Merge tag 'sched_ext-for-7.0-rc6-fixes-2' of git://git.kernel.org/pub/scm/linux/kernel/git/tj/sched_ext

Pull sched_ext fixes from Tejun Heo:
 "These are late but both fix subtle yet critical problems and the blast
  radius is limited strictly to sched_ext.

   - Fix stale direct dispatch state in ddsp_dsq_id which can cause
     spurious warnings in mark_direct_dispatch() on task wakeup

   - Fix is_bpf_migration_disabled() false negative on non-PREEMPT_RCU
     configs which can lead to incorrectly dispatching migration-
     disabled tasks to remote CPUs"

* tag 'sched_ext-for-7.0-rc6-fixes-2' of git://git.kernel.org/pub/scm/linux/kernel/git/tj/sched_ext:
  sched_ext: Fix stale direct dispatch state in ddsp_dsq_id
  sched_ext: Fix is_bpf_migration_disabled() false negative on non-PREEMPT_RCU

2 files changed, 54 insertions, 26 deletions

diff --git a/kernel/sched/ext.c b/kernel/sched/ext.c
index d5bdcdb3f700..064eaa76be4b 100644
--- a/kernel/sched/ext.c
+++ b/kernel/sched/ext.c
@@ -1110,15 +1110,6 @@ static void dispatch_enqueue(struct scx_sched *sch, struct scx_dispatch_q *dsq,
 	p->scx.dsq = dsq;
 
 	/*
-	 * scx.ddsp_dsq_id and scx.ddsp_enq_flags are only relevant on the
-	 * direct dispatch path, but we clear them here because the direct
-	 * dispatch verdict may be overridden on the enqueue path during e.g.
-	 * bypass.
-	 */
-	p->scx.ddsp_dsq_id = SCX_DSQ_INVALID;
-	p->scx.ddsp_enq_flags = 0;
-
-	/*
 	 * We're transitioning out of QUEUEING or DISPATCHING. store_release to
 	 * match waiters' load_acquire.
 	 */
@@ -1283,12 +1274,34 @@ static void mark_direct_dispatch(struct scx_sched *sch,
 	p->scx.ddsp_enq_flags = enq_flags;
 }
 
+/*
+ * Clear @p direct dispatch state when leaving the scheduler.
+ *
+ * Direct dispatch state must be cleared in the following cases:
+ *  - direct_dispatch(): cleared on the synchronous enqueue path, deferred
+ *    dispatch keeps the state until consumed
+ *  - process_ddsp_deferred_locals(): cleared after consuming deferred state,
+ *  - do_enqueue_task(): cleared on enqueue fallbacks where the dispatch
+ *    verdict is ignored (local/global/bypass)
+ *  - dequeue_task_scx(): cleared after dispatch_dequeue(), covering deferred
+ *    cancellation and holding_cpu races
+ *  - scx_disable_task(): cleared for queued wakeup tasks, which are excluded by
+ *    the scx_bypass() loop, so that stale state is not reused by a subsequent
+ *    scheduler instance
+ */
+static inline void clear_direct_dispatch(struct task_struct *p)
+{
+	p->scx.ddsp_dsq_id = SCX_DSQ_INVALID;
+	p->scx.ddsp_enq_flags = 0;
+}
+
 static void direct_dispatch(struct scx_sched *sch, struct task_struct *p,
 			    u64 enq_flags)
 {
 	struct rq *rq = task_rq(p);
 	struct scx_dispatch_q *dsq =
 		find_dsq_for_dispatch(sch, rq, p->scx.ddsp_dsq_id, p);
+	u64 ddsp_enq_flags;
 
 	touch_core_sched_dispatch(rq, p);
 
@@ -1329,8 +1342,10 @@ static void direct_dispatch(struct scx_sched *sch, struct task_struct *p,
 		return;
 	}
 
-	dispatch_enqueue(sch, dsq, p,
-			 p->scx.ddsp_enq_flags | SCX_ENQ_CLEAR_OPSS);
+	ddsp_enq_flags = p->scx.ddsp_enq_flags;
+	clear_direct_dispatch(p);
+
+	dispatch_enqueue(sch, dsq, p, ddsp_enq_flags | SCX_ENQ_CLEAR_OPSS);
 }
 
 static bool scx_rq_online(struct rq *rq)
@@ -1439,6 +1454,7 @@ enqueue:
 	 */
 	touch_core_sched(rq, p);
 	refill_task_slice_dfl(sch, p);
+	clear_direct_dispatch(p);
 	dispatch_enqueue(sch, dsq, p, enq_flags);
 }
 
