PM / devfreq: Core updates to support devices which can idle
Rajagopal Venkat [Thu, 25 Oct 2012 23:50:09 +0000 (01:50 +0200)]
Prepare devfreq core framework to support devices which
can idle. When device idleness is detected perhaps through
runtime-pm, need some mechanism to suspend devfreq load
monitoring and resume back when device is online. Present
code continues monitoring unless device is removed from
devfreq core.

This patch introduces following design changes,

 - use per device work instead of global work to monitor device
   load. This enables suspend/resume of device devfreq and
   reduces monitoring code complexity.
 - decouple delayed work based load monitoring logic from core
   by introducing helpers functions to be used by governors. This
   provides flexibility for governors either to use delayed work
   based monitoring functions or to implement their own mechanism.
 - devfreq core interacts with governors via events to perform
   specific actions. These events include start/stop devfreq.
   This sets ground for adding suspend/resume events.

The devfreq apis are not modified and are kept intact.

Signed-off-by: Rajagopal Venkat <rajagopal.venkat@linaro.org>
Acked-by: MyungJoo Ham <myungjoo.ham@samsung.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>

Documentation/ABI/testing/sysfs-class-devfreq
drivers/devfreq/devfreq.c
drivers/devfreq/governor.h
drivers/devfreq/governor_performance.c
drivers/devfreq/governor_powersave.c
drivers/devfreq/governor_simpleondemand.c
drivers/devfreq/governor_userspace.c
include/linux/devfreq.h

index 23d78b5..89283b1 100644 (file)
@@ -21,14 +21,6 @@ Description:
                The /sys/class/devfreq/.../cur_freq shows the current
                frequency of the corresponding devfreq object.
 
-What:          /sys/class/devfreq/.../central_polling
-Date:          September 2011
-Contact:       MyungJoo Ham <myungjoo.ham@samsung.com>
-Description:
-               The /sys/class/devfreq/.../central_polling shows whether
-               the devfreq ojbect is using devfreq-provided central
-               polling mechanism or not.
-
 What:          /sys/class/devfreq/.../polling_interval
 Date:          September 2011
 Contact:       MyungJoo Ham <myungjoo.ham@samsung.com>
index b146d76..1aaf1ae 100644 (file)
 struct class *devfreq_class;
 
 /*
- * devfreq_work periodically monitors every registered device.
- * The minimum polling interval is one jiffy. The polling interval is
- * determined by the minimum polling period among all polling devfreq
- * devices. The resolution of polling interval is one jiffy.
+ * devfreq core provides delayed work based load monitoring helper
+ * functions. Governors can use these or can implement their own
+ * monitoring mechanism.
  */
-static bool polling;
 static struct workqueue_struct *devfreq_wq;
-static struct delayed_work devfreq_work;
-
-/* wait removing if this is to be removed */
-static struct devfreq *wait_remove_device;
 
 /* The list of all device-devfreq */
 static LIST_HEAD(devfreq_list);
@@ -72,6 +66,8 @@ static struct devfreq *find_device_devfreq(struct device *dev)
        return ERR_PTR(-ENODEV);
 }
 
+/* Load monitoring helper functions for governors use */
+
 /**
  * update_devfreq() - Reevaluate the device and configure frequency.
  * @devfreq:   the devfreq instance.
@@ -121,6 +117,152 @@ int update_devfreq(struct devfreq *devfreq)
 }
 
