PM / Sleep: Add user space interface for manipulating wakeup sources, v3
[linux-2.6.git] / drivers / base / power / runtime.c
index 4bafef8..5989487 100644 (file)
 /*
- * drivers/base/power/runtime.c - Handling dynamic device power management.
+ * drivers/base/power/runtime.c - Helper functions for device runtime PM
  *
- * Copyright (c) 2003 Patrick Mochel
- * Copyright (c) 2003 Open Source Development Lab
+ * Copyright (c) 2009 Rafael J. Wysocki <rjw@sisk.pl>, Novell Inc.
+ * Copyright (C) 2010 Alan Stern <stern@rowland.harvard.edu>
  *
+ * This file is released under the GPLv2.
  */
 
-#include <linux/device.h>
+#include <linux/sched.h>
+#include <linux/export.h>
+#include <linux/pm_runtime.h>
+#include <trace/events/rpm.h>
 #include "power.h"
 
+static int rpm_resume(struct device *dev, int rpmflags);
+static int rpm_suspend(struct device *dev, int rpmflags);
 
-static void runtime_resume(struct device * dev)
+/**
+ * update_pm_runtime_accounting - Update the time accounting of power states
+ * @dev: Device to update the accounting for
+ *
+ * In order to be able to have time accounting of the various power states
+ * (as used by programs such as PowerTOP to show the effectiveness of runtime
+ * PM), we need to track the time spent in each state.
+ * update_pm_runtime_accounting must be called each time before the
+ * runtime_status field is updated, to account the time in the old state
+ * correctly.
+ */
+void update_pm_runtime_accounting(struct device *dev)
 {
-       dev_dbg(dev, "resuming\n");
-       if (!dev->power.power_state.event)
+       unsigned long now = jiffies;
+       unsigned long delta;
+
+       delta = now - dev->power.accounting_timestamp;
+
+       dev->power.accounting_timestamp = now;
+
+       if (dev->power.disable_depth > 0)
                return;
-       if (!resume_device(dev))
-               dev->power.power_state = PMSG_ON;
+
+       if (dev->power.runtime_status == RPM_SUSPENDED)
+               dev->power.suspended_jiffies += delta;
+       else
+               dev->power.active_jiffies += delta;
 }
 
+static void __update_runtime_status(struct device *dev, enum rpm_status status)
+{
+       update_pm_runtime_accounting(dev);
+       dev->power.runtime_status = status;
+}
 
 /**
- *     dpm_runtime_resume - Power one device back on.
- *     @dev:   Device.
+ * pm_runtime_deactivate_timer - Deactivate given device's suspend timer.
+ * @dev: Device to handle.
+ */
+static void pm_runtime_deactivate_timer(struct device *dev)
+{
+       if (dev->power.timer_expires > 0) {
+               del_timer(&dev->power.suspend_timer);
+               dev->power.timer_expires = 0;
+       }
+}
+
+/**
+ * pm_runtime_cancel_pending - Deactivate suspend timer and cancel requests.
+ * @dev: Device to handle.
+ */
+static void pm_runtime_cancel_pending(struct device *dev)
+{
+       pm_runtime_deactivate_timer(dev);
+       /*
+        * In case there's a request pending, make sure its work function will
+        * return without doing anything.
+        */
+       dev->power.request = RPM_REQ_NONE;
+}
+
+/*
+ * pm_runtime_autosuspend_expiration - Get a device's autosuspend-delay expiration time.
+ * @dev: Device to handle.
  *
- *     Bring one device back to the on state by first powering it
- *     on, then restoring state. We only operate on devices that aren't
- *     already on.
- *     FIXME: We need to handle devices that are in an unknown state.
+ * Compute the autosuspend-delay expiration time based on the device's
+ * power.last_busy time.  If the delay has already expired or is disabled
+ * (negative) or the power.use_autosuspend flag isn't set, return 0.
+ * Otherwise return the expiration time in jiffies (adjusted to be nonzero).
+ *
+ * This function may be called either with or without dev->power.lock held.
+ * Either way it can be racy, since power.last_busy may be updated at any time.
+ */
+unsigned long pm_runtime_autosuspend_expiration(struct device *dev)
+{
+       int autosuspend_delay;
+       long elapsed;
+       unsigned long last_busy;
+       unsigned long expires = 0;
+
+       if (!dev->power.use_autosuspend)
+               goto out;
+
+       autosuspend_delay = ACCESS_ONCE(dev->power.autosuspend_delay);
+       if (autosuspend_delay < 0)
+               goto out;
+
+       last_busy = ACCESS_ONCE(dev->power.last_busy);
+       elapsed = jiffies - last_busy;
+       if (elapsed < 0)
+               goto out;       /* jiffies has wrapped around. */
+
+       /*
+        * If the autosuspend_delay is >= 1 second, align the timer by rounding
+        * up to the nearest second.
+        */
+       expires = last_busy + msecs_to_jiffies(autosuspend_delay);
+       if (autosuspend_delay >= 1000)
+               expires = round_jiffies(expires);
+       expires += !expires;
+       if (elapsed >= expires - last_busy)
+               expires = 0;    /* Already expired. */
+
+ out:
+       return expires;
+}
+EXPORT_SYMBOL_GPL(pm_runtime_autosuspend_expiration);
+
+/**
+ * rpm_check_suspend_allowed - Test whether a device may be suspended.
+ * @dev: Device to test.
  */
+static int rpm_check_suspend_allowed(struct device *dev)
+{
+       int retval = 0;
+
+       if (dev->power.runtime_error)
+               retval = -EINVAL;
+       else if (dev->power.disable_depth > 0)
+               retval = -EACCES;
+       else if (atomic_read(&dev->power.usage_count) > 0)
+               retval = -EAGAIN;
+       else if (!pm_children_suspended(dev))
+               retval = -EBUSY;
 
-void dpm_runtime_resume(struct device * dev)
+       /* Pending resume requests take precedence over suspends. */
+       else if ((dev->power.deferred_resume
+                       && dev->power.runtime_status == RPM_SUSPENDING)
+           || (dev->power.request_pending
+                       && dev->power.request == RPM_REQ_RESUME))
+               retval = -EAGAIN;
+       else if (dev->power.runtime_status == RPM_SUSPENDED)
+               retval = 1;
+
+       return retval;
+}
+
+/**
+ * __rpm_callback - Run a given runtime PM callback for a given device.
+ * @cb: Runtime PM callback to run.
+ * @dev: Device to run the callback for.
+ */
+static int __rpm_callback(int (*cb)(struct device *), struct device *dev)
+       __releases(&dev->power.lock) __acquires(&dev->power.lock)
 {
-       down(&dpm_sem);
-       runtime_resume(dev);
-       up(&dpm_sem);
+       int retval;
+
+       if (dev->power.irq_safe)
+               spin_unlock(&dev->power.lock);
+       else
+               spin_unlock_irq(&dev->power.lock);
+
+       retval = cb(dev);
+
+       if (dev->power.irq_safe)
+               spin_lock(&dev->power.lock);
+       else
+               spin_lock_irq(&dev->power.lock);
+
+       return retval;
 }
-EXPORT_SYMBOL(dpm_runtime_resume);
 
+/**
+ * rpm_idle - Notify device bus type if the device can be suspended.
+ * @dev: Device to notify the bus type about.
+ * @rpmflags: Flag bits.
+ *
+ * Check if the device's runtime PM status allows it to be suspended.  If
+ * another idle notification has been started earlier, return immediately.  If
+ * the RPM_ASYNC flag is set then queue an idle-notification request; otherwise
+ * run the ->runtime_idle() callback directly.
+ *
+ * This function must be called under dev->power.lock with interrupts disabled.
+ */
+static int rpm_idle(struct device *dev, int rpmflags)
+{
+       int (*callback)(struct device *);
+       int retval;
+
+       trace_rpm_idle(dev, rpmflags);
+       retval = rpm_check_suspend_allowed(dev);
+       if (retval < 0)
+               ;       /* Conditions are wrong. */
+
+       /* Idle notifications are allowed only in the RPM_ACTIVE state. */
+       else if (dev->power.runtime_status != RPM_ACTIVE)
+               retval = -EAGAIN;
+
+       /*
+        * Any pending request other than an idle notification takes
+        * precedence over us, except that the timer may be running.
+        */
+       else if (dev->power.request_pending &&
+           dev->power.request > RPM_REQ_IDLE)
+               retval = -EAGAIN;
+
+       /* Act as though RPM_NOWAIT is always set. */
+       else if (dev->power.idle_notification)
+               retval = -EINPROGRESS;
+       if (retval)
+               goto out;
+
+       /* Pending requests need to be canceled. */
+       dev->power.request = RPM_REQ_NONE;
+
+       if (dev->power.no_callbacks) {
+               /* Assume ->runtime_idle() callback would have suspended. */
+               retval = rpm_suspend(dev, rpmflags);
+               goto out;
+       }
+
+       /* Carry out an asynchronous or a synchronous idle notification. */
+       if (rpmflags & RPM_ASYNC) {
+               dev->power.request = RPM_REQ_IDLE;
+               if (!dev->power.request_pending) {
+                       dev->power.request_pending = true;
+                       queue_work(pm_wq, &dev->power.work);
+               }
+               goto out;
+       }
+
+       dev->power.idle_notification = true;
+
+       if (dev->pm_domain)
+               callback = dev->pm_domain->ops.runtime_idle;
+       else if (dev->type && dev->type->pm)
+               callback = dev->type->pm->runtime_idle;
+       else if (dev->class && dev->class->pm)
+               callback = dev->class->pm->runtime_idle;
+       else if (dev->bus && dev->bus->pm)
+               callback = dev->bus->pm->runtime_idle;
+       else
+               callback = NULL;
+
+       if (!callback && dev->driver && dev->driver->pm)
+               callback = dev->driver->pm->runtime_idle;
+
+       if (callback)
+               __rpm_callback(callback, dev);
+
+       dev->power.idle_notification = false;
+       wake_up_all(&dev->power.wait_queue);
+
+ out:
+       trace_rpm_return_int(dev, _THIS_IP_, retval);
+       return retval;
+}
 
 /**
- *     dpm_runtime_suspend - Put one device in low-power state.
- *     @dev:   Device.
- *     @state: State to enter.
+ * rpm_callback - Run a given runtime PM callback for a given device.
+ * @cb: Runtime PM callback to run.
+ * @dev: Device to run the callback for.
  */
+static int rpm_callback(int (*cb)(struct device *), struct device *dev)
+{
+       int retval;
+
+       if (!cb)
+               return -ENOSYS;
+
+       retval = __rpm_callback(cb, dev);
 
-int dpm_runtime_suspend(struct device * dev, pm_message_t state)
+       dev->power.runtime_error = retval;
+       return retval != -EACCES ? retval : -EIO;
+}
+
+/**
+ * rpm_suspend - Carry out runtime suspend of given device.
+ * @dev: Device to suspend.
+ * @rpmflags: Flag bits.
+ *
+ * Check if the device's runtime PM status allows it to be suspended.
+ * Cancel a pending idle notification, autosuspend or suspend. If
+ * another suspend has been started earlier, either return immediately
+ * or wait for it to finish, depending on the RPM_NOWAIT and RPM_ASYNC
+ * flags. If the RPM_ASYNC flag is set then queue a suspend request;
+ * otherwise run the ->runtime_suspend() callback directly. When
+ * ->runtime_suspend succeeded, if a deferred resume was requested while
+ * the callback was running then carry it out, otherwise send an idle
+ * notification for its parent (if the suspend succeeded and both
+ * ignore_children of parent->power and irq_safe of dev->power are not set).
+ * If ->runtime_suspend failed with -EAGAIN or -EBUSY, and if the RPM_AUTO
+ * flag is set and the next autosuspend-delay expiration time is in the
+ * future, schedule another autosuspend attempt.
+ *
+ * This function must be called under dev->power.lock with interrupts disabled.
