[PATCH] sched_ext: idle: use active idle domain in WAKE_SYNC check

[PATCH] sched_ext: idle: use active idle domain in WAKE_SYNC check

[Joseph Salisbury]

The WAKE_SYNC fast path checks whether there are idle CPUs before
biasing wakeups to the waker CPU.  After per-node idle masks were
introduced, this path started indexing idle_cpumask() with
cpu_to_node().

When per-node idle tracking is disabled, per-node masks are not kept in
sync and may be stale.  Reading them here can mis-detect system
utilization and lead to incorrect WAKE_SYNC wakeup placement.

Use scx_cpu_node_if_enabled() for the waker node lookup so this check
uses the global idle mask when per-node tracking is disabled and the
per-node mask only when that mode is enabled.

Fixes: 48849271e661 ("sched_ext: idle: Per-node idle cpumasks")
Cc: stable@vger.kernel.org # v6.15+
Assisted-by: Codex:GPT-5
Signed-off-by: Joseph Salisbury <joseph.salisbury@oracle.com>
---
 kernel/sched/ext_idle.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/kernel/sched/ext_idle.c b/kernel/sched/ext_idle.c
index 8cfe0e9a7319..6b4520ed563d 100644
--- a/kernel/sched/ext_idle.c
+++ b/kernel/sched/ext_idle.c
@@ -543,7 +543,7 @@ s32 scx_select_cpu_dfl(struct task_struct *p, s32 prev_cpu, u64 wake_flags,
 		 * piled up on it even if there is an idle core elsewhere on
 		 * the system.
 		 */
-		waker_node = cpu_to_node(cpu);
+		waker_node = scx_cpu_node_if_enabled(cpu);
 		if (!(current->flags & PF_EXITING) &&
 		    cpu_rq(cpu)->scx.local_dsq.nr == 0 &&
 		    (!(flags & SCX_PICK_IDLE_IN_NODE) || (waker_node == node)) &&
-- 
2.47.3

Re: [PATCH] sched_ext: idle: use active idle domain in WAKE_SYNC check

[Andrea Righi]

Hi Joe,

On Tue, Mar 24, 2026 at 03:58:36PM -0400, Joseph Salisbury wrote:
> The WAKE_SYNC fast path checks whether there are idle CPUs before
> biasing wakeups to the waker CPU.  After per-node idle masks were
> introduced, this path started indexing idle_cpumask() with
> cpu_to_node().
> 
> When per-node idle tracking is disabled, per-node masks are not kept in
> sync and may be stale.  Reading them here can mis-detect system
> utilization and lead to incorrect WAKE_SYNC wakeup placement.
> 
> Use scx_cpu_node_if_enabled() for the waker node lookup so this check
> uses the global idle mask when per-node tracking is disabled and the
> per-node mask only when that mode is enabled.
> 
> Fixes: 48849271e661 ("sched_ext: idle: Per-node idle cpumasks")
> Cc: stable@vger.kernel.org # v6.15+
> Assisted-by: Codex:GPT-5
> Signed-off-by: Joseph Salisbury <joseph.salisbury@oracle.com>

This is good, but it's already fixed:
https://lore.kernel.org/all/20260321105503.869337-1-yphbchou0911@gmail.com

You can probably use Tejun's branch to search for new issues/bugs:

 git://git.kernel.org/pub/scm/linux/kernel/git/tj/sched_ext.git

Thanks,
-Andrea

> ---
>  kernel/sched/ext_idle.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/kernel/sched/ext_idle.c b/kernel/sched/ext_idle.c
> index 8cfe0e9a7319..6b4520ed563d 100644
> --- a/kernel/sched/ext_idle.c
> +++ b/kernel/sched/ext_idle.c
> @@ -543,7 +543,7 @@ s32 scx_select_cpu_dfl(struct task_struct *p, s32 prev_cpu, u64 wake_flags,
>  		 * piled up on it even if there is an idle core elsewhere on
>  		 * the system.
>  		 */
> -		waker_node = cpu_to_node(cpu);
> +		waker_node = scx_cpu_node_if_enabled(cpu);
>  		if (!(current->flags & PF_EXITING) &&
>  		    cpu_rq(cpu)->scx.local_dsq.nr == 0 &&
>  		    (!(flags & SCX_PICK_IDLE_IN_NODE) || (waker_node == node)) &&
> -- 
> 2.47.3

Re: [External] : Re: [PATCH] sched_ext: idle: use active idle domain in WAKE_SYNC check

[Joseph Salisbury]

On 3/25/26 7:24 PM, Andrea Righi wrote:
> Hi Joe,
>
> On Tue, Mar 24, 2026 at 03:58:36PM -0400, Joseph Salisbury wrote:
>> The WAKE_SYNC fast path checks whether there are idle CPUs before
>> biasing wakeups to the waker CPU.  After per-node idle masks were
>> introduced, this path started indexing idle_cpumask() with
>> cpu_to_node().
>>
>> When per-node idle tracking is disabled, per-node masks are not kept in
>> sync and may be stale.  Reading them here can mis-detect system
>> utilization and lead to incorrect WAKE_SYNC wakeup placement.
>>
>> Use scx_cpu_node_if_enabled() for the waker node lookup so this check
>> uses the global idle mask when per-node tracking is disabled and the
>> per-node mask only when that mode is enabled.
>>
>> Fixes: 48849271e661 ("sched_ext: idle: Per-node idle cpumasks")
>> Cc: stable@vger.kernel.org # v6.15+
>> Assisted-by: Codex:GPT-5
>> Signed-off-by: Joseph Salisbury <joseph.salisbury@oracle.com>
> This is good, but it's already fixed:
> https://urldefense.com/v3/__https://lore.kernel.org/all/20260321105503.869337-1-yphbchou0911@gmail.com__;!!ACWV5N9M2RV99hQ!M88Tr7SYG3FHedTvBFVj_5kAPCVI-WVHWNdHhvaysPHoUD3qbj9bZ8kNj8TwcjHXNvVTnhx0BvGh7pM5R6fC$
>
> You can probably use Tejun's branch to search for new issues/bugs:
>
>   git://git.kernel.org/pub/scm/linux/kernel/git/tj/sched_ext.git
Thanks for the pointer!  I'll review this branch prior to sending/fixing 
any more bugs.
>
> Thanks,
> -Andrea
>
>> ---
>>   kernel/sched/ext_idle.c | 2 +-
>>   1 file changed, 1 insertion(+), 1 deletion(-)
>>
>> diff --git a/kernel/sched/ext_idle.c b/kernel/sched/ext_idle.c
>> index 8cfe0e9a7319..6b4520ed563d 100644
>> --- a/kernel/sched/ext_idle.c
>> +++ b/kernel/sched/ext_idle.c
>> @@ -543,7 +543,7 @@ s32 scx_select_cpu_dfl(struct task_struct *p, s32 prev_cpu, u64 wake_flags,
>>   		 * piled up on it even if there is an idle core elsewhere on
>>   		 * the system.
>>   		 */
>> -		waker_node = cpu_to_node(cpu);
>> +		waker_node = scx_cpu_node_if_enabled(cpu);
>>   		if (!(current->flags & PF_EXITING) &&
>>   		    cpu_rq(cpu)->scx.local_dsq.nr == 0 &&
>>   		    (!(flags & SCX_PICK_IDLE_IN_NODE) || (waker_node == node)) &&
>> -- 
>> 2.47.3