1 files changed, 298 insertions, 67 deletions

diff --git a/drivers/base/power/main.c b/drivers/base/power/main.c
index c8b0a9e29ed8..dbf5456cd891 100644
--- a/drivers/base/power/main.c
+++ b/drivers/base/power/main.c
@@ -63,8 +63,23 @@ static LIST_HEAD(dpm_noirq_list);
 static DEFINE_MUTEX(dpm_list_mtx);
 static pm_message_t pm_transition;
 
+static DEFINE_MUTEX(async_wip_mtx);
 static int async_error;
 
+/**
+ * pm_hibernate_is_recovering - if recovering from hibernate due to error.
+ *
+ * Used to query if dev_pm_ops.thaw() is called for normal hibernation case or
+ * recovering from some error.
+ *
+ * Return: true for error case, false for normal case.
+ */
+bool pm_hibernate_is_recovering(void)
+{
+	return pm_transition.event == PM_EVENT_RECOVER;
+}
+EXPORT_SYMBOL_GPL(pm_hibernate_is_recovering);
+
 static const char *pm_verb(int event)
 {
 	switch (event) {
@@ -512,7 +527,7 @@ struct dpm_watchdog {
  */
 static void dpm_watchdog_handler(struct timer_list *t)
 {
-	struct dpm_watchdog *wd = from_timer(wd, t, timer);
+	struct dpm_watchdog *wd = timer_container_of(wd, t, timer);
 	struct timer_list *timer = &wd->timer;
 	unsigned int time_left;
 
@@ -560,7 +575,7 @@ static void dpm_watchdog_clear(struct dpm_watchdog *wd)
 	struct timer_list *timer = &wd->timer;
 
 	timer_delete_sync(timer);
-	destroy_timer_on_stack(timer);
+	timer_destroy_on_stack(timer);
 }
 #else
 #define DECLARE_DPM_WATCHDOG_ON_STACK(wd)
@@ -597,8 +612,11 @@ static bool is_async(struct device *dev)
 		&& !pm_trace_is_enabled();
 }
 
-static bool dpm_async_fn(struct device *dev, async_func_t func)
+static bool __dpm_async(struct device *dev, async_func_t func)
 {
+	if (dev->power.work_in_progress)
+		return true;
+
 	if (!is_async(dev))
 		return false;
 
@@ -611,14 +629,57 @@ static bool dpm_async_fn(struct device *dev, async_func_t func)
 
 	put_device(dev);
 
+	return false;
+}
+
+static bool dpm_async_fn(struct device *dev, async_func_t func)
+{
+	guard(mutex)(&async_wip_mtx);
+
+	return __dpm_async(dev, func);
+}
+
+static int dpm_async_with_cleanup(struct device *dev, void *fn)
+{
+	guard(mutex)(&async_wip_mtx);
+
+	if (!__dpm_async(dev, fn))
+		dev->power.work_in_progress = false;
+
+	return 0;
+}
+
+static void dpm_async_resume_children(struct device *dev, async_func_t func)
+{
 	/*
-	 * async_schedule_dev_nocall() above has returned false, so func() is
-	 * not running and it is safe to update power.work_in_progress without
-	 * extra synchronization.
+	 * Prevent racing with dpm_clear_async_state() during initial list
+	 * walks in dpm_noirq_resume_devices(), dpm_resume_early(), and
+	 * dpm_resume().
 	 */
-	dev->power.work_in_progress = false;
+	guard(mutex)(&dpm_list_mtx);
 
-	return false;
+	/*
+	 * Start processing "async" children of the device unless it's been
+	 * started already for them.
+	 */
+	device_for_each_child(dev, func, dpm_async_with_cleanup);
+}
+
+static void dpm_async_resume_subordinate(struct device *dev, async_func_t func)
+{
+	struct device_link *link;
+	int idx;
+
+	dpm_async_resume_children(dev, func);
+
+	idx = device_links_read_lock();
+
+	/* Start processing the device's "async" consumers. */
+	list_for_each_entry_rcu(link, &dev->links.consumers, s_node)
+		if (READ_ONCE(link->status) != DL_STATE_DORMANT)
+			dpm_async_with_cleanup(link->consumer, func);
+
+	device_links_read_unlock(idx);
 }
 
 static void dpm_clear_async_state(struct device *dev)
@@ -627,6 +688,20 @@ static void dpm_clear_async_state(struct device *dev)
 	dev->power.work_in_progress = false;
 }
 
