thermal: core: Fix race between zone registration and system suspend

If the registration of a thermal zone takes place at the time when
system suspend is started, thermal_pm_notify() can run before the new
thermal zone is added to thermal_tz_list and its "suspended" flag will
not be set.  Consequently, if __thermal_zone_device_update() is called
for that thermal zone, it will not return early as expected which may
cause some destructive interference with the system suspend or resume
flow to occur.

To avoid that, make thermal_zone_init_complete() introduced previously
set the "suspended" flag for new thermal zones if it runs during system
suspend or resume.

Fixes: 4e81417 ("thermal: core: Fix thermal zone suspend-resume synchronization")
Signed-off-by: Rafael J. Wysocki <[email protected]>
Link: https://patch.msgid.link/[email protected]
Reviewed-by: Lukasz Luba <[email protected]>

Author: rafaeljw

drivers/thermal/thermal_core.c

@@ -40,6 +40,8 @@ static DEFINE_MUTEX(thermal_governor_lock);
 
 static struct thermal_governor *def_governor;
 
+static bool thermal_pm_suspended;
+
 /*
  * Governor section: set of functions to handle thermal governors
  *
@@ -1339,6 +1341,14 @@ static void thermal_zone_init_complete(struct thermal_zone_device *tz)
 	mutex_lock(&tz->lock);
 
 	tz->state &= ~TZ_STATE_FLAG_INIT;
+	/*
+	 * If system suspend or resume is in progress at this point, the
+	 * new thermal zone needs to be marked as suspended because
+	 * thermal_pm_notify() has run already.
+	 */
+	if (thermal_pm_suspended)
+		tz->state |= TZ_STATE_FLAG_SUSPENDED;
+
 	__thermal_zone_device_update(tz, THERMAL_EVENT_UNSPECIFIED);
 
 	mutex_unlock(&tz->lock);
@@ -1520,10 +1530,10 @@ thermal_zone_device_register_with_trips(const char *type,
 	list_for_each_entry(cdev, &thermal_cdev_list, node)
 		thermal_zone_cdev_bind(tz, cdev);
 
-	mutex_unlock(&thermal_list_lock);
-
 	thermal_zone_init_complete(tz);
 
+	mutex_unlock(&thermal_list_lock);
+
 	thermal_notify_tz_create(tz);
 
 	thermal_debug_tz_add(tz);
@@ -1746,6 +1756,8 @@ static int thermal_pm_notify(struct notifier_block *nb,
 	case PM_SUSPEND_PREPARE:
 		mutex_lock(&thermal_list_lock);
 
+		thermal_pm_suspended = true;
+
 		list_for_each_entry(tz, &thermal_tz_list, node)
 			thermal_zone_pm_prepare(tz);
 
@@ -1756,6 +1768,8 @@ static int thermal_pm_notify(struct notifier_block *nb,
 	case PM_POST_SUSPEND:
 		mutex_lock(&thermal_list_lock);
 
+		thermal_pm_suspended = false;
+
 		list_for_each_entry(tz, &thermal_tz_list, node)
 			thermal_zone_pm_complete(tz);