[PATCH 1/2] sched_ext: idle: Honor SCX_PICK_IDLE_CORE in CPU selection shortcuts

[Hui Su]

SCX_PICK_IDLE_CORE requires selecting a CPU whose SMT siblings are all
idle. The restricted select_cpu kfunc path now derives its candidate from
task state, but it does not check that condition before claiming the CPU.

The WAKE_SYNC shortcuts in scx_select_cpu_dfl() can bypass strict-core
selection on SMT systems as well. Skip those shortcuts when
SCX_PICK_IDLE_CORE is requested, and validate restricted candidates against
their idle SMT mask before consuming the idle bit.

Preserve individual-idle selection when the flag is not set and on
non-SMT systems.

Fixes: 683d2d0faba1 ("sched_ext: idle: Introduce scx_bpf_select_cpu_and()")
Cc: stable@xxxxxxxxxxxxxxx
Signed-off-by: Hui Su <sh_def@xxxxxxx>
---
 kernel/sched/ext/idle.c | 6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)

diff --git a/kernel/sched/ext/idle.c b/kernel/sched/ext/idle.c
index 28ed9e327085..09e40928aef4 100644
--- a/kernel/sched/ext/idle.c
+++ b/kernel/sched/ext/idle.c
@@ -523,7 +523,7 @@ s32 scx_select_cpu_dfl(struct task_struct *p, s32 prev_cpu, u64 wake_flags,
 	/*
 	 * If WAKE_SYNC, try to migrate the wakee to the waker's CPU.
 	 */
-	if (wake_flags & SCX_WAKE_SYNC) {
+	if ((wake_flags & SCX_WAKE_SYNC) && !(sched_smt_active() && (flags & SCX_PICK_IDLE_CORE))) {
 		int waker_node;
 
 		/*
@@ -976,6 +976,10 @@ static s32 select_cpu_restricted(struct task_struct *p, s32 prev_cpu,
 			return -EBUSY;
 	}
 
+	if ((flags & SCX_PICK_IDLE_CORE) && sched_smt_active() &&
+	    !cpumask_test_cpu(cpu, idle_cpumask(scx_cpu_node_if_enabled(cpu))->smt))
+		return -EBUSY;
+
 	if (!scx_idle_test_and_clear_cpu(cpu))
 		return -EBUSY;
 
-- 
2.55.0