PM / Sleep: Add user space interface for manipulating wakeup sources, v3
[linux-2.6.git] / kernel / power / main.c
index b4d2190..54ec071 100644 (file)
@@ -3,33 +3,23 @@
  *
  * Copyright (c) 2003 Patrick Mochel
  * Copyright (c) 2003 Open Source Development Lab
- * 
+ *
  * This file is released under the GPLv2
  *
  */
 
-#include <linux/module.h>
-#include <linux/suspend.h>
+#include <linux/export.h>
 #include <linux/kobject.h>
 #include <linux/string.h>
-#include <linux/delay.h>
-#include <linux/errno.h>
-#include <linux/kmod.h>
-#include <linux/init.h>
-#include <linux/console.h>
-#include <linux/cpu.h>
 #include <linux/resume-trace.h>
-#include <linux/freezer.h>
-#include <linux/vmstat.h>
-#include <linux/syscalls.h>
+#include <linux/workqueue.h>
+#include <linux/debugfs.h>
+#include <linux/seq_file.h>
 
 #include "power.h"
 
 DEFINE_MUTEX(pm_mutex);
 
-unsigned int pm_flags;
-EXPORT_SYMBOL(pm_flags);
-
 #ifdef CONFIG_PM_SLEEP
 
 /* Routines for PM-transition notifications */
@@ -50,10 +40,37 @@ EXPORT_SYMBOL_GPL(unregister_pm_notifier);
 
 int pm_notifier_call_chain(unsigned long val)
 {
-       return (blocking_notifier_call_chain(&pm_chain_head, val, NULL)
-                       == NOTIFY_BAD) ? -EINVAL : 0;
+       int ret = blocking_notifier_call_chain(&pm_chain_head, val, NULL);
+
+       return notifier_to_errno(ret);
+}
+
+/* If set, devices may be suspended and resumed asynchronously. */
+int pm_async_enabled = 1;
+
+static ssize_t pm_async_show(struct kobject *kobj, struct kobj_attribute *attr,
+                            char *buf)
+{
+       return sprintf(buf, "%d\n", pm_async_enabled);
+}
+
+static ssize_t pm_async_store(struct kobject *kobj, struct kobj_attribute *attr,
+                             const char *buf, size_t n)
+{
+       unsigned long val;
+
+       if (strict_strtoul(buf, 10, &val))
+               return -EINVAL;
+
+       if (val > 1)
+               return -EINVAL;
+
+       pm_async_enabled = val;
+       return n;
 }
 
+power_attr(pm_async);
+
 #ifdef CONFIG_PM_DEBUG
 int pm_test_level = TEST_NONE;
 
@@ -99,7 +116,7 @@ static ssize_t pm_test_store(struct kobject *kobj, struct kobj_attribute *attr,
        p = memchr(buf, '\n', n);
        len = p ? p - buf : n;
 
-       mutex_lock(&pm_mutex);
+       lock_system_sleep();
 
        level = TEST_FIRST;
        for (s = &pm_tests[level]; level <= TEST_MAX; s++, level++)
@@ -109,7 +126,7 @@ static ssize_t pm_test_store(struct kobject *kobj, struct kobj_attribute *attr,
                        break;
                }
 
-       mutex_unlock(&pm_mutex);
+       unlock_system_sleep();
 
        return error ? error : n;
 }
@@ -117,424 +134,341 @@ static ssize_t pm_test_store(struct kobject *kobj, struct kobj_attribute *attr,
 power_attr(pm_test);
 #endif /* CONFIG_PM_DEBUG */
 
-#endif /* CONFIG_PM_SLEEP */
-
-#ifdef CONFIG_SUSPEND
-
-static int suspend_test(int level)
+#ifdef CONFIG_DEBUG_FS
+static char *suspend_step_name(enum suspend_stat_step step)
 {
-#ifdef CONFIG_PM_DEBUG
-       if (pm_test_level == level) {
-               printk(KERN_INFO "suspend debug: Waiting for 5 seconds.\n");
-               mdelay(5000);
-               return 1;
+       switch (step) {
+       case SUSPEND_FREEZE:
+               return "freeze";
+       case SUSPEND_PREPARE:
+               return "prepare";
+       case SUSPEND_SUSPEND:
+               return "suspend";
+       case SUSPEND_SUSPEND_NOIRQ:
+               return "suspend_noirq";
+       case SUSPEND_RESUME_NOIRQ:
+               return "resume_noirq";
+       case SUSPEND_RESUME:
+               return "resume";
+       default:
+               return "";
        }
-#endif /* !CONFIG_PM_DEBUG */
-       return 0;
 }
 