 /**
+ * devfreq_monitor() - Periodically poll devfreq objects.
+ * @work:      the work struct used to run devfreq_monitor periodically.
+ *
+ */
+static void devfreq_monitor(struct work_struct *work)
+{
+       int err;
+       struct devfreq *devfreq = container_of(work,
+                                       struct devfreq, work.work);
+
+       mutex_lock(&devfreq->lock);
+       err = update_devfreq(devfreq);
+       if (err)
+               dev_err(&devfreq->dev, "dvfs failed with (%d) error\n", err);
+
+       queue_delayed_work(devfreq_wq, &devfreq->work,
+                               msecs_to_jiffies(devfreq->profile->polling_ms));
+       mutex_unlock(&devfreq->lock);
+}
+
+/**
+ * devfreq_monitor_start() - Start load monitoring of devfreq instance
+ * @devfreq:   the devfreq instance.
+ *
+ * Helper function for starting devfreq device load monitoing. By
+ * default delayed work based monitoring is supported. Function
+ * to be called from governor in response to DEVFREQ_GOV_START
+ * event when device is added to devfreq framework.
+ */
+void devfreq_monitor_start(struct devfreq *devfreq)
+{
+       INIT_DEFERRABLE_WORK(&devfreq->work, devfreq_monitor);
+       if (devfreq->profile->polling_ms)
+               queue_delayed_work(devfreq_wq, &devfreq->work,
+                       msecs_to_jiffies(devfreq->profile->polling_ms));
+}
+
+/**
+ * devfreq_monitor_stop() - Stop load monitoring of a devfreq instance
+ * @devfreq:   the devfreq instance.
+ *
+ * Helper function to stop devfreq device load monitoing. Function
+ * to be called from governor in response to DEVFREQ_GOV_STOP
+ * event when device is removed from devfreq framework.
+ */
+void devfreq_monitor_stop(struct devfreq *devfreq)
+{
+       cancel_delayed_work_sync(&devfreq->work);
+}
+
+/**
+ * devfreq_monitor_suspend() - Suspend load monitoring of a devfreq instance
+ * @devfreq:   the devfreq instance.
+ *
+ * Helper function to suspend devfreq device load monitoing. Function
+ * to be called from governor in response to DEVFREQ_GOV_SUSPEND
+ * event or when polling interval is set to zero.
+ *
+ * Note: Though this function is same as devfreq_monitor_stop(),
+ * intentionally kept separate to provide hooks for collecting
+ * transition statistics.
+ */
+void devfreq_monitor_suspend(struct devfreq *devfreq)
+{
+       mutex_lock(&devfreq->lock);
+       if (devfreq->stop_polling) {
+               mutex_unlock(&devfreq->lock);
+               return;
+       }
+
+       devfreq->stop_polling = true;
+       mutex_unlock(&devfreq->lock);
+       cancel_delayed_work_sync(&devfreq->work);
+}
+
+/**
+ * devfreq_monitor_resume() - Resume load monitoring of a devfreq instance
+ * @devfreq:    the devfreq instance.
+ *
+ * Helper function to resume devfreq device load monitoing. Function
+ * to be called from governor in response to DEVFREQ_GOV_RESUME
+ * event or when polling interval is set to non-zero.
+ */
+void devfreq_monitor_resume(struct devfreq *devfreq)
+{
+       mutex_lock(&devfreq->lock);
+       if (!devfreq->stop_polling)
+               goto out;
+
+       if (!delayed_work_pending(&devfreq->work) &&
+                       devfreq->profile->polling_ms)
+               queue_delayed_work(devfreq_wq, &devfreq->work,
+                       msecs_to_jiffies(devfreq->profile->polling_ms));
+       devfreq->stop_polling = false;
+
+out:
+       mutex_unlock(&devfreq->lock);
+}
+
+/**
+ * devfreq_interval_update() - Update device devfreq monitoring interval
+ * @devfreq:    the devfreq instance.
+ * @delay:      new polling interval to be set.
+ *
+ * Helper function to set new load monitoring polling interval. Function
+ * to be called from governor in response to DEVFREQ_GOV_INTERVAL event.
+ */
+void devfreq_interval_update(struct devfreq *devfreq, unsigned int *delay)
+{
+       unsigned int cur_delay = devfreq->profile->polling_ms;
+       unsigned int new_delay = *delay;
+
+       mutex_lock(&devfreq->lock);
+       devfreq->profile->polling_ms = new_delay;
+
+       if (devfreq->stop_polling)
+               goto out;
+
+       /* if new delay is zero, stop polling */
+       if (!new_delay) {
+               mutex_unlock(&devfreq->lock);
+               cancel_delayed_work_sync(&devfreq->work);
+               return;
+       }
+
+       /* if current delay is zero, start polling with new delay */
+       if (!cur_delay) {
+               queue_delayed_work(devfreq_wq, &devfreq->work,
+                       msecs_to_jiffies(devfreq->profile->polling_ms));
+               goto out;
+       }
+
+       /* if current delay is greater than new delay, restart polling */
+       if (cur_delay > new_delay) {
+               mutex_unlock(&devfreq->lock);
+               cancel_delayed_work_sync(&devfreq->work);
+               mutex_lock(&devfreq->lock);
+               if (!devfreq->stop_polling)
+                       queue_delayed_work(devfreq_wq, &devfreq->work,
+                             msecs_to_jiffies(devfreq->profile->polling_ms));
+       }
+out:
+       mutex_unlock(&devfreq->lock);
+}
+
+/**
  * devfreq_notifier_call() - Notify that the device frequency requirements
  *                        has been changed out of devfreq framework.
  * @nb         the notifier_block (supposed to be devfreq->nb)
@@ -143,59 +285,32 @@ static int devfreq_notifier_call(struct notifier_block *nb, unsigned long type,
 }
 
