From: Rafael J. Wysocki <[email protected]>
After commit 5a5efdaffda5 ("thermal: core: Resume thermal zones
asynchronously") it is theoretically possible that, if a system suspend
starts immediately after a system resume, thermal_zone_device_resume()
spawned by the thermal PM notifier for one of the thermal zones at the
end of the system resume will run after the PM thermal notifier for the
suspend-prepare action. If that happens, tz->suspended set by the latter
will be reset by the former which may lead to unexpected consequences.
To avoid that race, synchronize thermal_zone_device_resume() with the
suspend-prepare thermal PM notifier with the help of additional bool
field and completion in struct thermal_zone_device.
Note that this also ensures running __thermal_zone_device_update() at
least once for each thermal zone between system resume and the following
system suspend in case it is needed to start thermal mitigation.
Fixes: 5a5efdaffda5 ("thermal: core: Resume thermal zones asynchronously")
Signed-off-by: Rafael J. Wysocki <[email protected]>
---
drivers/thermal/thermal_core.c | 21 +++++++++++++++++++++
drivers/thermal/thermal_core.h | 4 ++++
2 files changed, 25 insertions(+)
Index: linux-pm/drivers/thermal/thermal_core.c
===================================================================
--- linux-pm.orig/drivers/thermal/thermal_core.c
+++ linux-pm/drivers/thermal/thermal_core.c
@@ -1406,6 +1406,7 @@ thermal_zone_device_register_with_trips(
ida_init(&tz->ida);
mutex_init(&tz->lock);
init_completion(&tz->removal);
+ init_completion(&tz->resume);
id = ida_alloc(&thermal_tz_ida, GFP_KERNEL);
if (id < 0) {
result = id;
@@ -1652,6 +1653,9 @@ static void thermal_zone_device_resume(s
thermal_zone_device_init(tz);
__thermal_zone_device_update(tz, THERMAL_EVENT_UNSPECIFIED);
+ complete(&tz->resume);
+ tz->resuming = false;
+
mutex_unlock(&tz->lock);
}
@@ -1669,6 +1673,20 @@ static int thermal_pm_notify(struct noti
list_for_each_entry(tz, &thermal_tz_list, node) {
mutex_lock(&tz->lock);
+ if (tz->resuming) {
+ /*
+ * thermal_zone_device_resume() queued up for
+ * this zone has not acquired the lock yet, so
+ * release it to let the function run and wait
+ * util it has done the work.
+ */
+ mutex_unlock(&tz->lock);
+
+ wait_for_completion(&tz->resume);
+
+ mutex_lock(&tz->lock);
+ }
+
tz->suspended = true;
mutex_unlock(&tz->lock);
@@ -1686,6 +1704,9 @@ static int thermal_pm_notify(struct noti
cancel_delayed_work(&tz->poll_queue);
+ reinit_completion(&tz->resume);
+ tz->resuming = true;
+
/*
* Replace the work function with the resume one, which
* will restore the original work function and schedule
Index: linux-pm/drivers/thermal/thermal_core.h
===================================================================
--- linux-pm.orig/drivers/thermal/thermal_core.h
+++ linux-pm/drivers/thermal/thermal_core.h
@@ -55,6 +55,7 @@ struct thermal_governor {
* @type: the thermal zone device type
* @device: &struct device for this thermal zone
* @removal: removal completion
+ * @resume: resume completion
* @trip_temp_attrs: attributes for trip points for sysfs: trip temperature
* @trip_type_attrs: attributes for trip points for sysfs: trip type
* @trip_hyst_attrs: attributes for trip points for sysfs: trip hysteresis
@@ -89,6 +90,7 @@ struct thermal_governor {
* @poll_queue: delayed work for polling
* @notify_event: Last notification event
* @suspended: thermal zone suspend indicator
+ * @resuming: indicates whether or not thermal zone resume is in progress
* @trips: array of struct thermal_trip objects
*/
struct thermal_zone_device {
@@ -96,6 +98,7 @@ struct thermal_zone_device {
char type[THERMAL_NAME_LENGTH];
struct device device;
struct completion removal;
+ struct completion resume;
struct attribute_group trips_attribute_group;
struct thermal_attr *trip_temp_attrs;
struct thermal_attr *trip_type_attrs;
@@ -123,6 +126,7 @@ struct thermal_zone_device {
struct delayed_work poll_queue;
enum thermal_notify_event notify_event;
bool suspended;
+ bool resuming;
#ifdef CONFIG_THERMAL_DEBUGFS
struct thermal_debugfs *debugfs;
#endif