Skip to contentSkip to conversation

rt: Increase/decrease the nr of migratory tasks when enabling/disabling migration

PATCH Superseded Daniel Bristot de Oliveira 2 participants 11 messages Jun 21, 2017 – Jun 26, 2017 Scheduler

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

cover rt: Increase/decrease the nr of migratory tasks when enabling/disabling migration

This should be the v2 of the:
[RFC] rt: Some fixes for migrate_disable/enable

However, migrate_disable/enable was reworked during the 4.11-rt window, so it turns out that 2 of 3 problems were fixed. Good! :-)

But there is still one problem, which is the dl/rt_nr_migratory inc/dec.

The problem is reproducible with the following command [in a 4 CPU box]:

  # chrt -f 1 taskset -c 3 cat /dev/full | taskset -c 0-2 grep 'batman'

By applying only the patch 1/2, it is possible to see the problem with the following command:

  # cat /proc/sched_debug | grep rt_nr_migratory
    .rt_nr_migratory               : 18446744073709542849
    .rt_nr_migratory               : 18446744073709538566
    .rt_nr_migratory               : 18446744073709548257
    .rt_nr_migratory               : 0

The detailed description of the bug, and the fix, is in the log of the patch 2/2.

Changes from RFC:

- The problems addressed in the patches:
x rt: Update nr_cpus_allowed if the affinity of a task changes while its
migration is disabled
x rt: Checks if task needs migration when re-enabling migration

were fixed, so these patches are not needed anymore, while patch:

x rt: Increase/decrease the nr of migratory tasks when
enabling/disabling migration

is still needed, so it was reworked for the new implementation.

- The patch showing the rt/dl_nr_migratory was added.

Daniel Bristot de Oliveira (2):
sched/debug: Inform the number of rt/dl task that can migrate
rt: Increase/decrease the nr of migratory tasks when
enabling/disabling migration

 kernel/sched/core.c  | 40 ++++++++++++++++++++++++++++++++--------
 kernel/sched/debug.c | 15 +++++++++++++--
 2 files changed, 45 insertions(+), 10 deletions(-)
Signature · 1 line
2.9.4
#1 lore raw

PATCH 1/2 sched/debug: Inform the number of rt/dl task that can migrate

Add the value of the rt_rq.rt_nr_migratory and dl_rq.dl_nr_migratory to the sched_debug output, for instance:

rt_rq[0]:

  .rt_nr_running                 : 2
  .rt_nr_migratory               : 1     <--- Like this
  .rt_throttled                  : 0
  .rt_time                       : 828.645877
  .rt_runtime                    : 1000.000000

This is useful to debug problems related to the dl/rt schedulers.

This also fixes the format of some variables, that were unsigned, rather than signed.

[ I am sending this patch to be able to demonstrate       ]
[ the problem in the kernel-rt, however, this patch can   ]
[ be merged in the vanilla kernel, if people decide it    ]
[ is worth having it.                                     ]
Cc Luis Claudio R. Goncalves <[email protected]>
Cc Clark Williams <[email protected]>
Cc Luiz Capitulino <[email protected]>
Cc Sebastian Andrzej Siewior <[email protected]>
Cc Thomas Gleixner <[email protected]>
Cc Steven Rostedt <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc linux-rt-users <[email protected]>
---
 kernel/sched/debug.c | 15 +++++++++++++--
 1 file changed, 13 insertions(+), 2 deletions(-)
kernel/sched/debug.c +13 -2
index 0e2af53..2bdeda8 100644
@@ -552,15 +552,19 @@ void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq)
 
 #define P(x) \
 	SEQ_printf(m, "  .%-30s: %Ld\n", #x, (long long)(rt_rq->x))
+#define PU(x) \
+	SEQ_printf(m, "  .%-30s: %lu\n", #x, (unsigned long)(rt_rq->x))
 #define PN(x) \
 	SEQ_printf(m, "  .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x))
 