-#ifdef CONFIG_PM_TEST_SUSPEND
-
-/*
- * We test the system suspend code by setting an RTC wakealarm a short
- * time in the future, then suspending.  Suspending the devices won't
- * normally take long ... some systems only need a few milliseconds.
- *
- * The time it takes is system-specific though, so when we test this
- * during system bootup we allow a LOT of time.
- */
-#define TEST_SUSPEND_SECONDS   5
-
-static unsigned long suspend_test_start_time;
-
-static void suspend_test_start(void)
+static int suspend_stats_show(struct seq_file *s, void *unused)
 {
-       /* FIXME Use better timebase than "jiffies", ideally a clocksource.
-        * What we want is a hardware counter that will work correctly even
-        * during the irqs-are-off stages of the suspend/resume cycle...
-        */
-       suspend_test_start_time = jiffies;
-}
+       int i, index, last_dev, last_errno, last_step;
+
+       last_dev = suspend_stats.last_failed_dev + REC_FAILED_NUM - 1;
+       last_dev %= REC_FAILED_NUM;
+       last_errno = suspend_stats.last_failed_errno + REC_FAILED_NUM - 1;
+       last_errno %= REC_FAILED_NUM;
+       last_step = suspend_stats.last_failed_step + REC_FAILED_NUM - 1;
+       last_step %= REC_FAILED_NUM;
+       seq_printf(s, "%s: %d\n%s: %d\n%s: %d\n%s: %d\n%s: %d\n"
+                       "%s: %d\n%s: %d\n%s: %d\n%s: %d\n%s: %d\n",
+                       "success", suspend_stats.success,
+                       "fail", suspend_stats.fail,
+                       "failed_freeze", suspend_stats.failed_freeze,
+                       "failed_prepare", suspend_stats.failed_prepare,
+                       "failed_suspend", suspend_stats.failed_suspend,
+                       "failed_suspend_late",
+                               suspend_stats.failed_suspend_late,
+                       "failed_suspend_noirq",
+                               suspend_stats.failed_suspend_noirq,
+                       "failed_resume", suspend_stats.failed_resume,
+                       "failed_resume_early",
+                               suspend_stats.failed_resume_early,
+                       "failed_resume_noirq",
+                               suspend_stats.failed_resume_noirq);
+       seq_printf(s,   "failures:\n  last_failed_dev:\t%-s\n",
+                       suspend_stats.failed_devs[last_dev]);
+       for (i = 1; i < REC_FAILED_NUM; i++) {
+               index = last_dev + REC_FAILED_NUM - i;
+               index %= REC_FAILED_NUM;
+               seq_printf(s, "\t\t\t%-s\n",
+                       suspend_stats.failed_devs[index]);
+       }
+       seq_printf(s,   "  last_failed_errno:\t%-d\n",
+                       suspend_stats.errno[last_errno]);
+       for (i = 1; i < REC_FAILED_NUM; i++) {
+               index = last_errno + REC_FAILED_NUM - i;
+               index %= REC_FAILED_NUM;
+               seq_printf(s, "\t\t\t%-d\n",
+                       suspend_stats.errno[index]);
+       }
+       seq_printf(s,   "  last_failed_step:\t%-s\n",
+                       suspend_step_name(
+                               suspend_stats.failed_steps[last_step]));
+       for (i = 1; i < REC_FAILED_NUM; i++) {
+               index = last_step + REC_FAILED_NUM - i;
+               index %= REC_FAILED_NUM;
+               seq_printf(s, "\t\t\t%-s\n",
+                       suspend_step_name(
+                               suspend_stats.failed_steps[index]));
+       }
 
-static void suspend_test_finish(const char *label)
-{
-       long nj = jiffies - suspend_test_start_time;
-       unsigned msec;
-
-       msec = jiffies_to_msecs(abs(nj));
-       pr_info("PM: %s took %d.%03d seconds\n", label,
-                       msec / 1000, msec % 1000);
-
-       /* Warning on suspend means the RTC alarm period needs to be
-        * larger -- the system was sooo slooowwww to suspend that the
-        * alarm (should have) fired before the system went to sleep!
-        *
-        * Warning on either suspend or resume also means the system
-        * has some performance issues.  The stack dump of a WARN_ON
-        * is more likely to get the right attention than a printk...
-        */
-       WARN(msec > (TEST_SUSPEND_SECONDS * 1000), "Component: %s\n", label);
+       return 0;
 }
 
