sched: Optimize rq_lockp() usage

rq_lockp() includes a static_branch(), which is asm-goto, which is
asm volatile which defeats regular CSE. This means that:

	if (!static_branch(&foo))
		return simple;

	if (static_branch(&foo) && cond)
		return complex;

Doesn't fold and we get horrible code. Introduce __rq_lockp() without
the static_branch() on.

Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Tested-by: Don Hiatt <dhiatt@digitalocean.com>
Tested-by: Hongyu Ning <hongyu.ning@linux.intel.com>
Tested-by: Vincent Guittot <vincent.guittot@linaro.org>
Link: https://lkml.kernel.org/r/20210422123308.316696988@infradead.org

4 files changed, 36 insertions, 19 deletions

diff --git a/kernel/sched/core.c b/kernel/sched/core.c
index 384b79363a39..42c1c88741c0 100644
--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -300,9 +300,9 @@ void raw_spin_rq_lock_nested(struct rq *rq, int subclass)
 	}
 
 	for (;;) {
-		lock = rq_lockp(rq);
+		lock = __rq_lockp(rq);
 		raw_spin_lock_nested(lock, subclass);
-		if (likely(lock == rq_lockp(rq))) {
+		if (likely(lock == __rq_lockp(rq))) {
 			/* preempt_count *MUST* be > 1 */
 			preempt_enable_no_resched();
 			return;
@@ -325,9 +325,9 @@ bool raw_spin_rq_trylock(struct rq *rq)
 	}
 
 	for (;;) {
-		lock = rq_lockp(rq);
+		lock = __rq_lockp(rq);
 		ret = raw_spin_trylock(lock);
-		if (!ret || (likely(lock == rq_lockp(rq)))) {
+		if (!ret || (likely(lock == __rq_lockp(rq)))) {
 			preempt_enable();
 			return ret;
 		}
@@ -352,7 +352,7 @@ void double_rq_lock(struct rq *rq1, struct rq *rq2)
 		swap(rq1, rq2);
 
 	raw_spin_rq_lock(rq1);
-	if (rq_lockp(rq1) == rq_lockp(rq2))
+	if (__rq_lockp(rq1) == __rq_lockp(rq2))
 		return;
 
 	raw_spin_rq_lock_nested(rq2, SINGLE_DEPTH_NESTING);
@@ -2622,7 +2622,7 @@ void set_task_cpu(struct task_struct *p, unsigned int new_cpu)
 	 * task_rq_lock().
 	 */
 	WARN_ON_ONCE(debug_locks && !(lockdep_is_held(&p->pi_lock) ||
-				      lockdep_is_held(rq_lockp(task_rq(p)))));
+				      lockdep_is_held(__rq_lockp(task_rq(p)))));
 #endif
 	/*
 	 * Clearly, migrating tasks to offline CPUs is a fairly daft thing.
@@ -4248,7 +4248,7 @@ prepare_lock_switch(struct rq *rq, struct task_struct *next, struct rq_flags *rf
 	 * do an early lockdep release here:
 	 */
 	rq_unpin_lock(rq, rf);
-	spin_release(&rq_lockp(rq)->dep_map, _THIS_IP_);
+	spin_release(&__rq_lockp(rq)->dep_map, _THIS_IP_);
 #ifdef CONFIG_DEBUG_SPINLOCK
 	/* this is a valid case when another task releases the spinlock */
 	rq_lockp(rq)->owner = next;
@@ -4262,7 +4262,7 @@ static inline void finish_lock_switch(struct rq *rq)
 	 * fix up the runqueue lock - which gets 'carried over' from
 	 * prev into current:
 	 */
-	spin_acquire(&rq_lockp(rq)->dep_map, 0, 0, _THIS_IP_);
+	spin_acquire(&__rq_lockp(rq)->dep_map, 0, 0, _THIS_IP_);
 	__balance_callbacks(rq);
 	raw_spin_rq_unlock_irq(rq);
 }
diff --git a/kernel/sched/deadline.c b/kernel/sched/deadline.c
index 6e99b8b37c8c..fb0eb9a19fc7 100644
--- a/kernel/sched/deadline.c
+++ b/kernel/sched/deadline.c
@@ -1097,9 +1097,9 @@ static enum hrtimer_restart dl_task_timer(struct hrtimer *timer)
 		 * If the runqueue is no longer available, migrate the
 		 * task elsewhere. This necessarily changes rq.
 		 */
-		lockdep_unpin_lock(rq_lockp(rq), rf.cookie);
+		lockdep_unpin_lock(__rq_lockp(rq), rf.cookie);
 		rq = dl_task_offline_migration(rq, p);
-		rf.cookie = lockdep_pin_lock(rq_lockp(rq));
+		rf.cookie = lockdep_pin_lock(__rq_lockp(rq));
 		update_rq_clock(rq);
 