-	P(rt_nr_running);
+	PU(rt_nr_running);
+	PU(rt_nr_migratory);
 	P(rt_throttled);
 	PN(rt_time);
 	PN(rt_runtime);
 
 #undef PN
+#undef PU
 #undef P
 }
 
@@ -569,7 +573,12 @@ void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq)
 	struct dl_bw *dl_bw;
 
 	SEQ_printf(m, "\ndl_rq[%d]:\n", cpu);
-	SEQ_printf(m, "  .%-30s: %ld\n", "dl_nr_running", dl_rq->dl_nr_running);
+
+#define PU(x) \
+	SEQ_printf(m, "  .%-30s: %lu\n", #x, (unsigned long)(dl_rq->x))
+
+	PU(dl_nr_running);
+	PU(dl_nr_migratory);
 #ifdef CONFIG_SMP
 	dl_bw = &cpu_rq(cpu)->rd->dl_bw;
 #else
@@ -577,6 +586,8 @@ void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq)
 #endif
 	SEQ_printf(m, "  .%-30s: %lld\n", "dl_bw->bw", dl_bw->bw);
 	SEQ_printf(m, "  .%-30s: %lld\n", "dl_bw->total_bw", dl_bw->total_bw);
+
+#undef PU
 }
 
 extern __read_mostly int sched_clock_running;
Signature · 1 line
2.9.4
#2 lore raw

Re: [PATCH 1/2] sched/debug: Inform the number of rt/dl task that can migrate

Quoted 12 lines · Daniel Bristot de Oliveira

Add the value of the rt_rq.rt_nr_migratory and dl_rq.dl_nr_migratory to the sched_debug output, for instance:

rt_rq[0]:

  .rt_nr_running                 : 2
  .rt_nr_migratory               : 1     <--- Like this
  .rt_throttled                  : 0
  .rt_time                       : 828.645877
  .rt_runtime                    : 1000.000000

This is useful to debug problems related to the dl/rt schedulers.

Quoted 7 lines

This also fixes the format of some variables, that were unsigned, rather than signed.

[ I am sending this patch to be able to demonstrate       ]
[ the problem in the kernel-rt, however, this patch can   ]
[ be merged in the vanilla kernel, if people decide it    ]
[ is worth having it.                                     ]

Note that this patch breaks in various cross-builds - here's the Alpha defconfig build:

/home/mingo/tip/kernel/sched/debug.c: In function 'print_rt_rq':
/home/mingo/tip/kernel/sched/debug.c:556:60: error: 'struct rt_rq' has no member named 'rt_nr_migratory'
SEQ_printf(m, " .%-30s: %lu\n", #x, (unsigned long)(rt_rq->x))

                                                            ^

/home/mingo/tip/kernel/sched/debug.c:33:17: note: in definition of macro 'SEQ_printf'
seq_printf(m, x); \
...

Thanks,

Ingo

#3 lore raw

Re: [PATCH 1/2] sched/debug: Inform the number of rt/dl task that can migrate

Quoted 7 lines · Ingo Molnar

Note that this patch breaks in various cross-builds - here's the Alpha defconfig build:

/home/mingo/tip/kernel/sched/debug.c: In function 'print_rt_rq':
/home/mingo/tip/kernel/sched/debug.c:556:60: error: 'struct rt_rq' has no member named 'rt_nr_migratory'
SEQ_printf(m, " .%-30s: %lu\n", #x, (unsigned long)(rt_rq->x))

Quoted 5 lines
                                                            ^

/home/mingo/tip/kernel/sched/debug.c:33:17: note: in definition of macro 'SEQ_printf'
seq_printf(m, x); \
...

Oops...

The rt_nr_migratory is conditioned to:

#ifdef CONFIG_SMP

Sorry for the miss attention, I will fix this.

-- Daniel

#4 lore raw

PATCH 2/2 rt: Increase/decrease the nr of migratory tasks when enabling/disabling migration

In reply to Daniel Bristot de Oliveira

There is a problem in the migrate_disable()/enable() implementation regarding the number of migratory tasks in the rt/dl RQs. The problem is the following:

When a task is attached to the rt runqueue, it is checked if it either can run in more than one CPU, or if it is with migration disable. If either check is true, the rt_rq->rt_nr_migratory counter is not increased. The counter increases otherwise.

