[PATCH 0/2] sched: Clean up newidle_balance() and pick_next_task()

[PATCH 0/2] sched: Clean up newidle_balance() and pick_next_task()

[Chen Yu]

As Peter suggested, here are two minor code clean up in the scheduler:
1. turn newidle_balance() into a static function.
2. code extraction in pick_next_task() for future re-use.

Chen Yu (2):
  sched: Make newidle_balance() static again
  sched: Extract the code to put previous task in pick_next_task()

 kernel/sched/core.c  | 39 +++++++++++++++++++++++----------------
 kernel/sched/fair.c  |  4 +++-
 kernel/sched/sched.h |  4 ----
 3 files changed, 26 insertions(+), 21 deletions(-)

-- 
2.20.1

[PATCH 1/2] sched: Make newidle_balance() static again

[Chen Yu]

After Commit 6e2df0581f56 ("sched: Fix pick_next_task() vs 'change'
pattern race"), there is no need to expose newidle_balance() as it
is only used within fair.c file. Change this function back to static again.

No functional change.

Suggested-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Chen Yu <yu.c.chen@intel.com>
---
 kernel/sched/fair.c  | 4 +++-
 kernel/sched/sched.h | 4 ----
 2 files changed, 3 insertions(+), 5 deletions(-)

diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index 02f323b85b6d..da7a505c28a6 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -86,6 +86,8 @@ static unsigned int normalized_sysctl_sched_wakeup_granularity	= 1000000UL;
 
 const_debug unsigned int sysctl_sched_migration_cost	= 500000UL;
 
+static int newidle_balance(struct rq *this_rq, struct rq_flags *rf);
+
 int sched_thermal_decay_shift;
 static int __init setup_sched_thermal_decay_shift(char *str)
 {
@@ -10425,7 +10427,7 @@ static inline void nohz_newidle_balance(struct rq *this_rq) { }
  *     0 - failed, no new tasks
  *   > 0 - success, new (fair) tasks present
  */
-int newidle_balance(struct rq *this_rq, struct rq_flags *rf)
+static int newidle_balance(struct rq *this_rq, struct rq_flags *rf)
 {
 	unsigned long next_balance = jiffies + HZ;
 	int this_cpu = this_rq->cpu;
diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h
index db3a57675ccf..be83f88495fb 100644
--- a/kernel/sched/sched.h
+++ b/kernel/sched/sched.h
@@ -1504,14 +1504,10 @@ static inline void unregister_sched_domain_sysctl(void)
 }
 #endif
 
-extern int newidle_balance(struct rq *this_rq, struct rq_flags *rf);
-
 #else
 
 static inline void sched_ttwu_pending(void) { }
 
-static inline int newidle_balance(struct rq *this_rq, struct rq_flags *rf) { return 0; }
-
 #endif /* CONFIG_SMP */
 
 #include "stats.h"
-- 
2.20.1

[PATCH 2/2] sched: Extract the task putting code from pick_next_task()

[Chen Yu]

Introduce a new function finish_prev_task() to do the balance
when necessary, and then put previous task back to the run queue.
This function is extracted from pick_next_task() to prepare for
future usage by other type of task picking logic.

No functional change.

Suggested-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Chen Yu <yu.c.chen@intel.com>
---
 kernel/sched/core.c | 39 +++++++++++++++++++++++----------------
 1 file changed, 23 insertions(+), 16 deletions(-)

diff --git a/kernel/sched/core.c b/kernel/sched/core.c
index 3a61a3b8eaa9..bf59a5cf030c 100644
--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -3904,6 +3904,28 @@ static inline void schedule_debug(struct task_struct *prev, bool preempt)
 	schedstat_inc(this_rq()->sched_count);
 }
 
+static void finish_prev_task(struct rq *rq, struct task_struct *prev,
+			     struct rq_flags *rf)
+{
+	const struct sched_class *class;
+#ifdef CONFIG_SMP
+	/*
+	 * We must do the balancing pass before put_next_task(), such
+	 * that when we release the rq->lock the task is in the same
+	 * state as before we took rq->lock.
+	 *
+	 * We can terminate the balance pass as soon as we know there is
+	 * a runnable task of @class priority or higher.
+	 */
+	for_class_range(class, prev->sched_class, &idle_sched_class) {
+		if (class->balance(rq, prev, rf))
+			break;
+	}
+#endif
+
+	put_prev_task(rq, prev);
+}
+
 /*
  * Pick up the highest-prio task:
  */
@@ -3937,22 +3959,7 @@ pick_next_task(struct rq *rq, struct task_struct *prev, struct rq_flags *rf)
 	}
 
