Skip to contentSkip to conversation

rework #2 of "prioritize non-migratable tasks over migratable ones"

sched-devel, patch-rfc Applied Dmitry Adamushko 4 participants 6 messages Jul 1, 2008 – Jul 18, 2008

lore
Share & tools
lore
#0 lore raw

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 <[email protected]>
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

---

kernel/sched_rt.c +39 -27
index 7c96147..7bc73e8 100644
@@ -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

#1 lore raw

2008/7/1 Dmitry Adamushko <[email protected]>:

Quoted 7 lines

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,

Quoted 149 lines

---
From: Dmitry Adamushko <[email protected]>
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

---

kernel/sched_rt.c +39 -29
index 7c96147..7bc73e8 100644
@@ -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
Signature · 2 lines
Best regards,
Dmitry Adamushko
#2 lore raw
Quoted 6 lines · Dmitry Adamushko

2008/7/1 Dmitry Adamushko <[email protected]>:

Quoted 4 lines

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

Quoted 8 lines
Quoted 3 lines

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

#3 lore raw
In reply to Dmitry Adamushko
Quoted 1 line
Quoted 1 line
Quoted 1 line

On Mon, Jul 14, 2008 at 10:50 AM, in message

<[email protected]>, "Dmitry

Quoted 6 lines · Adamushko

2008/7/1 Dmitry Adamushko <[email protected]>:

Quoted 2 lines

Finally, this new version compiles and boots (applied a minor compilation

fix).

Quoted 2 lines

Functionality is not yet fully tested though.

Quoted 7 lines
Quoted 3 lines

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

Quoted 161 lines

TIA,

Quoted 4 lines

---
From: Dmitry Adamushko <[email protected]>
Subject: prioritize non-migratable tasks over migratable ones in a generic

way

kernel/sched_rt.c static void __enqueue_rt_entity(struct sched_rt_entity

Quoted 21 lines

(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

---

kernel/sched_rt.c +0 -0
index 7c96147..7bc73e8 100644
@@ -599,11 +599,7 @@ static void __enqueue_rt_entity(struct sched_rt_entity 

*rt_se)

kernel/sched_rt.c static void dequeue_task_rt(struct rq *rq, struct

Quoted 13 lines
kernel/sched_rt.c +1 -5
       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)

kernel/sched_rt.c static void dequeue_task_rt(struct rq *rq, struct task_struct *p, int sleep)

Quoted 5 lines
kernel/sched_rt.c +1 -1
 * 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)

kernel/sched_rt.c static int select_task_rq_rt(struct task_struct *p, int

Quoted 38 lines
kernel/sched_rt.c +10 -8
 {
-       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)

kernel/sched_rt.c static void check_preempt_curr_rt(struct rq *rq, struct

Quoted 31 lines
kernel/sched_rt.c +24 -0
        */
       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)

kernel/sched_rt.c static void check_preempt_curr_rt(struct rq *rq, struct task_struct *p)

Quoted 12 lines
kernel/sched_rt.c +0 -9
        * 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

kernel/sched_rt.c static void task_tick_rt(struct rq *rq, struct

Quoted 9 lines
kernel/sched_rt.c +2 -3
-                        */
-                       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)

Quoted 16 lines
kernel/sched_rt.c +1 -3
        * 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
#4 lore raw

2008/7/15 Gregory Haskins <[email protected]>:

Quoted 13 lines
Quoted 1 line
Quoted 1 line
Quoted 1 line

On Mon, Jul 14, 2008 at 10:50 AM, in message

<[email protected]>, "Dmitry

Quoted 10 lines · Adamushko

2008/7/1 Dmitry Adamushko <[email protected]>:

Quoted 2 lines

Finally, this new version compiles and boots (applied a minor compilation

fix).

Quoted 5 lines

Functionality is not yet fully tested though.

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

Any objections to this approach?

Quoted 7 lines
Quoted 3 lines

[ 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!

Quoted 3 lines

Regards,
-Greg

Signature · 2 lines
Best regards,
Dmitry Adamushko
Applied
#5 lore raw
In reply to Dmitry Adamushko
Quoted 6 lines · 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?

applied to tip/sched/devel - thanks Dmitry.

Ingo