[sched-devel, patch-rfc] rework #2 of "prioritize non-migratable tasks over migratable ones"

[sched-devel, patch-rfc] rework #2 of "prioritize non-migratable tasks over migratable ones"

[Dmitry Adamushko]

[-- Attachment #1: Type: text/plain, Size: 4580 bytes --]

Hi,


this is a continuation of another thread:
http://www.ussg.iu.edu/hypermail/linux/kernel/0807.0/0134.html


The following patch is just an illustration, not tested even compilation wise.


(1) handle in a generic way all cases when a newly woken-up task is
not migratable (not just a corner case when "rt_se->nr_cpus_allowed ==
1")

(2) if current is to be preempted, then make sure "p" will be picked
up by pick_next_task_rt().
i.e. move task's group at the head of its list as well.

currently, it's not a case for group-scheduling as described here:
http://www.ussg.iu.edu/hypermail/linux/kernel/0807.0/0134.html


what do you think?


(non white-space-damaged version is enclosed)


---

--- sched_rt-old.c      2008-07-01 11:42:30.000000000 +0200
+++ sched_rt.c  2008-07-01 15:00:55.000000000 +0200
@@ -599,11 +599,7 @@ static void __enqueue_rt_entity(struct s
        if (group_rq && (rt_rq_throttled(group_rq) ||
!group_rq->rt_nr_running))                return;

-       if (rt_se->nr_cpus_allowed == 1)
-               list_add(&rt_se->run_list, queue);
-       else
-               list_add_tail(&rt_se->run_list, queue);
-
+       list_add_tail(&rt_se->run_list, queue);
        __set_bit(rt_se_prio(rt_se), array->bitmap);

        inc_rt_tasks(rt_se, rt_rq);
@@ -689,31 +685,33 @@ static void dequeue_task_rt(struct rq *r
  * followed by enqueue.
  */
 static
-void requeue_rt_entity(struct rt_rq *rt_rq, struct sched_rt_entity *rt_se)
+void requeue_rt_entity(struct rt_rq *rt_rq, struct sched_rt_entity
*rt_se, int head)
 {
-       struct rt_prio_array *array = &rt_rq->active;
-
        if (on_rt_rq(rt_se)) {
-               list_del_init(&rt_se->run_list);
-               list_add_tail(&rt_se->run_list,
-                             array->queue + rt_se_prio(rt_se));
+               struct rt_prio_array *array = &rt_rq->active;
+               struct list_head *queue = array->queue + rt_se_prio(rt_se);
+
+               if (head)
+                       list_move(&rt_se->run_list, queue);
+               else
+                       list_move_tail(&rt_se->run_list, queue);
        }
 }

-static void requeue_task_rt(struct rq *rq, struct task_struct *p)
+static void requeue_task_rt(struct rq *rq, struct task_struct *p, int head)
 {
        struct sched_rt_entity *rt_se = &p->rt;
        struct rt_rq *rt_rq;

        for_each_sched_rt_entity(rt_se) {
                rt_rq = rt_rq_of_se(rt_se);
-               requeue_rt_entity(rt_rq, rt_se);
+               requeue_rt_entity(rt_rq, rt_se, head);
        }
 }

 static void yield_task_rt(struct rq *rq)
 {
-       requeue_task_rt(rq, rq->curr);
+       requeue_task_rt(rq, rq->curr, 0);
 }

 #ifdef CONFIG_SMP
@@ -753,6 +751,29 @@ static int select_task_rq_rt(struct task
         */
        return task_cpu(p);
 }
+
+static void check_preempt_equal_prio(struct rq *rq, struct task_struct *p)
+{
+       cpumask_t mask;
+
+       if (rq->curr.rt.nr_cpus_allowed == 1 || p->rt.nr_cpus_allowed == 1)
+               return;
+
+       if (cpupri_find(&rq->rd->cpupri, p, &mask))
+               return;
+
+       if (!cpupri_find(&rq->rd->cpupri, rq->curr, &mask))
+               return;
+
+       /*
+        * There appears to be other cpus that can accept
+        * current and none to run 'p', so lets reschedule
+        * to try and push current away:
+        */
+       requeue_task_rt(rq, p, 1);
+       resched_task(rq->curr);
+}
+
 #endif /* CONFIG_SMP */

 /*
@@ -778,18 +799,8 @@ static void check_preempt_curr_rt(struct
         * to move current somewhere else, making room for our non-migratable
         * task.
         */
-       if((p->prio == rq->curr->prio)
-          && p->rt.nr_cpus_allowed == 1
-          && rq->curr->rt.nr_cpus_allowed != 1) {
-               cpumask_t mask;
-
-               if (cpupri_find(&rq->rd->cpupri, rq->curr, &mask))
-                       /*
-                        * There appears to be other cpus that can accept
-                        * current, so lets reschedule to try and push it away
-                        */
-                       resched_task(rq->curr);
-       }
+       if (p->prio == rq->curr->prio && !need_resched())
+               check_preempt_equal_prio(rq, p);
 #endif
 }

@@ -1415,7 +1426,7 @@ static void task_tick_rt(struct rq *rq,
         * on the queue:
         */
        if (p->rt.run_list.prev != p->rt.run_list.next) {
-               requeue_task_rt(rq, p);
+               requeue_task_rt(rq, p, 0);
                set_tsk_need_resched(p);
        }
 }

---



-- 
Best regards,
Dmitry Adamushko

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #2: resched-eqial-prio.patch --]
[-- Type: text/x-patch; name=resched-eqial-prio.patch, Size: 3172 bytes --]