-#else
-
-static void suspend_test_start(void)
+static int suspend_stats_open(struct inode *inode, struct file *file)
 {
+       return single_open(file, suspend_stats_show, NULL);
 }
 
-static void suspend_test_finish(const char *label)
+static const struct file_operations suspend_stats_operations = {
+       .open           = suspend_stats_open,
+       .read           = seq_read,
+       .llseek         = seq_lseek,
+       .release        = single_release,
+};
+
+static int __init pm_debugfs_init(void)
 {
+       debugfs_create_file("suspend_stats", S_IFREG | S_IRUGO,
+                       NULL, NULL, &suspend_stats_operations);
+       return 0;
 }
 
-#endif
-
-/* This is just an arbitrary number */
-#define FREE_PAGE_NUMBER (100)
+late_initcall(pm_debugfs_init);
+#endif /* CONFIG_DEBUG_FS */
 
-static struct platform_suspend_ops *suspend_ops;
-
-/**
- *     suspend_set_ops - Set the global suspend method table.
- *     @ops:   Pointer to ops structure.
- */
+#endif /* CONFIG_PM_SLEEP */
 
-void suspend_set_ops(struct platform_suspend_ops *ops)
-{
-       mutex_lock(&pm_mutex);
-       suspend_ops = ops;
-       mutex_unlock(&pm_mutex);
-}
+struct kobject *power_kobj;
 
 /**
- * suspend_valid_only_mem - generic memory-only valid callback
+ *     state - control system power state.
  *
- * Platform drivers that implement mem suspend only and only need
- * to check for that in their .valid callback can use this instead
- * of rolling their own .valid callback.
- */
-int suspend_valid_only_mem(suspend_state_t state)
-{
-       return state == PM_SUSPEND_MEM;
-}
-
-/**
- *     suspend_prepare - Do prep work before entering low-power state.
+ *     show() returns what states are supported, which is hard-coded to
+ *     'standby' (Power-On Suspend), 'mem' (Suspend-to-RAM), and
+ *     'disk' (Suspend-to-Disk).
  *
- *     This is common code that is called for each state that we're entering.
- *     Run suspend notifiers, allocate a console and stop all processes.
+ *     store() accepts one of those strings, translates it into the
+ *     proper enumerated value, and initiates a suspend transition.
  */
-static int suspend_prepare(void)
+static ssize_t state_show(struct kobject *kobj, struct kobj_attribute *attr,
+                         char *buf)
 {
-       int error;
-       unsigned int free_pages;
-
-       if (!suspend_ops || !suspend_ops->enter)
-               return -EPERM;
-
-       pm_prepare_console();
-
-       error = pm_notifier_call_chain(PM_SUSPEND_PREPARE);
-       if (error)
-               goto Finish;
-
-       error = usermodehelper_disable();
-       if (error)
-               goto Finish;
-
-       if (suspend_freeze_processes()) {
-               error = -EAGAIN;
-               goto Thaw;
-       }
+       char *s = buf;
+#ifdef CONFIG_SUSPEND
+       int i;
 
-       free_pages = global_page_state(NR_FREE_PAGES);
-       if (free_pages < FREE_PAGE_NUMBER) {
-               pr_debug("PM: free some memory\n");
-               shrink_all_memory(FREE_PAGE_NUMBER - free_pages);
-               if (nr_free_pages() < FREE_PAGE_NUMBER) {
-                       error = -ENOMEM;
-                       printk(KERN_ERR "PM: No enough memory\n");
-               }
+       for (i = 0; i < PM_SUSPEND_MAX; i++) {
+               if (pm_states[i] && valid_state(i))
+                       s += sprintf(s,"%s ", pm_states[i]);
        }
-       if (!error)
-               return 0;
-
- Thaw:
-       suspend_thaw_processes();
-       usermodehelper_enable();
- Finish:
-       pm_notifier_call_chain(PM_POST_SUSPEND);
-       pm_restore_console();
-       return error;
-}
-
-/* default implementation */
-void __attribute__ ((weak)) arch_suspend_disable_irqs(void)
-{
-       local_irq_disable();
-}
-
-/* default implementation */
-void __attribute__ ((weak)) arch_suspend_enable_irqs(void)
-{
-       local_irq_enable();
+#endif
+#ifdef CONFIG_HIBERNATION
+       s += sprintf(s, "%s\n", "disk");
+#else
+       if (s != buf)
+               /* convert the last space to a newline */
+               *(s-1) = '\n';
+#endif
+       return (s - buf);
 }
 
