[PATCH 0/2] sched: make the scheduler (even) more modular

[PATCH 0/2] sched: make the scheduler (even) more modular

[Henrik Austad]

Several of the functions in the core scheduler tests explicitly for class and 
then to class-specific operations. This should be handled by the sched_class 
interface.

This patch-series moves 2 such elements over to sched_entity, or more precisely, 
the priority part.

Ingo, Peter: I forgot to CC lkml the last time around, sorry about this resend.

[PATCH 1/2] sched: Move priority calculation into sched_class.

[Henrik Austad]

When calculating the normal priority of a task, sched.c tested to see if the
task was an RT-task, before doing scheduling-class specific operations. 

By moving this into sched_class, the calculation becomes more modular, and 
the core scheduler becomes less aware of sched-class' internals.

Signed-off-by: Henrik Austad <henrik@austad.us>
Acked-by: Peter Zijlstra <peterz@infradead.org>
---
 include/linux/sched.h   |    1 +
 kernel/sched.c          |   16 +---------------
 kernel/sched_fair.c     |    7 +++++++
 kernel/sched_idletask.c |    6 ++++++
 kernel/sched_rt.c       |    6 ++++++
 5 files changed, 21 insertions(+), 15 deletions(-)

diff --git a/include/linux/sched.h b/include/linux/sched.h
index b94f354..6dc6591 100644
--- a/include/linux/sched.h
+++ b/include/linux/sched.h
@@ -1014,6 +1014,7 @@ struct sched_class {
 			     int running);
 	void (*prio_changed) (struct rq *this_rq, struct task_struct *task,
 			     int oldprio, int running);
+	int (*normal_prio) (struct task_struct *task);
 
 #ifdef CONFIG_FAIR_GROUP_SCHED
 	void (*moved_group) (struct task_struct *p);
diff --git a/kernel/sched.c b/kernel/sched.c
index 6cc1fd5..da13af5 100644
--- a/kernel/sched.c
+++ b/kernel/sched.c
@@ -1794,14 +1794,6 @@ static void dequeue_task(struct rq *rq, struct task_struct *p, int sleep)
 }
 
 /*
- * __normal_prio - return the priority that is based on the static prio
- */
-static inline int __normal_prio(struct task_struct *p)
-{
-	return p->static_prio;
-}
-
-/*
  * Calculate the expected normal priority: i.e. priority
  * without taking RT-inheritance into account. Might be
  * boosted by interactivity modifiers. Changes upon fork,
@@ -1810,13 +1802,7 @@ static inline int __normal_prio(struct task_struct *p)
  */
 static inline int normal_prio(struct task_struct *p)
 {
-	int prio;
-
-	if (task_has_rt_policy(p))
-		prio = MAX_RT_PRIO-1 - p->rt_priority;
-	else
-		prio = __normal_prio(p);
-	return prio;
+	return p->sched_class->normal_prio(p);
 }
 
 /*
diff --git a/kernel/sched_fair.c b/kernel/sched_fair.c
index 3816f21..c94519a 100644
--- a/kernel/sched_fair.c
+++ b/kernel/sched_fair.c
@@ -1749,6 +1749,12 @@ static void prio_changed_fair(struct rq *rq, struct task_struct *p,
 		check_preempt_curr(rq, p, 0);
 }
 
+/* Return the expected, normal priority */
+static int normal_prio_fair(struct task_struct *p)
+{
+	return p->static_prio;
+}
+
 /*
  * We switched to the sched_fair class.
  */
@@ -1815,6 +1821,7 @@ static const struct sched_class fair_sched_class = {
 	.task_new		= task_new_fair,
 
 	.prio_changed		= prio_changed_fair,
+	.normal_prio		= normal_prio_fair,
 	.switched_to		= switched_to_fair,
 
 #ifdef CONFIG_FAIR_GROUP_SCHED
diff --git a/kernel/sched_idletask.c b/kernel/sched_idletask.c
index 8a21a2e..e52ae7c 100644
--- a/kernel/sched_idletask.c
+++ b/kernel/sched_idletask.c
@@ -96,6 +96,11 @@ static void prio_changed_idle(struct rq *rq, struct task_struct *p,
 		check_preempt_curr(rq, p, 0);
 }
 
+static int normal_prio_idle(struct task_struct *p)
+{
+	return p->static_prio;
+}
+
 /*
  * Simple, special scheduling class for the per-CPU idle tasks:
  */
@@ -122,6 +127,7 @@ static const struct sched_class idle_sched_class = {
 	.task_tick		= task_tick_idle,
 
 	.prio_changed		= prio_changed_idle,
+	.normal_prio		= normal_prio_idle,
 	.switched_to		= switched_to_idle,
 
 	/* no .task_new for idle tasks */
diff --git a/kernel/sched_rt.c b/kernel/sched_rt.c
index 299d012..ec966b7 100644
--- a/kernel/sched_rt.c
+++ b/kernel/sched_rt.c
@@ -1671,6 +1671,11 @@ static void prio_changed_rt(struct rq *rq, struct task_struct *p,
 	}
 }
 
+static int normal_prio_rt(struct task_struct *p)
+{
+	return MAX_RT_PRIO-1 - p->rt_priority;
+}
+
 static void watchdog(struct rq *rq, struct task_struct *p)
 {
 	unsigned long soft, hard;
@@ -1759,6 +1764,7 @@ static const struct sched_class rt_sched_class = {
 	.task_tick		= task_tick_rt,
 
 	.prio_changed		= prio_changed_rt,
+	.normal_prio		= normal_prio_rt,
 	.switched_to		= switched_to_rt,
 };
 
-- 
1.6.1

[PATCH 2/2] sched: Let sched_class handle effective_prio calculation

[Henrik Austad]

Instead of testing for sched_class in the core scheduler, the calculation of the
class-specific effective priority should be handled by the scheduling class.