--- sched_rt-old.c	2008-07-01 11:42:30.000000000 +0200
+++ sched_rt.c	2008-07-01 15:00:55.000000000 +0200
@@ -599,11 +599,7 @@ static void __enqueue_rt_entity(struct s
 	if (group_rq && (rt_rq_throttled(group_rq) || !group_rq->rt_nr_running))
 		return;
 
-	if (rt_se->nr_cpus_allowed == 1)
-		list_add(&rt_se->run_list, queue);
-	else
-		list_add_tail(&rt_se->run_list, queue);
-
+	list_add_tail(&rt_se->run_list, queue);
 	__set_bit(rt_se_prio(rt_se), array->bitmap);
 
 	inc_rt_tasks(rt_se, rt_rq);
@@ -689,31 +685,33 @@ static void dequeue_task_rt(struct rq *r
  * followed by enqueue.
  */
 static
-void requeue_rt_entity(struct rt_rq *rt_rq, struct sched_rt_entity *rt_se)
+void requeue_rt_entity(struct rt_rq *rt_rq, struct sched_rt_entity *rt_se, int head)
 {
-	struct rt_prio_array *array = &rt_rq->active;
-
 	if (on_rt_rq(rt_se)) {
-		list_del_init(&rt_se->run_list);
-		list_add_tail(&rt_se->run_list,
-			      array->queue + rt_se_prio(rt_se));
+		struct rt_prio_array *array = &rt_rq->active;
+		struct list_head *queue = array->queue + rt_se_prio(rt_se);
+
+		if (head)
+			list_move(&rt_se->run_list, queue);
+		else
+			list_move_tail(&rt_se->run_list, queue);
 	}
 }
 
-static void requeue_task_rt(struct rq *rq, struct task_struct *p)
+static void requeue_task_rt(struct rq *rq, struct task_struct *p, int head)
 {
 	struct sched_rt_entity *rt_se = &p->rt;
 	struct rt_rq *rt_rq;
 
 	for_each_sched_rt_entity(rt_se) {
 		rt_rq = rt_rq_of_se(rt_se);
-		requeue_rt_entity(rt_rq, rt_se);
+		requeue_rt_entity(rt_rq, rt_se, head);
 	}
 }
 
 static void yield_task_rt(struct rq *rq)
 {
-	requeue_task_rt(rq, rq->curr);
+	requeue_task_rt(rq, rq->curr, 0);
 }
 
 #ifdef CONFIG_SMP
@@ -753,6 +751,29 @@ static int select_task_rq_rt(struct task
 	 */
 	return task_cpu(p);
 }
+
+static void check_preempt_equal_prio(struct rq *rq, struct task_struct *p)
+{
+	cpumask_t mask;
+
+	if (rq->curr.rt.nr_cpus_allowed == 1 || p->rt.nr_cpus_allowed == 1)
+		return;
+
+	if (cpupri_find(&rq->rd->cpupri, p, &mask))
+		return;
+	
+	if (!cpupri_find(&rq->rd->cpupri, rq->curr, &mask))
+		return;
+
+	/*
+	 * There appears to be other cpus that can accept
+	 * current and none to run 'p', so lets reschedule
+	 * to try and push current away:
+	 */
+	requeue_task_rt(rq, p, 1);
+	resched_task(rq->curr);
+}
+
 #endif /* CONFIG_SMP */
 
 /*
@@ -778,18 +799,8 @@ static void check_preempt_curr_rt(struct
 	 * to move current somewhere else, making room for our non-migratable
 	 * task.
 	 */
-	if((p->prio == rq->curr->prio)
-	   && p->rt.nr_cpus_allowed == 1
-	   && rq->curr->rt.nr_cpus_allowed != 1) {
-		cpumask_t mask;
-
-		if (cpupri_find(&rq->rd->cpupri, rq->curr, &mask))
-			/*
-			 * There appears to be other cpus that can accept
-			 * current, so lets reschedule to try and push it away
-			 */
-			resched_task(rq->curr);
-	}
+	if (p->prio == rq->curr->prio && !need_resched())
+		check_preempt_equal_prio(rq, p);
 #endif
 }
 
@@ -1415,7 +1426,7 @@ static void task_tick_rt(struct rq *rq, 
 	 * on the queue:
 	 */
 	if (p->rt.run_list.prev != p->rt.run_list.next) {
-		requeue_task_rt(rq, p);
+		requeue_task_rt(rq, p, 0);
 		set_tsk_need_resched(p);
 	}
 }

Re: [sched-devel, patch-rfc] rework #2 of "prioritize non-migratable tasks over migratable ones"

[Steven Rostedt]

On Tue, 1 Jul 2008, Dmitry Adamushko wrote:

>
> ---
>
> --- sched_rt-old.c      2008-07-01 11:42:30.000000000 +0200
> +++ sched_rt.c  2008-07-01 15:00:55.000000000 +0200
> @@ -599,11 +599,7 @@ static void __enqueue_rt_entity(struct s
>         if (group_rq && (rt_rq_throttled(group_rq) ||
> !group_rq->rt_nr_running))                return;
>
> -       if (rt_se->nr_cpus_allowed == 1)
> -               list_add(&rt_se->run_list, queue);
> -       else
> -               list_add_tail(&rt_se->run_list, queue);
> -
> +       list_add_tail(&rt_se->run_list, queue);
>         __set_bit(rt_se_prio(rt_se), array->bitmap);
>
>         inc_rt_tasks(rt_se, rt_rq);
> @@ -689,31 +685,33 @@ static void dequeue_task_rt(struct rq *r
>   * followed by enqueue.
>   */
>  static
> -void requeue_rt_entity(struct rt_rq *rt_rq, struct sched_rt_entity *rt_se)
> +void requeue_rt_entity(struct rt_rq *rt_rq, struct sched_rt_entity
> *rt_se, int head)
>  {
> -       struct rt_prio_array *array = &rt_rq->active;
> -
>         if (on_rt_rq(rt_se)) {
> -               list_del_init(&rt_se->run_list);
> -               list_add_tail(&rt_se->run_list,
> -                             array->queue + rt_se_prio(rt_se));
> +               struct rt_prio_array *array = &rt_rq->active;
> +               struct list_head *queue = array->queue + rt_se_prio(rt_se);
> +
> +               if (head)
> +                       list_move(&rt_se->run_list, queue);
> +               else
> +                       list_move_tail(&rt_se->run_list, queue);
>         }
>  }
>
> -static void requeue_task_rt(struct rq *rq, struct task_struct *p)
> +static void requeue_task_rt(struct rq *rq, struct task_struct *p, int head)
>  {
>         struct sched_rt_entity *rt_se = &p->rt;
>         struct rt_rq *rt_rq;
>
>         for_each_sched_rt_entity(rt_se) {
>                 rt_rq = rt_rq_of_se(rt_se);
> -               requeue_rt_entity(rt_rq, rt_se);
> +               requeue_rt_entity(rt_rq, rt_se, head);
>         }
>  }
>
>  static void yield_task_rt(struct rq *rq)
>  {
> -       requeue_task_rt(rq, rq->curr);
> +       requeue_task_rt(rq, rq->curr, 0);
>  }
>
>  #ifdef CONFIG_SMP
> @@ -753,6 +751,29 @@ static int select_task_rq_rt(struct task
>          */
>         return task_cpu(p);
>  }
> +
> +static void check_preempt_equal_prio(struct rq *rq, struct task_struct *p)
> +{
> +       cpumask_t mask;
> +
> +       if (rq->curr.rt.nr_cpus_allowed == 1 || p->rt.nr_cpus_allowed == 1)
> +               return;

I can understand why we break out when current is bound to a single CPU,
but why do we break out when p is bound to a single CPU. Don't we want to
push curr off if we can find a CPU for it?

-- Steve

> +
> +       if (cpupri_find(&rq->rd->cpupri, p, &mask))
> +               return;
> +
> +       if (!cpupri_find(&rq->rd->cpupri, rq->curr, &mask))
> +               return;
> +
> +       /*
> +        * There appears to be other cpus that can accept
> +        * current and none to run 'p', so lets reschedule
> +        * to try and push current away:
> +        */
> +       requeue_task_rt(rq, p, 1);
> +       resched_task(rq->curr);
> +}
> +
>  #endif /* CONFIG_SMP */
>

Re: [sched-devel, patch-rfc] rework #2 of "prioritize non-migratable tasks over migratable ones"

[Dmitry Adamushko]

[-- Attachment #1: Type: text/plain, Size: 4388 bytes --]

2008/7/1 Steven Rostedt <rostedt@goodmis.org>:
>
> [ ... ]
>
>> +
>> +static void check_preempt_equal_prio(struct rq *rq, struct task_struct *p)
>> +{
>> +       cpumask_t mask;
>> +
>> +       if (rq->curr.rt.nr_cpus_allowed == 1 || p->rt.nr_cpus_allowed == 1)
>> +               return;
>
> I can understand why we break out when current is bound to a single CPU,
> but why do we break out when p is bound to a single CPU.

yes, my fault. Thanks, Steven!

This version should be better:

---
--- sched_rt-old.c      2008-07-01 11:42:30.000000000 +0200
+++ sched_rt.c  2008-07-01 16:16:48.000000000 +0200
@@ -599,11 +599,7 @@ static void __enqueue_rt_entity(struct s
        if (group_rq && (rt_rq_throttled(group_rq) ||
!group_rq->rt_nr_running))                return;

-       if (rt_se->nr_cpus_allowed == 1)
-               list_add(&rt_se->run_list, queue);
-       else
-               list_add_tail(&rt_se->run_list, queue);
-
+       list_add_tail(&rt_se->run_list, queue);
        __set_bit(rt_se_prio(rt_se), array->bitmap);

        inc_rt_tasks(rt_se, rt_rq);
@@ -689,31 +685,33 @@ static void dequeue_task_rt(struct rq *r
  * followed by enqueue.
  */
 static
-void requeue_rt_entity(struct rt_rq *rt_rq, struct sched_rt_entity *rt_se)
+void requeue_rt_entity(struct rt_rq *rt_rq, struct sched_rt_entity
*rt_se, int head)
 {
-       struct rt_prio_array *array = &rt_rq->active;
-
        if (on_rt_rq(rt_se)) {
-               list_del_init(&rt_se->run_list);
-               list_add_tail(&rt_se->run_list,
-                             array->queue + rt_se_prio(rt_se));
+               struct rt_prio_array *array = &rt_rq->active;
+               struct list_head *queue = array->queue + rt_se_prio(rt_se);
+
+               if (head)
+                       list_move(&rt_se->run_list, queue);
+               else
+                       list_move_tail(&rt_se->run_list, queue);
        }
 }

-static void requeue_task_rt(struct rq *rq, struct task_struct *p)
+static void requeue_task_rt(struct rq *rq, struct task_struct *p, int head)
 {
        struct sched_rt_entity *rt_se = &p->rt;
        struct rt_rq *rt_rq;

        for_each_sched_rt_entity(rt_se) {
                rt_rq = rt_rq_of_se(rt_se);
-               requeue_rt_entity(rt_rq, rt_se);
+               requeue_rt_entity(rt_rq, rt_se, head);
        }
 }

 static void yield_task_rt(struct rq *rq)
 {
-       requeue_task_rt(rq, rq->curr);
+       requeue_task_rt(rq, rq->curr, 0);
 }

