Trinabh Gupta [off-list ref] writes:

Cpuidle subsystem only suggests the state to enter and does not
guarantee if the suggested state is entered. The actual entered state
may be different because of software or hardware demotion. Software
demotion is done by the back-end cpuidle driver and can be accounted
correctly. Current cpuidle code uses last_state field to capture the
actual state entered and based on that updates the statistics for the
state entered.

Ideally the driver enter routine should update the counters,
and it should return the state actually entered rather than the time
spent there. 

OK, the return type was changed to return the state index instead of the
time, but since the governors are still relying on dev->last_residency,
drivers are required to update it.

Because of that, why not keep the update of the  time/usage counters
in common code rather than duplicating it (9 times in this patch) into 
all the drivers?

Something like the patch below should suffice.

Kevin

diff --git a/drivers/cpuidle/cpuidle.c b/drivers/cpuidle/cpuidle.c
index 845d3ef..875d241 100644
--- a/drivers/cpuidle/cpuidle.c
+++ b/drivers/cpuidle/cpuidle.c

@@ -91,6 +91,11 @@ static void cpuidle_idle_call(void)
 
 	entered_state = target_state->enter(dev, drv, next_state);
 
+	/* Update cpuidle counters */
+	dev->states_usage[entered_state].time += 
+		(unsigned long long)dev->last_residency;
+	dev->states_usage[entered_state].usage++;
+
 	trace_power_end(dev->cpu);
 	trace_cpu_idle(PWR_EVENT_EXIT, dev->cpu);