_device *candidate = to_rtc_device(dev);
657 
658 	if (!candidate->ops->set_alarm)
659 		return 0;
660 	if (!device_may_wakeup(candidate->dev.parent))
661 		return 0;
662 
663 	*(const char **)name_ptr = dev_name(dev);
664 	return 1;
665 }
666 
667 /*
668  * Kernel options like "test_suspend=mem" force suspend/resume sanity tests
669  * at startup time.  They're normally disabled, for faster boot and because
670  * we can't know which states really work on this particular system.
671  */
672 static suspend_state_t test_state __initdata = PM_SUSPEND_ON;
673 
674 static char warn_bad_state[] __initdata =
675 	KERN_WARNING "PM: can't test '%s' suspend state\n";
676 
677 static int __init setup_test_suspend(char *value)
678 {
679 	unsigned i;
680 
681 	/* "=mem" ==> "mem" */
682 	value++;
683 	for (i = 0; i < PM_SUSPEND_MAX; i++) {
684 		if (!pm_states[i])
685 			continue;
686 		if (strcmp(pm_states[i], value) != 0)
687 			continue;
688 		test_state = (__force suspend_state_t) i;
689 		return 0;
690 	}
691 	printk(warn_bad_state, value);
692 	return 0;
693 }
694 __setup("test_suspend", setup_test_suspend);
695 
696 static int __init test_suspend(void)
697 {
698 	static char		warn_no_rtc[] __initdata =
699 		KERN_WARNING "PM: no wakealarm-capable RTC driver is ready\n";
700 
701 	char			*pony = NULL;
702 	struct rtc_device	*rtc = NULL;
703 
704 	/* PM is initialized by now; is that state testable? */
705 	if (test_state == PM_SUSPEND_ON)
706 		goto done;
707 	if (!valid_state(test_state)) {
708 		printk(warn_bad_state, pm_states[test_state]);
709 		goto done;
710 	}
711 
712 	/* RTCs have initialized by now too ... can we use one? */
713 	class_find_device(rtc_class, NULL, &pony, has_wakealarm);
714 	if (pony)
715 		rtc = rtc_class_open(pony);
716 	if (!rtc) {
717 		printk(warn_no_rtc);
718 		goto done;
719 	}
720 
721 	/* go for it */
722 	test_wakealarm(rtc, test_state);
723 	rtc_class_close(rtc);
724 done:
725 	return 0;
726 }
727 late_initcall(test_suspend);
728 
729 #endif /* CONFIG_PM_TEST_SUSPEND */
730