 restart:
-#ifdef CONFIG_SMP
-	/*
-	 * We must do the balancing pass before put_next_task(), such
-	 * that when we release the rq->lock the task is in the same
-	 * state as before we took rq->lock.
-	 *
-	 * We can terminate the balance pass as soon as we know there is
-	 * a runnable task of @class priority or higher.
-	 */
-	for_class_range(class, prev->sched_class, &idle_sched_class) {
-		if (class->balance(rq, prev, rf))
-			break;
-	}
-#endif
-
-	put_prev_task(rq, prev);
+	finish_prev_task(rq, prev, rf);
 
 	for_each_class(class) {
 		p = class->pick_next_task(rq);
-- 
2.20.1

Re: [PATCH 2/2] sched: Extract the task putting code from pick_next_task()

[Steven Rostedt]

On Mon, 20 Apr 2020 00:31:52 +0800
Chen Yu <yu.c.chen@intel.com> wrote:

> Introduce a new function finish_prev_task() to do the balance
> when necessary, and then put previous task back to the run queue.
> This function is extracted from pick_next_task() to prepare for
> future usage by other type of task picking logic.
> 
> No functional change.
> 
> Suggested-by: Peter Zijlstra <peterz@infradead.org>
> Signed-off-by: Chen Yu <yu.c.chen@intel.com>
> ---
>  kernel/sched/core.c | 39 +++++++++++++++++++++++----------------
>  1 file changed, 23 insertions(+), 16 deletions(-)
> 
> diff --git a/kernel/sched/core.c b/kernel/sched/core.c
> index 3a61a3b8eaa9..bf59a5cf030c 100644
> --- a/kernel/sched/core.c
> +++ b/kernel/sched/core.c
> @@ -3904,6 +3904,28 @@ static inline void schedule_debug(struct task_struct *prev, bool preempt)
>  	schedstat_inc(this_rq()->sched_count);
>  }
>  
> +static void finish_prev_task(struct rq *rq, struct task_struct *prev,
> +			     struct rq_flags *rf)
> +{
> +	const struct sched_class *class;
> +#ifdef CONFIG_SMP
> +	/*
> +	 * We must do the balancing pass before put_next_task(), such

I know this is just a cut and paste move, but I'm thinking that this
comment is wrong. Shouldn't this be "put_prev_task()" as we have no
"put_next_task()" function.


> +	 * that when we release the rq->lock the task is in the same
> +	 * state as before we took rq->lock.
> +	 *
> +	 * We can terminate the balance pass as soon as we know there is
> +	 * a runnable task of @class priority or higher.
> +	 */
> +	for_class_range(class, prev->sched_class, &idle_sched_class) {
> +		if (class->balance(rq, prev, rf))
> +			break;
> +	}
> +#endif
> +
> +	put_prev_task(rq, prev);
> +}
> +
>  /*
>   * Pick up the highest-prio task:
>   */
> @@ -3937,22 +3959,7 @@ pick_next_task(struct rq *rq, struct task_struct *prev, struct rq_flags *rf)
>  	}
>  
>  restart:
> -#ifdef CONFIG_SMP
> -	/*
> -	 * We must do the balancing pass before put_next_task(), such
> -	 * that when we release the rq->lock the task is in the same
> -	 * state as before we took rq->lock.
> -	 *
> -	 * We can terminate the balance pass as soon as we know there is
> -	 * a runnable task of @class priority or higher.
> -	 */
> -	for_class_range(class, prev->sched_class, &idle_sched_class) {
> -		if (class->balance(rq, prev, rf))
> -			break;
> -	}
> -#endif
> -
> -	put_prev_task(rq, prev);
> +	finish_prev_task(rq, prev, rf);

I'm not sure I like the name of this function. Perhaps
"balance_and_put_prev_task()"? Something more in kind to what the function
does.

-- Steve

>  
>  	for_each_class(class) {
>  		p = class->pick_next_task(rq);

Re: [PATCH 2/2] sched: Extract the task putting code from pick_next_task()

[Valentin Schneider]

(There's a v2 at cover.1587393807.git.yu.c.chen@intel.com but I think this
still applies)

On 20/04/20 23:32, Steven Rostedt wrote:
>> @@ -3904,6 +3904,28 @@ static inline void schedule_debug(struct task_struct *prev, bool preempt)
>>      schedstat_inc(this_rq()->sched_count);
>>  }
>>
>> +static void finish_prev_task(struct rq *rq, struct task_struct *prev,
>> +			     struct rq_flags *rf)
>> +{
>> +	const struct sched_class *class;
>> +#ifdef CONFIG_SMP
>> +	/*
>> +	 * We must do the balancing pass before put_next_task(), such
>
> I know this is just a cut and paste move, but I'm thinking that this
> comment is wrong. Shouldn't this be "put_prev_task()" as we have no
> "put_next_task()" function.
>

Oh, I think you're right.

