On 08/09, Daniel Lezcano wrote:

quoted

yes, but at least the broadcast mechanism should send an IPI to cpu0 to
wake it up, no ? As Stephen stated this kind of configuration should has
never been tested before so the tick broadcast code is not handling this
case properly IMHO.

If you have a per-cpu tick device that isn't suffering from
FEAT_C3_STOP why wouldn't you use that for the tick versus a
per-cpu tick device that has FEAT_C3_STOP? It sounds like there
is a bug in the preference logic or you should boost the rating
of the arm global timer above the twd. Does this patch help? It
should make the arm global timer the tick device and whatever the
cadence timer you have into the broadcast device.

---8<----

diff --git a/kernel/time/tick-broadcast.c b/kernel/time/tick-broadcast.c
index 218bcb5..d3539e5 100644
--- a/kernel/time/tick-broadcast.c
+++ b/kernel/time/tick-broadcast.c

@@ -77,6 +77,9 @@ static bool tick_check_broadcast_device(struct clock_event_device *curdev,
 	    !(newdev->features & CLOCK_EVT_FEAT_ONESHOT))
 		return false;
 
+	if (cpumask_equal(newdev->cpumask, cpumask_of(smp_processor_id())))
+		return false;

 	return !curdev || newdev->rating > curdev->rating;
 }
 

diff --git a/kernel/time/tick-common.c b/kernel/time/tick-common.c
index 64522ec..1628b9f 100644
--- a/kernel/time/tick-common.c
+++ b/kernel/time/tick-common.c

@@ -245,6 +245,15 @@ static bool tick_check_preferred(struct clock_event_device *curdev,
 	}
 
 	/*
+	 * Prefer tick devices that don't suffer from FEAT_C3_STOP
+	 * regardless of their rating
+	 */
+	if (curdev && cpumask_equal(curdev->cpumask, newdev->cpumask) &&
+	    !(newdev->features & CLOCK_EVT_FEAT_C3STOP) &&
+	    (curdev->features & CLOCK_EVT_FEAT_C3STOP))
+		return true;