 /**
- * _remove_devfreq() - Remove devfreq from the device.
+ * _remove_devfreq() - Remove devfreq from the list and release its resources.
  * @devfreq:   the devfreq struct
  * @skip:      skip calling device_unregister().
- *
- * Note that the caller should lock devfreq->lock before calling
- * this. _remove_devfreq() will unlock it and free devfreq
- * internally. devfreq_list_lock should be locked by the caller
- * as well (not relased at return)
- *
- * Lock usage:
- * devfreq->lock: locked before call.
- *               unlocked at return (and freed)
- * devfreq_list_lock: locked before call.
- *                   kept locked at return.
- *                   if devfreq is centrally polled.
- *
- * Freed memory:
- * devfreq
  */
 static void _remove_devfreq(struct devfreq *devfreq, bool skip)
 {
-       if (!mutex_is_locked(&devfreq->lock)) {
-               WARN(true, "devfreq->lock must be locked by the caller.\n");
-               return;
-       }
-       if (!devfreq->governor->no_central_polling &&
-           !mutex_is_locked(&devfreq_list_lock)) {
-               WARN(true, "devfreq_list_lock must be locked by the caller.\n");
+       mutex_lock(&devfreq_list_lock);
+       if (IS_ERR(find_device_devfreq(devfreq->dev.parent))) {
+               mutex_unlock(&devfreq_list_lock);
+               dev_warn(&devfreq->dev, "releasing devfreq which doesn't exist\n");
                return;
        }
+       list_del(&devfreq->node);
+       mutex_unlock(&devfreq_list_lock);
 
-       if (devfreq->being_removed)
-               return;
-
-       devfreq->being_removed = true;
+       devfreq->governor->event_handler(devfreq, DEVFREQ_GOV_STOP, NULL);
 
        if (devfreq->profile->exit)
                devfreq->profile->exit(devfreq->dev.parent);
 
-       if (devfreq->governor->exit)
-               devfreq->governor->exit(devfreq);
-
        if (!skip && get_device(&devfreq->dev)) {
                device_unregister(&devfreq->dev);
                put_device(&devfreq->dev);
        }
 
-       if (!devfreq->governor->no_central_polling)
-               list_del(&devfreq->node);
-
-       mutex_unlock(&devfreq->lock);
        mutex_destroy(&devfreq->lock);
-
        kfree(devfreq);
 }
 
@@ -210,130 +325,8 @@ static void _remove_devfreq(struct devfreq *devfreq, bool skip)
 static void devfreq_dev_release(struct device *dev)
 {
        struct devfreq *devfreq = to_devfreq(dev);
-       bool central_polling = !devfreq->governor->no_central_polling;
-
-       /*
-        * If devfreq_dev_release() was called by device_unregister() of
-        * _remove_devfreq(), we cannot mutex_lock(&devfreq->lock) and
-        * being_removed is already set. This also partially checks the case
-        * where devfreq_dev_release() is called from a thread other than
-        * the one called _remove_devfreq(); however, this case is
-        * dealt completely with another following being_removed check.
-        *
-        * Because being_removed is never being
-        * unset, we do not need to worry about race conditions on
-        * being_removed.
-        */
-       if (devfreq->being_removed)
-               return;
 