+ */
+static int rpm_suspend(struct device *dev, int rpmflags)
+       __releases(&dev->power.lock) __acquires(&dev->power.lock)
 {
+       int (*callback)(struct device *);
+       struct device *parent = NULL;
+       int retval;
+
+       trace_rpm_suspend(dev, rpmflags);
+
+ repeat:
+       retval = rpm_check_suspend_allowed(dev);
+
+       if (retval < 0)
+               ;       /* Conditions are wrong. */
+
+       /* Synchronous suspends are not allowed in the RPM_RESUMING state. */
+       else if (dev->power.runtime_status == RPM_RESUMING &&
+           !(rpmflags & RPM_ASYNC))
+               retval = -EAGAIN;
+       if (retval)
+               goto out;
+
+       /* If the autosuspend_delay time hasn't expired yet, reschedule. */
+       if ((rpmflags & RPM_AUTO)
+           && dev->power.runtime_status != RPM_SUSPENDING) {
+               unsigned long expires = pm_runtime_autosuspend_expiration(dev);
+
+               if (expires != 0) {
+                       /* Pending requests need to be canceled. */
+                       dev->power.request = RPM_REQ_NONE;
+
+                       /*
+                        * Optimization: If the timer is already running and is
+                        * set to expire at or before the autosuspend delay,
+                        * avoid the overhead of resetting it.  Just let it
+                        * expire; pm_suspend_timer_fn() will take care of the
+                        * rest.
+                        */
+                       if (!(dev->power.timer_expires && time_before_eq(
+                           dev->power.timer_expires, expires))) {
+                               dev->power.timer_expires = expires;
+                               mod_timer(&dev->power.suspend_timer, expires);
+                       }
+                       dev->power.timer_autosuspends = 1;
+                       goto out;
+               }
+       }
+
+       /* Other scheduled or pending requests need to be canceled. */
+       pm_runtime_cancel_pending(dev);
+
+       if (dev->power.runtime_status == RPM_SUSPENDING) {
+               DEFINE_WAIT(wait);
+
+               if (rpmflags & (RPM_ASYNC | RPM_NOWAIT)) {
+                       retval = -EINPROGRESS;
+                       goto out;
+               }
+
+               if (dev->power.irq_safe) {
+                       spin_unlock(&dev->power.lock);
+
+                       cpu_relax();
+
+                       spin_lock(&dev->power.lock);
+                       goto repeat;
+               }
+
+               /* Wait for the other suspend running in parallel with us. */
+               for (;;) {
+                       prepare_to_wait(&dev->power.wait_queue, &wait,
+                                       TASK_UNINTERRUPTIBLE);
+                       if (dev->power.runtime_status != RPM_SUSPENDING)
+                               break;
+
+                       spin_unlock_irq(&dev->power.lock);
+
+                       schedule();
+
+                       spin_lock_irq(&dev->power.lock);
+               }
+               finish_wait(&dev->power.wait_queue, &wait);
+               goto repeat;
+       }
+
+       dev->power.deferred_resume = false;
+       if (dev->power.no_callbacks)
+               goto no_callback;       /* Assume success. */
+
+       /* Carry out an asynchronous or a synchronous suspend. */
+       if (rpmflags & RPM_ASYNC) {
+               dev->power.request = (rpmflags & RPM_AUTO) ?
+                   RPM_REQ_AUTOSUSPEND : RPM_REQ_SUSPEND;
+               if (!dev->power.request_pending) {
+                       dev->power.request_pending = true;
+                       queue_work(pm_wq, &dev->power.work);
+               }
+               goto out;
+       }
+
+       if (__dev_pm_qos_read_value(dev) < 0) {
+               /* Negative PM QoS constraint means "never suspend". */
+               retval = -EPERM;
+               goto out;
+       }
+
+       __update_runtime_status(dev, RPM_SUSPENDING);
+
+       if (dev->pm_domain)
+               callback = dev->pm_domain->ops.runtime_suspend;
+       else if (dev->type && dev->type->pm)
+               callback = dev->type->pm->runtime_suspend;
+       else if (dev->class && dev->class->pm)
+               callback = dev->class->pm->runtime_suspend;
+       else if (dev->bus && dev->bus->pm)
+               callback = dev->bus->pm->runtime_suspend;
+       else
+               callback = NULL;
+
+       if (!callback && dev->driver && dev->driver->pm)
+               callback = dev->driver->pm->runtime_suspend;
+
+       retval = rpm_callback(callback, dev);
+       if (retval)
+               goto fail;
+
+ no_callback:
+       __update_runtime_status(dev, RPM_SUSPENDED);
+       pm_runtime_deactivate_timer(dev);
+
+       if (dev->parent) {
+               parent = dev->parent;
+               atomic_add_unless(&parent->power.child_count, -1, 0);
+       }
+       wake_up_all(&dev->power.wait_queue);
+
+       if (dev->power.deferred_resume) {
+               rpm_resume(dev, 0);
+               retval = -EAGAIN;
+               goto out;
+       }
+
+       /* Maybe the parent is now able to suspend. */
+       if (parent && !parent->power.ignore_children && !dev->power.irq_safe) {
+               spin_unlock(&dev->power.lock);
+
+               spin_lock(&parent->power.lock);
+               rpm_idle(parent, RPM_ASYNC);
+               spin_unlock(&parent->power.lock);
+
+               spin_lock(&dev->power.lock);
+       }
+
+ out:
+       trace_rpm_return_int(dev, _THIS_IP_, retval);
+
+       return retval;
+
+ fail:
+       __update_runtime_status(dev, RPM_ACTIVE);
+       dev->power.deferred_resume = false;
+       wake_up_all(&dev->power.wait_queue);
+
+       if (retval == -EAGAIN || retval == -EBUSY) {
+               dev->power.runtime_error = 0;
+
+               /*
+                * If the callback routine failed an autosuspend, and
+                * if the last_busy time has been updated so that there
+                * is a new autosuspend expiration time, automatically
+                * reschedule another autosuspend.
+                */
+               if ((rpmflags & RPM_AUTO) &&
+                   pm_runtime_autosuspend_expiration(dev) != 0)
+                       goto repeat;
+       } else {
+               pm_runtime_cancel_pending(dev);
+       }
+       goto out;
+}
+
+/**
+ * rpm_resume - Carry out runtime resume of given device.
