PM: Do not hold dpm_list_mtx while disabling/enabling nonboot CPUs
[linux-2.6/verdex.git] / kernel / power / main.c
blob868028280d13474811810098cfa0f1923d212b32
1 /*
2 * kernel/power/main.c - PM subsystem core functionality.
4 * Copyright (c) 2003 Patrick Mochel
5 * Copyright (c) 2003 Open Source Development Lab
6 *
7 * This file is released under the GPLv2
9 */
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>
26 #include "power.h"
28 DEFINE_MUTEX(pm_mutex);
30 unsigned int pm_flags;
31 EXPORT_SYMBOL(pm_flags);
33 #ifdef CONFIG_PM_SLEEP
35 /* Routines for PM-transition notifications */
37 static BLOCKING_NOTIFIER_HEAD(pm_chain_head);
39 int register_pm_notifier(struct notifier_block *nb)
41 return blocking_notifier_chain_register(&pm_chain_head, nb);
43 EXPORT_SYMBOL_GPL(register_pm_notifier);
45 int unregister_pm_notifier(struct notifier_block *nb)
47 return blocking_notifier_chain_unregister(&pm_chain_head, nb);
49 EXPORT_SYMBOL_GPL(unregister_pm_notifier);
51 int pm_notifier_call_chain(unsigned long val)
53 return (blocking_notifier_call_chain(&pm_chain_head, val, NULL)
54 == NOTIFY_BAD) ? -EINVAL : 0;
57 #ifdef CONFIG_PM_DEBUG
58 int pm_test_level = TEST_NONE;
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",
69 static ssize_t pm_test_show(struct kobject *kobj, struct kobj_attribute *attr,
70 char *buf)
72 char *s = buf;
73 int level;
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]);
83 if (s != buf)
84 /* convert the last space to a newline */
85 *(s-1) = '\n';
87 return (s - buf);
90 static ssize_t pm_test_store(struct kobject *kobj, struct kobj_attribute *attr,
91 const char *buf, size_t n)
93 const char * const *s;
94 int level;
95 char *p;
96 int len;
97 int error = -EINVAL;
99 p = memchr(buf, '\n', n);
100 len = p ? p - buf : n;
102 mutex_lock(&pm_mutex);
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;
112 mutex_unlock(&pm_mutex);
114 return error ? error : n;
117 power_attr(pm_test);
118 #endif /* CONFIG_PM_DEBUG */
120 #endif /* CONFIG_PM_SLEEP */
122 #ifdef CONFIG_SUSPEND
124 static int suspend_test(int level)
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;
132 #endif /* !CONFIG_PM_DEBUG */
133 return 0;
136 #ifdef CONFIG_PM_TEST_SUSPEND
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.
143 * The time it takes is system-specific though, so when we test this
144 * during system bootup we allow a LOT of time.
146 #define TEST_SUSPEND_SECONDS 5
148 static unsigned long suspend_test_start_time;
150 static void suspend_test_start(void)
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...
156 suspend_test_start_time = jiffies;
159 static void suspend_test_finish(const char *label)
161 long nj = jiffies - suspend_test_start_time;
162 unsigned msec;
164 msec = jiffies_to_msecs(abs(nj));
165 pr_info("PM: %s took %d.%03d seconds\n", label,
166 msec / 1000, msec % 1000);
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!
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...
176 WARN(msec > (TEST_SUSPEND_SECONDS * 1000), "Component: %s\n", label);
179 #else
181 static void suspend_test_start(void)
185 static void suspend_test_finish(const char *label)
189 #endif
191 /* This is just an arbitrary number */
192 #define FREE_PAGE_NUMBER (100)
194 static struct platform_suspend_ops *suspend_ops;
197 * suspend_set_ops - Set the global suspend method table.
198 * @ops: Pointer to ops structure.
201 void suspend_set_ops(struct platform_suspend_ops *ops)
203 mutex_lock(&pm_mutex);
204 suspend_ops = ops;
205 mutex_unlock(&pm_mutex);
209 * suspend_valid_only_mem - generic memory-only valid callback
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.
215 int suspend_valid_only_mem(suspend_state_t state)
217 return state == PM_SUSPEND_MEM;
221 * suspend_prepare - Do prep work before entering low-power state.
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.
