[PATCH] sched_ext: Use the resched_cpu() to replace resched_curr() in the bypass_lb_node()

[PATCH] sched_ext: Use the resched_cpu() to replace resched_curr() in the bypass_lb_node()

[Zqiang]

For the PREEMPT_RT kernels, the scx_bypass_lb_timerfn() running in the
preemptible per-CPU ktimer kthread context, this means that the following
scenarios will occur(for x86 platform):

       cpu1                          cpu2
				 ktimer kthread:
                                 ->scx_bypass_lb_timerfn
                                   ->bypass_lb_node
                                     ->for_each_cpu(cpu, resched_mask)

    migration/1:                       by preempt by migration/2:
    multi_cpu_stop()                     multi_cpu_stop()
    ->take_cpu_down()
      ->__cpu_disable()
	->set cpu1 offline

                                       ->rq1 = cpu_rq(cpu1)
                                       ->resched_curr(rq1)
                                         ->smp_send_reschedule(cpu1)
					   ->native_smp_send_reschedule(cpu1)
					     ->if(unlikely(cpu_is_offline(cpu))) {
                					WARN(1, "sched: Unexpected
							reschedule of offline CPU#%d!\n", cpu);
                					return;
        					}

This commit therefore use the resched_cpu() to replace resched_curr()
in the bypass_lb_node() to avoid send-ipi to offline CPUs.

Signed-off-by: Zqiang <qiang.zhang@linux.dev>
---
 kernel/sched/ext.c | 9 ++-------
 1 file changed, 2 insertions(+), 7 deletions(-)

diff --git a/kernel/sched/ext.c b/kernel/sched/ext.c
index 5ebf8a740847..8f6d8d7f895c 100644
--- a/kernel/sched/ext.c
+++ b/kernel/sched/ext.c
@@ -3956,13 +3956,8 @@ static void bypass_lb_node(struct scx_sched *sch, int node)
 					     nr_donor_target, nr_target);
 	}
 
-	for_each_cpu(cpu, resched_mask) {
-		struct rq *rq = cpu_rq(cpu);
-
-		raw_spin_rq_lock_irq(rq);
-		resched_curr(rq);
-		raw_spin_rq_unlock_irq(rq);
-	}
+	for_each_cpu(cpu, resched_mask)
+		resched_cpu(cpu);
 
 	for_each_cpu_and(cpu, cpu_online_mask, node_mask) {
 		u32 nr = READ_ONCE(cpu_rq(cpu)->scx.bypass_dsq.nr);
-- 
2.17.1

[PATCH] sched_ext: Avoid multiple irq_work_queue() calls in destroy_dsq()

[Zqiang]

This commit only make irq_work_queue() to be called when the
llist_add() returns true.

Signed-off-by: Zqiang <qiang.zhang@linux.dev>
---
 kernel/sched/ext.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/kernel/sched/ext.c b/kernel/sched/ext.c
index 8f6d8d7f895c..136b01950a62 100644
--- a/kernel/sched/ext.c
+++ b/kernel/sched/ext.c
@@ -3439,8 +3439,8 @@ static void destroy_dsq(struct scx_sched *sch, u64 dsq_id)
 	 * operations inside scheduler locks.
 	 */
 	dsq->id = SCX_DSQ_INVALID;
-	llist_add(&dsq->free_node, &dsqs_to_free);
-	irq_work_queue(&free_dsq_irq_work);
+	if (llist_add(&dsq->free_node, &dsqs_to_free))
+		irq_work_queue(&free_dsq_irq_work);
 
 out_unlock_dsq:
 	raw_spin_unlock_irqrestore(&dsq->lock, flags);
-- 
2.17.1

Re: [PATCH] sched_ext: Use the resched_cpu() to replace resched_curr() in the bypass_lb_node()

[Andrea Righi]

On Mon, Dec 22, 2025 at 07:53:17PM +0800, Zqiang wrote:
> For the PREEMPT_RT kernels, the scx_bypass_lb_timerfn() running in the
> preemptible per-CPU ktimer kthread context, this means that the following
> scenarios will occur(for x86 platform):
> 
>        cpu1                          cpu2
> 				 ktimer kthread:
>                                  ->scx_bypass_lb_timerfn
>                                    ->bypass_lb_node
>                                      ->for_each_cpu(cpu, resched_mask)
> 
>     migration/1:                       by preempt by migration/2:
>     multi_cpu_stop()                     multi_cpu_stop()
>     ->take_cpu_down()
>       ->__cpu_disable()
> 	->set cpu1 offline
> 
>                                        ->rq1 = cpu_rq(cpu1)
>                                        ->resched_curr(rq1)
>                                          ->smp_send_reschedule(cpu1)
> 					   ->native_smp_send_reschedule(cpu1)
> 					     ->if(unlikely(cpu_is_offline(cpu))) {
>                 					WARN(1, "sched: Unexpected
> 							reschedule of offline CPU#%d!\n", cpu);
>                 					return;
>         					}
> 
> This commit therefore use the resched_cpu() to replace resched_curr()
> in the bypass_lb_node() to avoid send-ipi to offline CPUs.
> 
> Signed-off-by: Zqiang <qiang.zhang@linux.dev>

Good catch, resched_cpu() checks for online CPUs, so makes sense to me.