+ * @dev: Device to resume.
+ * @rpmflags: Flag bits.
+ *
+ * Check if the device's runtime PM status allows it to be resumed.  Cancel
+ * any scheduled or pending requests.  If another resume has been started
+ * earlier, either return immediately or wait for it to finish, depending on the
+ * RPM_NOWAIT and RPM_ASYNC flags.  Similarly, if there's a suspend running in
+ * parallel with this function, either tell the other process to resume after
+ * suspending (deferred_resume) or wait for it to finish.  If the RPM_ASYNC
+ * flag is set then queue a resume request; otherwise run the
+ * ->runtime_resume() callback directly.  Queue an idle notification for the
+ * device if the resume succeeded.
+ *
+ * This function must be called under dev->power.lock with interrupts disabled.
+ */
+static int rpm_resume(struct device *dev, int rpmflags)
+       __releases(&dev->power.lock) __acquires(&dev->power.lock)
+{
+       int (*callback)(struct device *);
+       struct device *parent = NULL;
+       int retval = 0;
+
+       trace_rpm_resume(dev, rpmflags);
+
+ repeat:
+       if (dev->power.runtime_error)
+               retval = -EINVAL;
+       else if (dev->power.disable_depth > 0)
+               retval = -EACCES;
+       if (retval)
+               goto out;
+
+       /*
+        * Other scheduled or pending requests need to be canceled.  Small
+        * optimization: If an autosuspend timer is running, leave it running
+        * rather than cancelling it now only to restart it again in the near
+        * future.
+        */
+       dev->power.request = RPM_REQ_NONE;
+       if (!dev->power.timer_autosuspends)
+               pm_runtime_deactivate_timer(dev);
+
+       if (dev->power.runtime_status == RPM_ACTIVE) {
+               retval = 1;
+               goto out;
+       }
+
+       if (dev->power.runtime_status == RPM_RESUMING
+           || dev->power.runtime_status == RPM_SUSPENDING) {
+               DEFINE_WAIT(wait);
+
+               if (rpmflags & (RPM_ASYNC | RPM_NOWAIT)) {
+                       if (dev->power.runtime_status == RPM_SUSPENDING)
+                               dev->power.deferred_resume = true;
+                       else
+                               retval = -EINPROGRESS;
+                       goto out;
+               }
+
+               if (dev->power.irq_safe) {
+                       spin_unlock(&dev->power.lock);
+
+                       cpu_relax();
+
+                       spin_lock(&dev->power.lock);
+                       goto repeat;
+               }
+
+               /* Wait for the operation carried out in parallel with us. */
+               for (;;) {
+                       prepare_to_wait(&dev->power.wait_queue, &wait,
+                                       TASK_UNINTERRUPTIBLE);
+                       if (dev->power.runtime_status != RPM_RESUMING
+                           && dev->power.runtime_status != RPM_SUSPENDING)
+                               break;
+
+                       spin_unlock_irq(&dev->power.lock);
+
+                       schedule();
+
+                       spin_lock_irq(&dev->power.lock);
+               }
+               finish_wait(&dev->power.wait_queue, &wait);
+               goto repeat;
+       }
+
+       /*
+        * See if we can skip waking up the parent.  This is safe only if
+        * power.no_callbacks is set, because otherwise we don't know whether
+        * the resume will actually succeed.
+        */
+       if (dev->power.no_callbacks && !parent && dev->parent) {
+               spin_lock_nested(&dev->parent->power.lock, SINGLE_DEPTH_NESTING);
+               if (dev->parent->power.disable_depth > 0
+                   || dev->parent->power.ignore_children
+                   || dev->parent->power.runtime_status == RPM_ACTIVE) {
+                       atomic_inc(&dev->parent->power.child_count);
+                       spin_unlock(&dev->parent->power.lock);
+                       goto no_callback;       /* Assume success. */
+               }
+               spin_unlock(&dev->parent->power.lock);
+       }
+
+       /* Carry out an asynchronous or a synchronous resume. */
+       if (rpmflags & RPM_ASYNC) {
+               dev->power.request = RPM_REQ_RESUME;
+               if (!dev->power.request_pending) {
+                       dev->power.request_pending = true;
+                       queue_work(pm_wq, &dev->power.work);
+               }
+               retval = 0;
+               goto out;
+       }
+
+       if (!parent && dev->parent) {
+               /*
+                * Increment the parent's usage counter and resume it if
+                * necessary.  Not needed if dev is irq-safe; then the
+                * parent is permanently resumed.
+                */
+               parent = dev->parent;
+               if (dev->power.irq_safe)
+                       goto skip_parent;
+               spin_unlock(&dev->power.lock);
+
+               pm_runtime_get_noresume(parent);
+
+               spin_lock(&parent->power.lock);
+               /*
+                * We can resume if the parent's runtime PM is disabled or it
+                * is set to ignore children.
+                */
+               if (!parent->power.disable_depth
+                   && !parent->power.ignore_children) {
+                       rpm_resume(parent, 0);
+                       if (parent->power.runtime_status != RPM_ACTIVE)
+                               retval = -EBUSY;
+               }
+               spin_unlock(&parent->power.lock);
+
+               spin_lock(&dev->power.lock);
+               if (retval)
+                       goto out;
+               goto repeat;
+       }
+ skip_parent:
+
+       if (dev->power.no_callbacks)
+               goto no_callback;       /* Assume success. */
+
+       __update_runtime_status(dev, RPM_RESUMING);
+
+       if (dev->pm_domain)
+               callback = dev->pm_domain->ops.runtime_resume;
+       else if (dev->type && dev->type->pm)
+               callback = dev->type->pm->runtime_resume;
+       else if (dev->class && dev->class->pm)
+               callback = dev->class->pm->runtime_resume;
+       else if (dev->bus && dev->bus->pm)
+               callback = dev->bus->pm->runtime_resume;
+       else
+               callback = NULL;
+
+       if (!callback && dev->driver && dev->driver->pm)
+               callback = dev->driver->pm->runtime_resume;
+
+       retval = rpm_callback(callback, dev);
+       if (retval) {
+               __update_runtime_status(dev, RPM_SUSPENDED);
+               pm_runtime_cancel_pending(dev);
+       } else {
+ no_callback:
+               __update_runtime_status(dev, RPM_ACTIVE);
+               if (parent)
+                       atomic_inc(&parent->power.child_count);
+       }
+       wake_up_all(&dev->power.wait_queue);
+
+       if (!retval)
+               rpm_idle(dev, RPM_ASYNC);
+
+ out:
+       if (parent && !dev->power.irq_safe) {
+               spin_unlock_irq(&dev->power.lock);
+
+               pm_runtime_put(parent);
+
+               spin_lock_irq(&dev->power.lock);
+       }
+
+       trace_rpm_return_int(dev, _THIS_IP_, retval);
+
+       return retval;
+}
+
+/**
+ * pm_runtime_work - Universal runtime PM work function.