226 static int suspend_prepare(void)
228 int error;
229 unsigned int free_pages;
231 if (!suspend_ops || !suspend_ops->enter)
232 return -EPERM;
234 pm_prepare_console();
236 error = pm_notifier_call_chain(PM_SUSPEND_PREPARE);
237 if (error)
238 goto Finish;
240 error = usermodehelper_disable();
241 if (error)
242 goto Finish;
244 if (suspend_freeze_processes()) {
245 error = -EAGAIN;
246 goto Thaw;
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");
258 if (!error)
259 return 0;
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;
270 /* default implementation */
271 void __attribute__ ((weak)) arch_suspend_disable_irqs(void)
273 local_irq_disable();
276 /* default implementation */
277 void __attribute__ ((weak)) arch_suspend_enable_irqs(void)
279 local_irq_enable();
283 * suspend_enter - enter the desired system sleep state.
284 * @state: state to enter
286 * This function should be called after devices have been suspended.
288 static int suspend_enter(suspend_state_t state)
290 int error;
292 if (suspend_ops->prepare) {
293 error = suspend_ops->prepare();
294 if (error)
295 return error;
298 error = device_power_down(PMSG_SUSPEND);
299 if (error) {
300 printk(KERN_ERR "PM: Some devices failed to power down\n");
301 goto Platfrom_finish;
304 if (suspend_ops->prepare_late) {
305 error = suspend_ops->prepare_late();
306 if (error)
307 goto Power_up_devices;
310 if (suspend_test(TEST_PLATFORM))
311 goto Platform_wake;
313 error = disable_nonboot_cpus();
314 if (error || suspend_test(TEST_CPUS))
315 goto Enable_cpus;
317 arch_suspend_disable_irqs();
318 BUG_ON(!irqs_disabled());
320 error = sysdev_suspend(PMSG_SUSPEND);
321 if (!error) {
322 if (!suspend_test(TEST_CORE))
323 error = suspend_ops->enter(state);
324 sysdev_resume();
327 arch_suspend_enable_irqs();
328 BUG_ON(irqs_disabled());
330 Enable_cpus:
331 enable_nonboot_cpus();
333 Platform_wake:
334 if (suspend_ops->wake)
335 suspend_ops->wake();
337 Power_up_devices:
338 device_power_up(PMSG_RESUME);
340 Platfrom_finish:
341 if (suspend_ops->finish)
342 suspend_ops->finish();
344 return error;
348 * suspend_devices_and_enter - suspend devices and enter the desired system
349 * sleep state.
350 * @state: state to enter
352 int suspend_devices_and_enter(suspend_state_t state)
354 int error;
356 if (!suspend_ops)
357 return -ENOSYS;
359 if (suspend_ops->begin) {
360 error = suspend_ops->begin(state);
361 if (error)
362 goto Close;
364 suspend_console();
365 suspend_test_start();
366 error = device_suspend(PMSG_SUSPEND);
367 if (error) {
368 printk(KERN_ERR "PM: Some devices failed to suspend\n");
369 goto Recover_platform;
371 suspend_test_finish("suspend devices");
372 if (suspend_test(TEST_DEVICES))
373 goto Recover_platform;
375 suspend_enter(state);
377 Resume_devices:
378 suspend_test_start();
379 device_resume(PMSG_RESUME);
380 suspend_test_finish("resume devices");
381 resume_console();
382 Close:
383 if (suspend_ops->end)
384 suspend_ops->end();
385 return error;
387 Recover_platform:
388 if (suspend_ops->recover)
389 suspend_ops->recover();
390 goto Resume_devices;
394 * suspend_finish - Do final work before exiting suspend sequence.
396 * Call platform code to clean up, restart processes, and free the
397 * console that we've allocated. This is not called for suspend-to-disk.
399 static void suspend_finish(void)
401 suspend_thaw_processes();
402 usermodehelper_enable();
403 pm_notifier_call_chain(PM_POST_SUSPEND);
404 pm_restore_console();
410 static const char * const pm_states[PM_SUSPEND_MAX] = {
411 [PM_SUSPEND_STANDBY] = "standby",
412 [PM_SUSPEND_MEM] = "mem",
415 static inline int valid_state(suspend_state_t state)
417 /* All states need lowlevel support and need to be valid
418 * to the lowlevel implementation, no valid callback
419 * implies that none are valid. */
420 if (!suspend_ops || !suspend_ops->valid || !suspend_ops->valid(state))
421 return 0;
422 return 1;
427 * enter_state - Do common work of entering low-power state.
428 * @state: pm_state structure for state we're entering.
430 * Make sure we're the only ones trying to enter a sleep state. Fail
431 * if someone has beat us to it, since we don't want anything weird to
432 * happen when we wake up.
433 * Then, do the setup for suspend, enter the state, and cleaup (after
434 * we've woken up).