-       if (central_polling)
-               mutex_lock(&devfreq_list_lock);
-
-       mutex_lock(&devfreq->lock);
-
-       /*
-        * Check being_removed flag again for the case where
-        * devfreq_dev_release() was called in a thread other than the one
-        * possibly called _remove_devfreq().
-        */
-       if (devfreq->being_removed) {
-               mutex_unlock(&devfreq->lock);
-               goto out;
-       }
-
-       /* devfreq->lock is unlocked and removed in _removed_devfreq() */
        _remove_devfreq(devfreq, true);
-
-out:
-       if (central_polling)
-               mutex_unlock(&devfreq_list_lock);
-}
-
-/**
- * devfreq_monitor() - Periodically poll devfreq objects.
- * @work: the work struct used to run devfreq_monitor periodically.
- *
- */
-static void devfreq_monitor(struct work_struct *work)
-{
-       static unsigned long last_polled_at;
-       struct devfreq *devfreq, *tmp;
-       int error;
-       unsigned long jiffies_passed;
-       unsigned long next_jiffies = ULONG_MAX, now = jiffies;
-       struct device *dev;
-
-       /* Initially last_polled_at = 0, polling every device at bootup */
-       jiffies_passed = now - last_polled_at;
-       last_polled_at = now;
-       if (jiffies_passed == 0)
-               jiffies_passed = 1;
-
-       mutex_lock(&devfreq_list_lock);
-       list_for_each_entry_safe(devfreq, tmp, &devfreq_list, node) {
-               mutex_lock(&devfreq->lock);
-               dev = devfreq->dev.parent;
-
-               /* Do not remove tmp for a while */
-               wait_remove_device = tmp;
-
-               if (devfreq->governor->no_central_polling ||
-                   devfreq->next_polling == 0) {
-                       mutex_unlock(&devfreq->lock);
-                       continue;
-               }
-               mutex_unlock(&devfreq_list_lock);
-
-               /*
-                * Reduce more next_polling if devfreq_wq took an extra
-                * delay. (i.e., CPU has been idled.)
-                */
-               if (devfreq->next_polling <= jiffies_passed) {
-                       error = update_devfreq(devfreq);
-
-                       /* Remove a devfreq with an error. */
-                       if (error && error != -EAGAIN) {
-
-                               dev_err(dev, "Due to update_devfreq error(%d), devfreq(%s) is removed from the device\n",
-                                       error, devfreq->governor->name);
-
-                               /*
-                                * Unlock devfreq before locking the list
-                                * in order to avoid deadlock with
-                                * find_device_devfreq or others
-                                */
-                               mutex_unlock(&devfreq->lock);
-                               mutex_lock(&devfreq_list_lock);
-                               /* Check if devfreq is already removed */
-                               if (IS_ERR(find_device_devfreq(dev)))
-                                       continue;
-                               mutex_lock(&devfreq->lock);
-                               /* This unlocks devfreq->lock and free it */
-                               _remove_devfreq(devfreq, false);
-                               continue;
-                       }
-                       devfreq->next_polling = devfreq->polling_jiffies;
-               } else {
-                       devfreq->next_polling -= jiffies_passed;
-               }
-
-               if (devfreq->next_polling)
-                       next_jiffies = (next_jiffies > devfreq->next_polling) ?
-                                       devfreq->next_polling : next_jiffies;
-
-               mutex_unlock(&devfreq->lock);
-               mutex_lock(&devfreq_list_lock);
-       }
-       wait_remove_device = NULL;
-       mutex_unlock(&devfreq_list_lock);
-
-       if (next_jiffies > 0 && next_jiffies < ULONG_MAX) {
-               polling = true;
-               queue_delayed_work(devfreq_wq, &devfreq_work, next_jiffies);
-       } else {
-               polling = false;
-       }
 }
 