+ * @work: Work structure used for scheduling the execution of this function.
+ *
+ * Use @work to get the device object the work is to be done for, determine what
+ * is to be done and execute the appropriate runtime PM function.
+ */
+static void pm_runtime_work(struct work_struct *work)
+{
+       struct device *dev = container_of(work, struct device, power.work);
+       enum rpm_request req;
+
+       spin_lock_irq(&dev->power.lock);
+
+       if (!dev->power.request_pending)
+               goto out;
+
+       req = dev->power.request;
+       dev->power.request = RPM_REQ_NONE;
+       dev->power.request_pending = false;
+
+       switch (req) {
+       case RPM_REQ_NONE:
+               break;
+       case RPM_REQ_IDLE:
+               rpm_idle(dev, RPM_NOWAIT);
+               break;
+       case RPM_REQ_SUSPEND:
+               rpm_suspend(dev, RPM_NOWAIT);
+               break;
+       case RPM_REQ_AUTOSUSPEND:
+               rpm_suspend(dev, RPM_NOWAIT | RPM_AUTO);
+               break;
+       case RPM_REQ_RESUME:
+               rpm_resume(dev, RPM_NOWAIT);
+               break;
+       }
+
+ out:
+       spin_unlock_irq(&dev->power.lock);
+}
+
+/**
+ * pm_suspend_timer_fn - Timer function for pm_schedule_suspend().
+ * @data: Device pointer passed by pm_schedule_suspend().
+ *
+ * Check if the time is right and queue a suspend request.
+ */
+static void pm_suspend_timer_fn(unsigned long data)
+{
+       struct device *dev = (struct device *)data;
+       unsigned long flags;
+       unsigned long expires;
+
+       spin_lock_irqsave(&dev->power.lock, flags);
+
+       expires = dev->power.timer_expires;
+       /* If 'expire' is after 'jiffies' we've been called too early. */
+       if (expires > 0 && !time_after(expires, jiffies)) {
+               dev->power.timer_expires = 0;
+               rpm_suspend(dev, dev->power.timer_autosuspends ?
+                   (RPM_ASYNC | RPM_AUTO) : RPM_ASYNC);
+       }
+
+       spin_unlock_irqrestore(&dev->power.lock, flags);
+}
+
+/**
+ * pm_schedule_suspend - Set up a timer to submit a suspend request in future.
+ * @dev: Device to suspend.
+ * @delay: Time to wait before submitting a suspend request, in milliseconds.
+ */
+int pm_schedule_suspend(struct device *dev, unsigned int delay)
+{
+       unsigned long flags;
+       int retval;
+
+       spin_lock_irqsave(&dev->power.lock, flags);
+
+       if (!delay) {
+               retval = rpm_suspend(dev, RPM_ASYNC);
+               goto out;
+       }
+
+       retval = rpm_check_suspend_allowed(dev);
+       if (retval)
+               goto out;
+
+       /* Other scheduled or pending requests need to be canceled. */
+       pm_runtime_cancel_pending(dev);
+
+       dev->power.timer_expires = jiffies + msecs_to_jiffies(delay);
+       dev->power.timer_expires += !dev->power.timer_expires;
+       dev->power.timer_autosuspends = 0;
+       mod_timer(&dev->power.suspend_timer, dev->power.timer_expires);
+
+ out:
+       spin_unlock_irqrestore(&dev->power.lock, flags);
+
+       return retval;
+}
+EXPORT_SYMBOL_GPL(pm_schedule_suspend);
+
+/**
+ * __pm_runtime_idle - Entry point for runtime idle operations.
+ * @dev: Device to send idle notification for.
+ * @rpmflags: Flag bits.
+ *
+ * If the RPM_GET_PUT flag is set, decrement the device's usage count and
+ * return immediately if it is larger than zero.  Then carry out an idle
+ * notification, either synchronous or asynchronous.
+ *
+ * This routine may be called in atomic context if the RPM_ASYNC flag is set,
+ * or if pm_runtime_irq_safe() has been called.
+ */
+int __pm_runtime_idle(struct device *dev, int rpmflags)
+{
+       unsigned long flags;
+       int retval;
+
+       might_sleep_if(!(rpmflags & RPM_ASYNC) && !dev->power.irq_safe);
+
+       if (rpmflags & RPM_GET_PUT) {
+               if (!atomic_dec_and_test(&dev->power.usage_count))
+                       return 0;
+       }
+
+       spin_lock_irqsave(&dev->power.lock, flags);
+       retval = rpm_idle(dev, rpmflags);
+       spin_unlock_irqrestore(&dev->power.lock, flags);
+
+       return retval;
+}
+EXPORT_SYMBOL_GPL(__pm_runtime_idle);
+
+/**
+ * __pm_runtime_suspend - Entry point for runtime put/suspend operations.
+ * @dev: Device to suspend.
+ * @rpmflags: Flag bits.
+ *
+ * If the RPM_GET_PUT flag is set, decrement the device's usage count and
+ * return immediately if it is larger than zero.  Then carry out a suspend,
+ * either synchronous or asynchronous.
+ *
+ * This routine may be called in atomic context if the RPM_ASYNC flag is set,
+ * or if pm_runtime_irq_safe() has been called.