When the task is detached, the same check is done. If either check is true, the rt_rq->rt_nr_migratory counter is not decreased. The counter decreases otherwise. The same check is done in the dl scheduler.

One important thing is that, migrate disable/enable does not touch this counter for tasks attached to the rt rq. So suppose the following chain of events.

Assumptions:

Task A is the only runnable task in A      Task B runs on the CPU B
Task A runs on CFS (non-rt)                Task B has RT priority
Thus, rt_nr_migratory is 0                 B is running

Task A can run on all CPUS.

Timeline:
CPU A/TASK A CPU B/TASK B
A takes the rt mutex X .
A disables migration .
. B tries to take the rt mutex X
. As it is held by A {
. A inherits the rt priority of B
. A is dequeued from CFS RQ of CPU A
. A is enqueued in the RT RQ of CPU A
. As migration is disabled
. rt_nr_migratory in A is not increased
.
A enables migration
A releases the rt mutex X {
A returns to its original priority
A ask to be dequeued from RT RQ {
As migration is now enabled and it can run on all CPUS {
rt_nr_migratory should be decreased
As rt_nr_migratory is 0, rt_nr_migratory under flows
}
}

This variable is important because it notifies if there are more than one runnable & migratory task in the runqueue. If there are more than one tasks, the rt_rq is set as overloaded, and then tries to migrate some tasks. This rule is important to keep the scheduler working conserving, that is, in a system with M CPUs, the M highest priority tasks should be running.

As rt_nr_migratory is unsigned, it will become > 0, notifying that the RQ is overloaded, activating pushing mechanism without need.

This patch fixes this problem by decreasing/increasing the rt/dl_nr_migratory in the migrate disable/enable operations.

Cc Luis Claudio R. Goncalves <[email protected]>
Cc Clark Williams <[email protected]>
Cc Luiz Capitulino <[email protected]>
Cc Sebastian Andrzej Siewior <[email protected]>
Cc Thomas Gleixner <[email protected]>
Cc Steven Rostedt <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc linux-rt-users <[email protected]>
---
 kernel/sched/core.c | 40 ++++++++++++++++++++++++++++++++--------
 1 file changed, 32 insertions(+), 8 deletions(-)
kernel/sched/core.c +32 -8
index ce34e4f..2b78189 100644
@@ -7569,6 +7569,9 @@ const u32 sched_prio_to_wmult[40] = {
 void migrate_disable(void)
 {
 	struct task_struct *p = current;
+	struct rq *rq;
+	struct rq_flags rf;
+
 
 	if (in_atomic() || irqs_disabled()) {
 #ifdef CONFIG_SCHED_DEBUG
@@ -7593,10 +7596,21 @@ void migrate_disable(void)
 	preempt_disable();
 	preempt_lazy_disable();
 	pin_current_cpu();
-	p->migrate_disable = 1;
 
-	p->cpus_ptr = cpumask_of(smp_processor_id());
+	rq = task_rq_lock(p, &rf);
+	if (unlikely((p->sched_class == &rt_sched_class ||
+		      p->sched_class == &dl_sched_class) &&
+		      p->nr_cpus_allowed > 1)) {
+		if (p->sched_class == &rt_sched_class)
+			task_rq(p)->rt.rt_nr_migratory--;
+		else
+			task_rq(p)->dl.dl_nr_migratory--;
+	}
 	p->nr_cpus_allowed = 1;
+	task_rq_unlock(rq, p, &rf);
+	p->cpus_ptr = cpumask_of(smp_processor_id());
+	p->migrate_disable = 1;
+
 
 	preempt_enable();
 }
@@ -7605,6 +7619,9 @@ EXPORT_SYMBOL(migrate_disable);
 void migrate_enable(void)
 {
 	struct task_struct *p = current;
+	struct rq *rq;
+	struct rq_flags rf;
+
 
 	if (in_atomic() || irqs_disabled()) {
 #ifdef CONFIG_SCHED_DEBUG
@@ -7628,17 +7645,24 @@ void migrate_enable(void)
 
 	preempt_disable();
 