 /**
@@ -357,16 +350,13 @@ struct devfreq *devfreq_add_device(struct device *dev,
                return ERR_PTR(-EINVAL);
        }
 
-
-       if (!governor->no_central_polling) {
-               mutex_lock(&devfreq_list_lock);
-               devfreq = find_device_devfreq(dev);
-               mutex_unlock(&devfreq_list_lock);
-               if (!IS_ERR(devfreq)) {
-                       dev_err(dev, "%s: Unable to create devfreq for the device. It already has one.\n", __func__);
-                       err = -EINVAL;
-                       goto err_out;
-               }
+       mutex_lock(&devfreq_list_lock);
+       devfreq = find_device_devfreq(dev);
+       mutex_unlock(&devfreq_list_lock);
+       if (!IS_ERR(devfreq)) {
+               dev_err(dev, "%s: Unable to create devfreq for the device. It already has one.\n", __func__);
+               err = -EINVAL;
+               goto err_out;
        }
 
        devfreq = kzalloc(sizeof(struct devfreq), GFP_KERNEL);
@@ -386,48 +376,41 @@ struct devfreq *devfreq_add_device(struct device *dev,
        devfreq->governor = governor;
        devfreq->previous_freq = profile->initial_freq;
        devfreq->data = data;
-       devfreq->next_polling = devfreq->polling_jiffies
-                             = msecs_to_jiffies(devfreq->profile->polling_ms);
        devfreq->nb.notifier_call = devfreq_notifier_call;
 
        dev_set_name(&devfreq->dev, dev_name(dev));
        err = device_register(&devfreq->dev);
        if (err) {
                put_device(&devfreq->dev);
+               mutex_unlock(&devfreq->lock);
                goto err_dev;
        }
 
-       if (governor->init)
-               err = governor->init(devfreq);
-       if (err)
-               goto err_init;
-
        mutex_unlock(&devfreq->lock);
 
-       if (governor->no_central_polling)
-               goto out;
-
        mutex_lock(&devfreq_list_lock);
-
        list_add(&devfreq->node, &devfreq_list);
+       mutex_unlock(&devfreq_list_lock);
 
-       if (devfreq_wq && devfreq->next_polling && !polling) {
-               polling = true;
-               queue_delayed_work(devfreq_wq, &devfreq_work,
-                                  devfreq->next_polling);
+       err = devfreq->governor->event_handler(devfreq,
+                               DEVFREQ_GOV_START, NULL);
+       if (err) {
+               dev_err(dev, "%s: Unable to start governor for the device\n",
+                       __func__);
+               goto err_init;
        }
-       mutex_unlock(&devfreq_list_lock);
-out:
+
        return devfreq;
 
 err_init:
+       list_del(&devfreq->node);
        device_unregister(&devfreq->dev);
 err_dev:
-       mutex_unlock(&devfreq->lock);
        kfree(devfreq);
 err_out:
        return ERR_PTR(err);
 }
+EXPORT_SYMBOL(devfreq_add_device);
 
 /**
  * devfreq_remove_device() - Remove devfreq feature from a device.
@@ -435,30 +418,14 @@ err_out:
  */
 int devfreq_remove_device(struct devfreq *devfreq)
 {
-       bool central_polling;
-
        if (!devfreq)
                return -EINVAL;
 
-       central_polling = !devfreq->governor->no_central_polling;
-
-       if (central_polling) {
-               mutex_lock(&devfreq_list_lock);
-               while (wait_remove_device == devfreq) {
-                       mutex_unlock(&devfreq_list_lock);
-                       schedule();
-                       mutex_lock(&devfreq_list_lock);
-               }
-       }
-
-       mutex_lock(&devfreq->lock);
-       _remove_devfreq(devfreq, false); /* it unlocks devfreq->lock */
-
-       if (central_polling)
-               mutex_unlock(&devfreq_list_lock);
+       _remove_devfreq(devfreq, false);
 
        return 0;
 }
+EXPORT_SYMBOL(devfreq_remove_device);
 
 static ssize_t show_governor(struct device *dev,
                             struct device_attribute *attr, char *buf)
@@ -490,35 +457,13 @@ static ssize_t store_polling_interval(struct device *dev,
        if (ret != 1)
                goto out;
 