 #ifdef CONFIG_SMP
@@ -753,6 +751,30 @@ static int select_task_rq_rt(struct task
         */
        return task_cpu(p);
 }
+
+static void check_preempt_equal_prio(struct rq *rq, struct task_struct *p)
+{
+       cpumask_t mask;
+
+       if (rq->curr.rt.nr_cpus_allowed == 1)
+               return;
+
+       if (p->rt.nr_cpus_allowed != 1
+           && cpupri_find(&rq->rd->cpupri, p, &mask))
+               return;
+
+       if (!cpupri_find(&rq->rd->cpupri, rq->curr, &mask))
+               return;
+
+       /*
+        * There appears to be other cpus that can accept
+        * current and none to run 'p', so lets reschedule
+        * to try and push current away:
+        */
+       requeue_task_rt(rq, p, 1);
+       resched_task(rq->curr);
+}
+
 #endif /* CONFIG_SMP */

 /*
@@ -778,18 +800,8 @@ static void check_preempt_curr_rt(struct
         * to move current somewhere else, making room for our non-migratable
         * task.
         */
-       if((p->prio == rq->curr->prio)
-          && p->rt.nr_cpus_allowed == 1
-          && rq->curr->rt.nr_cpus_allowed != 1) {
-               cpumask_t mask;
-
-               if (cpupri_find(&rq->rd->cpupri, rq->curr, &mask))
-                       /*
-                        * There appears to be other cpus that can accept
-                        * current, so lets reschedule to try and push it away
-                        */
-                       resched_task(rq->curr);
-       }
+       if (p->prio == rq->curr->prio && !need_resched())
+               check_preempt_equal_prio(rq, p);
 #endif
 }

@@ -1415,7 +1427,7 @@ static void task_tick_rt(struct rq *rq,
         * on the queue:
         */
        if (p->rt.run_list.prev != p->rt.run_list.next) {
-               requeue_task_rt(rq, p);
+               requeue_task_rt(rq, p, 0);
                set_tsk_need_resched(p);
        }
 }

---


-- 
Best regards,
Dmitry Adamushko

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #2: resched-eqial-prio.patch --]
[-- Type: text/x-patch; name=resched-eqial-prio.patch, Size: 3178 bytes --]

--- sched_rt-old.c	2008-07-01 11:42:30.000000000 +0200
+++ sched_rt.c	2008-07-01 16:16:48.000000000 +0200
@@ -599,11 +599,7 @@ static void __enqueue_rt_entity(struct s
 	if (group_rq && (rt_rq_throttled(group_rq) || !group_rq->rt_nr_running))
 		return;
 
-	if (rt_se->nr_cpus_allowed == 1)
-		list_add(&rt_se->run_list, queue);
-	else
-		list_add_tail(&rt_se->run_list, queue);
-
+	list_add_tail(&rt_se->run_list, queue);
 	__set_bit(rt_se_prio(rt_se), array->bitmap);
 
 	inc_rt_tasks(rt_se, rt_rq);
@@ -689,31 +685,33 @@ static void dequeue_task_rt(struct rq *r
  * followed by enqueue.
  */
 static
-void requeue_rt_entity(struct rt_rq *rt_rq, struct sched_rt_entity *rt_se)
+void requeue_rt_entity(struct rt_rq *rt_rq, struct sched_rt_entity *rt_se, int head)
 {
-	struct rt_prio_array *array = &rt_rq->active;
-
 	if (on_rt_rq(rt_se)) {
-		list_del_init(&rt_se->run_list);
-		list_add_tail(&rt_se->run_list,
-			      array->queue + rt_se_prio(rt_se));
+		struct rt_prio_array *array = &rt_rq->active;
+		struct list_head *queue = array->queue + rt_se_prio(rt_se);
+
+		if (head)
+			list_move(&rt_se->run_list, queue);
+		else
+			list_move_tail(&rt_se->run_list, queue);
 	}
 }
 
-static void requeue_task_rt(struct rq *rq, struct task_struct *p)
+static void requeue_task_rt(struct rq *rq, struct task_struct *p, int head)
 {
 	struct sched_rt_entity *rt_se = &p->rt;
 	struct rt_rq *rt_rq;
 
 	for_each_sched_rt_entity(rt_se) {
 		rt_rq = rt_rq_of_se(rt_se);
-		requeue_rt_entity(rt_rq, rt_se);
+		requeue_rt_entity(rt_rq, rt_se, head);
 	}
 }
 
 static void yield_task_rt(struct rq *rq)
 {
-	requeue_task_rt(rq, rq->curr);
+	requeue_task_rt(rq, rq->curr, 0);
 }
 
 #ifdef CONFIG_SMP
@@ -753,6 +751,30 @@ static int select_task_rq_rt(struct task
 	 */
 	return task_cpu(p);
 }
+
+static void check_preempt_equal_prio(struct rq *rq, struct task_struct *p)
+{
+	cpumask_t mask;
+
+	if (rq->curr.rt.nr_cpus_allowed == 1)
+		return;
+
+	if (p->rt.nr_cpus_allowed != 1
+	    && cpupri_find(&rq->rd->cpupri, p, &mask))
+		return;
+	
+	if (!cpupri_find(&rq->rd->cpupri, rq->curr, &mask))
+		return;
+
+	/*
+	 * There appears to be other cpus that can accept
+	 * current and none to run 'p', so lets reschedule
+	 * to try and push current away:
+	 */
+	requeue_task_rt(rq, p, 1);
+	resched_task(rq->curr);
+}
+
 #endif /* CONFIG_SMP */
 
 /*
@@ -778,18 +800,8 @@ static void check_preempt_curr_rt(struct
 	 * to move current somewhere else, making room for our non-migratable
 	 * task.
 	 */
-	if((p->prio == rq->curr->prio)
-	   && p->rt.nr_cpus_allowed == 1
-	   && rq->curr->rt.nr_cpus_allowed != 1) {
-		cpumask_t mask;
-
-		if (cpupri_find(&rq->rd->cpupri, rq->curr, &mask))
-			/*
-			 * There appears to be other cpus that can accept
-			 * current, so lets reschedule to try and push it away
-			 */
-			resched_task(rq->curr);
-	}
+	if (p->prio == rq->curr->prio && !need_resched())
+		check_preempt_equal_prio(rq, p);
 #endif
 }
 
@@ -1415,7 +1427,7 @@ static void task_tick_rt(struct rq *rq, 
 	 * on the queue:
 	 */
 	if (p->rt.run_list.prev != p->rt.run_list.next) {
-		requeue_task_rt(rq, p);
+		requeue_task_rt(rq, p, 0);
 		set_tsk_need_resched(p);
 	}
 }

Re: [sched-devel, patch-rfc] rework #2 of "prioritize non-migratable tasks over migratable ones"

[Dmitry Adamushko]

Finally, this new version compiles and boots (applied a minor compilation fix).
Functionality is not yet fully tested though.

It's on top of today's tip tree.

Any objections to this approach?


---
From: Dmitry Adamushko <dmitry.adamushko@gmail.com>
Subject: prioritize non-migratable tasks over migratable ones in a generic way

(1) handle in a generic way all cases when a newly woken-up task is
not migratable (not just a corner case when "rt_se->nr_cpus_allowed ==
1")

(2) if current is to be preempted, then make sure "p" will be picked
up by pick_next_task_rt().
i.e. move task's group at the head of its list as well.

currently, it's not a case for the group-scheduling case as described
here: http://www.ussg.iu.edu/hypermail/linux/kernel/0807.0/0134.html


Signed-off-by: Dmitry Adamushko <dmitry.adamushko@gmail.com>

---
diff --git a/kernel/sched_rt.c b/kernel/sched_rt.c
index 7c96147..7bc73e8 100644
--- a/kernel/sched_rt.c
+++ b/kernel/sched_rt.c
@@ -599,11 +599,7 @@ static void __enqueue_rt_entity(struct sched_rt_entity *rt_se)
 	if (group_rq && (rt_rq_throttled(group_rq) || !group_rq->rt_nr_running))
 		return;
 
-	if (rt_se->nr_cpus_allowed == 1)
-		list_add(&rt_se->run_list, queue);
-	else
-		list_add_tail(&rt_se->run_list, queue);
-
+	list_add_tail(&rt_se->run_list, queue);
 	__set_bit(rt_se_prio(rt_se), array->bitmap);
 
 	inc_rt_tasks(rt_se, rt_rq);
@@ -689,31 +685,33 @@ static void dequeue_task_rt(struct rq *rq, struct task_struct *p, int sleep)
  * followed by enqueue.
  */
 static
-void requeue_rt_entity(struct rt_rq *rt_rq, struct sched_rt_entity *rt_se)
+void requeue_rt_entity(struct rt_rq *rt_rq, struct sched_rt_entity *rt_se, int head)
 {
-	struct rt_prio_array *array = &rt_rq->active;
-
 	if (on_rt_rq(rt_se)) {
-		list_del_init(&rt_se->run_list);
-		list_add_tail(&rt_se->run_list,
-			      array->queue + rt_se_prio(rt_se));
+		struct rt_prio_array *array = &rt_rq->active;
+		struct list_head *queue = array->queue + rt_se_prio(rt_se);
+
+		if (head)
+			list_move(&rt_se->run_list, queue);
+		else
+			list_move_tail(&rt_se->run_list, queue);
 	}
 }
 
-static void requeue_task_rt(struct rq *rq, struct task_struct *p)
+static void requeue_task_rt(struct rq *rq, struct task_struct *p, int head)
 {
 	struct sched_rt_entity *rt_se = &p->rt;
 	struct rt_rq *rt_rq;
 
 	for_each_sched_rt_entity(rt_se) {
 		rt_rq = rt_rq_of_se(rt_se);
-		requeue_rt_entity(rt_rq, rt_se);
+		requeue_rt_entity(rt_rq, rt_se, head);
 	}
 }
 
 static void yield_task_rt(struct rq *rq)
 {
-	requeue_task_rt(rq, rq->curr);
+	requeue_task_rt(rq, rq->curr, 0);
 }
 
 #ifdef CONFIG_SMP
@@ -753,6 +751,30 @@ static int select_task_rq_rt(struct task_struct *p, int sync)
 	 */
 	return task_cpu(p);
 }
+
+static void check_preempt_equal_prio(struct rq *rq, struct task_struct *p)
+{
+	cpumask_t mask;
+
+	if (rq->curr->rt.nr_cpus_allowed == 1)
+		return;
+
+	if (p->rt.nr_cpus_allowed != 1
+	    && cpupri_find(&rq->rd->cpupri, p, &mask))
+		return;
+	
+	if (!cpupri_find(&rq->rd->cpupri, rq->curr, &mask))
+		return;
+
+	/*
+	 * There appears to be other cpus that can accept
+	 * current and none to run 'p', so lets reschedule
+	 * to try and push current away:
+	 */
+	requeue_task_rt(rq, p, 1);
+	resched_task(rq->curr);
+}
+
 #endif /* CONFIG_SMP */
 
 /*
@@ -778,18 +800,8 @@ static void check_preempt_curr_rt(struct rq *rq, struct task_struct *p)
 	 * to move current somewhere else, making room for our non-migratable
 	 * task.
 	 */
-	if((p->prio == rq->curr->prio)
-	   && p->rt.nr_cpus_allowed == 1
-	   && rq->curr->rt.nr_cpus_allowed != 1) {
-		cpumask_t mask;
-
-		if (cpupri_find(&rq->rd->cpupri, rq->curr, &mask))
-			/*
-			 * There appears to be other cpus that can accept
-			 * current, so lets reschedule to try and push it away
-			 */
-			resched_task(rq->curr);
-	}
+	if (p->prio == rq->curr->prio && !need_resched())
+		check_preempt_equal_prio(rq, p);
 #endif
 }
 
@@ -1415,7 +1427,7 @@ static void task_tick_rt(struct rq *rq, struct task_struct *p, int queued)
 	 * on the queue:
 	 */
 	if (p->rt.run_list.prev != p->rt.run_list.next) {
-		requeue_task_rt(rq, p);
+		requeue_task_rt(rq, p, 0);
 		set_tsk_need_resched(p);
 	}
 }

---


--Dmitry

Re: [sched-devel, patch-rfc] rework #2 of "prioritize non-migratable tasks over migratable ones"

[Dmitry Adamushko]

2008/7/1 Dmitry Adamushko <dmitry.adamushko@gmail.com>:
>
> Finally, this new version compiles and boots (applied a minor compilation fix).
> Functionality is not yet fully tested though.
>
> It's on top of today's tip tree.
>
> Any objections to this approach?

[ ping ]

Gregory or anyone else, any feedback on this one?

TIA,


>
> ---
> From: Dmitry Adamushko <dmitry.adamushko@gmail.com>
> Subject: prioritize non-migratable tasks over migratable ones in a generic way
>
> (1) handle in a generic way all cases when a newly woken-up task is
> not migratable (not just a corner case when "rt_se->nr_cpus_allowed ==
> 1")
>
> (2) if current is to be preempted, then make sure "p" will be picked
> up by pick_next_task_rt().
> i.e. move task's group at the head of its list as well.
>
> currently, it's not a case for the group-scheduling case as described
> here: http://www.ussg.iu.edu/hypermail/linux/kernel/0807.0/0134.html
>
>
> Signed-off-by: Dmitry Adamushko <dmitry.adamushko@gmail.com>
>
> ---
> diff --git a/kernel/sched_rt.c b/kernel/sched_rt.c
> index 7c96147..7bc73e8 100644
> --- a/kernel/sched_rt.c
> +++ b/kernel/sched_rt.c
> @@ -599,11 +599,7 @@ static void __enqueue_rt_entity(struct sched_rt_entity *rt_se)
>        if (group_rq && (rt_rq_throttled(group_rq) || !group_rq->rt_nr_running))
>                return;
>
> -       if (rt_se->nr_cpus_allowed == 1)
> -               list_add(&rt_se->run_list, queue);
> -       else
> -               list_add_tail(&rt_se->run_list, queue);
> -
> +       list_add_tail(&rt_se->run_list, queue);
>        __set_bit(rt_se_prio(rt_se), array->bitmap);
>
>        inc_rt_tasks(rt_se, rt_rq);
> @@ -689,31 +685,33 @@ static void dequeue_task_rt(struct rq *rq, struct task_struct *p, int sleep)
>  * followed by enqueue.
>  */
>  static
> -void requeue_rt_entity(struct rt_rq *rt_rq, struct sched_rt_entity *rt_se)
> +void requeue_rt_entity(struct rt_rq *rt_rq, struct sched_rt_entity *rt_se, int head)
>  {
> -       struct rt_prio_array *array = &rt_rq->active;
> -
>        if (on_rt_rq(rt_se)) {
> -               list_del_init(&rt_se->run_list);
> -               list_add_tail(&rt_se->run_list,
> -                             array->queue + rt_se_prio(rt_se));
> +               struct rt_prio_array *array = &rt_rq->active;
> +               struct list_head *queue = array->queue + rt_se_prio(rt_se);
> +
> +               if (head)
> +                       list_move(&rt_se->run_list, queue);
> +               else
> +                       list_move_tail(&rt_se->run_list, queue);
>        }
>  }
>
> -static void requeue_task_rt(struct rq *rq, struct task_struct *p)
> +static void requeue_task_rt(struct rq *rq, struct task_struct *p, int head)
>  {
>        struct sched_rt_entity *rt_se = &p->rt;
>        struct rt_rq *rt_rq;
>
>        for_each_sched_rt_entity(rt_se) {
>                rt_rq = rt_rq_of_se(rt_se);
> -               requeue_rt_entity(rt_rq, rt_se);
> +               requeue_rt_entity(rt_rq, rt_se, head);
>        }
>  }
>
>  static void yield_task_rt(struct rq *rq)
>  {
> -       requeue_task_rt(rq, rq->curr);
> +       requeue_task_rt(rq, rq->curr, 0);
>  }
>
>  #ifdef CONFIG_SMP
> @@ -753,6 +751,30 @@ static int select_task_rq_rt(struct task_struct *p, int sync)
>         */
>        return task_cpu(p);
>  }
> +
> +static void check_preempt_equal_prio(struct rq *rq, struct task_struct *p)
> +{
> +       cpumask_t mask;
> +
> +       if (rq->curr->rt.nr_cpus_allowed == 1)
> +               return;
> +
> +       if (p->rt.nr_cpus_allowed != 1
> +           && cpupri_find(&rq->rd->cpupri, p, &mask))
> +               return;
> +
> +       if (!cpupri_find(&rq->rd->cpupri, rq->curr, &mask))
> +               return;
> +
> +       /*
> +        * There appears to be other cpus that can accept
> +        * current and none to run 'p', so lets reschedule
> +        * to try and push current away:
> +        */
> +       requeue_task_rt(rq, p, 1);
> +       resched_task(rq->curr);
> +}
> +
>  #endif /* CONFIG_SMP */
>
>  /*
> @@ -778,18 +800,8 @@ static void check_preempt_curr_rt(struct rq *rq, struct task_struct *p)
>         * to move current somewhere else, making room for our non-migratable
>         * task.
>         */
> -       if((p->prio == rq->curr->prio)
> -          && p->rt.nr_cpus_allowed == 1
> -          && rq->curr->rt.nr_cpus_allowed != 1) {
> -               cpumask_t mask;
> -
> -               if (cpupri_find(&rq->rd->cpupri, rq->curr, &mask))
> -                       /*
> -                        * There appears to be other cpus that can accept
> -                        * current, so lets reschedule to try and push it away
> -                        */
> -                       resched_task(rq->curr);
> -       }
> +       if (p->prio == rq->curr->prio && !need_resched())
> +               check_preempt_equal_prio(rq, p);
>  #endif
>  }
>
> @@ -1415,7 +1427,7 @@ static void task_tick_rt(struct rq *rq, struct task_struct *p, int queued)
>         * on the queue:
>         */
>        if (p->rt.run_list.prev != p->rt.run_list.next) {
> -               requeue_task_rt(rq, p);
> +               requeue_task_rt(rq, p, 0);
>                set_tsk_need_resched(p);
>        }
>  }
>
> ---
>
>
> --Dmitry
>
>
>

-- 
Best regards,
Dmitry Adamushko

Re: [sched-devel, patch-rfc] rework #2 of "prioritize non-migratable tasks over migratable ones"

[Steven Rostedt]

On Mon, 14 Jul 2008, Dmitry Adamushko wrote:

>
> 2008/7/1 Dmitry Adamushko <dmitry.adamushko@gmail.com>:
> >
> > Finally, this new version compiles and boots (applied a minor compilation fix).
> > Functionality is not yet fully tested though.
> >
> > It's on top of today's tip tree.
> >
> > Any objections to this approach?
>
> [ ping ]
>
> Gregory or anyone else, any feedback on this one?
>

A quick look at this, looks fine with me.

-- Steve

Re: [sched-devel, patch-rfc] rework #2 of "prioritize non-migratable tasks over migratable ones"

[Gregory Haskins]

>>> On Mon, Jul 14, 2008 at 10:50 AM, in message
<b647ffbd0807140750r54e5139aj7b25a28ef7ca247e@mail.gmail.com>, "Dmitry
Adamushko" <dmitry.adamushko@gmail.com> wrote: 
> 2008/7/1 Dmitry Adamushko <dmitry.adamushko@gmail.com>:
>>
>> Finally, this new version compiles and boots (applied a minor compilation 
> fix).
>> Functionality is not yet fully tested though.
>>
>> It's on top of today's tip tree.
>>
>> Any objections to this approach?
> 
> [ ping ]
> 
> Gregory or anyone else, any feedback on this one?

Hi Dmitry,
  A brief inspection of the patch looks good.  I am still not convinced we need to worry
about this condition anymore, however.  The fact that we cannot guarantee that the
migratable task will start running seems that it violates *something*, even if it is not
technically called out in posix. :)

But if we are going to solve that problem, this seems like a decent way to do it.

Regards,
-Greg

> 
> TIA,
> 
> 
>>
>> ---
>> From: Dmitry Adamushko <dmitry.adamushko@gmail.com>
>> Subject: prioritize non-migratable tasks over migratable ones in a generic 
> way
>>
>> (1) handle in a generic way all cases when a newly woken-up task is
>> not migratable (not just a corner case when "rt_se->nr_cpus_allowed ==
>> 1")
>>
>> (2) if current is to be preempted, then make sure "p" will be picked
>> up by pick_next_task_rt().
>> i.e. move task's group at the head of its list as well.
>>
>> currently, it's not a case for the group-scheduling case as described
>> here: http://www.ussg.iu.edu/hypermail/linux/kernel/0807.0/0134.html
>>
>>
>> Signed-off-by: Dmitry Adamushko <dmitry.adamushko@gmail.com>
>>
>> ---
>> diff --git a/kernel/sched_rt.c b/kernel/sched_rt.c
>> index 7c96147..7bc73e8 100644
>> --- a/kernel/sched_rt.c
>> +++ b/kernel/sched_rt.c
>> @@ -599,11 +599,7 @@ static void __enqueue_rt_entity(struct sched_rt_entity 
> *rt_se)
>>        if (group_rq && (rt_rq_throttled(group_rq) || !group_rq->rt_nr_running))
>>                return;
>>
>> -       if (rt_se->nr_cpus_allowed == 1)
>> -               list_add(&rt_se->run_list, queue);
>> -       else
>> -               list_add_tail(&rt_se->run_list, queue);
>> -
>> +       list_add_tail(&rt_se->run_list, queue);
>>        __set_bit(rt_se_prio(rt_se), array->bitmap);
>>
>>        inc_rt_tasks(rt_se, rt_rq);
>> @@ -689,31 +685,33 @@ static void dequeue_task_rt(struct rq *rq, struct 
> task_struct *p, int sleep)
>>  * followed by enqueue.
>>  */
>>  static
>> -void requeue_rt_entity(struct rt_rq *rt_rq, struct sched_rt_entity *rt_se)
>> +void requeue_rt_entity(struct rt_rq *rt_rq, struct sched_rt_entity *rt_se, 
> int head)
>>  {
>> -       struct rt_prio_array *array = &rt_rq->active;
>> -
>>        if (on_rt_rq(rt_se)) {
>> -               list_del_init(&rt_se->run_list);
>> -               list_add_tail(&rt_se->run_list,
>> -                             array->queue + rt_se_prio(rt_se));
>> +               struct rt_prio_array *array = &rt_rq->active;
>> +               struct list_head *queue = array->queue + rt_se_prio(rt_se);
>> +
>> +               if (head)
>> +                       list_move(&rt_se->run_list, queue);
>> +               else
>> +                       list_move_tail(&rt_se->run_list, queue);
>>        }
>>  }
>>
>> -static void requeue_task_rt(struct rq *rq, struct task_struct *p)
>> +static void requeue_task_rt(struct rq *rq, struct task_struct *p, int head)
>>  {
>>        struct sched_rt_entity *rt_se = &p->rt;
>>        struct rt_rq *rt_rq;
>>
>>        for_each_sched_rt_entity(rt_se) {
>>                rt_rq = rt_rq_of_se(rt_se);
>> -               requeue_rt_entity(rt_rq, rt_se);
>> +               requeue_rt_entity(rt_rq, rt_se, head);
>>        }
>>  }
>>
>>  static void yield_task_rt(struct rq *rq)
>>  {
>> -       requeue_task_rt(rq, rq->curr);
>> +       requeue_task_rt(rq, rq->curr, 0);
>>  }
>>
>>  #ifdef CONFIG_SMP
>> @@ -753,6 +751,30 @@ static int select_task_rq_rt(struct task_struct *p, int 
> sync)
>>         */
>>        return task_cpu(p);
>>  }
>> +
>> +static void check_preempt_equal_prio(struct rq *rq, struct task_struct *p)
>> +{
>> +       cpumask_t mask;
>> +
>> +       if (rq->curr->rt.nr_cpus_allowed == 1)
>> +               return;
>> +
>> +       if (p->rt.nr_cpus_allowed != 1
>> +           && cpupri_find(&rq->rd->cpupri, p, &mask))
>> +               return;
>> +
>> +       if (!cpupri_find(&rq->rd->cpupri, rq->curr, &mask))
>> +               return;
>> +
>> +       /*
>> +        * There appears to be other cpus that can accept
>> +        * current and none to run 'p', so lets reschedule
>> +        * to try and push current away:
>> +        */
>> +       requeue_task_rt(rq, p, 1);
>> +       resched_task(rq->curr);
>> +}
>> +
>>  #endif /* CONFIG_SMP */
>>
>>  /*
>> @@ -778,18 +800,8 @@ static void check_preempt_curr_rt(struct rq *rq, struct 
> task_struct *p)
>>         * to move current somewhere else, making room for our non-migratable
>>         * task.
>>         */
>> -       if((p->prio == rq->curr->prio)
>> -          && p->rt.nr_cpus_allowed == 1
>> -          && rq->curr->rt.nr_cpus_allowed != 1) {
>> -               cpumask_t mask;
>> -
>> -               if (cpupri_find(&rq->rd->cpupri, rq->curr, &mask))
>> -                       /*
>> -                        * There appears to be other cpus that can accept
>> -                        * current, so lets reschedule to try and push it 
> away
>> -                        */
>> -                       resched_task(rq->curr);
>> -       }
>> +       if (p->prio == rq->curr->prio && !need_resched())
>> +               check_preempt_equal_prio(rq, p);
>>  #endif
>>  }
>>
>> @@ -1415,7 +1427,7 @@ static void task_tick_rt(struct rq *rq, struct 
> task_struct *p, int queued)
>>         * on the queue:
>>         */
>>        if (p->rt.run_list.prev != p->rt.run_list.next) {
>> -               requeue_task_rt(rq, p);
>> +               requeue_task_rt(rq, p, 0);
>>                set_tsk_need_resched(p);
>>        }
>>  }
>>
>> ---
>>
>>
>> --Dmitry
>>
>>
>>

Re: [sched-devel, patch-rfc] rework #2 of "prioritize non-migratable tasks over migratable ones"

[Dmitry Adamushko]

2008/7/15 Gregory Haskins <ghaskins@novell.com>:
>>>> On Mon, Jul 14, 2008 at 10:50 AM, in message
> <b647ffbd0807140750r54e5139aj7b25a28ef7ca247e@mail.gmail.com>, "Dmitry
> Adamushko" <dmitry.adamushko@gmail.com> wrote:
>> 2008/7/1 Dmitry Adamushko <dmitry.adamushko@gmail.com>:
>>>
>>> Finally, this new version compiles and boots (applied a minor compilation
>> fix).
>>> Functionality is not yet fully tested though.
>>>
>>> It's on top of today's tip tree.
>>>
>>> Any objections to this approach?
>>
>> [ ping ]
>>
>> Gregory or anyone else, any feedback on this one?
>
> Hi Dmitry,
>  A brief inspection of the patch looks good.  I am still not convinced we need to worry
> about this condition anymore, however.

Neither am I.

It's just that (I think):

(a) either we remove this whole new 'concept' (as it's implemented
now) completely;
(b) or do it in a more generic way (like with this patch).

Thanks for the feedback!

>
> Regards,
> -Greg

-- 
Best regards,
Dmitry Adamushko

Re: [sched-devel, patch-rfc] rework #2 of "prioritize non-migratable tasks over migratable ones"

[Ingo Molnar]

* Dmitry Adamushko <dmitry.adamushko@gmail.com> wrote:

> Finally, this new version compiles and boots (applied a minor 
> compilation fix). Functionality is not yet fully tested though.
> 
> It's on top of today's tip tree.
> 
> Any objections to this approach?

applied to tip/sched/devel - thanks Dmitry.

	Ingo