[PATCH v2] sched_ext: Replace hardcoding with macro and minor typo fix

[PATCH v2] sched_ext: Replace hardcoding with macro and minor typo fix

[Zhao Mengmeng]

From: Zhao Mengmeng <zhaomengmeng@kylinos.cn>

1. replace hardcoding with SCX_KF_UNLOCKED.
2. scx_next_task_picked() has been replaced with siwtch_class().
3. minor typo fixes.

Signed-off-by: Zhao Mengmeng <zhaomengmeng@kylinos.cn>
---
 kernel/sched/ext.c             | 6 +++---
 tools/sched_ext/scx_qmap.bpf.c | 2 +-
 2 files changed, 4 insertions(+), 4 deletions(-)

diff --git a/kernel/sched/ext.c b/kernel/sched/ext.c
index 1b1c33f12dd7..832f77d1d318 100644
--- a/kernel/sched/ext.c
+++ b/kernel/sched/ext.c
@@ -2759,10 +2759,10 @@ static int balance_one(struct rq *rq, struct task_struct *prev)
 		 * If the previous sched_class for the current CPU was not SCX,
 		 * notify the BPF scheduler that it again has control of the
 		 * core. This callback complements ->cpu_release(), which is
-		 * emitted in scx_next_task_picked().
+		 * emitted in switch_class().
 		 */
 		if (SCX_HAS_OP(cpu_acquire))
-			SCX_CALL_OP(0, cpu_acquire, cpu_of(rq), NULL);
+			SCX_CALL_OP(SCX_KF_UNLOCKED, cpu_acquire, cpu_of(rq), NULL);
 		rq->scx.cpu_released = false;
 	}
 
@@ -6096,7 +6096,7 @@ static void kick_cpus_irq_workfn(struct irq_work *irq_work)
 		if (cpu != cpu_of(this_rq)) {
 			/*
 			 * Pairs with smp_store_release() issued by this CPU in
-			 * scx_next_task_picked() on the resched path.
+			 * switch_class() on the resched path.
 			 *
 			 * We busy-wait here to guarantee that no other task can
 			 * be scheduled on our core before the target CPU has
diff --git a/tools/sched_ext/scx_qmap.bpf.c b/tools/sched_ext/scx_qmap.bpf.c
index ee264947e0c3..f230641929ec 100644
--- a/tools/sched_ext/scx_qmap.bpf.c
+++ b/tools/sched_ext/scx_qmap.bpf.c
@@ -5,7 +5,7 @@
  * There are five FIFOs implemented using BPF_MAP_TYPE_QUEUE. A task gets
  * assigned to one depending on its compound weight. Each CPU round robins
  * through the FIFOs and dispatches more from FIFOs with higher indices - 1 from
- * queue0, 2 from queue1, 4 from queue2 and so on.
+ * queue0, 2 from queue1, 3 from queue2 and so on.
  *
  * This scheduler demonstrates:
  *
-- 
2.43.0

Re: [PATCH v2] sched_ext: Replace hardcoding with macro and minor typo fix

[Tejun Heo]

Hello,

On Wed, Nov 13, 2024 at 10:59:08AM +0800, Zhao Mengmeng wrote:
> From: Zhao Mengmeng <zhaomengmeng@kylinos.cn>
> 
> 1. replace hardcoding with SCX_KF_UNLOCKED.
> 2. scx_next_task_picked() has been replaced with siwtch_class().
> 3. minor typo fixes.
> 
> Signed-off-by: Zhao Mengmeng <zhaomengmeng@kylinos.cn>
> ---
>  kernel/sched/ext.c             | 6 +++---
>  tools/sched_ext/scx_qmap.bpf.c | 2 +-
>  2 files changed, 4 insertions(+), 4 deletions(-)
> 
> diff --git a/kernel/sched/ext.c b/kernel/sched/ext.c
> index 1b1c33f12dd7..832f77d1d318 100644
> --- a/kernel/sched/ext.c
> +++ b/kernel/sched/ext.c
> @@ -2759,10 +2759,10 @@ static int balance_one(struct rq *rq, struct task_struct *prev)
>  		 * If the previous sched_class for the current CPU was not SCX,
>  		 * notify the BPF scheduler that it again has control of the
>  		 * core. This callback complements ->cpu_release(), which is
> -		 * emitted in scx_next_task_picked().
> +		 * emitted in switch_class().
>  		 */
>  		if (SCX_HAS_OP(cpu_acquire))
> -			SCX_CALL_OP(0, cpu_acquire, cpu_of(rq), NULL);
> +			SCX_CALL_OP(SCX_KF_UNLOCKED, cpu_acquire, cpu_of(rq), NULL);