+static bool dpm_root_device(struct device *dev)
+{
+	lockdep_assert_held(&dpm_list_mtx);
+
+	/*
+	 * Since this function is required to run under dpm_list_mtx, the
+	 * list_empty() below will only return true if the device's list of
+	 * consumers is actually empty before calling it.
+	 */
+	return !dev->parent && list_empty(&dev->links.suppliers);
+}
+
+static void async_resume_noirq(void *data, async_cookie_t cookie);
+
 /**
  * device_resume_noirq - Execute a "noirq resume" callback for given device.
  * @dev: Device to handle.
@@ -706,10 +781,12 @@ Out:
 	TRACE_RESUME(error);
 
 	if (error) {
-		async_error = error;
+		WRITE_ONCE(async_error, error);
 		dpm_save_failed_dev(dev_name(dev));
 		pm_dev_err(dev, state, async ? " async noirq" : " noirq", error);
 	}
+
+	dpm_async_resume_subordinate(dev, async_resume_noirq);
 }
 
 static void async_resume_noirq(void *data, async_cookie_t cookie)
@@ -733,19 +810,20 @@ static void dpm_noirq_resume_devices(pm_message_t state)
 	mutex_lock(&dpm_list_mtx);
 
 	/*
-	 * Trigger the resume of "async" devices upfront so they don't have to
-	 * wait for the "non-async" ones they don't depend on.
+	 * Start processing "async" root devices upfront so they don't wait for
+	 * the "sync" devices they don't depend on.
 	 */
 	list_for_each_entry(dev, &dpm_noirq_list, power.entry) {
 		dpm_clear_async_state(dev);
-		dpm_async_fn(dev, async_resume_noirq);
+		if (dpm_root_device(dev))
+			dpm_async_with_cleanup(dev, async_resume_noirq);
 	}
 
 	while (!list_empty(&dpm_noirq_list)) {
 		dev = to_device(dpm_noirq_list.next);
 		list_move_tail(&dev->power.entry, &dpm_late_early_list);
 
-		if (!dev->power.work_in_progress) {
+		if (!dpm_async_fn(dev, async_resume_noirq)) {
 			get_device(dev);
 
 			mutex_unlock(&dpm_list_mtx);
@@ -760,7 +838,7 @@ static void dpm_noirq_resume_devices(pm_message_t state)
 	mutex_unlock(&dpm_list_mtx);
 	async_synchronize_full();
 	dpm_show_time(starttime, state, 0, "noirq");
-	if (async_error)
+	if (READ_ONCE(async_error))
 		dpm_save_failed_step(SUSPEND_RESUME_NOIRQ);
 
 	trace_suspend_resume(TPS("dpm_resume_noirq"), state.event, false);
@@ -781,6 +859,8 @@ void dpm_resume_noirq(pm_message_t state)
 	device_wakeup_disarm_wake_irqs();
 }
 
+static void async_resume_early(void *data, async_cookie_t cookie);
+
 /**
  * device_resume_early - Execute an "early resume" callback for given device.
  * @dev: Device to handle.
@@ -844,10 +924,12 @@ Out:
 	complete_all(&dev->power.completion);
 
 	if (error) {
-		async_error = error;
+		WRITE_ONCE(async_error, error);
 		dpm_save_failed_dev(dev_name(dev));
 		pm_dev_err(dev, state, async ? " async early" : " early", error);
 	}
+
+	dpm_async_resume_subordinate(dev, async_resume_early);
 }
 
 static void async_resume_early(void *data, async_cookie_t cookie)
@@ -875,19 +957,20 @@ void dpm_resume_early(pm_message_t state)
 	mutex_lock(&dpm_list_mtx);
 
 	/*
-	 * Trigger the resume of "async" devices upfront so they don't have to
-	 * wait for the "non-async" ones they don't depend on.
+	 * Start processing "async" root devices upfront so they don't wait for
+	 * the "sync" devices they don't depend on.
 	 */
 	list_for_each_entry(dev, &dpm_late_early_list, power.entry) {
 		dpm_clear_async_state(dev);
-		dpm_async_fn(dev, async_resume_early);
+		if (dpm_root_device(dev))
+			dpm_async_with_cleanup(dev, async_resume_early);
 	}
 
 	while (!list_empty(&dpm_late_early_list)) {
 		dev = to_device(dpm_late_early_list.next);
 		list_move_tail(&dev->power.entry, &dpm_suspended_list);
 
