[PATCH v2] sched/core: Preserve wake flags across queued wakeups

[Shubhang Kaushik (Ampere)]

The queued wakeup path only remembers whether the wakee migrated. When
the target CPU drains the wakelist, sched_ttwu_pending() turns that back
into either WF_MIGRATED or 0 and passes it to ttwu_do_activate().

That drops the rest of the wake flags seen by the direct wakeup path.
Some of those flags can still matter when the queued wakeup is drained
on a non-idle target CPU, so queued and direct wakeups can make different
preemption decisions.

Store the relevant wake_flags bits instead of a single boolean and pass
them through when the wakelist is drained. Keep WF_CURRENT_CPU out of
the saved mask as it is a placement hint for CPU selection, not an
activation flag.

Signed-off-by: Shubhang Kaushik (Ampere) <sh@xxxxxxxxxx>
---
This aims to preserve activation-relevant wake flags across queued
remote wakeups, keeping the queued and direct wakeup paths consistent when
calling ttwu_do_activate().

Tested on an Ampere Altra 80 CPU system with perf bench sched messaging and
hackbench. Results were within run-to-run noise of the unpatched mainline
baseline.

Baseline: mainline origin/master at d96fcfe1b7f9.
---
Changes in v2:
  - Move sched_remote_wakeup_flags to a standalone u8 outside the scheduler
    bitfields.
  - Drop the unnecessary reset in sched_ttwu_pending().
  - Add WF_TTWU_QUEUE_MASK for the wake flags preserved across the wakelist.
  - Tighten the changelog around direct-vs-queued wakeup consistency.

Link to v1: https://lore.kernel.org/r/20260710-b4-sched-ttwu-wake-flags-v1-1-23182a4c5367@xxxxxxxxxx
---
 include/linux/sched.h | 17 +----------------
 kernel/sched/core.c   |  7 +++++--
 kernel/sched/sched.h  |  1 +
 3 files changed, 7 insertions(+), 18 deletions(-)

diff --git a/include/linux/sched.h b/include/linux/sched.h
index 373bcc0598d10b4256a11f8c8373ece78fa5e0e5..a2064476a6bf84a2513c9aad3644f88cbc09a298 100644
--- a/include/linux/sched.h
+++ b/include/linux/sched.h
@@ -983,6 +983,7 @@ struct task_struct {
 
 	/* Used for emulating ABI behavior of previous Linux versions: */
 	unsigned int			personality;
+	u8				sched_remote_wakeup_flags;
 
 	/* Scheduler bits, serialized by scheduler locks: */
 	unsigned			sched_reset_on_fork:1;
@@ -993,22 +994,6 @@ struct task_struct {
 	/* Force alignment to the next boundary: */
 	unsigned			:0;
 
-	/* Unserialized, strictly 'current' */
-
-	/*
-	 * This field must not be in the scheduler word above due to wakelist
-	 * queueing no longer being serialized by p->on_cpu. However:
-	 *
-	 * p->XXX = X;			ttwu()
-	 * schedule()			  if (p->on_rq && ..) // false
-	 *   smp_mb__after_spinlock();	  if (smp_load_acquire(&p->on_cpu) && //true
-	 *   deactivate_task()		      ttwu_queue_wakelist())
-	 *     p->on_rq = 0;			p->sched_remote_wakeup = Y;
-	 *
-	 * guarantees all stores of 'current' are visible before
-	 * ->sched_remote_wakeup gets used, so it can be in this word.
-	 */
-	unsigned			sched_remote_wakeup:1;
 #ifdef CONFIG_RT_MUTEXES
 	unsigned			sched_rt_mutex:1;
 #endif
diff --git a/kernel/sched/core.c b/kernel/sched/core.c
index 96226707c2f6135341aa779b8262f113e103d8ad..62708634259f0cf870897fd5bfe5c15ff88e40d0 100644
--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -3902,13 +3902,16 @@ void sched_ttwu_pending(void *arg)
 	update_rq_clock(rq);
 
 	llist_for_each_entry_safe(p, t, llist, wake_entry.llist) {
+		int wake_flags;
+
 		if (WARN_ON_ONCE(p->on_cpu))
 			smp_cond_load_acquire(&p->on_cpu, !VAL);
 
 		if (WARN_ON_ONCE(task_cpu(p) != cpu_of(rq)))
 			set_task_cpu(p, cpu_of(rq));
 
-		ttwu_do_activate(rq, p, p->sched_remote_wakeup ? WF_MIGRATED : 0, &rf);
+		wake_flags = p->sched_remote_wakeup_flags;
+		ttwu_do_activate(rq, p, wake_flags, &rf);
 	}
 
 	/*
@@ -3951,7 +3954,7 @@ static void __ttwu_queue_wakelist(struct task_struct *p, int cpu, int wake_flags
 {
 	struct rq *rq = cpu_rq(cpu);
 
-	p->sched_remote_wakeup = !!(wake_flags & WF_MIGRATED);
+	p->sched_remote_wakeup_flags = wake_flags & WF_TTWU_QUEUE_MASK;
 
 	WRITE_ONCE(rq->ttwu_pending, 1);
 #ifdef CONFIG_SMP
diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h
index 56acf502ba260ab18bacd7a4c2efdec612d50125..50cd0141aeb49af4beb1af51862e319f096ec5f2 100644
--- a/kernel/sched/sched.h
+++ b/kernel/sched/sched.h
@@ -2494,6 +2494,7 @@ static inline int task_on_rq_migrating(struct task_struct *p)
 #define WF_MIGRATED		0x20 /* Internal use, task got migrated */
 #define WF_CURRENT_CPU		0x40 /* Prefer to move the wakee to the current CPU. */
 #define WF_RQ_SELECTED		0x80 /* ->select_task_rq() was called */
+#define WF_TTWU_QUEUE_MASK	(WF_TTWU | WF_SYNC | WF_MIGRATED | WF_RQ_SELECTED)
 
 static_assert(WF_EXEC == SD_BALANCE_EXEC);
 static_assert(WF_FORK == SD_BALANCE_FORK);

---
base-commit: d96fcfe1b7f94ac742984ae7986b94a116abff1b
change-id: 20260710-b4-sched-ttwu-wake-flags-e87ada0e65e1

Best regards,
-- 
Shubhang Kaushik (Ampere) <sh@xxxxxxxxxx>