PM / Runtime: Don't enable interrupts while running in_interrupt

This patch (as1445) fixes a bug in the runtime PM core left over from
the addition of the no_callbacks flag.  If this flag is set then it is
possible for rpm_suspend() to be called in_interrupt, so when
releasing spinlocks it's important not to re-enable interrupts.

To avoid an unnecessary save-and-restore of the interrupt flag, the
patch also inlines a pm_request_idle() call.

This fixes Bugzilla #27482.

(The offending code was added in 2.6.37, so it's not necessary to apply
this to any earlier stable kernels.)

Signed-off-by: Alan Stern <stern@rowland.harvard.edu>
Reported-by: tim blechmann <tim@klingt.org>
CC: <stable@kernel.org>
Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>

drivers/base/power/runtime.c

@@ -407,12 +407,15 @@ static int rpm_suspend(struct device *dev, int rpmflags)
 		goto out;
 	}
 
+	/* Maybe the parent is now able to suspend. */
 	if (parent && !parent->power.ignore_children && !dev->power.irq_safe) {
-		spin_unlock_irq(&dev->power.lock);
+		spin_unlock(&dev->power.lock);
 
-		pm_request_idle(parent);
+		spin_lock(&parent->power.lock);
+		rpm_idle(parent, RPM_ASYNC);
+		spin_unlock(&parent->power.lock);
 
-		spin_lock_irq(&dev->power.lock);
+		spin_lock(&dev->power.lock);
 	}
 
  out: