]> git.karo-electronics.de Git - linux-beck.git/commitdiff
thermal: fix race condition when updating cooling device
authorMichele Di Giorgio <michele.digiorgio@arm.com>
Thu, 2 Jun 2016 14:25:31 +0000 (15:25 +0100)
committerZhang Rui <rui.zhang@intel.com>
Mon, 8 Aug 2016 02:57:39 +0000 (10:57 +0800)
When multiple thermal zones are bound to the same cooling device, multiple
kernel threads may want to update the cooling device state by calling
thermal_cdev_update(). Having cdev not protected by a mutex can lead to a race
condition. Consider the following situation with two kernel threads k1 and k2:

    Thread k1 Thread k2
                                    ||
                                    ||  call thermal_cdev_update()
                                    ||      ...
                                    ||      set_cur_state(cdev, target);
    call power_actor_set_power()    ||
        ...                         ||
        instance->target = state;   ||
        cdev->updated = false;      ||
                                    ||      cdev->updated = true;
                                    ||      // completes execution
    call thermal_cdev_update()      ||
        // cdev->updated == true    ||
        return;                     ||
                                    \/
                                    time

k2 has already looped through the thermal instances looking for the deepest
cooling device state and is preempted right before setting cdev->updated to
true. Now, k1 runs, modifies the thermal instance state and sets cdev->updated
to false. Then, k1 is preempted and k2 continues the execution by setting
cdev->updated to true, therefore preventing k1 from performing the update.
Notice that this is not an issue if k2 looks at the instance->target modified by
k1 "after" it is assigned by k1. In fact, in this case the update will happen
anyway and k1 can safely return immediately from thermal_cdev_update().

This may lead to a situation where a thermal governor never updates the cooling
device. For example, this is the case for the step_wise governor: when calling
the function thermal_zone_trip_update(), the governor may always get a new state
equal to the old one (which, however, wasn't notified to the cooling device) and
will therefore skip the update.

CC: Zhang Rui <rui.zhang@intel.com>
CC: Eduardo Valentin <edubezval@gmail.com>
CC: Peter Feuerer <peter@piie.net>
Reported-by: Toby Huang <toby.huang@arm.com>
Signed-off-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Reviewed-by: Javi Merino <javi.merino@arm.com>
Signed-off-by: Zhang Rui <rui.zhang@intel.com>
drivers/thermal/fair_share.c
drivers/thermal/gov_bang_bang.c
drivers/thermal/power_allocator.c
drivers/thermal/step_wise.c
drivers/thermal/thermal_core.c

index 34fe36504a552cdaf112a6982483655f25dc2238..68bd1b56911850ee41dd7c508c7cad7f7c278ddf 100644 (file)
@@ -116,7 +116,9 @@ static int fair_share_throttle(struct thermal_zone_device *tz, int trip)
                instance->target = get_target_state(tz, cdev, percentage,
                                                    cur_trip_level);
 
+               mutex_lock(&instance->cdev->lock);
                instance->cdev->updated = false;
+               mutex_unlock(&instance->cdev->lock);
                thermal_cdev_update(cdev);
        }
        return 0;
index fc52016d4e85be590e15f4e1976b8378d2c456d2..bb118a152cbbde3c570f90fb5355bc04afe54beb 100644 (file)
@@ -71,7 +71,9 @@ static void thermal_zone_trip_update(struct thermal_zone_device *tz, int trip)
                dev_dbg(&instance->cdev->device, "target=%d\n",
                                        (int)instance->target);
 
+               mutex_lock(&instance->cdev->lock);
                instance->cdev->updated = false; /* cdev needs update */
+               mutex_unlock(&instance->cdev->lock);
        }
 
        mutex_unlock(&tz->lock);
index 2f1a863a8e15bc834e007fea264d262d180a04cf..b4d3116cfdafe81767b2b1c91fcab4a034f29041 100644 (file)
@@ -529,7 +529,9 @@ static void allow_maximum_power(struct thermal_zone_device *tz)
                        continue;
 
                instance->target = 0;
+               mutex_lock(&instance->cdev->lock);
                instance->cdev->updated = false;
+               mutex_unlock(&instance->cdev->lock);
                thermal_cdev_update(instance->cdev);
        }
 }
index ea9366ad3e6bb285e52e368691a0d495cbb3429f..bcef2e7c4ec96f1cfc662019ccf0440d52d26328 100644 (file)
@@ -175,7 +175,9 @@ static void thermal_zone_trip_update(struct thermal_zone_device *tz, int trip)
                        update_passive_instance(tz, trip_type, -1);
 
                instance->initialized = true;
+               mutex_lock(&instance->cdev->lock);
                instance->cdev->updated = false; /* cdev needs update */
+               mutex_unlock(&instance->cdev->lock);
        }
 
        mutex_unlock(&tz->lock);
index 5133cd1e10b7ae99d838823af2eef272aa00ab76..e2fc6161dded9650c300cf829a381c0a2d14b38c 100644 (file)
@@ -1093,7 +1093,9 @@ int power_actor_set_power(struct thermal_cooling_device *cdev,
                return ret;
 
        instance->target = state;
+       mutex_lock(&cdev->lock);
        cdev->updated = false;
+       mutex_unlock(&cdev->lock);
        thermal_cdev_update(cdev);
 
        return 0;
@@ -1623,11 +1625,13 @@ void thermal_cdev_update(struct thermal_cooling_device *cdev)
        struct thermal_instance *instance;
        unsigned long target = 0;
 
+       mutex_lock(&cdev->lock);
        /* cooling device is updated*/
-       if (cdev->updated)
+       if (cdev->updated) {
+               mutex_unlock(&cdev->lock);
                return;
+       }
 
-       mutex_lock(&cdev->lock);
        /* Make sure cdev enters the deepest cooling state */
        list_for_each_entry(instance, &cdev->thermal_instances, cdev_node) {
                dev_dbg(&cdev->device, "zone%d->target=%lu\n",
@@ -1637,9 +1641,9 @@ void thermal_cdev_update(struct thermal_cooling_device *cdev)
                if (instance->target > target)
                        target = instance->target;
        }
-       mutex_unlock(&cdev->lock);
        cdev->ops->set_cur_state(cdev, target);
        cdev->updated = true;
+       mutex_unlock(&cdev->lock);
        trace_cdev_update(cdev, target);
        dev_dbg(&cdev->device, "set to state %lu\n", target);
 }