-/**
- *     suspend_enter - enter the desired system sleep state.
- *     @state:         state to enter
- *
- *     This function should be called after devices have been suspended.
- */
-static int suspend_enter(suspend_state_t state)
+static suspend_state_t decode_state(const char *buf, size_t n)
 {
-       int error = 0;
+#ifdef CONFIG_SUSPEND
+       suspend_state_t state = PM_SUSPEND_STANDBY;
+       const char * const *s;
+#endif
+       char *p;
+       int len;
 
-       device_pm_lock();
-       arch_suspend_disable_irqs();
-       BUG_ON(!irqs_disabled());
+       p = memchr(buf, '\n', n);
+       len = p ? p - buf : n;
 
-       if ((error = device_power_down(PMSG_SUSPEND))) {
-               printk(KERN_ERR "PM: Some devices failed to power down\n");
-               goto Done;
-       }
+       /* Check hibernation first. */
+       if (len == 4 && !strncmp(buf, "disk", len))
+               return PM_SUSPEND_MAX;
 
-       if (!suspend_test(TEST_CORE))
-               error = suspend_ops->enter(state);
+#ifdef CONFIG_SUSPEND
+       for (s = &pm_states[state]; state < PM_SUSPEND_MAX; s++, state++)
+               if (*s && len == strlen(*s) && !strncmp(buf, *s, len))
+                       return state;
+#endif
 
-       device_power_up(PMSG_RESUME);
- Done:
-       arch_suspend_enable_irqs();
-       BUG_ON(irqs_disabled());
-       device_pm_unlock();
-       return error;
+       return PM_SUSPEND_ON;
 }
 
-/**
- *     suspend_devices_and_enter - suspend devices and enter the desired system
- *                                 sleep state.
- *     @state:           state to enter
- */
-int suspend_devices_and_enter(suspend_state_t state)
+static ssize_t state_store(struct kobject *kobj, struct kobj_attribute *attr,
+                          const char *buf, size_t n)
 {
+       suspend_state_t state;
        int error;
 
-       if (!suspend_ops)
-               return -ENOSYS;
+       error = pm_autosleep_lock();
+       if (error)
+               return error;
 
-       if (suspend_ops->begin) {
-               error = suspend_ops->begin(state);
-               if (error)
-                       goto Close;
-       }
-       suspend_console();
-       suspend_test_start();
-       error = device_suspend(PMSG_SUSPEND);
-       if (error) {
-               printk(KERN_ERR "PM: Some devices failed to suspend\n");
-               goto Recover_platform;
-       }
-       suspend_test_finish("suspend devices");
-       if (suspend_test(TEST_DEVICES))
-               goto Recover_platform;
-
-       if (suspend_ops->prepare) {
-               error = suspend_ops->prepare();
-               if (error)
-                       goto Resume_devices;
+       if (pm_autosleep_state() > PM_SUSPEND_ON) {
+               error = -EBUSY;
+               goto out;
        }
 
-       if (suspend_test(TEST_PLATFORM))
-               goto Finish;
-
-       error = disable_nonboot_cpus();
-       if (!error && !suspend_test(TEST_CPUS))
-               suspend_enter(state);
-
-       enable_nonboot_cpus();
- Finish:
-       if (suspend_ops->finish)
-               suspend_ops->finish();
- Resume_devices:
-       suspend_test_start();
-       device_resume(PMSG_RESUME);
-       suspend_test_finish("resume devices");
-       resume_console();
- Close:
-       if (suspend_ops->end)
-               suspend_ops->end();
-       return error;
+       state = decode_state(buf, n);
+       if (state < PM_SUSPEND_MAX)
+               error = pm_suspend(state);
+       else if (state == PM_SUSPEND_MAX)
+               error = hibernate();
+       else
+               error = -EINVAL;
 
- Recover_platform:
-       if (suspend_ops->recover)
-               suspend_ops->recover();
-       goto Resume_devices;
+ out:
+       pm_autosleep_unlock();
+       return error ? error : n;
 }
 
-/**
- *     suspend_finish - Do final work before exiting suspend sequence.
+power_attr(state);
+
+#ifdef CONFIG_PM_SLEEP
+/*
+ * The 'wakeup_count' attribute, along with the functions defined in
+ * drivers/base/power/wakeup.c, provides a means by which wakeup events can be
+ * handled in a non-racy way.
  *
- *     Call platform code to clean up, restart processes, and free the 
- *     console that we've allocated. This is not called for suspend-to-disk.
+ * If a wakeup event occurs when the system is in a sleep state, it simply is
+ * woken up.  In turn, if an event that would wake the system up from a sleep
+ * state occurs when it is undergoing a transition to that sleep state, the
+ * transition should be aborted.  Moreover, if such an event occurs when the
+ * system is in the working state, an attempt to start a transition to the
+ * given sleep state should fail during certain period after the detection of
+ * the event.  Using the 'state' attribute alone is not sufficient to satisfy
+ * these requirements, because a wakeup event may occur exactly when 'state'
+ * is being written to and may be delivered to user space right before it is
+ * frozen, so the event will remain only partially processed until the system is
+ * woken up by another event.  In particular, it won't cause the transition to
+ * a sleep state to be aborted.
+ *
+ * This difficulty may be overcome if user space uses 'wakeup_count' before
+ * writing to 'state'.  It first should read from 'wakeup_count' and store
+ * the read value.  Then, after carrying out its own preparations for the system
+ * transition to a sleep state, it should write the stored value to
+ * 'wakeup_count'.  If that fails, at least one wakeup event has occurred since
+ * 'wakeup_count' was read and 'state' should not be written to.  Otherwise, it
+ * is allowed to write to 'state', but the transition will be aborted if there
+ * are any wakeup events detected after 'wakeup_count' was written to.
  */
-static void suspend_finish(void)
-{
-       suspend_thaw_processes();
-       usermodehelper_enable();
-       pm_notifier_call_chain(PM_POST_SUSPEND);
-       pm_restore_console();
-}
-
-
 
-
-static const char * const pm_states[PM_SUSPEND_MAX] = {
-       [PM_SUSPEND_STANDBY]    = "standby",
-       [PM_SUSPEND_MEM]        = "mem",
-};
-
-static inline int valid_state(suspend_state_t state)
+static ssize_t wakeup_count_show(struct kobject *kobj,
+                               struct kobj_attribute *attr,
+                               char *buf)
 {
-       /* All states need lowlevel support and need to be valid
-        * to the lowlevel implementation, no valid callback
-        * implies that none are valid. */
-       if (!suspend_ops || !suspend_ops->valid || !suspend_ops->valid(state))
-               return 0;
-       return 1;
-}
+       unsigned int val;
 
+       return pm_get_wakeup_count(&val, true) ?
+               sprintf(buf, "%u\n", val) : -EINTR;
+}
 
-/**
- *     enter_state - Do common work of entering low-power state.
- *     @state:         pm_state structure for state we're entering.
- *
- *     Make sure we're the only ones trying to enter a sleep state. Fail
- *     if someone has beat us to it, since we don't want anything weird to
- *     happen when we wake up.
- *     Then, do the setup for suspend, enter the state, and cleaup (after
- *     we've woken up).
- */
-static int enter_state(suspend_state_t state)
+static ssize_t wakeup_count_store(struct kobject *kobj,
+                               struct kobj_attribute *attr,
+                               const char *buf, size_t n)
 {
+       unsigned int val;
        int error;
 
-       if (!valid_state(state))
-               return -ENODEV;
-
-       if (!mutex_trylock(&pm_mutex))
-               return -EBUSY;
-
-       printk(KERN_INFO "PM: Syncing filesystems ... ");
-       sys_sync();
-       printk("done.\n");
-
-       pr_debug("PM: Preparing system for %s sleep\n", pm_states[state]);
-       error = suspend_prepare();
+       error = pm_autosleep_lock();
        if (error)
-               goto Unlock;
+               return error;
 
-       if (suspend_test(TEST_FREEZER))
-               goto Finish;
+       if (pm_autosleep_state() > PM_SUSPEND_ON) {
+               error = -EBUSY;
+               goto out;
+       }
 
-       pr_debug("PM: Entering %s sleep\n", pm_states[state]);
-       error = suspend_devices_and_enter(state);
+       error = -EINVAL;
+       if (sscanf(buf, "%u", &val) == 1) {
+               if (pm_save_wakeup_count(val))
+                       error = n;
+       }
 
- Finish:
-       pr_debug("PM: Finishing wakeup.\n");
-       suspend_finish();
- Unlock:
-       mutex_unlock(&pm_mutex);
+ out:
+       pm_autosleep_unlock();
        return error;
 }
 
+power_attr(wakeup_count);
 
-/**
- *     pm_suspend - Externally visible function for suspending system.
- *     @state:         Enumerated value of state to enter.
- *
- *     Determine whether or not value is within range, get state 
- *     structure, and enter (above).
- */
-
-int pm_suspend(suspend_state_t state)
+#ifdef CONFIG_PM_AUTOSLEEP
+static ssize_t autosleep_show(struct kobject *kobj,
+                             struct kobj_attribute *attr,
+                             char *buf)
 {
-       if (state > PM_SUSPEND_ON && state <= PM_SUSPEND_MAX)
-               return enter_state(state);
-       return -EINVAL;
-}
+       suspend_state_t state = pm_autosleep_state();
 
-EXPORT_SYMBOL(pm_suspend);
+       if (state == PM_SUSPEND_ON)
+               return sprintf(buf, "off\n");
 
-#endif /* CONFIG_SUSPEND */
-
-struct kobject *power_kobj;
-
-/**
- *     state - control system power state.
- *
- *     show() returns what states are supported, which is hard-coded to
- *     'standby' (Power-On Suspend), 'mem' (Suspend-to-RAM), and
- *     'disk' (Suspend-to-Disk).
- *
- *     store() accepts one of those strings, translates it into the 
- *     proper enumerated value, and initiates a suspend transition.
- */
-
-static ssize_t state_show(struct kobject *kobj, struct kobj_attribute *attr,
-                         char *buf)
-{
-       char *s = buf;
 #ifdef CONFIG_SUSPEND
-       int i;
-
-       for (i = 0; i < PM_SUSPEND_MAX; i++) {
-               if (pm_states[i] && valid_state(i))
-                       s += sprintf(s,"%s ", pm_states[i]);
-       }
+       if (state < PM_SUSPEND_MAX)
+               return sprintf(buf, "%s\n", valid_state(state) ?
+                                               pm_states[state] : "error");
 #endif
 #ifdef CONFIG_HIBERNATION
-       s += sprintf(s, "%s\n", "disk");
+       return sprintf(buf, "disk\n");
 #else
-       if (s != buf)
-               /* convert the last space to a newline */
-               *(s-1) = '\n';
+       return sprintf(buf, "error");
 #endif
-       return (s - buf);
 }
 
-static ssize_t state_store(struct kobject *kobj, struct kobj_attribute *attr,
-                          const char *buf, size_t n)
+static ssize_t autosleep_store(struct kobject *kobj,
+                              struct kobj_attribute *attr,
+                              const char *buf, size_t n)
 {
-#ifdef CONFIG_SUSPEND
-       suspend_state_t state = PM_SUSPEND_STANDBY;
-       const char * const *s;
-#endif
-       char *p;
-       int len;
-       int error = -EINVAL;
+       suspend_state_t state = decode_state(buf, n);
+       int error;
 
-       p = memchr(buf, '\n', n);
-       len = p ? p - buf : n;
+       if (state == PM_SUSPEND_ON
+           && !(strncmp(buf, "off", 3) && strncmp(buf, "off\n", 4)))
+               return -EINVAL;
 
-       /* First, check if we are requested to hibernate */
-       if (len == 4 && !strncmp(buf, "disk", len)) {
-               error = hibernate();
-  goto Exit;
-       }
+       error = pm_autosleep_set_state(state);
+       return error ? error : n;
+}
 
-#ifdef CONFIG_SUSPEND
-       for (s = &pm_states[state]; state < PM_SUSPEND_MAX; s++, state++) {
-               if (*s && len == strlen(*s) && !strncmp(buf, *s, len))
-                       break;
-       }
-       if (state < PM_SUSPEND_MAX && *s)
-               error = enter_state(state);
-#endif
+power_attr(autosleep);
+#endif /* CONFIG_PM_AUTOSLEEP */
+
+#ifdef CONFIG_PM_WAKELOCKS
+static ssize_t wake_lock_show(struct kobject *kobj,
+                             struct kobj_attribute *attr,
+                             char *buf)
+{
+       return pm_show_wakelocks(buf, true);
+}
 