+ */
+int __pm_runtime_suspend(struct device *dev, int rpmflags)
+{
+       unsigned long flags;
+       int retval;
+
+       might_sleep_if(!(rpmflags & RPM_ASYNC) && !dev->power.irq_safe);
+
+       if (rpmflags & RPM_GET_PUT) {
+               if (!atomic_dec_and_test(&dev->power.usage_count))
+                       return 0;
+       }
+
+       spin_lock_irqsave(&dev->power.lock, flags);
+       retval = rpm_suspend(dev, rpmflags);
+       spin_unlock_irqrestore(&dev->power.lock, flags);
+
+       return retval;
+}
+EXPORT_SYMBOL_GPL(__pm_runtime_suspend);
+
+/**
+ * __pm_runtime_resume - Entry point for runtime resume operations.
+ * @dev: Device to resume.
+ * @rpmflags: Flag bits.
+ *
+ * If the RPM_GET_PUT flag is set, increment the device's usage count.  Then
+ * carry out a resume, either synchronous or asynchronous.
+ *
+ * This routine may be called in atomic context if the RPM_ASYNC flag is set,
+ * or if pm_runtime_irq_safe() has been called.
+ */
+int __pm_runtime_resume(struct device *dev, int rpmflags)
+{
+       unsigned long flags;
+       int retval;
+
+       might_sleep_if(!(rpmflags & RPM_ASYNC) && !dev->power.irq_safe);
+
+       if (rpmflags & RPM_GET_PUT)
+               atomic_inc(&dev->power.usage_count);
+
+       spin_lock_irqsave(&dev->power.lock, flags);
+       retval = rpm_resume(dev, rpmflags);
+       spin_unlock_irqrestore(&dev->power.lock, flags);
+
+       return retval;
+}
+EXPORT_SYMBOL_GPL(__pm_runtime_resume);
+
+/**
+ * __pm_runtime_set_status - Set runtime PM status of a device.
+ * @dev: Device to handle.
+ * @status: New runtime PM status of the device.
+ *
+ * If runtime PM of the device is disabled or its power.runtime_error field is
+ * different from zero, the status may be changed either to RPM_ACTIVE, or to
+ * RPM_SUSPENDED, as long as that reflects the actual state of the device.
+ * However, if the device has a parent and the parent is not active, and the
+ * parent's power.ignore_children flag is unset, the device's status cannot be
+ * set to RPM_ACTIVE, so -EBUSY is returned in that case.
+ *
+ * If successful, __pm_runtime_set_status() clears the power.runtime_error field
+ * and the device parent's counter of unsuspended children is modified to
+ * reflect the new status.  If the new status is RPM_SUSPENDED, an idle
+ * notification request for the parent is submitted.
+ */
+int __pm_runtime_set_status(struct device *dev, unsigned int status)
+{
+       struct device *parent = dev->parent;
+       unsigned long flags;
+       bool notify_parent = false;
        int error = 0;
 
-       down(&dpm_sem);
-       if (dev->power.power_state.event == state.event)
-               goto Done;
+       if (status != RPM_ACTIVE && status != RPM_SUSPENDED)
+               return -EINVAL;
+
+       spin_lock_irqsave(&dev->power.lock, flags);
+
+       if (!dev->power.runtime_error && !dev->power.disable_depth) {
+               error = -EAGAIN;
+               goto out;
+       }
+
+       if (dev->power.runtime_status == status)
+               goto out_set;
+
+       if (status == RPM_SUSPENDED) {
+               /* It always is possible to set the status to 'suspended'. */
+               if (parent) {
+                       atomic_add_unless(&parent->power.child_count, -1, 0);
+                       notify_parent = !parent->power.ignore_children;
+               }
+               goto out_set;
+       }
+
+       if (parent) {
+               spin_lock_nested(&parent->power.lock, SINGLE_DEPTH_NESTING);
+
+               /*
+                * It is invalid to put an active child under a parent that is
+                * not active, has runtime PM enabled and the
+                * 'power.ignore_children' flag unset.
+                */
+               if (!parent->power.disable_depth
+                   && !parent->power.ignore_children
+                   && parent->power.runtime_status != RPM_ACTIVE)
+                       error = -EBUSY;
+               else if (dev->power.runtime_status == RPM_SUSPENDED)
+                       atomic_inc(&parent->power.child_count);
+
+               spin_unlock(&parent->power.lock);
 
-       if (dev->power.power_state.event)
-               runtime_resume(dev);
+               if (error)
+                       goto out;
+       }
+
+ out_set:
+       __update_runtime_status(dev, status);
+       dev->power.runtime_error = 0;
+ out:
+       spin_unlock_irqrestore(&dev->power.lock, flags);
+
+       if (notify_parent)
+               pm_request_idle(parent);
 
-       if (!(error = suspend_device(dev, state)))
-               dev->power.power_state = state;
- Done:
-       up(&dpm_sem);
        return error;
 }
+EXPORT_SYMBOL_GPL(__pm_runtime_set_status);
+
+/**
+ * __pm_runtime_barrier - Cancel pending requests and wait for completions.
+ * @dev: Device to handle.
+ *
+ * Flush all pending requests for the device from pm_wq and wait for all
+ * runtime PM operations involving the device in progress to complete.
+ *
+ * Should be called under dev->power.lock with interrupts disabled.
+ */
+static void __pm_runtime_barrier(struct device *dev)
+{
+       pm_runtime_deactivate_timer(dev);
+
+       if (dev->power.request_pending) {
+               dev->power.request = RPM_REQ_NONE;
+               spin_unlock_irq(&dev->power.lock);
+
+               cancel_work_sync(&dev->power.work);
+
+               spin_lock_irq(&dev->power.lock);
+               dev->power.request_pending = false;
+       }
+
+       if (dev->power.runtime_status == RPM_SUSPENDING
+           || dev->power.runtime_status == RPM_RESUMING
+           || dev->power.idle_notification) {
+               DEFINE_WAIT(wait);
+
+               /* Suspend, wake-up or idle notification in progress. */
+               for (;;) {
+                       prepare_to_wait(&dev->power.wait_queue, &wait,
+                                       TASK_UNINTERRUPTIBLE);
+                       if (dev->power.runtime_status != RPM_SUSPENDING
+                           && dev->power.runtime_status != RPM_RESUMING
+                           && !dev->power.idle_notification)
+                               break;
+                       spin_unlock_irq(&dev->power.lock);
+
+                       schedule();
+
+                       spin_lock_irq(&dev->power.lock);
+               }
+               finish_wait(&dev->power.wait_queue, &wait);
+       }
+}
+
+/**
+ * pm_runtime_barrier - Flush pending requests and wait for completions.