-       mutex_lock(&df->lock);
-       df->profile->polling_ms = value;
-       df->next_polling = df->polling_jiffies
-                        = msecs_to_jiffies(value);
-       mutex_unlock(&df->lock);
-
+       df->governor->event_handler(df, DEVFREQ_GOV_INTERVAL, &value);
        ret = count;
 
-       if (df->governor->no_central_polling)
-               goto out;
-
-       mutex_lock(&devfreq_list_lock);
-       if (df->next_polling > 0 && !polling) {
-               polling = true;
-               queue_delayed_work(devfreq_wq, &devfreq_work,
-                                  df->next_polling);
-       }
-       mutex_unlock(&devfreq_list_lock);
 out:
        return ret;
 }
 
-static ssize_t show_central_polling(struct device *dev,
-                                   struct device_attribute *attr, char *buf)
-{
-       return sprintf(buf, "%d\n",
-                      !to_devfreq(dev)->governor->no_central_polling);
-}
-
 static ssize_t store_min_freq(struct device *dev, struct device_attribute *attr,
                              const char *buf, size_t count)
 {
@@ -590,7 +535,6 @@ static ssize_t show_max_freq(struct device *dev, struct device_attribute *attr,
 static struct device_attribute devfreq_attrs[] = {
        __ATTR(governor, S_IRUGO, show_governor, NULL),
        __ATTR(cur_freq, S_IRUGO, show_freq, NULL),
-       __ATTR(central_polling, S_IRUGO, show_central_polling, NULL),
        __ATTR(polling_interval, S_IRUGO | S_IWUSR, show_polling_interval,
               store_polling_interval),
        __ATTR(min_freq, S_IRUGO | S_IWUSR, show_min_freq, store_min_freq),
@@ -598,23 +542,6 @@ static struct device_attribute devfreq_attrs[] = {
        { },
 };
 
-/**
- * devfreq_start_polling() - Initialize data structure for devfreq framework and
- *                        start polling registered devfreq devices.
- */
-static int __init devfreq_start_polling(void)
-{
-       mutex_lock(&devfreq_list_lock);
-       polling = false;
-       devfreq_wq = create_freezable_workqueue("devfreq_wq");
-       INIT_DEFERRABLE_WORK(&devfreq_work, devfreq_monitor);
-       mutex_unlock(&devfreq_list_lock);
-
-       devfreq_monitor(&devfreq_work.work);
-       return 0;
-}
-late_initcall(devfreq_start_polling);
-
 static int __init devfreq_init(void)
 {
        devfreq_class = class_create(THIS_MODULE, "devfreq");
@@ -622,7 +549,15 @@ static int __init devfreq_init(void)
                pr_err("%s: couldn't create class\n", __FILE__);
                return PTR_ERR(devfreq_class);
        }
+
+       devfreq_wq = create_freezable_workqueue("devfreq_wq");
+       if (IS_ERR(devfreq_wq)) {
+               class_destroy(devfreq_class);
+               pr_err("%s: couldn't create workqueue\n", __FILE__);
+               return PTR_ERR(devfreq_wq);
+       }
        devfreq_class->dev_attrs = devfreq_attrs;
+
        return 0;
 }
 subsys_initcall(devfreq_init);
@@ -630,6 +565,7 @@ subsys_initcall(devfreq_init);
 static void __exit devfreq_exit(void)
 {
        class_destroy(devfreq_class);
+       destroy_workqueue(devfreq_wq);
 }
 module_exit(devfreq_exit);
 
index ea7f13c..bb3aff3 100644 (file)
 