- Exit:
+static ssize_t wake_lock_store(struct kobject *kobj,
+                              struct kobj_attribute *attr,
+                              const char *buf, size_t n)
+{
+       int error = pm_wake_lock(buf);
        return error ? error : n;
 }
 
-power_attr(state);
+power_attr(wake_lock);
+
+static ssize_t wake_unlock_show(struct kobject *kobj,
+                               struct kobj_attribute *attr,
+                               char *buf)
+{
+       return pm_show_wakelocks(buf, false);
+}
+
+static ssize_t wake_unlock_store(struct kobject *kobj,
+                                struct kobj_attribute *attr,
+                                const char *buf, size_t n)
+{
+       int error = pm_wake_unlock(buf);
+       return error ? error : n;
+}
+
+power_attr(wake_unlock);
+
+#endif /* CONFIG_PM_WAKELOCKS */
+#endif /* CONFIG_PM_SLEEP */
 
 #ifdef CONFIG_PM_TRACE
 int pm_trace_enabled;
@@ -559,16 +493,45 @@ pm_trace_store(struct kobject *kobj, struct kobj_attribute *attr,
 }
 
 power_attr(pm_trace);
+
+static ssize_t pm_trace_dev_match_show(struct kobject *kobj,
+                                      struct kobj_attribute *attr,
+                                      char *buf)
+{
+       return show_trace_dev_match(buf, PAGE_SIZE);
+}
+
+static ssize_t
+pm_trace_dev_match_store(struct kobject *kobj, struct kobj_attribute *attr,
+                        const char *buf, size_t n)
+{
+       return -EINVAL;
+}
+
+power_attr(pm_trace_dev_match);
+
 #endif /* CONFIG_PM_TRACE */
 
 static struct attribute * g[] = {
        &state_attr.attr,
 #ifdef CONFIG_PM_TRACE
        &pm_trace_attr.attr,
+       &pm_trace_dev_match_attr.attr,
+#endif
+#ifdef CONFIG_PM_SLEEP
+       &pm_async_attr.attr,
+       &wakeup_count_attr.attr,
+#ifdef CONFIG_PM_AUTOSLEEP
+       &autosleep_attr.attr,
+#endif
+#ifdef CONFIG_PM_WAKELOCKS
+       &wake_lock_attr.attr,
+       &wake_unlock_attr.attr,
 #endif
-#if defined(CONFIG_PM_SLEEP) && defined(CONFIG_PM_DEBUG)
+#ifdef CONFIG_PM_DEBUG
        &pm_test_attr.attr,
 #endif
+#endif
        NULL,
 };
 
@@ -576,154 +539,34 @@ static struct attribute_group attr_group = {
        .attrs = g,
 };
 
+#ifdef CONFIG_PM_RUNTIME
+struct workqueue_struct *pm_wq;
+EXPORT_SYMBOL_GPL(pm_wq);
+
+static int __init pm_start_workqueue(void)
+{
+       pm_wq = alloc_workqueue("pm", WQ_FREEZABLE, 0);
+
+       return pm_wq ? 0 : -ENOMEM;
+}
+#else
+static inline int pm_start_workqueue(void) { return 0; }
+#endif
 
 static int __init pm_init(void)
 {
+       int error = pm_start_workqueue();
+       if (error)
+               return error;
+       hibernate_image_size_init();
+       hibernate_reserved_size_init();
        power_kobj = kobject_create_and_add("power", NULL);
        if (!power_kobj)
                return -ENOMEM;
-       return sysfs_create_group(power_kobj, &attr_group);
+       error = sysfs_create_group(power_kobj, &attr_group);
+       if (error)
+               return error;
+       return pm_autosleep_init();
 }
 