-	p->cpus_ptr = &p->cpus_mask;
-	p->nr_cpus_allowed = cpumask_weight(&p->cpus_mask);
 	p->migrate_disable = 0;
+	p->cpus_ptr = &p->cpus_mask;
 
-	if (p->migrate_disable_update) {
-		struct rq *rq;
-		struct rq_flags rf;
+	rq = task_rq_lock(p, &rf);
+	p->nr_cpus_allowed = cpumask_weight(&p->cpus_mask);
+	if (unlikely((p->sched_class == &rt_sched_class ||
+		      p->sched_class == &dl_sched_class) &&
+		      p->nr_cpus_allowed > 1)) {
+		if (p->sched_class == &rt_sched_class)
+			task_rq(p)->rt.rt_nr_migratory++;
+		else
+			task_rq(p)->dl.dl_nr_migratory++;
+	}
+	task_rq_unlock(rq, p, &rf);
 
+	if (unlikely(p->migrate_disable_update)) {
 		rq = task_rq_lock(p, &rf);
 		update_rq_clock(rq);
-
 		__do_set_cpus_allowed_tail(p, &p->cpus_mask);
 		task_rq_unlock(rq, p, &rf);
 
Signature · 1 line
2.9.4
#5 lore raw
Quoted 56 lines · Daniel Bristot de Oliveira

void migrate_disable(void)
{

kernel/sched/core.c +21 -7
 	struct task_struct *p = current;
+	struct rq *rq;
+	struct rq_flags rf;
+
 
 	if (in_atomic() || irqs_disabled()) {
 #ifdef CONFIG_SCHED_DEBUG
@@ -7593,10 +7596,21 @@ void migrate_disable(void)
 	preempt_disable();
 	preempt_lazy_disable();
 	pin_current_cpu();
-	p->migrate_disable = 1;
 
-	p->cpus_ptr = cpumask_of(smp_processor_id());
+	rq = task_rq_lock(p, &rf);
+	if (unlikely((p->sched_class == &rt_sched_class ||
+		      p->sched_class == &dl_sched_class) &&
+		      p->nr_cpus_allowed > 1)) {
+		if (p->sched_class == &rt_sched_class)
+			task_rq(p)->rt.rt_nr_migratory--;
+		else
+			task_rq(p)->dl.dl_nr_migratory--;
+	}
 	p->nr_cpus_allowed = 1;
+	task_rq_unlock(rq, p, &rf);
+	p->cpus_ptr = cpumask_of(smp_processor_id());
+	p->migrate_disable = 1;
+
 
 	preempt_enable();
 }
@@ -7605,6 +7619,9 @@ EXPORT_SYMBOL(migrate_disable);
 void migrate_enable(void)
 {
 	struct task_struct *p = current;
+	struct rq *rq;
+	struct rq_flags rf;
+
 
 	if (in_atomic() || irqs_disabled()) {
 #ifdef CONFIG_SCHED_DEBUG
@@ -7628,17 +7645,24 @@ void migrate_enable(void)
 
 	preempt_disable();
 
-	p->cpus_ptr = &p->cpus_mask;
-	p->nr_cpus_allowed = cpumask_weight(&p->cpus_mask);
 	p->migrate_disable = 0;
+	p->cpus_ptr = &p->cpus_mask;
 
-	if (p->migrate_disable_update) {
-		struct rq *rq;
-		struct rq_flags rf;
+	rq = task_rq_lock(p, &rf);

kernel/sched/core.c void migrate_enable(void) · line 7661

Quoted 10 lines
kernel/sched/core.c +10 -0
+	p->nr_cpus_allowed = cpumask_weight(&p->cpus_mask);
+	if (unlikely((p->sched_class == &rt_sched_class ||
+		      p->sched_class == &dl_sched_class) &&
+		      p->nr_cpus_allowed > 1)) {
+		if (p->sched_class == &rt_sched_class)
+			task_rq(p)->rt.rt_nr_migratory++;
+		else
+			task_rq(p)->dl.dl_nr_migratory++;
+	}
+	task_rq_unlock(rq, p, &rf);

The fix looks good to me, but AFAICS the repeat pattern introduced here could be factored out into a helper function instead, right?

Thanks,

Ingo

#6 lore raw
Quoted 64 lines · Ingo Molnar
Quoted 61 lines · Daniel Bristot de Oliveira

void migrate_disable(void)
{

kernel/sched/core.c +26 -7
 	struct task_struct *p = current;
+	struct rq *rq;
+	struct rq_flags rf;
+
 
 	if (in_atomic() || irqs_disabled()) {
 #ifdef CONFIG_SCHED_DEBUG
@@ -7593,10 +7596,21 @@ void migrate_disable(void)
 	preempt_disable();
 	preempt_lazy_disable();
 	pin_current_cpu();
-	p->migrate_disable = 1;
 
-	p->cpus_ptr = cpumask_of(smp_processor_id());
+	rq = task_rq_lock(p, &rf);
+	if (unlikely((p->sched_class == &rt_sched_class ||
+		      p->sched_class == &dl_sched_class) &&
+		      p->nr_cpus_allowed > 1)) {
+		if (p->sched_class == &rt_sched_class)
+			task_rq(p)->rt.rt_nr_migratory--;
+		else
+			task_rq(p)->dl.dl_nr_migratory--;
+	}
 	p->nr_cpus_allowed = 1;
+	task_rq_unlock(rq, p, &rf);
+	p->cpus_ptr = cpumask_of(smp_processor_id());
+	p->migrate_disable = 1;
+
 
 	preempt_enable();
 }
@@ -7605,6 +7619,9 @@ EXPORT_SYMBOL(migrate_disable);
 void migrate_enable(void)
 {
 	struct task_struct *p = current;
+	struct rq *rq;
+	struct rq_flags rf;
+
 
 	if (in_atomic() || irqs_disabled()) {
 #ifdef CONFIG_SCHED_DEBUG
@@ -7628,17 +7645,24 @@ void migrate_enable(void)
 
 	preempt_disable();
 
-	p->cpus_ptr = &p->cpus_mask;
-	p->nr_cpus_allowed = cpumask_weight(&p->cpus_mask);
 	p->migrate_disable = 0;
+	p->cpus_ptr = &p->cpus_mask;
 
-	if (p->migrate_disable_update) {
-		struct rq *rq;
-		struct rq_flags rf;
+	rq = task_rq_lock(p, &rf);
+	p->nr_cpus_allowed = cpumask_weight(&p->cpus_mask);
+	if (unlikely((p->sched_class == &rt_sched_class ||
+		      p->sched_class == &dl_sched_class) &&
+		      p->nr_cpus_allowed > 1)) {
+		if (p->sched_class == &rt_sched_class)
Quoted 8 lines

kernel/sched/core.c void migrate_enable(void) · line 7661

Quoted 5 lines
kernel/sched/core.c +5 -0
+			task_rq(p)->rt.rt_nr_migratory++;
+		else
+			task_rq(p)->dl.dl_nr_migratory++;
+	}
+	task_rq_unlock(rq, p, &rf);

The fix looks good to me, but AFAICS the repeat pattern introduced here could be factored out into a helper function instead, right?

Like:

static inline int task_in_rt_class(struct task_struct *p)

{
	return p->sched_class == &rt_sched_class;
}

static inline int task_in_dl_class(struct task_struct *p)

{
	return p->sched_class == &dl_sched_class;
}

?

Thanks!

-- Daniel

#7 lore raw
Quoted 81 lines · Daniel Bristot de Oliveira
Quoted 64 lines · Ingo Molnar
Quoted 61 lines · Daniel Bristot de Oliveira

void migrate_disable(void)
{

kernel/sched/core.c +26 -7
 	struct task_struct *p = current;
+	struct rq *rq;
+	struct rq_flags rf;
+
 
 	if (in_atomic() || irqs_disabled()) {
 #ifdef CONFIG_SCHED_DEBUG
@@ -7593,10 +7596,21 @@ void migrate_disable(void)
 	preempt_disable();
 	preempt_lazy_disable();
 	pin_current_cpu();
-	p->migrate_disable = 1;
 
-	p->cpus_ptr = cpumask_of(smp_processor_id());
+	rq = task_rq_lock(p, &rf);
+	if (unlikely((p->sched_class == &rt_sched_class ||
+		      p->sched_class == &dl_sched_class) &&
+		      p->nr_cpus_allowed > 1)) {
+		if (p->sched_class == &rt_sched_class)
+			task_rq(p)->rt.rt_nr_migratory--;
+		else
+			task_rq(p)->dl.dl_nr_migratory--;
+	}
 	p->nr_cpus_allowed = 1;
+	task_rq_unlock(rq, p, &rf);
+	p->cpus_ptr = cpumask_of(smp_processor_id());
+	p->migrate_disable = 1;
+
 
 	preempt_enable();
 }
@@ -7605,6 +7619,9 @@ EXPORT_SYMBOL(migrate_disable);
 void migrate_enable(void)
 {
 	struct task_struct *p = current;
+	struct rq *rq;
+	struct rq_flags rf;
+
 
 	if (in_atomic() || irqs_disabled()) {
 #ifdef CONFIG_SCHED_DEBUG
@@ -7628,17 +7645,24 @@ void migrate_enable(void)
 
 	preempt_disable();
 
-	p->cpus_ptr = &p->cpus_mask;
-	p->nr_cpus_allowed = cpumask_weight(&p->cpus_mask);
 	p->migrate_disable = 0;
+	p->cpus_ptr = &p->cpus_mask;
 
-	if (p->migrate_disable_update) {
-		struct rq *rq;
-		struct rq_flags rf;
+	rq = task_rq_lock(p, &rf);
+	p->nr_cpus_allowed = cpumask_weight(&p->cpus_mask);
+	if (unlikely((p->sched_class == &rt_sched_class ||
+		      p->sched_class == &dl_sched_class) &&
+		      p->nr_cpus_allowed > 1)) {
+		if (p->sched_class == &rt_sched_class)
Quoted 8 lines

kernel/sched/core.c void migrate_enable(void) · line 7661

Quoted 5 lines
kernel/sched/core.c +5 -0
+			task_rq(p)->rt.rt_nr_migratory++;
+		else
+			task_rq(p)->dl.dl_nr_migratory++;
+	}
+	task_rq_unlock(rq, p, &rf);

The fix looks good to me, but AFAICS the repeat pattern introduced here could be factored out into a helper function instead, right?

Like:

static inline int task_in_rt_class(struct task_struct *p)

{
	return p->sched_class == &rt_sched_class;
}
Quoted 6 lines

static inline int task_in_dl_class(struct task_struct *p)

{
	return p->sched_class == &dl_sched_class;
}

?

So AFAICS it's this block that is used twice:

Quoted 11 lines
Quoted 11 lines

kernel/sched/core.c void migrate_enable(void) · line 7661

Quoted 11 lines
kernel/sched/core.c +11 -0
+	rq = task_rq_lock(p, &rf);
+	p->nr_cpus_allowed = cpumask_weight(&p->cpus_mask);
+	if (unlikely((p->sched_class == &rt_sched_class ||
+		      p->sched_class == &dl_sched_class) &&
+		      p->nr_cpus_allowed > 1)) {
+		if (p->sched_class == &rt_sched_class)
+			task_rq(p)->rt.rt_nr_migratory++;
+		else
+			task_rq(p)->dl.dl_nr_migratory++;
+	}
+	task_rq_unlock(rq, p, &rf);

or is there some difference I haven't noticed?

