Commit e0367b12 authored by Juri Lelli's avatar Juri Lelli Committed by Ingo Molnar
Browse files

sched/deadline: Move CPU frequency selection triggering points 



Since SCHED_DEADLINE doesn't track utilization signal (but reserves a
fraction of CPU bandwidth to tasks admitted to the system), there is no
point in evaluating frequency changes during each tick event.

Move frequency selection triggering points to where running_bw changes.

Co-authored-by: default avatarClaudio Scordino <claudio@evidence.eu.com>
Signed-off-by: default avatarJuri Lelli <juri.lelli@arm.com>
Signed-off-by: default avatarPeter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: default avatarViresh Kumar <viresh.kumar@linaro.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Luca Abeni <luca.abeni@santannapisa.it>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Rafael J . Wysocki <rafael.j.wysocki@intel.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: alessio.balsini@arm.com
Cc: bristot@redhat.com
Cc: dietmar.eggemann@arm.com
Cc: joelaf@google.com
Cc: juri.lelli@redhat.com
Cc: mathieu.poirier@linaro.org
Cc: morten.rasmussen@arm.com
Cc: patrick.bellasi@arm.com
Cc: rjw@rjwysocki.net
Cc: rostedt@goodmis.org
Cc: tkjos@android.com
Cc: tommaso.cucinotta@santannapisa.it
Cc: vincent.guittot@linaro.org
Link: http://lkml.kernel.org/r/20171204102325.5110-3-juri.lelli@redhat.com


Signed-off-by: default avatarIngo Molnar <mingo@kernel.org>
parent d4edd662

kernel/sched/deadline.c

+4 −3
Original line number Diff line number Diff line
@@ -86,6 +86,8 @@ void add_running_bw(u64 dl_bw, struct dl_rq *dl_rq) 
	dl_rq->running_bw += dl_bw;

	SCHED_WARN_ON(dl_rq->running_bw < old); /* overflow */

	SCHED_WARN_ON(dl_rq->running_bw > dl_rq->this_bw);

	/* kick cpufreq (see the comment in kernel/sched/sched.h). */

	cpufreq_update_util(rq_of_dl_rq(dl_rq), SCHED_CPUFREQ_DL);

}



static inline
@@ -98,6 +100,8 @@ void sub_running_bw(u64 dl_bw, struct dl_rq *dl_rq) 
	SCHED_WARN_ON(dl_rq->running_bw > old); /* underflow */

	if (dl_rq->running_bw > old)

		dl_rq->running_bw = 0;

	/* kick cpufreq (see the comment in kernel/sched/sched.h). */

	cpufreq_update_util(rq_of_dl_rq(dl_rq), SCHED_CPUFREQ_DL);

}



static inline
@@ -1134,9 +1138,6 @@ static void update_curr_dl(struct rq *rq) 
		return;

	}



	/* kick cpufreq (see the comment in kernel/sched/sched.h). */

	cpufreq_update_util(rq, SCHED_CPUFREQ_DL);



	schedstat_set(curr->se.statistics.exec_max,

		      max(curr->se.statistics.exec_max, delta_exec));

kernel/sched/sched.h

+6 −6
Original line number Diff line number Diff line
@@ -2055,14 +2055,14 @@ DECLARE_PER_CPU(struct update_util_data *, cpufreq_update_util_data); 
 * The way cpufreq is currently arranged requires it to evaluate the CPU

 * performance state (frequency/voltage) on a regular basis to prevent it from

 * being stuck in a completely inadequate performance level for too long.

 * That is not guaranteed to happen if the updates are only triggered from CFS,

 * though, because they may not be coming in if RT or deadline tasks are active

 * all the time (or there are RT and DL tasks only).

 * That is not guaranteed to happen if the updates are only triggered from CFS

 * and DL, though, because they may not be coming in if only RT tasks are

 * active all the time (or there are RT tasks only).

 *

 * As a workaround for that issue, this function is called by the RT and DL

 * sched classes to trigger extra cpufreq updates to prevent it from stalling,

 * As a workaround for that issue, this function is called periodically by the

 * RT sched class to trigger extra cpufreq updates to prevent it from stalling,

 * but that really is a band-aid.  Going forward it should be replaced with

 * solutions targeted more specifically at RT and DL tasks.

 * solutions targeted more specifically at RT tasks.

 */

static inline void cpufreq_update_util(struct rq *rq, unsigned int flags)

{

CRA Git | Maintained and supported by SUSTech CRA and CCSE