2020-04-11 09:23:27

by Peng Wang

Subject: [PATCH] sched/fair: Simplify the code of should_we_balance()

We only consider group_balance_cpu() after there is no idle
cpu. So, just do comparison before return at these two cases.

Signed-off-by: Peng Wang <[email protected]>
---
kernel/sched/fair.c | 16 +++++-----------
1 file changed, 5 insertions(+), 11 deletions(-)

diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index 1ea3ddd..81b2c647 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -9413,7 +9413,7 @@ static int active_load_balance_cpu_stop(void *data);
static int should_we_balance(struct lb_env *env)
{
struct sched_group *sg = env->sd->groups;
- int cpu, balance_cpu = -1;
+ int cpu;

/*
* Ensure the balancing environment is consistent; can happen
@@ -9434,18 +9434,12 @@ static int should_we_balance(struct lb_env *env)
if (!idle_cpu(cpu))
continue;

- balance_cpu = cpu;
- break;
+ /* Are we the first idle CPU? */
+ return cpu == env->dst_cpu;
}

- if (balance_cpu == -1)
- balance_cpu = group_balance_cpu(sg);
-
- /*
- * First idle CPU or the first CPU(busiest) in this sched group
- * is eligible for doing load balancing at this and above domains.
- */
- return balance_cpu == env->dst_cpu;
+ /* Are we the first balance CPU of this group? */
+ return group_balance_cpu(sg) == env->dst_cpu;
}

/*
--
2.9.5

2020-04-12 08:43:40

by Peng Wang

Subject: Re: [PATCH] sched/fair: Simplify the code of should_we_balance()

On 4/11/20 5:20 PM, Peng Wang wrote:
> We only consider group_balance_cpu() after there is no idle
> cpu. So, just do comparison before return at these two cases.
>
> Signed-off-by: Peng Wang <[email protected]>
> ---
> kernel/sched/fair.c | 16 +++++-----------
> 1 file changed, 5 insertions(+), 11 deletions(-)
>
> diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
> index 1ea3ddd..81b2c647 100644
> --- a/kernel/sched/fair.c
> +++ b/kernel/sched/fair.c
> @@ -9413,7 +9413,7 @@ static int active_load_balance_cpu_stop(void *data);
> static int should_we_balance(struct lb_env *env)
> {
> struct sched_group *sg = env->sd->groups;
> - int cpu, balance_cpu = -1;
> + int cpu;
>
> /*
> * Ensure the balancing environment is consistent; can happen
> @@ -9434,18 +9434,12 @@ static int should_we_balance(struct lb_env *env)
> if (!idle_cpu(cpu))
> continue;
>
> - balance_cpu = cpu;
> - break;
> + /* Are we the first idle CPU? */
> + return cpu == env->dst_cpu;
> }
>
> - if (balance_cpu == -1)
> - balance_cpu = group_balance_cpu(sg);
> -
> - /*
> - * First idle CPU or the first CPU(busiest) in this sched group
> - * is eligible for doing load balancing at this and above domains.
> - */
> - return balance_cpu == env->dst_cpu;
> + /* Are we the first balance CPU of this group? */
> + return group_balance_cpu(sg) == env->dst_cpu;
> }
>
> /*
>

[email protected]

2020-04-14 14:49:05

by Vincent Guittot

Subject: Re: [PATCH] sched/fair: Simplify the code of should_we_balance()

On Sat, 11 Apr 2020 at 11:21, Peng Wang <[email protected]> wrote:
>
> We only consider group_balance_cpu() after there is no idle
> cpu. So, just do comparison before return at these two cases.
>
> Signed-off-by: Peng Wang <[email protected]>

With the small fix in the comment below
Reviewed-by: Vincent Guittot <[email protected]>

> ---
> kernel/sched/fair.c | 16 +++++-----------
> 1 file changed, 5 insertions(+), 11 deletions(-)
>
> diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
> index 1ea3ddd..81b2c647 100644
> --- a/kernel/sched/fair.c
> +++ b/kernel/sched/fair.c
> @@ -9413,7 +9413,7 @@ static int active_load_balance_cpu_stop(void *data);
> static int should_we_balance(struct lb_env *env)
> {
> struct sched_group *sg = env->sd->groups;
> - int cpu, balance_cpu = -1;
> + int cpu;
>
> /*
> * Ensure the balancing environment is consistent; can happen
> @@ -9434,18 +9434,12 @@ static int should_we_balance(struct lb_env *env)
> if (!idle_cpu(cpu))
> continue;
>
> - balance_cpu = cpu;
> - break;
> + /* Are we the first idle CPU? */
> + return cpu == env->dst_cpu;
> }
>
> - if (balance_cpu == -1)
> - balance_cpu = group_balance_cpu(sg);
> -
> - /*
> - * First idle CPU or the first CPU(busiest) in this sched group
> - * is eligible for doing load balancing at this and above domains.
> - */
> - return balance_cpu == env->dst_cpu;
> + /* Are we the first balance CPU of this group? */

