[PATCH sched_ext/for-6.15] sched_ext: idle: Fix scx_bpf_pick_any_cpu_node() behavior

[PATCH sched_ext/for-6.15] sched_ext: idle: Fix scx_bpf_pick_any_cpu_node() behavior

[Andrea Righi]

When %SCX_PICK_IDLE_IN_NODE is specified, scx_bpf_pick_any_cpu_node()
should always return a CPU from the specified node, regardless of its
idle state.

Also clarify this logic in the function documentation.

Fixes: 01059219b0cfd ("sched_ext: idle: Introduce node-aware idle cpu kfunc helpers")
Signed-off-by: Andrea Righi <arighi@nvidia.com>
---
 kernel/sched/ext_idle.c | 10 +++++++---
 1 file changed, 7 insertions(+), 3 deletions(-)

diff --git a/kernel/sched/ext_idle.c b/kernel/sched/ext_idle.c
index 759a06774b5b3..dc40e0baf77c5 100644
--- a/kernel/sched/ext_idle.c
+++ b/kernel/sched/ext_idle.c
@@ -1029,8 +1029,9 @@ __bpf_kfunc s32 scx_bpf_pick_idle_cpu(const struct cpumask *cpus_allowed,
  * empty.
  *
  * The search starts from @node and proceeds to other online NUMA nodes in
- * order of increasing distance (unless SCX_PICK_IDLE_IN_NODE is specified,
- * in which case the search is limited to the target @node).
+ * order of increasing distance (unless %SCX_PICK_IDLE_IN_NODE is specified,
+ * in which case the search is limited to the target @node, regardless of
+ * the CPU idle state).
  *
  * If ops.update_idle() is implemented and %SCX_OPS_KEEP_BUILTIN_IDLE is not
  * set, this function can't tell which CPUs are idle and will always pick any
@@ -1049,7 +1050,10 @@ __bpf_kfunc s32 scx_bpf_pick_any_cpu_node(const struct cpumask *cpus_allowed,
 	if (cpu >= 0)
 		return cpu;
 
-	cpu = cpumask_any_distribute(cpus_allowed);
+	if (flags & SCX_PICK_IDLE_IN_NODE)
+		cpu = cpumask_any_and_distribute(cpumask_of_node(node), cpus_allowed);
+	else
+		cpu = cpumask_any_distribute(cpus_allowed);
 	if (cpu < nr_cpu_ids)
 		return cpu;
 	else
-- 
2.48.1

Re: [PATCH sched_ext/for-6.15] sched_ext: idle: Fix scx_bpf_pick_any_cpu_node() behavior

[Tejun Heo]

On Tue, Feb 25, 2025 at 10:37:42AM +0100, Andrea Righi wrote:
> When %SCX_PICK_IDLE_IN_NODE is specified, scx_bpf_pick_any_cpu_node()
> should always return a CPU from the specified node, regardless of its
> idle state.
> 
> Also clarify this logic in the function documentation.
> 
> Fixes: 01059219b0cfd ("sched_ext: idle: Introduce node-aware idle cpu kfunc helpers")
> Signed-off-by: Andrea Righi <arighi@nvidia.com>

Applied to sched_ext/for-6.15.

Thanks.

-- 
tejun