[PATCH sched_ext/for-7.3-fixes] sched_ext: Wait for SCX_OPSS_DISPATCHING before reenqueueing a task

[Tejun Heo]

ebf1ccff79c4 ("sched_ext: Fix ops.dequeue() semantics") moved the final
ops_state store in scx_dispatch_enqueue() after the DSQ unlock so that the
custody update and ops.dequeue() precede it. A task can thus be found on a
DSQ while still SCX_OPSS_DISPATCHING.

The dequeue and core-sched pick paths wait for the state to clear in
ops_dequeue() but the reenqueue paths don't. A reenqueue in that window runs
ops.enqueue() and sets SCX_OPSS_QUEUED before the dispatch has completed.
The dispatcher's final store then overwrites it with SCX_OPSS_NONE and
finish_dispatch() drops every later dispatch of the task.

Wait for SCX_OPSS_DISPATCHING to clear before dequeueing a task for
reenqueue, the same way ops_dequeue() does.

Fixes: ebf1ccff79c4 ("sched_ext: Fix ops.dequeue() semantics")
Cc: stable@xxxxxxxxxxxxxxx # v7.1+
Signed-off-by: Tejun Heo <tj@xxxxxxxxxx>
---
 kernel/sched/ext/ext.c      |   13 +++++++++++++
 kernel/sched/ext/internal.h |    1 +
 kernel/sched/ext/sub.c      |    1 +
 3 files changed, 15 insertions(+)

--- a/kernel/sched/ext/ext.c
+++ b/kernel/sched/ext/ext.c
@@ -4404,6 +4404,17 @@ static bool local_task_should_reenq(stru
 	return *reenq_flags & SCX_REENQ_ANY;
 }
 
+/*
+ * The dispatcher stores the final ops_state after dropping the DSQ lock, so @p
+ * can be found on a DSQ while still %SCX_OPSS_DISPATCHING. Reenqueueing @p
+ * before that store lands would have it clobber the new %SCX_OPSS_QUEUED.
+ */
+void scx_reenq_wait_dispatching(struct task_struct *p)
+{
+	if (unlikely(atomic_long_read_acquire(&p->scx.ops_state) == SCX_OPSS_DISPATCHING))
+		wait_ops_state(p, SCX_OPSS_DISPATCHING);
+}
+
 static u32 reenq_local(struct scx_sched *sch, struct rq *rq, u64 reenq_flags)
 {
 	LIST_HEAD(tasks);
@@ -4447,6 +4458,7 @@ static u32 reenq_local(struct scx_sched
 		if (!local_task_should_reenq(rq, p, &reenq_flags, &reason))
 			continue;
 
+		scx_reenq_wait_dispatching(p);
 		scx_dispatch_dequeue(rq, p);
 
 		if (WARN_ON_ONCE(p->scx.flags & SCX_TASK_REENQ_REASON_MASK))
@@ -4570,6 +4582,7 @@ static void reenq_user(struct rq *rq, st
 		}
 
 		/* @p is on @dsq, its rq and @dsq are locked */
+		scx_reenq_wait_dispatching(p);
 		dispatch_dequeue_locked(p, dsq);
 		raw_spin_unlock(&dsq->lock);
 
--- a/kernel/sched/ext/internal.h
+++ b/kernel/sched/ext/internal.h
@@ -2078,6 +2078,7 @@ void scx_kick_cpu(struct scx_sched *sch,
 u64 __scx_bpf_now(struct rq *rq);
 void schedule_dsq_reenq(struct scx_sched *sch, struct scx_dispatch_q *dsq,
 			u64 reenq_flags, struct rq *locked_rq);
+void scx_reenq_wait_dispatching(struct task_struct *p);
 int __scx_init_task(struct scx_sched *sch, struct task_struct *p,
 		    struct cgroup *cgrp, bool fork);
 void scx_enable_task(struct scx_sched *sch, struct task_struct *p);
--- a/kernel/sched/ext/sub.c
+++ b/kernel/sched/ext/sub.c
@@ -801,6 +801,7 @@ void scx_reenq_reject(struct rq *rq)
 		if (WARN_ON_ONCE(p->migration_pending))
 			continue;
 
+		scx_reenq_wait_dispatching(p);
 		scx_dispatch_dequeue(rq, p);
 
 		if (WARN_ON_ONCE(p->scx.flags & SCX_TASK_REENQ_REASON_MASK))