cpufreq: governor: Reset sample delay in store_sampling_rate()
authorRafael J. Wysocki <rafael.j.wysocki@intel.com>
Mon, 15 Feb 2016 01:20:11 +0000 (02:20 +0100)
committerRafael J. Wysocki <rafael.j.wysocki@intel.com>
Wed, 9 Mar 2016 13:41:04 +0000 (14:41 +0100)
If store_sampling_rate() updates the sample delay when the ondemand
governor is in the middle of its high/low dance (OD_SUB_SAMPLE sample
type is set), the governor will still do the bottom half of the
previous sample which may take too much time.

To prevent that from happening, change store_sampling_rate() to always
reset the sample delay to 0 which also is consistent with the new
behavior of cpufreq_governor_limits().

Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
drivers/cpufreq/cpufreq_governor.c

index 99d25af6485b5e31a78040d8399c0c252bf29a07..fd4cdc2db238bbb7baf2ede76a579ee296d179da 100644 (file)
@@ -38,10 +38,6 @@ EXPORT_SYMBOL_GPL(dbs_data_mutex);
  * reducing the sampling rate, we need to make the new value effective
  * immediately.
  *
- * On the other hand, if new rate is larger than the old, then we may evaluate
- * the load too soon, and it might we worth updating sample_delay_ns then as
- * well.
- *
  * This must be called with dbs_data->mutex held, otherwise traversing
  * policy_dbs_list isn't safe.
  */
@@ -69,18 +65,14 @@ ssize_t store_sampling_rate(struct dbs_data *dbs_data, const char *buf,
                 * really doesn't matter.  If the read returns a value that's
                 * too big, the sample will be skipped, but the next invocation
                 * of dbs_update_util_handler() (when the update has been
-                * completed) will take a sample.  If the returned value is too
-                * small, the sample will be taken immediately, but that isn't a
-                * problem, as we want the new rate to take effect immediately
-                * anyway.
+                * completed) will take a sample.
                 *
                 * If this runs in parallel with dbs_work_handler(), we may end
                 * up overwriting the sample_delay_ns value that it has just
-                * written, but the difference should not be too big and it will
-                * be corrected next time a sample is taken, so it shouldn't be
-                * significant.
+                * written, but it will be corrected next time a sample is
+                * taken, so it shouldn't be significant.
                 */
-               gov_update_sample_delay(policy_dbs, dbs_data->sampling_rate);
+               gov_update_sample_delay(policy_dbs, 0);
                mutex_unlock(&policy_dbs->timer_mutex);
        }