@@ -1610,6 +1626,7 @@ static bool dequeue_task_scx(struct rq *rq, struct task_struct *p, int deq_flags
 	sub_nr_running(rq, 1);
 
 	dispatch_dequeue(rq, p);
+	clear_direct_dispatch(p);
 	return true;
 }
 
@@ -2293,13 +2310,15 @@ static void process_ddsp_deferred_locals(struct rq *rq)
 				struct task_struct, scx.dsq_list.node))) {
 		struct scx_sched *sch = scx_root;
 		struct scx_dispatch_q *dsq;
+		u64 dsq_id = p->scx.ddsp_dsq_id;
+		u64 enq_flags = p->scx.ddsp_enq_flags;
 
 		list_del_init(&p->scx.dsq_list.node);
+		clear_direct_dispatch(p);
 
-		dsq = find_dsq_for_dispatch(sch, rq, p->scx.ddsp_dsq_id, p);
+		dsq = find_dsq_for_dispatch(sch, rq, dsq_id, p);
 		if (!WARN_ON_ONCE(dsq->id != SCX_DSQ_LOCAL))
-			dispatch_to_local_dsq(sch, rq, dsq, p,
-					      p->scx.ddsp_enq_flags);
+			dispatch_to_local_dsq(sch, rq, dsq, p, enq_flags);
 	}
 }
 
@@ -3015,6 +3034,8 @@ static void scx_disable_task(struct task_struct *p)
 	lockdep_assert_rq_held(rq);
 	WARN_ON_ONCE(scx_get_task_state(p) != SCX_TASK_ENABLED);
 
+	clear_direct_dispatch(p);
+
 	if (SCX_HAS_OP(sch, disable))
 		SCX_CALL_OP_TASK(sch, SCX_KF_REST, disable, rq, p);
 	scx_set_task_state(p, SCX_TASK_READY);
diff --git a/kernel/sched/ext_idle.c b/kernel/sched/ext_idle.c
index 0ae93cd64004..44c3a50c542c 100644
--- a/kernel/sched/ext_idle.c
+++ b/kernel/sched/ext_idle.c
@@ -860,25 +860,32 @@ static bool check_builtin_idle_enabled(struct scx_sched *sch)
  * code.
  *
  * We can't simply check whether @p->migration_disabled is set in a
- * sched_ext callback, because migration is always disabled for the current
- * task while running BPF code.
+ * sched_ext callback, because the BPF prolog (__bpf_prog_enter) may disable
+ * migration for the current task while running BPF code.
  *
- * The prolog (__bpf_prog_enter) and epilog (__bpf_prog_exit) respectively
- * disable and re-enable migration. For this reason, the current task
- * inside a sched_ext callback is always a migration-disabled task.
+ * Since the BPF prolog calls migrate_disable() only when CONFIG_PREEMPT_RCU
+ * is enabled (via rcu_read_lock_dont_migrate()), migration_disabled == 1 for
+ * the current task is ambiguous only in that case: it could be from the BPF
+ * prolog rather than a real migrate_disable() call.
  *
- * Therefore, when @p->migration_disabled == 1, check whether @p is the
- * current task or not: if it is, then migration was not disabled before
- * entering the callback, otherwise migration was disabled.
+ * Without CONFIG_PREEMPT_RCU, the BPF prolog never calls migrate_disable(),
+ * so migration_disabled == 1 always means the task is truly
+ * migration-disabled.
+ *
+ * Therefore, when migration_disabled == 1 and CONFIG_PREEMPT_RCU is enabled,
+ * check whether @p is the current task or not: if it is, then migration was
+ * not disabled before entering the callback, otherwise migration was disabled.
  *
  * Returns true if @p is migration-disabled, false otherwise.
  */
 static bool is_bpf_migration_disabled(const struct task_struct *p)
 {
-	if (p->migration_disabled == 1)
-		return p != current;
-	else
-		return p->migration_disabled;
+	if (p->migration_disabled == 1) {
+		if (IS_ENABLED(CONFIG_PREEMPT_RCU))
+			return p != current;
+		return true;
+	}
+	return p->migration_disabled;
 }
 
 static s32 select_cpu_from_kfunc(struct scx_sched *sch, struct task_struct *p,