436 static int enter_state(suspend_state_t state)
438 int error;
440 if (!valid_state(state))
441 return -ENODEV;
443 if (!mutex_trylock(&pm_mutex))
444 return -EBUSY;
446 printk(KERN_INFO "PM: Syncing filesystems ... ");
447 sys_sync();
448 printk("done.\n");
450 pr_debug("PM: Preparing system for %s sleep\n", pm_states[state]);
451 error = suspend_prepare();
452 if (error)
453 goto Unlock;
455 if (suspend_test(TEST_FREEZER))
456 goto Finish;
458 pr_debug("PM: Entering %s sleep\n", pm_states[state]);
459 error = suspend_devices_and_enter(state);
461 Finish:
462 pr_debug("PM: Finishing wakeup.\n");
463 suspend_finish();
464 Unlock:
465 mutex_unlock(&pm_mutex);
466 return error;
471 * pm_suspend - Externally visible function for suspending system.
472 * @state: Enumerated value of state to enter.
474 * Determine whether or not value is within range, get state
475 * structure, and enter (above).
478 int pm_suspend(suspend_state_t state)
480 if (state > PM_SUSPEND_ON && state <= PM_SUSPEND_MAX)
481 return enter_state(state);
482 return -EINVAL;
485 EXPORT_SYMBOL(pm_suspend);
487 #endif /* CONFIG_SUSPEND */
489 struct kobject *power_kobj;
492 * state - control system power state.
494 * show() returns what states are supported, which is hard-coded to
495 * 'standby' (Power-On Suspend), 'mem' (Suspend-to-RAM), and
496 * 'disk' (Suspend-to-Disk).
498 * store() accepts one of those strings, translates it into the
499 * proper enumerated value, and initiates a suspend transition.
502 static ssize_t state_show(struct kobject *kobj, struct kobj_attribute *attr,
503 char *buf)
505 char *s = buf;
506 #ifdef CONFIG_SUSPEND
507 int i;
509 for (i = 0; i < PM_SUSPEND_MAX; i++) {
510 if (pm_states[i] && valid_state(i))
511 s += sprintf(s,"%s ", pm_states[i]);
513 #endif
514 #ifdef CONFIG_HIBERNATION
515 s += sprintf(s, "%s\n", "disk");
516 #else
517 if (s != buf)
518 /* convert the last space to a newline */
519 *(s-1) = '\n';
520 #endif
521 return (s - buf);
524 static ssize_t state_store(struct kobject *kobj, struct kobj_attribute *attr,
525 const char *buf, size_t n)
527 #ifdef CONFIG_SUSPEND
528 suspend_state_t state = PM_SUSPEND_STANDBY;
529 const char * const *s;
530 #endif
531 char *p;
532 int len;
533 int error = -EINVAL;
535 p = memchr(buf, '\n', n);
536 len = p ? p - buf : n;
538 /* First, check if we are requested to hibernate */
539 if (len == 4 && !strncmp(buf, "disk", len)) {
540 error = hibernate();
541 goto Exit;
544 #ifdef CONFIG_SUSPEND
545 for (s = &pm_states[state]; state < PM_SUSPEND_MAX; s++, state++) {
546 if (*s && len == strlen(*s) && !strncmp(buf, *s, len))
547 break;
549 if (state < PM_SUSPEND_MAX && *s)
550 error = enter_state(state);
551 #endif
553 Exit:
554 return error ? error : n;
557 power_attr(state);
559 #ifdef CONFIG_PM_TRACE
560 int pm_trace_enabled;
562 static ssize_t pm_trace_show(struct kobject *kobj, struct kobj_attribute *attr,
563 char *buf)
565 return sprintf(buf, "%d\n", pm_trace_enabled);
568 static ssize_t
569 pm_trace_store(struct kobject *kobj, struct kobj_attribute *attr,
570 const char *buf, size_t n)
572 int val;
574 if (sscanf(buf, "%d", &val) == 1) {
575 pm_trace_enabled = !!val;
576 return n;
578 return -EINVAL;
581 power_attr(pm_trace);
582 #endif /* CONFIG_PM_TRACE */
584 static struct attribute * g[] = {
585 &state_attr.attr,
586 #ifdef CONFIG_PM_TRACE
587 &pm_trace_attr.attr,
588 #endif
589 #if defined(CONFIG_PM_SLEEP) && defined(CONFIG_PM_DEBUG)
590 &pm_test_attr.attr,
591 #endif
592 NULL,
595 static struct attribute_group attr_group = {
596 .attrs = g,
600 static int __init pm_init(void)
602 power_kobj = kobject_create_and_add("power", NULL);
603 if (!power_kobj)
604 return -ENOMEM;
605 return sysfs_create_group(power_kobj, &attr_group);
608 core_initcall(pm_init);
611 #ifdef CONFIG_PM_TEST_SUSPEND
613 #include <linux/rtc.h>
616 * To test system suspend, we need a hands-off mechanism to resume the
617 * system. RTCs wake alarms are a common self-contained mechanism.