I think this is actually a bug. David, shouldn't this be SCX_KF_REST?

>  		rq->scx.cpu_released = false;
>  	}
>  
> @@ -6096,7 +6096,7 @@ static void kick_cpus_irq_workfn(struct irq_work *irq_work)
>  		if (cpu != cpu_of(this_rq)) {
>  			/*
>  			 * Pairs with smp_store_release() issued by this CPU in
> -			 * scx_next_task_picked() on the resched path.
> +			 * switch_class() on the resched path.
>  			 *
>  			 * We busy-wait here to guarantee that no other task can
>  			 * be scheduled on our core before the target CPU has
> diff --git a/tools/sched_ext/scx_qmap.bpf.c b/tools/sched_ext/scx_qmap.bpf.c
> index ee264947e0c3..f230641929ec 100644
> --- a/tools/sched_ext/scx_qmap.bpf.c
> +++ b/tools/sched_ext/scx_qmap.bpf.c
> @@ -5,7 +5,7 @@
>   * There are five FIFOs implemented using BPF_MAP_TYPE_QUEUE. A task gets
>   * assigned to one depending on its compound weight. Each CPU round robins
>   * through the FIFOs and dispatches more from FIFOs with higher indices - 1 from
> - * queue0, 2 from queue1, 4 from queue2 and so on.
> + * queue0, 2 from queue1, 3 from queue2 and so on.

The number to dispatch is determined by:

  cpuc->dsp_cnt = 1 << cpuc->dsp_idx;

I think the existing comment is correct.

Thasnks.

-- 
tejun

Re: [PATCH v2] sched_ext: Replace hardcoding with macro and minor typo fix

[David Vernet]

[-- Attachment #1: Type: text/plain, Size: 1504 bytes --]

On Thu, Nov 14, 2024 at 07:12:31AM -1000, Tejun Heo wrote:
> Hello,
> 
> On Wed, Nov 13, 2024 at 10:59:08AM +0800, Zhao Mengmeng wrote:
> > From: Zhao Mengmeng <zhaomengmeng@kylinos.cn>
> > 
> > 1. replace hardcoding with SCX_KF_UNLOCKED.
> > 2. scx_next_task_picked() has been replaced with siwtch_class().
> > 3. minor typo fixes.
> > 
> > Signed-off-by: Zhao Mengmeng <zhaomengmeng@kylinos.cn>
> > ---
> >  kernel/sched/ext.c             | 6 +++---
> >  tools/sched_ext/scx_qmap.bpf.c | 2 +-
> >  2 files changed, 4 insertions(+), 4 deletions(-)
> > 
> > diff --git a/kernel/sched/ext.c b/kernel/sched/ext.c
> > index 1b1c33f12dd7..832f77d1d318 100644
> > --- a/kernel/sched/ext.c
> > +++ b/kernel/sched/ext.c
> > @@ -2759,10 +2759,10 @@ static int balance_one(struct rq *rq, struct task_struct *prev)
> >  		 * If the previous sched_class for the current CPU was not SCX,
> >  		 * notify the BPF scheduler that it again has control of the
> >  		 * core. This callback complements ->cpu_release(), which is
> > -		 * emitted in scx_next_task_picked().
> > +		 * emitted in switch_class().
> >  		 */
> >  		if (SCX_HAS_OP(cpu_acquire))
> > -			SCX_CALL_OP(0, cpu_acquire, cpu_of(rq), NULL);
> > +			SCX_CALL_OP(SCX_KF_UNLOCKED, cpu_acquire, cpu_of(rq), NULL);
> 
> I think this is actually a bug. David, shouldn't this be SCX_KF_REST?