>
>> +	 * that when we release the rq->lock the task is in the same
>> +	 * state as before we took rq->lock.
>> +	 *
>> +	 * We can terminate the balance pass as soon as we know there is
>> +	 * a runnable task of @class priority or higher.
>> +	 */
>> +	for_class_range(class, prev->sched_class, &idle_sched_class) {
>> +		if (class->balance(rq, prev, rf))
>> +			break;
>> +	}
>> +#endif
>> +
>> +	put_prev_task(rq, prev);
>> +}
>> +
>>  /*
>>   * Pick up the highest-prio task:
>>   */
>> @@ -3937,22 +3959,7 @@ pick_next_task(struct rq *rq, struct task_struct *prev, struct rq_flags *rf)
>>      }
>>
>>  restart:
>> -#ifdef CONFIG_SMP
>> -	/*
>> -	 * We must do the balancing pass before put_next_task(), such
>> -	 * that when we release the rq->lock the task is in the same
>> -	 * state as before we took rq->lock.
>> -	 *
>> -	 * We can terminate the balance pass as soon as we know there is
>> -	 * a runnable task of @class priority or higher.
>> -	 */
>> -	for_class_range(class, prev->sched_class, &idle_sched_class) {
>> -		if (class->balance(rq, prev, rf))
>> -			break;
>> -	}
>> -#endif
>> -
>> -	put_prev_task(rq, prev);
>> +	finish_prev_task(rq, prev, rf);
>
> I'm not sure I like the name of this function. Perhaps
> "balance_and_put_prev_task()"? Something more in kind to what the function
> does.
>

The 'finish' thing isn't too far from the truth; it's the last thing we
need to do with the prev task (in terms of sched bookkeeping, I mean) -
and in Chen's defence ISTR Peter suggested that name.

Seeing as it's a "supercharged" put_prev_task(), I could live with the
marginally shorter "put_prev_task_balance()".

> -- Steve
>
>>
>>      for_each_class(class) {
>>              p = class->pick_next_task(rq);

Re: [PATCH 2/2] sched: Extract the task putting code from pick_next_task()

[Peter Zijlstra]

On Mon, Apr 20, 2020 at 11:55:21PM +0100, Valentin Schneider wrote:

> >> +	finish_prev_task(rq, prev, rf);
> >
> > I'm not sure I like the name of this function. Perhaps
> > "balance_and_put_prev_task()"? Something more in kind to what the function
> > does.
> >
> 
> The 'finish' thing isn't too far from the truth; it's the last thing we
> need to do with the prev task (in terms of sched bookkeeping, I mean) -
> and in Chen's defence ISTR Peter suggested that name.
> 
> Seeing as it's a "supercharged" put_prev_task(), I could live with the
> marginally shorter "put_prev_task_balance()".

What Valentin said; it's the last put we do before picking a new task.
Also, I don't like long names. That said, I'm open to short and
appropriate suggestions.

Re: [PATCH 2/2] sched: Extract the task putting code from pick_next_task()

[Steven Rostedt]

On Tue, 21 Apr 2020 01:13:55 +0200
Peter Zijlstra <peterz@infradead.org> wrote:

> > The 'finish' thing isn't too far from the truth; it's the last thing we
> > need to do with the prev task (in terms of sched bookkeeping, I mean) -
> > and in Chen's defence ISTR Peter suggested that name.
> > 
> > Seeing as it's a "supercharged" put_prev_task(), I could live with the
> > marginally shorter "put_prev_task_balance()".  
> 
> What Valentin said; it's the last put we do before picking a new task.
> Also, I don't like long names. That said, I'm open to short and
> appropriate suggestions.

I wont bikeshed this too much.

Is the "finish" more appropriate with the other use cases that are
coming. I do like that "put_prev_task_balance()" too.

