On Thursday, November 22, 2012 06:02:37 PM Fabio Baltieri wrote:

Hello Rafael,

thanks for the review!  I only have one concern before sending a v4:

On Thu, Nov 22, 2012 at 01:10:15AM +0100, Rafael J. Wysocki wrote:

quoted

quoted

@@ -627,32 +659,41 @@ static void do_dbs_timer(struct work_struct *work)
 				delay -= jiffies % delay;
 		}
 	} else {
-		__cpufreq_driver_target(dbs_info->cur_policy,
-			dbs_info->freq_lo, CPUFREQ_RELATION_H);
+		if (sample)
+			__cpufreq_driver_target(dbs_info->cur_policy,
+						dbs_info->freq_lo,
+						CPUFREQ_RELATION_H);
 		delay = dbs_info->freq_lo_jiffies;
 	}
-	schedule_delayed_work_on(cpu, &dbs_info->work, delay);
+	schedule_delayed_work_on(smp_processor_id(), dw, delay);

We're not supposed to be using smp_processor_id() any more.
get_cpu()/put_cpu() should be used instead.

That's going to add preemption protection, do I need that?  The function
is called from a kworker with the affinity set on a specific CPU, so it
should not migrate to a different one during execution.

Yes, you're right, in that case it should be OK.

I agree with you for all the other comments.

Cool. :-)

Thanks,
Rafael


-- 
I speak only for myself.
Rafael J. Wysocki, Intel Open Source Technology Center.