 #define to_devfreq(DEV)        container_of((DEV), struct devfreq, dev)
 
+/* Devfreq events */
+#define DEVFREQ_GOV_START                      0x1
+#define DEVFREQ_GOV_STOP                       0x2
+#define DEVFREQ_GOV_INTERVAL                   0x3
+
 /* Caution: devfreq->lock must be locked before calling update_devfreq */
 extern int update_devfreq(struct devfreq *devfreq);
 
+extern void devfreq_monitor_start(struct devfreq *devfreq);
+extern void devfreq_monitor_stop(struct devfreq *devfreq);
+extern void devfreq_monitor_suspend(struct devfreq *devfreq);
+extern void devfreq_monitor_resume(struct devfreq *devfreq);
+extern void devfreq_interval_update(struct devfreq *devfreq,
+                                       unsigned int *delay);
 #endif /* _GOVERNOR_H */
index af75ddd..eea3f9b 100644 (file)
@@ -26,14 +26,22 @@ static int devfreq_performance_func(struct devfreq *df,
        return 0;
 }
 
-static int performance_init(struct devfreq *devfreq)
+static int devfreq_performance_handler(struct devfreq *devfreq,
+                               unsigned int event, void *data)
 {
-       return update_devfreq(devfreq);
+       int ret = 0;
+
+       if (event == DEVFREQ_GOV_START) {
+               mutex_lock(&devfreq->lock);
+               ret = update_devfreq(devfreq);
+               mutex_unlock(&devfreq->lock);
+       }
+
+       return ret;
 }
 
 const struct devfreq_governor devfreq_performance = {
        .name = "performance",
-       .init = performance_init,
        .get_target_freq = devfreq_performance_func,
-       .no_central_polling = true,
+       .event_handler = devfreq_performance_handler,
 };
index fec0cdb..2868d98 100644 (file)
@@ -23,14 +23,22 @@ static int devfreq_powersave_func(struct devfreq *df,
        return 0;
 }
 
-static int powersave_init(struct devfreq *devfreq)
+static int devfreq_powersave_handler(struct devfreq *devfreq,
+                               unsigned int event, void *data)
 {
-       return update_devfreq(devfreq);
+       int ret = 0;
+
+       if (event == DEVFREQ_GOV_START) {
+               mutex_lock(&devfreq->lock);
+               ret = update_devfreq(devfreq);
+               mutex_unlock(&devfreq->lock);
+       }
+
+       return ret;
 }
 
 const struct devfreq_governor devfreq_powersave = {
        .name = "powersave",
-       .init = powersave_init,
        .get_target_freq = devfreq_powersave_func,
-       .no_central_polling = true,
+       .event_handler = devfreq_powersave_handler,
 };
index a2e3eae..3716a65 100644 (file)
@@ -12,6 +12,7 @@
 #include <linux/errno.h>
 #include <linux/devfreq.h>
 #include <linux/math64.h>
+#include "governor.h"
 
 /* Default constants for DevFreq-Simple-Ondemand (DFSO) */
 #define DFSO_UPTHRESHOLD       (90)
@@ -88,7 +89,30 @@ static int devfreq_simple_ondemand_func(struct devfreq *df,
        return 0;
 }
 
+static int devfreq_simple_ondemand_handler(struct devfreq *devfreq,
+                               unsigned int event, void *data)
+{
+       switch (event) {
+       case DEVFREQ_GOV_START:
+               devfreq_monitor_start(devfreq);
+               break;
+
+       case DEVFREQ_GOV_STOP:
+               devfreq_monitor_stop(devfreq);
+               break;
+
+       case DEVFREQ_GOV_INTERVAL:
+               devfreq_interval_update(devfreq, (unsigned int *)data);
+               break;
+       default:
+               break;
+       }
+
+       return 0;
+}
+
 const struct devfreq_governor devfreq_simple_ondemand = {
        .name = "simple_ondemand",
        .get_target_freq = devfreq_simple_ondemand_func,
+       .event_handler = devfreq_simple_ondemand_handler,
 };
index 0681246..7067555 100644 (file)
@@ -116,10 +116,27 @@ static void userspace_exit(struct devfreq *devfreq)
        devfreq->data = NULL;
 }
 
