PCI: document pci_target_state
[linux-2.6/x86.git] / kernel / power / main.c
blob0b7476f5d2a64fc848b88bfc51a06eb8d233eae1
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/init.h>
18 #include <linux/console.h>
19 #include <linux/cpu.h>
20 #include <linux/resume-trace.h>
21 #include <linux/freezer.h>
22 #include <linux/vmstat.h>
23 #include <linux/syscalls.h>
25 #include "power.h"
27 DEFINE_MUTEX(pm_mutex);
29 unsigned int pm_flags;
30 EXPORT_SYMBOL(pm_flags);
32 #ifdef CONFIG_PM_SLEEP
34 /* Routines for PM-transition notifications */
36 static BLOCKING_NOTIFIER_HEAD(pm_chain_head);
38 int register_pm_notifier(struct notifier_block *nb)
40 return blocking_notifier_chain_register(&pm_chain_head, nb);
42 EXPORT_SYMBOL_GPL(register_pm_notifier);
44 int unregister_pm_notifier(struct notifier_block *nb)
46 return blocking_notifier_chain_unregister(&pm_chain_head, nb);
48 EXPORT_SYMBOL_GPL(unregister_pm_notifier);
50 int pm_notifier_call_chain(unsigned long val)
52 return (blocking_notifier_call_chain(&pm_chain_head, val, NULL)
53 == NOTIFY_BAD) ? -EINVAL : 0;
56 #ifdef CONFIG_PM_DEBUG
57 int pm_test_level = TEST_NONE;
59 static int suspend_test(int level)
61 if (pm_test_level == level) {
62 printk(KERN_INFO "suspend debug: Waiting for 5 seconds.\n");
63 mdelay(5000);
64 return 1;
66 return 0;
69 static const char * const pm_tests[__TEST_AFTER_LAST] = {
70 [TEST_NONE] = "none",
71 [TEST_CORE] = "core",
72 [TEST_CPUS] = "processors",
73 [TEST_PLATFORM] = "platform",
74 [TEST_DEVICES] = "devices",
75 [TEST_FREEZER] = "freezer",
78 static ssize_t pm_test_show(struct kobject *kobj, struct kobj_attribute *attr,
79 char *buf)
81 char *s = buf;
82 int level;
84 for (level = TEST_FIRST; level <= TEST_MAX; level++)
85 if (pm_tests[level]) {
86 if (level == pm_test_level)
87 s += sprintf(s, "[%s] ", pm_tests[level]);
88 else
89 s += sprintf(s, "%s ", pm_tests[level]);
92 if (s != buf)
93 /* convert the last space to a newline */
94 *(s-1) = '\n';
96 return (s - buf);
99 static ssize_t pm_test_store(struct kobject *kobj, struct kobj_attribute *attr,
100 const char *buf, size_t n)
102 const char * const *s;
103 int level;
104 char *p;
105 int len;
106 int error = -EINVAL;
108 p = memchr(buf, '\n', n);
109 len = p ? p - buf : n;
111 mutex_lock(&pm_mutex);
113 level = TEST_FIRST;
114 for (s = &pm_tests[level]; level <= TEST_MAX; s++, level++)
115 if (*s && len == strlen(*s) && !strncmp(buf, *s, len)) {
116 pm_test_level = level;
117 error = 0;
118 break;
121 mutex_unlock(&pm_mutex);
123 return error ? error : n;
126 power_attr(pm_test);
127 #else /* !CONFIG_PM_DEBUG */
128 static inline int suspend_test(int level) { return 0; }
129 #endif /* !CONFIG_PM_DEBUG */
131 #endif /* CONFIG_PM_SLEEP */
133 #ifdef CONFIG_SUSPEND
135 #ifdef CONFIG_PM_TEST_SUSPEND
138 * We test the system suspend code by setting an RTC wakealarm a short
139 * time in the future, then suspending. Suspending the devices won't
140 * normally take long ... some systems only need a few milliseconds.
142 * The time it takes is system-specific though, so when we test this
143 * during system bootup we allow a LOT of time.