620 static void __init test_wakealarm(struct rtc_device *rtc, suspend_state_t state)
622 static char err_readtime[] __initdata =
623 KERN_ERR "PM: can't read %s time, err %d\n";
624 static char err_wakealarm [] __initdata =
625 KERN_ERR "PM: can't set %s wakealarm, err %d\n";
626 static char err_suspend[] __initdata =
627 KERN_ERR "PM: suspend test failed, error %d\n";
628 static char info_test[] __initdata =
629 KERN_INFO "PM: test RTC wakeup from '%s' suspend\n";
631 unsigned long now;
632 struct rtc_wkalrm alm;
633 int status;
635 /* this may fail if the RTC hasn't been initialized */
636 status = rtc_read_time(rtc, &alm.time);
637 if (status < 0) {
638 printk(err_readtime, dev_name(&rtc->dev), status);
639 return;
641 rtc_tm_to_time(&alm.time, &now);
643 memset(&alm, 0, sizeof alm);
644 rtc_time_to_tm(now + TEST_SUSPEND_SECONDS, &alm.time);
645 alm.enabled = true;
647 status = rtc_set_alarm(rtc, &alm);
648 if (status < 0) {
649 printk(err_wakealarm, dev_name(&rtc->dev), status);
650 return;
653 if (state == PM_SUSPEND_MEM) {
654 printk(info_test, pm_states[state]);
655 status = pm_suspend(state);
656 if (status == -ENODEV)
657 state = PM_SUSPEND_STANDBY;
659 if (state == PM_SUSPEND_STANDBY) {
660 printk(info_test, pm_states[state]);
661 status = pm_suspend(state);
663 if (status < 0)
664 printk(err_suspend, status);
666 /* Some platforms can't detect that the alarm triggered the
667 * wakeup, or (accordingly) disable it after it afterwards.
668 * It's supposed to give oneshot behavior; cope.
670 alm.enabled = false;
671 rtc_set_alarm(rtc, &alm);
674 static int __init has_wakealarm(struct device *dev, void *name_ptr)
676 struct rtc_device *candidate = to_rtc_device(dev);
678 if (!candidate->ops->set_alarm)
679 return 0;
680 if (!device_may_wakeup(candidate->dev.parent))
681 return 0;
683 *(const char **)name_ptr = dev_name(dev);
684 return 1;
688 * Kernel options like "test_suspend=mem" force suspend/resume sanity tests
689 * at startup time. They're normally disabled, for faster boot and because
690 * we can't know which states really work on this particular system.
692 static suspend_state_t test_state __initdata = PM_SUSPEND_ON;
694 static char warn_bad_state[] __initdata =
695 KERN_WARNING "PM: can't test '%s' suspend state\n";
697 static int __init setup_test_suspend(char *value)
699 unsigned i;
701 /* "=mem" ==> "mem" */
702 value++;
703 for (i = 0; i < PM_SUSPEND_MAX; i++) {
704 if (!pm_states[i])
705 continue;
706 if (strcmp(pm_states[i], value) != 0)
707 continue;
708 test_state = (__force suspend_state_t) i;
709 return 0;
711 printk(warn_bad_state, value);
712 return 0;
714 __setup("test_suspend", setup_test_suspend);
716 static int __init test_suspend(void)
718 static char warn_no_rtc[] __initdata =
719 KERN_WARNING "PM: no wakealarm-capable RTC driver is ready\n";
721 char *pony = NULL;
722 struct rtc_device *rtc = NULL;
724 /* PM is initialized by now; is that state testable? */
725 if (test_state == PM_SUSPEND_ON)
726 goto done;
727 if (!valid_state(test_state)) {
728 printk(warn_bad_state, pm_states[test_state]);
729 goto done;
732 /* RTCs have initialized by now too ... can we use one? */
733 class_find_device(rtc_class, NULL, &pony, has_wakealarm);
734 if (pony)
735 rtc = rtc_class_open(pony);
736 if (!rtc) {
737 printk(warn_no_rtc);
738 goto done;
741 /* go for it */
742 test_wakealarm(rtc, test_state);
743 rtc_class_close(rtc);
744 done:
745 return 0;
747 late_initcall(test_suspend);
749 #endif /* CONFIG_PM_TEST_SUSPEND */