+static int devfreq_userspace_handler(struct devfreq *devfreq,
+                       unsigned int event, void *data)
+{
+       int ret = 0;
+
+       switch (event) {
+       case DEVFREQ_GOV_START:
+               ret = userspace_init(devfreq);
+               break;
+       case DEVFREQ_GOV_STOP:
+               userspace_exit(devfreq);
+               break;
+       default:
+               break;
+       }
+
+       return ret;
+}
+
 const struct devfreq_governor devfreq_userspace = {
        .name = "userspace",
        .get_target_freq = devfreq_userspace_func,
-       .init = userspace_init,
-       .exit = userspace_exit,
-       .no_central_polling = true,
+       .event_handler = devfreq_userspace_handler,
 };
index 281c72a..9cdffde 100644 (file)
@@ -91,25 +91,18 @@ struct devfreq_dev_profile {
  *                     status of the device (load = busy_time / total_time).
  *                     If no_central_polling is set, this callback is called
  *                     only with update_devfreq() notified by OPP.
- * @init               Called when the devfreq is being attached to a device
- * @exit               Called when the devfreq is being removed from a
- *                     device. Governor should stop any internal routines
- *                     before return because related data may be
- *                     freed after exit().
- * @no_central_polling Do not use devfreq's central polling mechanism.
- *                     When this is set, devfreq will not call
- *                     get_target_freq with devfreq_monitor(). However,
- *                     devfreq will call get_target_freq with
- *                     devfreq_update() notified by OPP framework.
+ * @event_handler       Callback for devfreq core framework to notify events
+ *                      to governors. Events include per device governor
+ *                      init and exit, opp changes out of devfreq, suspend
+ *                      and resume of per device devfreq during device idle.
  *
  * Note that the callbacks are called with devfreq->lock locked by devfreq.
  */
 struct devfreq_governor {
        const char name[DEVFREQ_NAME_LEN];
        int (*get_target_freq)(struct devfreq *this, unsigned long *freq);
-       int (*init)(struct devfreq *this);
-       void (*exit)(struct devfreq *this);
-       const bool no_central_polling;
+       int (*event_handler)(struct devfreq *devfreq,
+                               unsigned int event, void *data);
 };
 
 /**
@@ -124,18 +117,13 @@ struct devfreq_governor {
  * @nb         notifier block used to notify devfreq object that it should
  *             reevaluate operable frequencies. Devfreq users may use
  *             devfreq.nb to the corresponding register notifier call chain.
- * @polling_jiffies    interval in jiffies.
+ * @work       delayed work for load monitoring.
  * @previous_freq      previously configured frequency value.
- * @next_polling       the number of remaining jiffies to poll with
- *                     "devfreq_monitor" executions to reevaluate
- *                     frequency/voltage of the device. Set by
- *                     profile's polling_ms interval.
  * @data       Private data of the governor. The devfreq framework does not
  *             touch this.
- * @being_removed      a flag to mark that this object is being removed in
- *                     order to prevent trying to remove the object multiple times.
  * @min_freq   Limit minimum frequency requested by user (0: none)
  * @max_freq   Limit maximum frequency requested by user (0: none)
+ * @stop_polling        devfreq polling status of a device.
  *
  * This structure stores the devfreq information for a give device.
  *
@@ -153,17 +141,15 @@ struct devfreq {
        struct devfreq_dev_profile *profile;
        const struct devfreq_governor *governor;
        struct notifier_block nb;
+       struct delayed_work work;
 
-       unsigned long polling_jiffies;
        unsigned long previous_freq;
-       unsigned int next_polling;
 
        void *data; /* private data for governors */
 
-       bool being_removed;
-
        unsigned long min_freq;
        unsigned long max_freq;
+       bool stop_polling;
 };
 
 #if defined(CONFIG_PM_DEVFREQ)