145 #define TEST_SUSPEND_SECONDS 5
147 static unsigned long suspend_test_start_time;
149 static void suspend_test_start(void)
151 /* FIXME Use better timebase than "jiffies", ideally a clocksource.
152 * What we want is a hardware counter that will work correctly even
153 * during the irqs-are-off stages of the suspend/resume cycle...
155 suspend_test_start_time = jiffies;
158 static void suspend_test_finish(const char *label)
160 long nj = jiffies - suspend_test_start_time;
161 unsigned msec;
163 msec = jiffies_to_msecs(abs(nj));
164 pr_info("PM: %s took %d.%03d seconds\n", label,
165 msec / 1000, msec % 1000);
167 /* Warning on suspend means the RTC alarm period needs to be
168 * larger -- the system was sooo slooowwww to suspend that the
169 * alarm (should have) fired before the system went to sleep!
171 * Warning on either suspend or resume also means the system
172 * has some performance issues. The stack dump of a WARN_ON
173 * is more likely to get the right attention than a printk...
175 WARN_ON(msec > (TEST_SUSPEND_SECONDS * 1000));
178 #else
180 static void suspend_test_start(void)
184 static void suspend_test_finish(const char *label)
188 #endif
190 /* This is just an arbitrary number */
191 #define FREE_PAGE_NUMBER (100)
193 static struct platform_suspend_ops *suspend_ops;
196 * suspend_set_ops - Set the global suspend method table.
197 * @ops: Pointer to ops structure.
200 void suspend_set_ops(struct platform_suspend_ops *ops)
202 mutex_lock(&pm_mutex);
203 suspend_ops = ops;
204 mutex_unlock(&pm_mutex);
208 * suspend_valid_only_mem - generic memory-only valid callback
210 * Platform drivers that implement mem suspend only and only need
211 * to check for that in their .valid callback can use this instead
212 * of rolling their own .valid callback.
214 int suspend_valid_only_mem(suspend_state_t state)
216 return state == PM_SUSPEND_MEM;
220 * suspend_prepare - Do prep work before entering low-power state.
222 * This is common code that is called for each state that we're entering.
223 * Run suspend notifiers, allocate a console and stop all processes.
225 static int suspend_prepare(void)
227 int error;
228 unsigned int free_pages;
230 if (!suspend_ops || !suspend_ops->enter)
231 return -EPERM;
233 pm_prepare_console();
235 error = pm_notifier_call_chain(PM_SUSPEND_PREPARE);
236 if (error)
237 goto Finish;
239 if (suspend_freeze_processes()) {
240 error = -EAGAIN;
241 goto Thaw;
244 free_pages = global_page_state(NR_FREE_PAGES);
245 if (free_pages < FREE_PAGE_NUMBER) {
246 pr_debug("PM: free some memory\n");
247 shrink_all_memory(FREE_PAGE_NUMBER - free_pages);
248 if (nr_free_pages() < FREE_PAGE_NUMBER) {
249 error = -ENOMEM;
250 printk(KERN_ERR "PM: No enough memory\n");
253 if (!error)
254 return 0;
256 Thaw:
257 suspend_thaw_processes();
258 Finish:
259 pm_notifier_call_chain(PM_POST_SUSPEND);
260 pm_restore_console();
261 return error;
264 /* default implementation */
265 void __attribute__ ((weak)) arch_suspend_disable_irqs(void)
267 local_irq_disable();
270 /* default implementation */
271 void __attribute__ ((weak)) arch_suspend_enable_irqs(void)
273 local_irq_enable();
277 * suspend_enter - enter the desired system sleep state.
278 * @state: state to enter
280 * This function should be called after devices have been suspended.
282 static int suspend_enter(suspend_state_t state)
284 int error = 0;
286 device_pm_lock();
287 arch_suspend_disable_irqs();
288 BUG_ON(!irqs_disabled());
290 if ((error = device_power_down(PMSG_SUSPEND))) {
291 printk(KERN_ERR "PM: Some devices failed to power down\n");
292 goto Done;
295 if (!suspend_test(TEST_CORE))
296 error = suspend_ops->enter(state);
298 device_power_up(PMSG_RESUME);
299 Done:
300 arch_suspend_enable_irqs();
301 BUG_ON(irqs_disabled());
302 device_pm_unlock();
303 return error;
307 * suspend_devices_and_enter - suspend devices and enter the desired system
308 * sleep state.