-		if (!dev->power.work_in_progress) {
+		if (!dpm_async_fn(dev, async_resume_early)) {
 			get_device(dev);
 
 			mutex_unlock(&dpm_list_mtx);
@@ -902,7 +985,7 @@ void dpm_resume_early(pm_message_t state)
 	mutex_unlock(&dpm_list_mtx);
 	async_synchronize_full();
 	dpm_show_time(starttime, state, 0, "early");
-	if (async_error)
+	if (READ_ONCE(async_error))
 		dpm_save_failed_step(SUSPEND_RESUME_EARLY);
 
 	trace_suspend_resume(TPS("dpm_resume_early"), state.event, false);
@@ -919,6 +1002,8 @@ void dpm_resume_start(pm_message_t state)
 }
 EXPORT_SYMBOL_GPL(dpm_resume_start);
 
+static void async_resume(void *data, async_cookie_t cookie);
+
 /**
  * device_resume - Execute "resume" callbacks for given device.
  * @dev: Device to handle.
@@ -941,6 +1026,8 @@ static void device_resume(struct device *dev, pm_message_t state, bool async)
 	if (!dev->power.is_suspended)
 		goto Complete;
 
+	dev->power.is_suspended = false;
+
 	if (dev->power.direct_complete) {
 		/*
 		 * Allow new children to be added under the device after this
@@ -1003,7 +1090,6 @@ static void device_resume(struct device *dev, pm_message_t state, bool async)
 
  End:
 	error = dpm_run_callback(callback, dev, state, info);
-	dev->power.is_suspended = false;
 
 	device_unlock(dev);
 	dpm_watchdog_clear(&wd);
@@ -1014,10 +1100,12 @@ static void device_resume(struct device *dev, pm_message_t state, bool async)
 	TRACE_RESUME(error);
 
 	if (error) {
-		async_error = error;
+		WRITE_ONCE(async_error, error);
 		dpm_save_failed_dev(dev_name(dev));
 		pm_dev_err(dev, state, async ? " async" : "", error);
 	}
+
+	dpm_async_resume_subordinate(dev, async_resume);
 }
 
 static void async_resume(void *data, async_cookie_t cookie)
@@ -1041,7 +1129,6 @@ void dpm_resume(pm_message_t state)
 	ktime_t starttime = ktime_get();
 
 	trace_suspend_resume(TPS("dpm_resume"), state.event, true);
-	might_sleep();
 
 	pm_transition = state;
 	async_error = 0;
@@ -1049,19 +1136,20 @@ void dpm_resume(pm_message_t state)
 	mutex_lock(&dpm_list_mtx);
 
 	/*
-	 * Trigger the resume of "async" devices upfront so they don't have to
-	 * wait for the "non-async" ones they don't depend on.
+	 * Start processing "async" root devices upfront so they don't wait for
+	 * the "sync" devices they don't depend on.
 	 */
 	list_for_each_entry(dev, &dpm_suspended_list, power.entry) {
 		dpm_clear_async_state(dev);
-		dpm_async_fn(dev, async_resume);
+		if (dpm_root_device(dev))
+			dpm_async_with_cleanup(dev, async_resume);
 	}
 
 	while (!list_empty(&dpm_suspended_list)) {
 		dev = to_device(dpm_suspended_list.next);
 		list_move_tail(&dev->power.entry, &dpm_prepared_list);
 
-		if (!dev->power.work_in_progress) {
+		if (!dpm_async_fn(dev, async_resume)) {
 			get_device(dev);
 
 			mutex_unlock(&dpm_list_mtx);
@@ -1076,7 +1164,7 @@ void dpm_resume(pm_message_t state)
 	mutex_unlock(&dpm_list_mtx);
 	async_synchronize_full();
 	dpm_show_time(starttime, state, 0, NULL);
-	if (async_error)
+	if (READ_ONCE(async_error))
 		dpm_save_failed_step(SUSPEND_RESUME);
 
 	cpufreq_resume();
@@ -1143,7 +1231,6 @@ void dpm_complete(pm_message_t state)
 	struct list_head list;
 
 	trace_suspend_resume(TPS("dpm_complete"), state.event, true);
-	might_sleep();
 