Yes, good catch. We're holding the rq lock so SCX_KF_UNLOCKED isn't safe. I
agree this should be SCX_KF_REST.

Thanks,
David

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 228 bytes --]

Re: [PATCH v2] sched_ext: Replace hardcoding with macro and minor typo fix

[Zhao Mengmeng]

On 2024/11/15 01:12, Tejun Heo wrote:
> Hello,
> 
> On Wed, Nov 13, 2024 at 10:59:08AM +0800, Zhao Mengmeng wrote:
>> From: Zhao Mengmeng <zhaomengmeng@kylinos.cn>
>>
>> 1. replace hardcoding with SCX_KF_UNLOCKED.
>> 2. scx_next_task_picked() has been replaced with siwtch_class().
>> 3. minor typo fixes.
>>
>> Signed-off-by: Zhao Mengmeng <zhaomengmeng@kylinos.cn>
>> ---
>>  kernel/sched/ext.c             | 6 +++---
>>  tools/sched_ext/scx_qmap.bpf.c | 2 +-
>>  2 files changed, 4 insertions(+), 4 deletions(-)
>>
>> diff --git a/kernel/sched/ext.c b/kernel/sched/ext.c
>> index 1b1c33f12dd7..832f77d1d318 100644
>> --- a/kernel/sched/ext.c
>> +++ b/kernel/sched/ext.c
>> @@ -2759,10 +2759,10 @@ static int balance_one(struct rq *rq, struct task_struct *prev)
>>  		 * If the previous sched_class for the current CPU was not SCX,
>>  		 * notify the BPF scheduler that it again has control of the
>>  		 * core. This callback complements ->cpu_release(), which is
>> -		 * emitted in scx_next_task_picked().
>> +		 * emitted in switch_class().
>>  		 */
>>  		if (SCX_HAS_OP(cpu_acquire))
>> -			SCX_CALL_OP(0, cpu_acquire, cpu_of(rq), NULL);
>> +			SCX_CALL_OP(SCX_KF_UNLOCKED, cpu_acquire, cpu_of(rq), NULL);
> 
> I think this is actually a bug. David, shouldn't this be SCX_KF_REST?
> 
>>  		rq->scx.cpu_released = false;
>>  	}

Got it.
>> @@ -6096,7 +6096,7 @@ static void kick_cpus_irq_workfn(struct irq_work *irq_work)
>>  		if (cpu != cpu_of(this_rq)) {
>>  			/*
>>  			 * Pairs with smp_store_release() issued by this CPU in
>> -			 * scx_next_task_picked() on the resched path.
>> +			 * switch_class() on the resched path.
>>  			 *
>>  			 * We busy-wait here to guarantee that no other task can
>>  			 * be scheduled on our core before the target CPU has
>> diff --git a/tools/sched_ext/scx_qmap.bpf.c b/tools/sched_ext/scx_qmap.bpf.c
>> index ee264947e0c3..f230641929ec 100644
>> --- a/tools/sched_ext/scx_qmap.bpf.c
>> +++ b/tools/sched_ext/scx_qmap.bpf.c
>> @@ -5,7 +5,7 @@
>>   * There are five FIFOs implemented using BPF_MAP_TYPE_QUEUE. A task gets
>>   * assigned to one depending on its compound weight. Each CPU round robins
>>   * through the FIFOs and dispatches more from FIFOs with higher indices - 1 from
>> - * queue0, 2 from queue1, 4 from queue2 and so on.
>> + * queue0, 2 from queue1, 3 from queue2 and so on.
> 
> The number to dispatch is determined by:
> 
>   cpuc->dsp_cnt = 1 << cpuc->dsp_idx;
> 
> I think the existing comment is correct.

You are right, I missed something. Will send a V3