/* Are we the first CPU of this group ? */

> + return group_balance_cpu(sg) == env->dst_cpu;
> }
>
> /*
> --
> 2.9.5
>

2020-04-15 00:58:22

by Valentin Schneider

Subject: Re: [PATCH] sched/fair: Simplify the code of should_we_balance()

On 12/04/20 09:42, Peng Wang wrote:
> On 4/11/20 5:20 PM, Peng Wang wrote:
>> We only consider group_balance_cpu() after there is no idle
>> cpu. So, just do comparison before return at these two cases.
>>

It's not really changing much, but if it helps making it a bit more
readable, why not. Small nit below.

Reviewed-by: Valentin Schneider <[email protected]>

>> Signed-off-by: Peng Wang <[email protected]>
>> ---
>> kernel/sched/fair.c | 16 +++++-----------
>> 1 file changed, 5 insertions(+), 11 deletions(-)
>>
>> diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
>> index 1ea3ddd..81b2c647 100644
>> --- a/kernel/sched/fair.c
>> +++ b/kernel/sched/fair.c
>> @@ -9413,7 +9413,7 @@ static int active_load_balance_cpu_stop(void *data);
>> static int should_we_balance(struct lb_env *env)
>> {
>> struct sched_group *sg = env->sd->groups;
>> - int cpu, balance_cpu = -1;
>> + int cpu;
>>
>> /*
>> * Ensure the balancing environment is consistent; can happen
>> @@ -9434,18 +9434,12 @@ static int should_we_balance(struct lb_env *env)
>> if (!idle_cpu(cpu))
>> continue;
>>
>> - balance_cpu = cpu;
>> - break;
>> + /* Are we the first idle CPU? */
>> + return cpu == env->dst_cpu;
>> }
>>
>> - if (balance_cpu == -1)
>> - balance_cpu = group_balance_cpu(sg);
>> -
>> - /*
>> - * First idle CPU or the first CPU(busiest) in this sched group
>> - * is eligible for doing load balancing at this and above domains.
>> - */
>> - return balance_cpu == env->dst_cpu;
>> + /* Are we the first balance CPU of this group? */

Nit: That should be either "the balance CPU" or "the first CPU in the group
balance mask"

>> + return group_balance_cpu(sg) == env->dst_cpu;
>> }
>>
>> /*
>>
>
> [email protected]

2020-04-15 17:40:26

by Peng Wang

Subject: [PATCH v2] sched/fair: Simplify the code of should_we_balance()

We only consider group_balance_cpu() after there is no idle
cpu. So, just do comparison before return at these two cases.

Reviewed-by: Valentin Schneider <[email protected]>
Reviewed-by: Vincent Guittot <[email protected]>
Signed-off-by: Peng Wang <[email protected]>
---
kernel/sched/fair.c | 16 +++++-----------
1 file changed, 5 insertions(+), 11 deletions(-)

diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index 02f323b..c3f57f4 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -9409,7 +9409,7 @@ static int active_load_balance_cpu_stop(void *data);
static int should_we_balance(struct lb_env *env)
{
struct sched_group *sg = env->sd->groups;
- int cpu, balance_cpu = -1;
+ int cpu;

/*
* Ensure the balancing environment is consistent; can happen
@@ -9430,18 +9430,12 @@ static int should_we_balance(struct lb_env *env)
if (!idle_cpu(cpu))
continue;

- balance_cpu = cpu;
- break;
+ /* Are we the first idle CPU? */
+ return cpu == env->dst_cpu;
}

- if (balance_cpu == -1)
- balance_cpu = group_balance_cpu(sg);
-
- /*
- * First idle CPU or the first CPU(busiest) in this sched group
- * is eligible for doing load balancing at this and above domains.
- */
- return balance_cpu == env->dst_cpu;
+ /* Are we the first CPU in the balance mask of this group? */
+ return group_balance_cpu(sg) == env->dst_cpu;
}

/*
--
2.9.5

2020-04-15 20:59:42

by Peter Zijlstra

Subject: Re: [PATCH] sched/fair: Simplify the code of should_we_balance()

On Tue, Apr 14, 2020 at 09:25:00AM +0200, Vincent Guittot wrote:
> On Sat, 11 Apr 2020 at 11:21, Peng Wang <[email protected]> wrote:
> >
> > We only consider group_balance_cpu() after there is no idle
> > cpu. So, just do comparison before return at these two cases.
> >
> > Signed-off-by: Peng Wang <[email protected]>
>
> With the small fix in the comment below

Done

> Reviewed-by: Vincent Guittot <[email protected]>

Thanks Guys!