 	INIT_LIST_HEAD(&list);
 	mutex_lock(&dpm_list_mtx);
@@ -1182,6 +1269,7 @@ void dpm_complete(pm_message_t state)
 void dpm_resume_end(pm_message_t state)
 {
 	dpm_resume(state);
+	pm_restore_gfp_mask();
 	dpm_complete(state);
 }
 EXPORT_SYMBOL_GPL(dpm_resume_end);
@@ -1189,6 +1277,82 @@ EXPORT_SYMBOL_GPL(dpm_resume_end);
 
 /*------------------------- Suspend routines -------------------------*/
 
+static bool dpm_leaf_device(struct device *dev)
+{
+	struct device *child;
+
+	lockdep_assert_held(&dpm_list_mtx);
+
+	child = device_find_any_child(dev);
+	if (child) {
+		put_device(child);
+
+		return false;
+	}
+
+	/*
+	 * Since this function is required to run under dpm_list_mtx, the
+	 * list_empty() below will only return true if the device's list of
+	 * consumers is actually empty before calling it.
+	 */
+	return list_empty(&dev->links.consumers);
+}
+
+static bool dpm_async_suspend_parent(struct device *dev, async_func_t func)
+{
+	guard(mutex)(&dpm_list_mtx);
+
+	/*
+	 * If the device is suspended asynchronously and the parent's callback
+	 * deletes both the device and the parent itself, the parent object may
+	 * be freed while this function is running, so avoid that by checking
+	 * if the device has been deleted already as the parent cannot be
+	 * deleted before it.
+	 */
+	if (!device_pm_initialized(dev))
+		return false;
+
+	/* Start processing the device's parent if it is "async". */
+	if (dev->parent)
+		dpm_async_with_cleanup(dev->parent, func);
+
+	return true;
+}
+
+static void dpm_async_suspend_superior(struct device *dev, async_func_t func)
+{
+	struct device_link *link;
+	int idx;
+
+	if (!dpm_async_suspend_parent(dev, func))
+		return;
+
+	idx = device_links_read_lock();
+
+	/* Start processing the device's "async" suppliers. */
+	list_for_each_entry_rcu(link, &dev->links.suppliers, c_node)
+		if (READ_ONCE(link->status) != DL_STATE_DORMANT)
+			dpm_async_with_cleanup(link->supplier, func);
+
+	device_links_read_unlock(idx);
+}
+
+static void dpm_async_suspend_complete_all(struct list_head *device_list)
+{
+	struct device *dev;
+
+	guard(mutex)(&async_wip_mtx);
+
+	list_for_each_entry_reverse(dev, device_list, power.entry) {
+		/*
+		 * In case the device is being waited for and async processing
+		 * has not started for it yet, let the waiters make progress.
+		 */
+		if (!dev->power.work_in_progress)
+			complete_all(&dev->power.completion);
+	}
+}
+
 /**
  * resume_event - Return a "resume" message for given "suspend" sleep state.
  * @sleep_state: PM message representing a sleep state.
@@ -1226,6 +1390,8 @@ static void dpm_superior_set_must_resume(struct device *dev)
 	device_links_read_unlock(idx);
 }
 
+static void async_suspend_noirq(void *data, async_cookie_t cookie);
+
 /**
  * device_suspend_noirq - Execute a "noirq suspend" callback for given device.
  * @dev: Device to handle.
@@ -1235,7 +1401,7 @@ static void dpm_superior_set_must_resume(struct device *dev)
  * The driver of @dev will not receive interrupts while this function is being
  * executed.
  */
-static int device_suspend_noirq(struct device *dev, pm_message_t state, bool async)
+static void device_suspend_noirq(struct device *dev, pm_message_t state, bool async)
 {
 	pm_callback_t callback = NULL;
 	const char *info = NULL;
@@ -1246,7 +1412,7 @@ static int device_suspend_noirq(struct device *dev, pm_message_t state, bool asy
 
 	dpm_wait_for_subordinate(dev, async);
 