Reviewed-by: Andrea Righi <arighi@nvidia.com>

Thanks,
-Andrea

> ---
>  kernel/sched/ext.c | 9 ++-------
>  1 file changed, 2 insertions(+), 7 deletions(-)
> 
> diff --git a/kernel/sched/ext.c b/kernel/sched/ext.c
> index 5ebf8a740847..8f6d8d7f895c 100644
> --- a/kernel/sched/ext.c
> +++ b/kernel/sched/ext.c
> @@ -3956,13 +3956,8 @@ static void bypass_lb_node(struct scx_sched *sch, int node)
>  					     nr_donor_target, nr_target);
>  	}
>  
> -	for_each_cpu(cpu, resched_mask) {
> -		struct rq *rq = cpu_rq(cpu);
> -
> -		raw_spin_rq_lock_irq(rq);
> -		resched_curr(rq);
> -		raw_spin_rq_unlock_irq(rq);
> -	}
> +	for_each_cpu(cpu, resched_mask)
> +		resched_cpu(cpu);
>  
>  	for_each_cpu_and(cpu, cpu_online_mask, node_mask) {
>  		u32 nr = READ_ONCE(cpu_rq(cpu)->scx.bypass_dsq.nr);
> -- 
> 2.17.1
> 

Re: [PATCH] sched_ext: Avoid multiple irq_work_queue() calls in destroy_dsq()

[Andrea Righi]

On Mon, Dec 22, 2025 at 07:53:18PM +0800, Zqiang wrote:
> This commit only make irq_work_queue() to be called when the
> llist_add() returns true.

Just to be more clear, we could rephrase the commit message as follows:

llist_add() returns true only when adding to an empty list, which indicates
that no IRQ work is currently queued or running. Therefore, we only need to
call irq_work_queue() when llist_add() returns true, to avoid unnecessarily
re-queueing IRQ work that is already pending or executing.

> 
> Signed-off-by: Zqiang <qiang.zhang@linux.dev>

But overall, looks good to me.

Reviewed-by: Andrea Righi <arighi@nvidia.com>

Thanks,
-Andrea

> ---
>  kernel/sched/ext.c | 4 ++--
>  1 file changed, 2 insertions(+), 2 deletions(-)
> 
> diff --git a/kernel/sched/ext.c b/kernel/sched/ext.c
> index 8f6d8d7f895c..136b01950a62 100644
> --- a/kernel/sched/ext.c
> +++ b/kernel/sched/ext.c
> @@ -3439,8 +3439,8 @@ static void destroy_dsq(struct scx_sched *sch, u64 dsq_id)
>  	 * operations inside scheduler locks.
>  	 */
>  	dsq->id = SCX_DSQ_INVALID;
> -	llist_add(&dsq->free_node, &dsqs_to_free);
> -	irq_work_queue(&free_dsq_irq_work);
> +	if (llist_add(&dsq->free_node, &dsqs_to_free))
> +		irq_work_queue(&free_dsq_irq_work);
>  
>  out_unlock_dsq:
>  	raw_spin_unlock_irqrestore(&dsq->lock, flags);
> -- 
> 2.17.1
> 

Re: [PATCH] sched_ext: Use the resched_cpu() to replace resched_curr() in the bypass_lb_node()

[Tejun Heo]

Applied to sched_ext/for-6.19-fixes.

Thanks.
--
tejun

Re: [PATCH] sched_ext: Avoid multiple irq_work_queue() calls in destroy_dsq()

[Tejun Heo]

Applied to sched_ext/for-6.20.

Note: Commit message updated per Andrea's suggestion.

Thanks.
--
tejun

Re: [PATCH] sched_ext: Avoid multiple irq_work_queue() calls in destroy_dsq()

[Zqiang]

> 
> On Mon, Dec 22, 2025 at 07:53:18PM +0800, Zqiang wrote:
> 
> > 
> > This commit only make irq_work_queue() to be called when the
> >  llist_add() returns true.
> > 
> Just to be more clear, we could rephrase the commit message as follows:
> 
> llist_add() returns true only when adding to an empty list, which indicates
> that no IRQ work is currently queued or running. Therefore, we only need to
> call irq_work_queue() when llist_add() returns true, to avoid unnecessarily
> re-queueing IRQ work that is already pending or executing.

Thank you for make the commit more clear and reviewed :) .

Thanks
Zqiang
 
> 
> > 
> > Signed-off-by: Zqiang <qiang.zhang@linux.dev>
> > 
> But overall, looks good to me.
> 
> Reviewed-by: Andrea Righi <arighi@nvidia.com>
> 
> Thanks,
> -Andrea
> 
> > 
> > ---
> >  kernel/sched/ext.c | 4 ++--
> >  1 file changed, 2 insertions(+), 2 deletions(-)
> >  
> >  diff --git a/kernel/sched/ext.c b/kernel/sched/ext.c
> >  index 8f6d8d7f895c..136b01950a62 100644
> >  --- a/kernel/sched/ext.c
> >  +++ b/kernel/sched/ext.c
> >  @@ -3439,8 +3439,8 @@ static void destroy_dsq(struct scx_sched *sch, u64 dsq_id)
> >  * operations inside scheduler locks.
> >  */
> >  dsq->id = SCX_DSQ_INVALID;
> >  - llist_add(&dsq->free_node, &dsqs_to_free);
> >  - irq_work_queue(&free_dsq_irq_work);
> >  + if (llist_add(&dsq->free_node, &dsqs_to_free))
> >  + irq_work_queue(&free_dsq_irq_work);
> >  
> >  out_unlock_dsq:
> >  raw_spin_unlock_irqrestore(&dsq->lock, flags);
> >  -- 
> >  2.17.1
> >
>

Re: [PATCH] sched_ext: Avoid multiple irq_work_queue() calls in destroy_dsq()

[Zqiang]

> 
> Applied to sched_ext/for-6.20.
> 
> Note: Commit message updated per Andrea's suggestion.

Thank you for update commit message.

Thanks
Zqiang

> 
> Thanks.
> --
> tejun
>