Thanks,

Ingo

#8 lore raw
Quoted 14 lines · Ingo Molnar

So AFAICS it's this block that is used twice:

Quoted 11 lines
Quoted 11 lines

Review

Quoted 11 lines
(quoted diff) +11 -0
+	rq = task_rq_lock(p, &rf);
+	p->nr_cpus_allowed = cpumask_weight(&p->cpus_mask);
+	if (unlikely((p->sched_class == &rt_sched_class ||
+		      p->sched_class == &dl_sched_class) &&
+		      p->nr_cpus_allowed > 1)) {
+		if (p->sched_class == &rt_sched_class)
+			task_rq(p)->rt.rt_nr_migratory++;
+		else
+			task_rq(p)->dl.dl_nr_migratory++;
+	}
+	task_rq_unlock(rq, p, &rf);

or is there some difference I haven't noticed?

One block increases the number of migratory tasks, and the other one decreases...

How about this version? (if it is good, I will polish it in a v2).

kernel/sched/core.c +51 -5
index ce34e4f..0f66376 100644
@@ -7566,10 +7566,57 @@ const u32 sched_prio_to_wmult[40] = {
 
 #if defined(CONFIG_PREEMPT_COUNT) && defined(CONFIG_SMP)
 
+enum inc_dec_migratory {
+	DEC_NR_MIGRATORY = -1,
+	INC_NR_MIGRATORY = 1,
+};
+
+static inline void
+inc_dec_nr_migratory(struct task_struct *p, enum inc_dec_migratory id)
+{
+	if (unlikely((p->sched_class == &rt_sched_class ||
+		      p->sched_class == &dl_sched_class) &&
+		      p->nr_cpus_allowed > 1)) {
+		if (p->sched_class == &rt_sched_class)
+			task_rq(p)->rt.rt_nr_migratory += id;
+		else
+			task_rq(p)->dl.dl_nr_migratory += id;
+	}
+}
+
+static inline void
+migrate_disable_update_cpus_allowed(struct task_struct *p)
+{
+	struct rq *rq;
+	struct rq_flags rf;
+
+	p->cpus_ptr = cpumask_of(smp_processor_id());
+
+	rq = task_rq_lock(p, &rf);
+	inc_dec_nr_migratory(p, DEC_NR_MIGRATORY);
+	p->nr_cpus_allowed = 1;
+	task_rq_unlock(rq, p, &rf);
+}
+
+static inline void
+migrate_enable_update_cpus_allowed(struct task_struct *p)
+{
+	struct rq *rq;
+	struct rq_flags rf;
+
+	p->cpus_ptr = &p->cpus_mask;
+
+	rq = task_rq_lock(p, &rf);
+	p->nr_cpus_allowed = cpumask_weight(&p->cpus_mask);
+	inc_dec_nr_migratory(p, INC_NR_MIGRATORY);
+	task_rq_unlock(rq, p, &rf);
+}
+
 void migrate_disable(void)
 {
 	struct task_struct *p = current;
 
+
 	if (in_atomic() || irqs_disabled()) {
 #ifdef CONFIG_SCHED_DEBUG
 		p->migrate_disable_atomic++;
@@ -7593,10 +7640,9 @@ void migrate_disable(void)
 	preempt_disable();
 	preempt_lazy_disable();
 	pin_current_cpu();
-	p->migrate_disable = 1;
 
-	p->cpus_ptr = cpumask_of(smp_processor_id());
-	p->nr_cpus_allowed = 1;
+	migrate_disable_update_cpus_allowed(p);
+	p->migrate_disable = 1;
 
 	preempt_enable();
 }
@@ -7606,6 +7652,7 @@ void migrate_enable(void)
 {
 	struct task_struct *p = current;
 
+
 	if (in_atomic() || irqs_disabled()) {
 #ifdef CONFIG_SCHED_DEBUG
 		p->migrate_disable_atomic--;
@@ -7628,9 +7675,8 @@ void migrate_enable(void)
 
 	preempt_disable();
 