-	if (async_error)
+	if (READ_ONCE(async_error))
 		goto Complete;
 
 	if (dev->power.syscore || dev->power.direct_complete)
@@ -1279,7 +1445,7 @@ static int device_suspend_noirq(struct device *dev, pm_message_t state, bool asy
 Run:
 	error = dpm_run_callback(callback, dev, state, info);
 	if (error) {
-		async_error = error;
+		WRITE_ONCE(async_error, error);
 		dpm_save_failed_dev(dev_name(dev));
 		pm_dev_err(dev, state, async ? " async noirq" : " noirq", error);
 		goto Complete;
@@ -1304,7 +1470,11 @@ Skip:
 Complete:
 	complete_all(&dev->power.completion);
 	TRACE_SUSPEND(error);
-	return error;
+
+	if (error || READ_ONCE(async_error))
+		return;
+
+	dpm_async_suspend_superior(dev, async_suspend_noirq);
 }
 
 static void async_suspend_noirq(void *data, async_cookie_t cookie)
@@ -1318,7 +1488,8 @@ static void async_suspend_noirq(void *data, async_cookie_t cookie)
 static int dpm_noirq_suspend_devices(pm_message_t state)
 {
 	ktime_t starttime = ktime_get();
-	int error = 0;
+	struct device *dev;
+	int error;
 
 	trace_suspend_resume(TPS("dpm_suspend_noirq"), state.event, true);
 
@@ -1327,12 +1498,21 @@ static int dpm_noirq_suspend_devices(pm_message_t state)
 
 	mutex_lock(&dpm_list_mtx);
 
+	/*
+	 * Start processing "async" leaf devices upfront so they don't need to
+	 * wait for the "sync" devices they don't depend on.
+	 */
+	list_for_each_entry_reverse(dev, &dpm_late_early_list, power.entry) {
+		dpm_clear_async_state(dev);
+		if (dpm_leaf_device(dev))
+			dpm_async_with_cleanup(dev, async_suspend_noirq);
+	}
+
 	while (!list_empty(&dpm_late_early_list)) {
-		struct device *dev = to_device(dpm_late_early_list.prev);
+		dev = to_device(dpm_late_early_list.prev);
 
 		list_move(&dev->power.entry, &dpm_noirq_list);
 
-		dpm_clear_async_state(dev);
 		if (dpm_async_fn(dev, async_suspend_noirq))
 			continue;
 
@@ -1340,22 +1520,28 @@ static int dpm_noirq_suspend_devices(pm_message_t state)
 
 		mutex_unlock(&dpm_list_mtx);
 
-		error = device_suspend_noirq(dev, state, false);
+		device_suspend_noirq(dev, state, false);
 
 		put_device(dev);
 
 		mutex_lock(&dpm_list_mtx);
 
-		if (error || async_error)
+		if (READ_ONCE(async_error)) {
+			dpm_async_suspend_complete_all(&dpm_late_early_list);
+			/*
+			 * Move all devices to the target list to resume them
+			 * properly.
+			 */
+			list_splice_init(&dpm_late_early_list, &dpm_noirq_list);
 			break;
+		}
 	}
 
 	mutex_unlock(&dpm_list_mtx);
 
 	async_synchronize_full();
-	if (!error)
-		error = async_error;
 
+	error = READ_ONCE(async_error);
 	if (error)
 		dpm_save_failed_step(SUSPEND_SUSPEND_NOIRQ);
 
@@ -1400,6 +1586,8 @@ static void dpm_propagate_wakeup_to_parent(struct device *dev)
 	spin_unlock_irq(&parent->power.lock);
 }
 
+static void async_suspend_late(void *data, async_cookie_t cookie);
+
 /**
  * device_suspend_late - Execute a "late suspend" callback for given device.
  * @dev: Device to handle.
@@ -1408,7 +1596,7 @@ static void dpm_propagate_wakeup_to_parent(struct device *dev)
  *
  * Runtime PM is disabled for @dev while this function is being executed.
  */
-static int device_suspend_late(struct device *dev, pm_message_t state, bool async)
+static void device_suspend_late(struct device *dev, pm_message_t state, bool async)
 {
 	pm_callback_t callback = NULL;
 	const char *info = NULL;
@@ -1425,11 +1613,11 @@ static int device_suspend_late(struct device *dev, pm_message_t state, bool asyn
 
 	dpm_wait_for_subordinate(dev, async);
 
-	if (async_error)
+	if (READ_ONCE(async_error))
 		goto Complete;
 
 	if (pm_wakeup_pending()) {
-		async_error = -EBUSY;
+		WRITE_ONCE(async_error, -EBUSY);
 		goto Complete;
 	}
 
@@ -1463,7 +1651,7 @@ static int device_suspend_late(struct device *dev, pm_message_t state, bool asyn
 Run:
 	error = dpm_run_callback(callback, dev, state, info);
 	if (error) {
-		async_error = error;
+		WRITE_ONCE(async_error, error);
 		dpm_save_failed_dev(dev_name(dev));
 		pm_dev_err(dev, state, async ? " async late" : " late", error);
 		goto Complete;
@@ -1476,7 +1664,11 @@ Skip:
 Complete:
 	TRACE_SUSPEND(error);
 	complete_all(&dev->power.completion);
-	return error;
+
+	if (error || READ_ONCE(async_error))
+		return;
+
+	dpm_async_suspend_superior(dev, async_suspend_late);
 }
 
 static void async_suspend_late(void *data, async_cookie_t cookie)
@@ -1494,7 +1686,8 @@ static void async_suspend_late(void *data, async_cookie_t cookie)
 int dpm_suspend_late(pm_message_t state)
 {
 	ktime_t starttime = ktime_get();
-	int error = 0;
+	struct device *dev;
+	int error;
 
 	trace_suspend_resume(TPS("dpm_suspend_late"), state.event, true);
 
@@ -1505,12 +1698,21 @@ int dpm_suspend_late(pm_message_t state)
 
 	mutex_lock(&dpm_list_mtx);
 
+	/*
+	 * Start processing "async" leaf devices upfront so they don't need to
+	 * wait for the "sync" devices they don't depend on.
+	 */
+	list_for_each_entry_reverse(dev, &dpm_suspended_list, power.entry) {
+		dpm_clear_async_state(dev);
+		if (dpm_leaf_device(dev))
+			dpm_async_with_cleanup(dev, async_suspend_late);
+	}
+
 	while (!list_empty(&dpm_suspended_list)) {
-		struct device *dev = to_device(dpm_suspended_list.prev);
+		dev = to_device(dpm_suspended_list.prev);
 
 		list_move(&dev->power.entry, &dpm_late_early_list);
 
-		dpm_clear_async_state(dev);
 		if (dpm_async_fn(dev, async_suspend_late))
 			continue;
 
@@ -1518,22 +1720,28 @@ int dpm_suspend_late(pm_message_t state)
 
 		mutex_unlock(&dpm_list_mtx);
 
-		error = device_suspend_late(dev, state, false);
+		device_suspend_late(dev, state, false);
 
 		put_device(dev);
 
 		mutex_lock(&dpm_list_mtx);
 
-		if (error || async_error)
+		if (READ_ONCE(async_error)) {
+			dpm_async_suspend_complete_all(&dpm_suspended_list);
+			/*
+			 * Move all devices to the target list to resume them
+			 * properly.
+			 */
+			list_splice_init(&dpm_suspended_list, &dpm_late_early_list);
 			break;
+		}
 	}
 
 	mutex_unlock(&dpm_list_mtx);
 
 	async_synchronize_full();
-	if (!error)
-		error = async_error;
 
+	error = READ_ONCE(async_error);
 	if (error) {
 		dpm_save_failed_step(SUSPEND_SUSPEND_LATE);
 		dpm_resume_early(resume_event(state));
@@ -1614,13 +1822,15 @@ static void dpm_clear_superiors_direct_complete(struct device *dev)
 	device_links_read_unlock(idx);
 }
 
+static void async_suspend(void *data, async_cookie_t cookie);
+
 /**
  * device_suspend - Execute "suspend" callbacks for given device.
  * @dev: Device to handle.
  * @state: PM transition of the system being carried out.
  * @async: If true, the device is being suspended asynchronously.
  */
-static int device_suspend(struct device *dev, pm_message_t state, bool async)
+static void device_suspend(struct device *dev, pm_message_t state, bool async)
 {
 	pm_callback_t callback = NULL;
 	const char *info = NULL;
@@ -1632,7 +1842,7 @@ static int device_suspend(struct device *dev, pm_message_t state, bool async)
 
 	dpm_wait_for_subordinate(dev, async);
 
-	if (async_error) {
+	if (READ_ONCE(async_error)) {
 		dev->power.direct_complete = false;
 		goto Complete;
 	}
@@ -1652,7 +1862,7 @@ static int device_suspend(struct device *dev, pm_message_t state, bool async)
 
 	if (pm_wakeup_pending()) {
 		dev->power.direct_complete = false;
-		async_error = -EBUSY;
+		WRITE_ONCE(async_error, -EBUSY);
 		goto Complete;
 	}
 