309 * @state: state to enter
311 int suspend_devices_and_enter(suspend_state_t state)
313 int error;
315 if (!suspend_ops)
316 return -ENOSYS;
318 if (suspend_ops->begin) {
319 error = suspend_ops->begin(state);
320 if (error)
321 goto Close;
323 suspend_console();
324 suspend_test_start();
325 error = device_suspend(PMSG_SUSPEND);
326 if (error) {
327 printk(KERN_ERR "PM: Some devices failed to suspend\n");
328 goto Recover_platform;
330 suspend_test_finish("suspend devices");
331 if (suspend_test(TEST_DEVICES))
332 goto Recover_platform;
334 if (suspend_ops->prepare) {
335 error = suspend_ops->prepare();
336 if (error)
337 goto Resume_devices;
340 if (suspend_test(TEST_PLATFORM))
341 goto Finish;
343 error = disable_nonboot_cpus();
344 if (!error && !suspend_test(TEST_CPUS))
345 suspend_enter(state);
347 enable_nonboot_cpus();
348 Finish:
349 if (suspend_ops->finish)
350 suspend_ops->finish();
351 Resume_devices:
352 suspend_test_start();
353 device_resume(PMSG_RESUME);
354 suspend_test_finish("resume devices");
355 resume_console();
356 Close:
357 if (suspend_ops->end)
358 suspend_ops->end();
359 return error;
361 Recover_platform:
362 if (suspend_ops->recover)
363 suspend_ops->recover();
364 goto Resume_devices;
368 * suspend_finish - Do final work before exiting suspend sequence.
370 * Call platform code to clean up, restart processes, and free the
371 * console that we've allocated. This is not called for suspend-to-disk.
373 static void suspend_finish(void)
375 suspend_thaw_processes();
376 pm_notifier_call_chain(PM_POST_SUSPEND);
377 pm_restore_console();
383 static const char * const pm_states[PM_SUSPEND_MAX] = {
384 [PM_SUSPEND_STANDBY] = "standby",
385 [PM_SUSPEND_MEM] = "mem",
388 static inline int valid_state(suspend_state_t state)
390 /* All states need lowlevel support and need to be valid
391 * to the lowlevel implementation, no valid callback
392 * implies that none are valid. */
393 if (!suspend_ops || !suspend_ops->valid || !suspend_ops->valid(state))
394 return 0;
395 return 1;
400 * enter_state - Do common work of entering low-power state.
401 * @state: pm_state structure for state we're entering.
403 * Make sure we're the only ones trying to enter a sleep state. Fail
404 * if someone has beat us to it, since we don't want anything weird to
405 * happen when we wake up.
406 * Then, do the setup for suspend, enter the state, and cleaup (after
407 * we've woken up).
409 static int enter_state(suspend_state_t state)
411 int error;
413 if (!valid_state(state))
414 return -ENODEV;
416 if (!mutex_trylock(&pm_mutex))
417 return -EBUSY;
419 printk(KERN_INFO "PM: Syncing filesystems ... ");
420 sys_sync();
421 printk("done.\n");
423 pr_debug("PM: Preparing system for %s sleep\n", pm_states[state]);
424 error = suspend_prepare();
425 if (error)
426 goto Unlock;
428 if (suspend_test(TEST_FREEZER))
429 goto Finish;
431 pr_debug("PM: Entering %s sleep\n", pm_states[state]);
432 error = suspend_devices_and_enter(state);
434 Finish:
435 pr_debug("PM: Finishing wakeup.\n");
436 suspend_finish();
437 Unlock:
438 mutex_unlock(&pm_mutex);
439 return error;
444 * pm_suspend - Externally visible function for suspending system.
445 * @state: Enumerated value of state to enter.
447 * Determine whether or not value is within range, get state
448 * structure, and enter (above).