-	p->cpus_ptr = &p->cpus_mask;
-	p->nr_cpus_allowed = cpumask_weight(&p->cpus_mask);
 	p->migrate_disable = 0;
+	migrate_enable_update_cpus_allowed(p);
 
 	if (p->migrate_disable_update) {
 		struct rq *rq;
#9 lore raw
Quoted 36 lines · Daniel Bristot de Oliveira
Quoted 14 lines · Ingo Molnar

So AFAICS it's this block that is used twice:

Quoted 11 lines
Quoted 11 lines

kernel/sched/core.c const u32 sched_prio_to_wmult[40] = { · line 7598

Quoted 11 lines
kernel/sched/core.c +11 -0
+	rq = task_rq_lock(p, &rf);
+	p->nr_cpus_allowed = cpumask_weight(&p->cpus_mask);
+	if (unlikely((p->sched_class == &rt_sched_class ||
+		      p->sched_class == &dl_sched_class) &&
+		      p->nr_cpus_allowed > 1)) {
+		if (p->sched_class == &rt_sched_class)
+			task_rq(p)->rt.rt_nr_migratory++;
+		else
+			task_rq(p)->dl.dl_nr_migratory++;
+	}
+	task_rq_unlock(rq, p, &rf);

or is there some difference I haven't noticed?

One block increases the number of migratory tasks, and the other one decreases...

How about this version? (if it is good, I will polish it in a v2).

kernel/sched/core.c +7 -0
index ce34e4f..0f66376 100644
@@ -7566,10 +7566,57 @@ const u32 sched_prio_to_wmult[40] = {
 
 #if defined(CONFIG_PREEMPT_COUNT) && defined(CONFIG_SMP)
 
+enum inc_dec_migratory {
+	DEC_NR_MIGRATORY = -1,
+	INC_NR_MIGRATORY = 1,
+};
+
+static inline void
+inc_dec_nr_migratory(struct task_struct *p, enum inc_dec_migratory id)

kernel/sched/core.c const u32 sched_prio_to_wmult[40] = { · line 7585

Quoted 10 lines
kernel/sched/core.c +10 -0
+{
+	if (unlikely((p->sched_class == &rt_sched_class ||
+		      p->sched_class == &dl_sched_class) &&
+		      p->nr_cpus_allowed > 1)) {
+		if (p->sched_class == &rt_sched_class)
+			task_rq(p)->rt.rt_nr_migratory += id;
+		else
+			task_rq(p)->dl.dl_nr_migratory += id;
+	}
+}

How about just 'long delta', pass in +1 or -1 and do away with the inc_dec_migratory complication?

Thanks,

Ingo

#10 lore raw
Quoted 11 lines · Ingo Molnar
Quoted 11 lines
kernel/sched/core.c +11 -0
+enum inc_dec_migratory {
+	DEC_NR_MIGRATORY = -1,
+	INC_NR_MIGRATORY = 1,
+};
+
+static inline void
+inc_dec_nr_migratory(struct task_struct *p, enum inc_dec_migratory id)
+{
+	if (unlikely((p->sched_class == &rt_sched_class ||
+		      p->sched_class == &dl_sched_class) &&
+		      p->nr_cpus_allowed > 1)) {
Quoted 8 lines

kernel/sched/core.c const u32 sched_prio_to_wmult[40] = { · line 7585

Quoted 6 lines
kernel/sched/core.c +6 -0
+		if (p->sched_class == &rt_sched_class)
+			task_rq(p)->rt.rt_nr_migratory += id;
+		else
+			task_rq(p)->dl.dl_nr_migratory += id;
+	}
+}

How about just 'long delta', pass in +1 or -1 and do away with the inc_dec_migratory complication?

Ack! I am cooking a v2 using a long delta, rather than the inc_dec_migratory complication (yeah, I exaggerated hehe).

Thanks!

-- Daniel