On Sun, Feb 26, 2012 at 8:47 PM, Robert Lee [off-list ref] wrote:

+/**
+ * cpuidle_enter_wrap - performing timekeeping and irq around enter function
+ * @dev: pointer to a valid cpuidle_device object
+ * @drv: pointer to a valid cpuidle_driver object
+ * @index: index of the target cpuidle state.
+ */
+static inline int cpuidle_wrap_enter(struct cpuidle_device *dev,
+ ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? struct cpuidle_driver *drv, int index,
+ ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? int (*enter)(struct cpuidle_device *dev,
+ ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? struct cpuidle_driver *drv, int index))
+{
+ ? ? ? ktime_t time_start, time_end;
+ ? ? ? s64 diff;
+
+ ? ? ? time_start = ktime_get();
+
+ ? ? ? index = enter(dev, drv, index);
+
+ ? ? ? time_end = ktime_get();
+
+ ? ? ? local_irq_enable();
+
+ ? ? ? diff = ktime_to_us(ktime_sub(time_end, time_start));
+ ? ? ? if (diff > INT_MAX)
+ ? ? ? ? ? ? ? diff = INT_MAX;
+
+ ? ? ? dev->last_residency = (int) diff;
+
+ ? ? ? return index;
+}

Hi Rob,

In a previous series I brought up the idea of not accounting for time
if a C-state transition fails.  My post on that thread can be found
here:
http://article.gmane.org/gmane.linux.ports.arm.kernel/149293/

How do you feel about adding something like the following?

if (IS_ERR(index))
        dev->last_residency = 0;
        return index;

Obviously it will up to the platforms to figure out how to propagate
that error up from their respective low power code.

Regards,
Mike