+ * @dev: Device to handle.
+ *
+ * Prevent the device from being suspended by incrementing its usage counter and
+ * if there's a pending resume request for the device, wake the device up.
+ * Next, make sure that all pending requests for the device have been flushed
+ * from pm_wq and wait for all runtime PM operations involving the device in
+ * progress to complete.
+ *
+ * Return value:
+ * 1, if there was a resume request pending and the device had to be woken up,
+ * 0, otherwise
+ */
+int pm_runtime_barrier(struct device *dev)
+{
+       int retval = 0;
+
+       pm_runtime_get_noresume(dev);
+       spin_lock_irq(&dev->power.lock);
+
+       if (dev->power.request_pending
+           && dev->power.request == RPM_REQ_RESUME) {
+               rpm_resume(dev, 0);
+               retval = 1;
+       }
+
+       __pm_runtime_barrier(dev);
+
+       spin_unlock_irq(&dev->power.lock);
+       pm_runtime_put_noidle(dev);
+
+       return retval;
+}
+EXPORT_SYMBOL_GPL(pm_runtime_barrier);
+
+/**
+ * __pm_runtime_disable - Disable runtime PM of a device.
+ * @dev: Device to handle.
+ * @check_resume: If set, check if there's a resume request for the device.
+ *
+ * Increment power.disable_depth for the device and if was zero previously,
+ * cancel all pending runtime PM requests for the device and wait for all
+ * operations in progress to complete.  The device can be either active or
+ * suspended after its runtime PM has been disabled.
+ *
+ * If @check_resume is set and there's a resume request pending when
+ * __pm_runtime_disable() is called and power.disable_depth is zero, the
+ * function will wake up the device before disabling its runtime PM.
+ */
+void __pm_runtime_disable(struct device *dev, bool check_resume)
+{
+       spin_lock_irq(&dev->power.lock);
+
+       if (dev->power.disable_depth > 0) {
+               dev->power.disable_depth++;
+               goto out;
+       }
+
+       /*
+        * Wake up the device if there's a resume request pending, because that
+        * means there probably is some I/O to process and disabling runtime PM
+        * shouldn't prevent the device from processing the I/O.
+        */
+       if (check_resume && dev->power.request_pending
+           && dev->power.request == RPM_REQ_RESUME) {
+               /*
+                * Prevent suspends and idle notifications from being carried
+                * out after we have woken up the device.
+                */
+               pm_runtime_get_noresume(dev);
+
+               rpm_resume(dev, 0);
+
+               pm_runtime_put_noidle(dev);
+       }
+
+       if (!dev->power.disable_depth++)
+               __pm_runtime_barrier(dev);
+
+ out:
+       spin_unlock_irq(&dev->power.lock);
+}
+EXPORT_SYMBOL_GPL(__pm_runtime_disable);
+
+/**
+ * pm_runtime_enable - Enable runtime PM of a device.
+ * @dev: Device to handle.
+ */
+void pm_runtime_enable(struct device *dev)
+{
+       unsigned long flags;
+
+       spin_lock_irqsave(&dev->power.lock, flags);
+
+       if (dev->power.disable_depth > 0)
+               dev->power.disable_depth--;
+       else
+               dev_warn(dev, "Unbalanced %s!\n", __func__);
+
+       spin_unlock_irqrestore(&dev->power.lock, flags);
+}
+EXPORT_SYMBOL_GPL(pm_runtime_enable);
+
+/**
+ * pm_runtime_forbid - Block runtime PM of a device.
+ * @dev: Device to handle.
+ *
+ * Increase the device's usage count and clear its power.runtime_auto flag,
+ * so that it cannot be suspended at run time until pm_runtime_allow() is called
+ * for it.
+ */
+void pm_runtime_forbid(struct device *dev)
+{
+       spin_lock_irq(&dev->power.lock);
+       if (!dev->power.runtime_auto)
+               goto out;
+
+       dev->power.runtime_auto = false;
+       atomic_inc(&dev->power.usage_count);
+       rpm_resume(dev, 0);
+
+ out:
+       spin_unlock_irq(&dev->power.lock);
+}
+EXPORT_SYMBOL_GPL(pm_runtime_forbid);
+
+/**
+ * pm_runtime_allow - Unblock runtime PM of a device.
+ * @dev: Device to handle.
+ *
+ * Decrease the device's usage count and set its power.runtime_auto flag.
+ */
+void pm_runtime_allow(struct device *dev)
+{
+       spin_lock_irq(&dev->power.lock);
+       if (dev->power.runtime_auto)
+               goto out;
 
+       dev->power.runtime_auto = true;
+       if (atomic_dec_and_test(&dev->power.usage_count))
+               rpm_idle(dev, RPM_AUTO);
+
+ out:
+       spin_unlock_irq(&dev->power.lock);
+}
+EXPORT_SYMBOL_GPL(pm_runtime_allow);
+
+/**
+ * pm_runtime_no_callbacks - Ignore runtime PM callbacks for a device.
+ * @dev: Device to handle.
+ *
+ * Set the power.no_callbacks flag, which tells the PM core that this
+ * device is power-managed through its parent and has no runtime PM
+ * callbacks of its own.  The runtime sysfs attributes will be removed.
+ */
+void pm_runtime_no_callbacks(struct device *dev)
+{
+       spin_lock_irq(&dev->power.lock);
+       dev->power.no_callbacks = 1;
+       spin_unlock_irq(&dev->power.lock);
+       if (device_is_registered(dev))
+               rpm_sysfs_remove(dev);
+}
+EXPORT_SYMBOL_GPL(pm_runtime_no_callbacks);
 
-#if 0
 /**
- *     dpm_set_power_state - Update power_state field.