 core_initcall(pm_init);
-
-
-#ifdef CONFIG_PM_TEST_SUSPEND
-
-#include <linux/rtc.h>
-
-/*
- * To test system suspend, we need a hands-off mechanism to resume the
- * system.  RTCs wake alarms are a common self-contained mechanism.
- */
-
-static void __init test_wakealarm(struct rtc_device *rtc, suspend_state_t state)
-{
-       static char err_readtime[] __initdata =
-               KERN_ERR "PM: can't read %s time, err %d\n";
-       static char err_wakealarm [] __initdata =
-               KERN_ERR "PM: can't set %s wakealarm, err %d\n";
-       static char err_suspend[] __initdata =
-               KERN_ERR "PM: suspend test failed, error %d\n";
-       static char info_test[] __initdata =
-               KERN_INFO "PM: test RTC wakeup from '%s' suspend\n";
-
-       unsigned long           now;
-       struct rtc_wkalrm       alm;
-       int                     status;
-
-       /* this may fail if the RTC hasn't been initialized */
-       status = rtc_read_time(rtc, &alm.time);
-       if (status < 0) {
-               printk(err_readtime, dev_name(&rtc->dev), status);
-               return;
-       }
-       rtc_tm_to_time(&alm.time, &now);
-
-       memset(&alm, 0, sizeof alm);
-       rtc_time_to_tm(now + TEST_SUSPEND_SECONDS, &alm.time);
-       alm.enabled = true;
-
-       status = rtc_set_alarm(rtc, &alm);
-       if (status < 0) {
-               printk(err_wakealarm, dev_name(&rtc->dev), status);
-               return;
-       }
-
-       if (state == PM_SUSPEND_MEM) {
-               printk(info_test, pm_states[state]);
-               status = pm_suspend(state);
-               if (status == -ENODEV)
-                       state = PM_SUSPEND_STANDBY;
-       }
-       if (state == PM_SUSPEND_STANDBY) {
-               printk(info_test, pm_states[state]);
-               status = pm_suspend(state);
-       }
-       if (status < 0)
-               printk(err_suspend, status);
-
-       /* Some platforms can't detect that the alarm triggered the
-        * wakeup, or (accordingly) disable it after it afterwards.
-        * It's supposed to give oneshot behavior; cope.
-        */
-       alm.enabled = false;
-       rtc_set_alarm(rtc, &alm);
-}
-
-static int __init has_wakealarm(struct device *dev, void *name_ptr)
-{
-       struct rtc_device *candidate = to_rtc_device(dev);
-
-       if (!candidate->ops->set_alarm)
-               return 0;
-       if (!device_may_wakeup(candidate->dev.parent))
-               return 0;
-
-       *(const char **)name_ptr = dev_name(dev);
-       return 1;
-}
-
-/*
- * Kernel options like "test_suspend=mem" force suspend/resume sanity tests
- * at startup time.  They're normally disabled, for faster boot and because
- * we can't know which states really work on this particular system.
- */
-static suspend_state_t test_state __initdata = PM_SUSPEND_ON;
-
-static char warn_bad_state[] __initdata =
-       KERN_WARNING "PM: can't test '%s' suspend state\n";
-
-static int __init setup_test_suspend(char *value)
-{
-       unsigned i;
-
-       /* "=mem" ==> "mem" */
-       value++;
-       for (i = 0; i < PM_SUSPEND_MAX; i++) {
-               if (!pm_states[i])
-                       continue;
-               if (strcmp(pm_states[i], value) != 0)
-                       continue;
-               test_state = (__force suspend_state_t) i;
-               return 0;
-       }
-       printk(warn_bad_state, value);
-       return 0;
-}
-__setup("test_suspend", setup_test_suspend);
-
-static int __init test_suspend(void)
-{
-       static char             warn_no_rtc[] __initdata =
-               KERN_WARNING "PM: no wakealarm-capable RTC driver is ready\n";
-
-       char                    *pony = NULL;
-       struct rtc_device       *rtc = NULL;
-
-       /* PM is initialized by now; is that state testable? */
-       if (test_state == PM_SUSPEND_ON)
-               goto done;
-       if (!valid_state(test_state)) {
-               printk(warn_bad_state, pm_states[test_state]);
-               goto done;
-       }
-
-       /* RTCs have initialized by now too ... can we use one? */
-       class_find_device(rtc_class, NULL, &pony, has_wakealarm);
-       if (pony)
-               rtc = rtc_class_open(pony);
-       if (!rtc) {
-               printk(warn_no_rtc);
-               goto done;
-       }
-
-       /* go for it */
-       test_wakealarm(rtc, test_state);
-       rtc_class_close(rtc);
-done:
-       return 0;
-}
-late_initcall(test_suspend);
-
-#endif /* CONFIG_PM_TEST_SUSPEND */