-- Steve

Re: [PATCH 2/2] sched: Extract the task putting code from pick_next_task()

[Vincent Guittot]

On Tue, 21 Apr 2020 at 04:23, Steven Rostedt <rostedt@goodmis.org> wrote:
>
> On Tue, 21 Apr 2020 01:13:55 +0200
> Peter Zijlstra <peterz@infradead.org> wrote:
>
> > > The 'finish' thing isn't too far from the truth; it's the last thing we
> > > need to do with the prev task (in terms of sched bookkeeping, I mean) -
> > > and in Chen's defence ISTR Peter suggested that name.
> > >
> > > Seeing as it's a "supercharged" put_prev_task(), I could live with the
> > > marginally shorter "put_prev_task_balance()".
> >
> > What Valentin said; it's the last put we do before picking a new task.
> > Also, I don't like long names. That said, I'm open to short and
> > appropriate suggestions.
>
> I wont bikeshed this too much.
>
> Is the "finish" more appropriate with the other use cases that are
> coming. I do like that "put_prev_task_balance()" too.

This name looks reasonnable

>
> -- Steve

Re: [PATCH 2/2] sched: Extract the task putting code from pick_next_task()

[Chen Yu]

On Tue, Apr 21, 2020 at 09:42:26AM +0200, Vincent Guittot wrote:
> On Tue, 21 Apr 2020 at 04:23, Steven Rostedt <rostedt@goodmis.org> wrote:
> >
> > On Tue, 21 Apr 2020 01:13:55 +0200
> > Peter Zijlstra <peterz@infradead.org> wrote:
> >
> > > > The 'finish' thing isn't too far from the truth; it's the last thing we
> > > > need to do with the prev task (in terms of sched bookkeeping, I mean) -
> > > > and in Chen's defence ISTR Peter suggested that name.
> > > >
> > > > Seeing as it's a "supercharged" put_prev_task(), I could live with the
> > > > marginally shorter "put_prev_task_balance()".
> > >
> > > What Valentin said; it's the last put we do before picking a new task.
> > > Also, I don't like long names. That said, I'm open to short and
> > > appropriate suggestions.
> >
> > I wont bikeshed this too much.
> >
> > Is the "finish" more appropriate with the other use cases that are
> > coming. I do like that "put_prev_task_balance()" too.
> 
> This name looks reasonnable
>
Okay, I'll change it to this name.

Thanks,
Chenyu
> >
> > -- Steve

Re: [PATCH 2/2] sched: Extract the task putting code from pick_next_task()

[Chen Yu]

On Mon, Apr 20, 2020 at 06:32:32PM -0400, Steven Rostedt wrote:
> On Mon, 20 Apr 2020 00:31:52 +0800
> Chen Yu <yu.c.chen@intel.com> wrote:
> 
> > Introduce a new function finish_prev_task() to do the balance
> > when necessary, and then put previous task back to the run queue.
> > This function is extracted from pick_next_task() to prepare for
> > future usage by other type of task picking logic.
> > 
> > No functional change.
> > 
> > Suggested-by: Peter Zijlstra <peterz@infradead.org>
> > Signed-off-by: Chen Yu <yu.c.chen@intel.com>
> > ---
> >  kernel/sched/core.c | 39 +++++++++++++++++++++++----------------
> >  1 file changed, 23 insertions(+), 16 deletions(-)
> > 
> > diff --git a/kernel/sched/core.c b/kernel/sched/core.c
> > index 3a61a3b8eaa9..bf59a5cf030c 100644
> > --- a/kernel/sched/core.c
> > +++ b/kernel/sched/core.c
> > @@ -3904,6 +3904,28 @@ static inline void schedule_debug(struct task_struct *prev, bool preempt)
> >  	schedstat_inc(this_rq()->sched_count);
> >  }
> >  
> > +static void finish_prev_task(struct rq *rq, struct task_struct *prev,
> > +			     struct rq_flags *rf)
> > +{
> > +	const struct sched_class *class;
> > +#ifdef CONFIG_SMP
> > +	/*
> > +	 * We must do the balancing pass before put_next_task(), such
> 
> I know this is just a cut and paste move, but I'm thinking that this
> comment is wrong. Shouldn't this be "put_prev_task()" as we have no
> "put_next_task()" function.
> 
>
Okay, I'll fix it in v3.
> > +	finish_prev_task(rq, prev, rf);
> 
> I'm not sure I like the name of this function. Perhaps
> "balance_and_put_prev_task()"? Something more in kind to what the function
> does.
>
Per the discussion, I think put_prev_task_balance() might be an
appropriate one.

Thanks,
Chenyu
> -- Steve
> 
> >  
> >  	for_each_class(class) {
> >  		p = class->pick_next_task(rq);
>