Merge tag 'pm-4.14-rc4' of git://git.kernel.org/pub/scm/linux/kernel/…

…git/rafael/linux-pm

Pull power management fix from Rafael Wysocki:
 "This fixes a code ordering issue in the main suspend-to-idle loop that
  causes some "low power S0 idle" conditions to be incorrectly reported
  as unmet with suspend/resume debug messages enabled"

* tag 'pm-4.14-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm:
  PM / s2idle: Invoke the ->wake() platform callback earlier

kernel/power/suspend.c

@@ -120,22 +120,26 @@ static void s2idle_loop(void)
 		 * frozen processes + suspended devices + idle processors.
 		 * Thus s2idle_enter() should be called right after
 		 * all devices have been suspended.
+		 *
+		 * Wakeups during the noirq suspend of devices may be spurious,
+		 * so prevent them from terminating the loop right away.
 		 */
 		error = dpm_noirq_suspend_devices(PMSG_SUSPEND);
 		if (!error)
 			s2idle_enter();
+		else if (error == -EBUSY && pm_wakeup_pending())
+			error = 0;
 
-		dpm_noirq_resume_devices(PMSG_RESUME);
-		if (error && (error != -EBUSY || !pm_wakeup_pending())) {
-			dpm_noirq_end();
-			break;
-		}
-
-		if (s2idle_ops && s2idle_ops->wake)
+		if (!error && s2idle_ops && s2idle_ops->wake)
 			s2idle_ops->wake();
 
+		dpm_noirq_resume_devices(PMSG_RESUME);
+
 		dpm_noirq_end();
 
+		if (error)
+			break;
+
 		if (s2idle_ops && s2idle_ops->sync)
 			s2idle_ops->sync();