Thread (29 messages) 29 messages, 4 authors, 2015-02-09

Re: [Update 2x] Re: [PATCH v3]PM/Sleep: Timer quiesce in freeze state

From: Peter Zijlstra <peterz@infradead.org>
Date: 2015-02-09 15:44:19
Also in: lkml

On Mon, Feb 09, 2015 at 03:54:22AM +0100, Rafael J. Wysocki wrote:
Complete patch with that modification is appended.  In the next few days I'm
going to split it into smaller parts and send along with cpuidle driver
patches implementing ->enter_freeze.

Please let me know what you think.
quoted hunk ↗ jump to hunk
@@ -104,6 +105,21 @@ static void cpuidle_idle_call(void)
 	rcu_idle_enter();
 
 	/*
+	 * Suspend-to-idle ("freeze") is a system state in which all user space
+	 * has been frozen, all I/O devices have been suspended and the only
+	 * activity happens here and in iterrupts (if any).  In that case bypass
+	 * the cpuidle governor and go stratight for the deepest idle state
+	 * available.  Possibly also suspend the local tick and the entire
+	 * timekeeping to prevent timer interrupts from kicking us out of idle
+	 * until a proper wakeup interrupt happens.
+	 */
+	if (idle_should_freeze()) {
+		cpuidle_enter_freeze();
+		local_irq_enable();
+		goto exit_idle;
+	}
+
+	/*
 	 * Ask the cpuidle framework to choose a convenient idle state.
 	 * Fall back to the default arch idle method on errors.
 	 */
I was hoping to not have to put that into the regular idle path; say
maybe share a single special branch with the play-dead call. People seem
to start complaining about the total amount of time it takes to just
'run' the idle path.

Now I don't think we can do that, because we need the
arch_cpu_idle_enter() nonsense for the one but not the other; also all
this really only makes sense in the cpuidle context, so nothing to be
done about that.

In any case, you could make that:

static inline bool idle_should_freeze(void)
{
	return unlikely(suspend_freeze_state == FREEZE_STATE_ENTER);
}

which should help a bit I suppose.
+static void enter_freeze_proper(struct cpuidle_driver *drv,
+                               struct cpuidle_device *dev, int index)
+{
+       tick_freeze();
+       drv->states[index].enter_freeze(dev, drv, index);
This is slightly different from cpuidle_enter() in that it does not
consider the coupled states nonsense, is that on purpose? And if so,
does that want a comment?
+       /*
+        * timekeeping_resume() that will be called by tick_unfreeze() for the
+        * last CPU executing it calls functions containing RCU read-side
+        * critical sections, so tell RCU about that.
+        */
+       RCU_NONIDLE(tick_unfreeze());
+}

But over all it looks fine to me.
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help