@@ -1736,14 +1946,18 @@ static int device_suspend(struct device *dev, pm_message_t state, bool async)
 
  Complete:
 	if (error) {
-		async_error = error;
+		WRITE_ONCE(async_error, error);
 		dpm_save_failed_dev(dev_name(dev));
 		pm_dev_err(dev, state, async ? " async" : "", error);
 	}
 
 	complete_all(&dev->power.completion);
 	TRACE_SUSPEND(error);
-	return error;
+
+	if (error || READ_ONCE(async_error))
+		return;
+
+	dpm_async_suspend_superior(dev, async_suspend);
 }
 
 static void async_suspend(void *data, async_cookie_t cookie)
@@ -1761,7 +1975,8 @@ static void async_suspend(void *data, async_cookie_t cookie)
 int dpm_suspend(pm_message_t state)
 {
 	ktime_t starttime = ktime_get();
-	int error = 0;
+	struct device *dev;
+	int error;
 
 	trace_suspend_resume(TPS("dpm_suspend"), state.event, true);
 	might_sleep();
@@ -1774,12 +1989,21 @@ int dpm_suspend(pm_message_t state)
 
 	mutex_lock(&dpm_list_mtx);
 
+	/*
+	 * Start processing "async" leaf devices upfront so they don't need to
+	 * wait for the "sync" devices they don't depend on.
+	 */
+	list_for_each_entry_reverse(dev, &dpm_prepared_list, power.entry) {
+		dpm_clear_async_state(dev);
+		if (dpm_leaf_device(dev))
+			dpm_async_with_cleanup(dev, async_suspend);
+	}
+
 	while (!list_empty(&dpm_prepared_list)) {
-		struct device *dev = to_device(dpm_prepared_list.prev);
+		dev = to_device(dpm_prepared_list.prev);
 
 		list_move(&dev->power.entry, &dpm_suspended_list);
 
-		dpm_clear_async_state(dev);
 		if (dpm_async_fn(dev, async_suspend))
 			continue;
 
@@ -1787,22 +2011,28 @@ int dpm_suspend(pm_message_t state)
 
 		mutex_unlock(&dpm_list_mtx);
 
-		error = device_suspend(dev, state, false);
+		device_suspend(dev, state, false);
 
 		put_device(dev);
 
 		mutex_lock(&dpm_list_mtx);
 
-		if (error || async_error)
+		if (READ_ONCE(async_error)) {
+			dpm_async_suspend_complete_all(&dpm_prepared_list);
+			/*
+			 * Move all devices to the target list to resume them
+			 * properly.
+			 */
+			list_splice_init(&dpm_prepared_list, &dpm_suspended_list);
 			break;
+		}
 	}
 
 	mutex_unlock(&dpm_list_mtx);
 
 	async_synchronize_full();
-	if (!error)
-		error = async_error;
 
+	error = READ_ONCE(async_error);
 	if (error)
 		dpm_save_failed_step(SUSPEND_SUSPEND);
 
@@ -1836,7 +2066,7 @@ static bool device_prepare_smart_suspend(struct device *dev)
 	idx = device_links_read_lock();
 
 	list_for_each_entry_rcu_locked(link, &dev->links.suppliers, c_node) {
-		if (!(link->flags & DL_FLAG_PM_RUNTIME))
+		if (!device_link_test(link, DL_FLAG_PM_RUNTIME))
 			continue;
 
 		if (!dev_pm_smart_suspend(link->supplier) &&
@@ -1947,7 +2177,6 @@ int dpm_prepare(pm_message_t state)
 	int error = 0;
 
 	trace_suspend_resume(TPS("dpm_prepare"), state.event, true);
-	might_sleep();
 
 	/*
 	 * Give a chance for the known devices to complete their probes, before
@@ -2014,8 +2243,10 @@ int dpm_suspend_start(pm_message_t state)
 	error = dpm_prepare(state);
 	if (error)
 		dpm_save_failed_step(SUSPEND_PREPARE);
-	else
+	else {
+		pm_restrict_gfp_mask();
 		error = dpm_suspend(state);
+	}
 
 	dpm_show_time(starttime, state, error, "start");
 	return error;