451 int pm_suspend(suspend_state_t state)
453 if (state > PM_SUSPEND_ON && state <= PM_SUSPEND_MAX)
454 return enter_state(state);
455 return -EINVAL;
458 EXPORT_SYMBOL(pm_suspend);
460 #endif /* CONFIG_SUSPEND */
462 struct kobject *power_kobj;
465 * state - control system power state.
467 * show() returns what states are supported, which is hard-coded to
468 * 'standby' (Power-On Suspend), 'mem' (Suspend-to-RAM), and
469 * 'disk' (Suspend-to-Disk).
471 * store() accepts one of those strings, translates it into the
472 * proper enumerated value, and initiates a suspend transition.
475 static ssize_t state_show(struct kobject *kobj, struct kobj_attribute *attr,
476 char *buf)
478 char *s = buf;
479 #ifdef CONFIG_SUSPEND
480 int i;
482 for (i = 0; i < PM_SUSPEND_MAX; i++) {
483 if (pm_states[i] && valid_state(i))
484 s += sprintf(s,"%s ", pm_states[i]);
486 #endif
487 #ifdef CONFIG_HIBERNATION
488 s += sprintf(s, "%s\n", "disk");
489 #else
490 if (s != buf)
491 /* convert the last space to a newline */
492 *(s-1) = '\n';
493 #endif
494 return (s - buf);
497 static ssize_t state_store(struct kobject *kobj, struct kobj_attribute *attr,
498 const char *buf, size_t n)
500 #ifdef CONFIG_SUSPEND
501 suspend_state_t state = PM_SUSPEND_STANDBY;
502 const char * const *s;
503 #endif
504 char *p;
505 int len;
506 int error = -EINVAL;
508 p = memchr(buf, '\n', n);
509 len = p ? p - buf : n;
511 /* First, check if we are requested to hibernate */
512 if (len == 4 && !strncmp(buf, "disk", len)) {
513 error = hibernate();
514 goto Exit;
517 #ifdef CONFIG_SUSPEND
518 for (s = &pm_states[state]; state < PM_SUSPEND_MAX; s++, state++) {
519 if (*s && len == strlen(*s) && !strncmp(buf, *s, len))
520 break;
522 if (state < PM_SUSPEND_MAX && *s)
523 error = enter_state(state);
524 #endif
526 Exit:
527 return error ? error : n;
530 power_attr(state);
532 #ifdef CONFIG_PM_TRACE
533 int pm_trace_enabled;
535 static ssize_t pm_trace_show(struct kobject *kobj, struct kobj_attribute *attr,
536 char *buf)
538 return sprintf(buf, "%d\n", pm_trace_enabled);
541 static ssize_t
542 pm_trace_store(struct kobject *kobj, struct kobj_attribute *attr,
543 const char *buf, size_t n)
545 int val;
547 if (sscanf(buf, "%d", &val) == 1) {
548 pm_trace_enabled = !!val;
549 return n;
551 return -EINVAL;
554 power_attr(pm_trace);
555 #endif /* CONFIG_PM_TRACE */
557 static struct attribute * g[] = {
558 &state_attr.attr,
559 #ifdef CONFIG_PM_TRACE
560 &pm_trace_attr.attr,
561 #endif
562 #if defined(CONFIG_PM_SLEEP) && defined(CONFIG_PM_DEBUG)
563 &pm_test_attr.attr,
564 #endif
565 NULL,
568 static struct attribute_group attr_group = {
569 .attrs = g,
573 static int __init pm_init(void)
575 power_kobj = kobject_create_and_add("power", NULL);
576 if (!power_kobj)
577 return -ENOMEM;
578 return sysfs_create_group(power_kobj, &attr_group);
581 core_initcall(pm_init);
584 #ifdef CONFIG_PM_TEST_SUSPEND
586 #include <linux/rtc.h>
589 * To test system suspend, we need a hands-off mechanism to resume the
590 * system. RTCs wake alarms are a common self-contained mechanism.