Signed-off-by: Henrik Austad <henrik@austad.us>
Acked-by: Peter Zijlstra <peterz@infradead.org>
---
 include/linux/sched.h   |    1 +
 kernel/sched.c          |   11 +++--------
 kernel/sched_fair.c     |    1 +
 kernel/sched_idletask.c |    1 +
 kernel/sched_rt.c       |    6 ++++++
 5 files changed, 12 insertions(+), 8 deletions(-)

diff --git a/include/linux/sched.h b/include/linux/sched.h
index 6dc6591..c98e912 100644
--- a/include/linux/sched.h
+++ b/include/linux/sched.h
@@ -1015,6 +1015,7 @@ struct sched_class {
 	void (*prio_changed) (struct rq *this_rq, struct task_struct *task,
 			     int oldprio, int running);
 	int (*normal_prio) (struct task_struct *task);
+	int (*effective_prio) (struct task_struct *task);
 
 #ifdef CONFIG_FAIR_GROUP_SCHED
 	void (*moved_group) (struct task_struct *p);
diff --git a/kernel/sched.c b/kernel/sched.c
index da13af5..5a6d4bd 100644
--- a/kernel/sched.c
+++ b/kernel/sched.c
@@ -1814,15 +1814,10 @@ static inline int normal_prio(struct task_struct *p)
  */
 static int effective_prio(struct task_struct *p)
 {
+	/* update normal prio */
 	p->normal_prio = normal_prio(p);
-	/*
-	 * If we are RT tasks or we were boosted to RT priority,
-	 * keep the priority unchanged. Otherwise, update priority
-	 * to the normal priority:
-	 */
-	if (!rt_prio(p->prio))
-		return p->normal_prio;
-	return p->prio;
+
+	return p->sched_class->effective_prio(p);
 }
 
 /*
diff --git a/kernel/sched_fair.c b/kernel/sched_fair.c
index c94519a..42af406 100644
--- a/kernel/sched_fair.c
+++ b/kernel/sched_fair.c
@@ -1822,6 +1822,7 @@ static const struct sched_class fair_sched_class = {
 
 	.prio_changed		= prio_changed_fair,
 	.normal_prio		= normal_prio_fair,
+	.effective_prio		= normal_prio_fair,
 	.switched_to		= switched_to_fair,
 
 #ifdef CONFIG_FAIR_GROUP_SCHED
diff --git a/kernel/sched_idletask.c b/kernel/sched_idletask.c
index e52ae7c..675de4e 100644
--- a/kernel/sched_idletask.c
+++ b/kernel/sched_idletask.c
@@ -128,6 +128,7 @@ static const struct sched_class idle_sched_class = {
 
 	.prio_changed		= prio_changed_idle,
 	.normal_prio		= normal_prio_idle,
+	.effective_prio		= normal_prio_idle,
 	.switched_to		= switched_to_idle,
 
 	/* no .task_new for idle tasks */
diff --git a/kernel/sched_rt.c b/kernel/sched_rt.c
index ec966b7..6426ced 100644
--- a/kernel/sched_rt.c
+++ b/kernel/sched_rt.c
@@ -1676,6 +1676,11 @@ static int normal_prio_rt(struct task_struct *p)
 	return MAX_RT_PRIO-1 - p->rt_priority;
 }
 
+static int effective_prio_rt(struct task_struct *p)
+{
+	return p->prio;
+}
+
 static void watchdog(struct rq *rq, struct task_struct *p)
 {
 	unsigned long soft, hard;
@@ -1765,6 +1770,7 @@ static const struct sched_class rt_sched_class = {
 
 	.prio_changed		= prio_changed_rt,
 	.normal_prio		= normal_prio_rt,
+	.effective_prio		= effective_prio_rt,
 	.switched_to		= switched_to_rt,
 };
 
-- 
1.6.1

Re: [PATCH 0/2] sched: make the scheduler (even) more modular

[Américo Wang]

On Tue, Apr 07, 2009 at 03:36:11PM +0200, Henrik Austad wrote:
>Several of the functions in the core scheduler tests explicitly for class and 
>then to class-specific operations. This should be handled by the sched_class 
>interface.
>
>This patch-series moves 2 such elements over to sched_entity, or more precisely, 
>the priority part.
>
>Ingo, Peter: I forgot to CC lkml the last time around, sorry about this resend.
>

I like these two patches! Really nice!

You can add my reviewed-by if you need:

Reviewed-by: WANG Cong <xiyou.wangcong@gmail.com>

Thanks.