- *     @dev:   Device.
- *     @state: Power state device is in.
+ * pm_runtime_irq_safe - Leave interrupts disabled during callbacks.
+ * @dev: Device to handle
  *
- *     This is an update mechanism for drivers to notify the core
- *     what power state a device is in. Device probing code may not
- *     always be able to tell, but we need accurate information to
- *     work reliably.
+ * Set the power.irq_safe flag, which tells the PM core that the
+ * ->runtime_suspend() and ->runtime_resume() callbacks for this device should
+ * always be invoked with the spinlock held and interrupts disabled.  It also
+ * causes the parent's usage counter to be permanently incremented, preventing
+ * the parent from runtime suspending -- otherwise an irq-safe child might have
+ * to wait for a non-irq-safe parent.
  */
-void dpm_set_power_state(struct device * dev, pm_message_t state)
+void pm_runtime_irq_safe(struct device *dev)
 {
-       down(&dpm_sem);
-       dev->power.power_state = state;
-       up(&dpm_sem);
+       if (dev->parent)
+               pm_runtime_get_sync(dev->parent);
+       spin_lock_irq(&dev->power.lock);
+       dev->power.irq_safe = 1;
+       spin_unlock_irq(&dev->power.lock);
+}
+EXPORT_SYMBOL_GPL(pm_runtime_irq_safe);
+
+/**
+ * update_autosuspend - Handle a change to a device's autosuspend settings.
+ * @dev: Device to handle.
+ * @old_delay: The former autosuspend_delay value.
+ * @old_use: The former use_autosuspend value.
+ *
+ * Prevent runtime suspend if the new delay is negative and use_autosuspend is
+ * set; otherwise allow it.  Send an idle notification if suspends are allowed.
+ *
+ * This function must be called under dev->power.lock with interrupts disabled.
+ */
+static void update_autosuspend(struct device *dev, int old_delay, int old_use)
+{
+       int delay = dev->power.autosuspend_delay;
+
+       /* Should runtime suspend be prevented now? */
+       if (dev->power.use_autosuspend && delay < 0) {
+
+               /* If it used to be allowed then prevent it. */
+               if (!old_use || old_delay >= 0) {
+                       atomic_inc(&dev->power.usage_count);
+                       rpm_resume(dev, 0);
+               }
+       }
+
+       /* Runtime suspend should be allowed now. */
+       else {
+
+               /* If it used to be prevented then allow it. */
+               if (old_use && old_delay < 0)
+                       atomic_dec(&dev->power.usage_count);
+
+               /* Maybe we can autosuspend now. */
+               rpm_idle(dev, RPM_AUTO);
+       }
+}
+
+/**
+ * pm_runtime_set_autosuspend_delay - Set a device's autosuspend_delay value.
+ * @dev: Device to handle.
+ * @delay: Value of the new delay in milliseconds.
+ *
+ * Set the device's power.autosuspend_delay value.  If it changes to negative
+ * and the power.use_autosuspend flag is set, prevent runtime suspends.  If it
+ * changes the other way, allow runtime suspends.
+ */
+void pm_runtime_set_autosuspend_delay(struct device *dev, int delay)
+{
+       int old_delay, old_use;
+
+       spin_lock_irq(&dev->power.lock);
+       old_delay = dev->power.autosuspend_delay;
+       old_use = dev->power.use_autosuspend;
+       dev->power.autosuspend_delay = delay;
+       update_autosuspend(dev, old_delay, old_use);
+       spin_unlock_irq(&dev->power.lock);
+}
+EXPORT_SYMBOL_GPL(pm_runtime_set_autosuspend_delay);
+
+/**
+ * __pm_runtime_use_autosuspend - Set a device's use_autosuspend flag.
+ * @dev: Device to handle.
+ * @use: New value for use_autosuspend.
+ *
+ * Set the device's power.use_autosuspend flag, and allow or prevent runtime
+ * suspends as needed.
+ */
+void __pm_runtime_use_autosuspend(struct device *dev, bool use)
+{
+       int old_delay, old_use;
+
+       spin_lock_irq(&dev->power.lock);
+       old_delay = dev->power.autosuspend_delay;
+       old_use = dev->power.use_autosuspend;
+       dev->power.use_autosuspend = use;
+       update_autosuspend(dev, old_delay, old_use);
+       spin_unlock_irq(&dev->power.lock);
+}
+EXPORT_SYMBOL_GPL(__pm_runtime_use_autosuspend);
+
+/**
+ * pm_runtime_init - Initialize runtime PM fields in given device object.
+ * @dev: Device object to initialize.
+ */
+void pm_runtime_init(struct device *dev)
+{
+       dev->power.runtime_status = RPM_SUSPENDED;
+       dev->power.idle_notification = false;
+
+       dev->power.disable_depth = 1;
+       atomic_set(&dev->power.usage_count, 0);
+
+       dev->power.runtime_error = 0;
+
+       atomic_set(&dev->power.child_count, 0);
+       pm_suspend_ignore_children(dev, false);
+       dev->power.runtime_auto = true;
+
+       dev->power.request_pending = false;
+       dev->power.request = RPM_REQ_NONE;
+       dev->power.deferred_resume = false;
+       dev->power.accounting_timestamp = jiffies;
+       INIT_WORK(&dev->power.work, pm_runtime_work);
+
+       dev->power.timer_expires = 0;
+       setup_timer(&dev->power.suspend_timer, pm_suspend_timer_fn,
+                       (unsigned long)dev);
+
+       init_waitqueue_head(&dev->power.wait_queue);
+}
+
+/**
+ * pm_runtime_remove - Prepare for removing a device from device hierarchy.
+ * @dev: Device object being removed from device hierarchy.
+ */
+void pm_runtime_remove(struct device *dev)
+{
+       __pm_runtime_disable(dev, false);
+
+       /* Change the status back to 'suspended' to match the initial status. */
+       if (dev->power.runtime_status == RPM_ACTIVE)
+               pm_runtime_set_suspended(dev);
+       if (dev->power.irq_safe && dev->parent)
+               pm_runtime_put_sync(dev->parent);
 }
-#endif  /*  0  */