593 static void __init test_wakealarm(struct rtc_device *rtc, suspend_state_t state)
595 static char err_readtime[] __initdata =
596 KERN_ERR "PM: can't read %s time, err %d\n";
597 static char err_wakealarm [] __initdata =
598 KERN_ERR "PM: can't set %s wakealarm, err %d\n";
599 static char err_suspend[] __initdata =
600 KERN_ERR "PM: suspend test failed, error %d\n";
601 static char info_test[] __initdata =
602 KERN_INFO "PM: test RTC wakeup from '%s' suspend\n";
604 unsigned long now;
605 struct rtc_wkalrm alm;
606 int status;
608 /* this may fail if the RTC hasn't been initialized */
609 status = rtc_read_time(rtc, &alm.time);
610 if (status < 0) {
611 printk(err_readtime, rtc->dev.bus_id, status);
612 return;
614 rtc_tm_to_time(&alm.time, &now);
616 memset(&alm, 0, sizeof alm);
617 rtc_time_to_tm(now + TEST_SUSPEND_SECONDS, &alm.time);
618 alm.enabled = true;
620 status = rtc_set_alarm(rtc, &alm);
621 if (status < 0) {
622 printk(err_wakealarm, rtc->dev.bus_id, status);
623 return;
626 if (state == PM_SUSPEND_MEM) {
627 printk(info_test, pm_states[state]);
628 status = pm_suspend(state);
629 if (status == -ENODEV)
630 state = PM_SUSPEND_STANDBY;
632 if (state == PM_SUSPEND_STANDBY) {
633 printk(info_test, pm_states[state]);
634 status = pm_suspend(state);
636 if (status < 0)
637 printk(err_suspend, status);
639 /* Some platforms can't detect that the alarm triggered the
640 * wakeup, or (accordingly) disable it after it afterwards.
641 * It's supposed to give oneshot behavior; cope.
643 alm.enabled = false;
644 rtc_set_alarm(rtc, &alm);
647 static int __init has_wakealarm(struct device *dev, void *name_ptr)
649 struct rtc_device *candidate = to_rtc_device(dev);
651 if (!candidate->ops->set_alarm)
652 return 0;
653 if (!device_may_wakeup(candidate->dev.parent))
654 return 0;
656 *(char **)name_ptr = dev->bus_id;
657 return 1;
661 * Kernel options like "test_suspend=mem" force suspend/resume sanity tests
662 * at startup time. They're normally disabled, for faster boot and because
663 * we can't know which states really work on this particular system.
665 static suspend_state_t test_state __initdata = PM_SUSPEND_ON;
667 static char warn_bad_state[] __initdata =
668 KERN_WARNING "PM: can't test '%s' suspend state\n";
670 static int __init setup_test_suspend(char *value)
672 unsigned i;
674 /* "=mem" ==> "mem" */
675 value++;
676 for (i = 0; i < PM_SUSPEND_MAX; i++) {
677 if (!pm_states[i])
678 continue;
679 if (strcmp(pm_states[i], value) != 0)
680 continue;
681 test_state = (__force suspend_state_t) i;
682 return 0;
684 printk(warn_bad_state, value);
685 return 0;
687 __setup("test_suspend", setup_test_suspend);
689 static int __init test_suspend(void)
691 static char warn_no_rtc[] __initdata =
692 KERN_WARNING "PM: no wakealarm-capable RTC driver is ready\n";
694 char *pony = NULL;
695 struct rtc_device *rtc = NULL;
697 /* PM is initialized by now; is that state testable? */
698 if (test_state == PM_SUSPEND_ON)
699 goto done;
700 if (!valid_state(test_state)) {
701 printk(warn_bad_state, pm_states[test_state]);
702 goto done;
705 /* RTCs have initialized by now too ... can we use one? */
706 class_find_device(rtc_class, NULL, &pony, has_wakealarm);
707 if (pony)
708 rtc = rtc_class_open(pony);
709 if (!rtc) {
710 printk(warn_no_rtc);
711 goto done;
714 /* go for it */
715 test_wakealarm(rtc, test_state);
716 rtc_class_close(rtc);
717 done:
718 return 0;
720 late_initcall(test_suspend);
722 #endif /* CONFIG_PM_TEST_SUSPEND */