PM: sleep: Make suspend of devices more asynchronous

In analogy with previous changes, make device_suspend_late() and
device_suspend_noirq() start the async suspend of the device's parent
after the device itself has been processed and make dpm_suspend_late()
and dpm_noirq_suspend_devices() start processing "async" leaf devices
(that is, devices without children) upfront so they don't need to wait
for the other devices they don't depend on.

This change reduces the total duration of device suspend on some systems
measurably, but not significantly.

Suggested-by: Saravana Kannan <[e-mail]>
Signed-off-by: Rafael J. Wysocki <[e-mail]>
Link: https://patch.msgid.link/[e-mail]

drivers/base/power/main.c

···
 	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.
···
 Complete:
 	complete_all(&dev->power.completion);
 	TRACE_SUSPEND(error);
-	return error;
+
+	if (error || async_error)
+		return error;
+
+	dpm_async_suspend_parent(dev, async_suspend_noirq);
+
+	return 0;
 }
 
 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();
+	struct device *dev;
 	int error = 0;
 
 	trace_suspend_resume(TPS("dpm_suspend_noirq"), state.event, true);
···
 
 	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;
 
···
 
 		mutex_lock(&dpm_list_mtx);
 
-		if (error || async_error)
+		if (error || async_error) {
+			/*
+			 * Move all devices to the target list to resume them
+			 * properly.
+			 */
+			list_splice(&dpm_late_early_list, &dpm_noirq_list);
 			break;
+		}
 	}
 
 	mutex_unlock(&dpm_list_mtx);
···
 
 	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.
···
 Complete:
 	TRACE_SUSPEND(error);
 	complete_all(&dev->power.completion);
-	return error;
+
+	if (error || async_error)
+		return error;
+
+	dpm_async_suspend_parent(dev, async_suspend_late);
+
+	return 0;
 }
 
 static void async_suspend_late(void *data, async_cookie_t cookie)
···
 int dpm_suspend_late(pm_message_t state)
 {
 	ktime_t starttime = ktime_get();
+	struct device *dev;
 	int error = 0;
 
 	trace_suspend_resume(TPS("dpm_suspend_late"), state.event, true);
···
 
 	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;
 
···
 
 		mutex_lock(&dpm_list_mtx);
 
-		if (error || async_error)
+		if (error || async_error) {
+			/*
+			 * Move all devices to the target list to resume them
+			 * properly.
+			 */
+			list_splice(&dpm_suspended_list, &dpm_late_early_list);
 			break;
+		}
 	}
 
 	mutex_unlock(&dpm_list_mtx);