 		/*
diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index e50bd75067d5..18960d00708a 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -1107,7 +1107,7 @@ struct numa_group {
 static struct numa_group *deref_task_numa_group(struct task_struct *p)
 {
 	return rcu_dereference_check(p->numa_group, p == current ||
-		(lockdep_is_held(rq_lockp(task_rq(p))) && !READ_ONCE(p->on_cpu)));
+		(lockdep_is_held(__rq_lockp(task_rq(p))) && !READ_ONCE(p->on_cpu)));
 }
 
 static struct numa_group *deref_curr_numa_group(struct task_struct *p)
diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h
index 29418b8c05dd..ca30af37b906 100644
--- a/kernel/sched/sched.h
+++ b/kernel/sched/sched.h
@@ -1133,6 +1133,10 @@ static inline bool sched_core_disabled(void)
 	return !static_branch_unlikely(&__sched_core_enabled);
 }
 
+/*
+ * Be careful with this function; not for general use. The return value isn't
+ * stable unless you actually hold a relevant rq->__lock.
+ */
 static inline raw_spinlock_t *rq_lockp(struct rq *rq)
 {
 	if (sched_core_enabled(rq))
@@ -1141,6 +1145,14 @@ static inline raw_spinlock_t *rq_lockp(struct rq *rq)
 	return &rq->__lock;
 }
 
+static inline raw_spinlock_t *__rq_lockp(struct rq *rq)
+{
+	if (rq->core_enabled)
+		return &rq->core->__lock;
+
+	return &rq->__lock;
+}
+
 #else /* !CONFIG_SCHED_CORE */
 
 static inline bool sched_core_enabled(struct rq *rq)
@@ -1158,11 +1170,16 @@ static inline raw_spinlock_t *rq_lockp(struct rq *rq)
 	return &rq->__lock;
 }
 
+static inline raw_spinlock_t *__rq_lockp(struct rq *rq)
+{
+	return &rq->__lock;
+}
+
 #endif /* CONFIG_SCHED_CORE */
 
 static inline void lockdep_assert_rq_held(struct rq *rq)
 {
-	lockdep_assert_held(rq_lockp(rq));
+	lockdep_assert_held(__rq_lockp(rq));
 }
 
 extern void raw_spin_rq_lock_nested(struct rq *rq, int subclass);
@@ -1346,7 +1363,7 @@ extern struct callback_head balance_push_callback;
  */
 static inline void rq_pin_lock(struct rq *rq, struct rq_flags *rf)
 {
-	rf->cookie = lockdep_pin_lock(rq_lockp(rq));
+	rf->cookie = lockdep_pin_lock(__rq_lockp(rq));
 
 #ifdef CONFIG_SCHED_DEBUG
 	rq->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP);
@@ -1364,12 +1381,12 @@ static inline void rq_unpin_lock(struct rq *rq, struct rq_flags *rf)
 		rf->clock_update_flags = RQCF_UPDATED;
 #endif
 
-	lockdep_unpin_lock(rq_lockp(rq), rf->cookie);
+	lockdep_unpin_lock(__rq_lockp(rq), rf->cookie);
 }
 
 static inline void rq_repin_lock(struct rq *rq, struct rq_flags *rf)
 {
-	lockdep_repin_lock(rq_lockp(rq), rf->cookie);
+	lockdep_repin_lock(__rq_lockp(rq), rf->cookie);
 
 #ifdef CONFIG_SCHED_DEBUG
 	/*
@@ -2338,7 +2355,7 @@ static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
 	__acquires(busiest->lock)
 	__acquires(this_rq->lock)
 {
-	if (rq_lockp(this_rq) == rq_lockp(busiest))
+	if (__rq_lockp(this_rq) == __rq_lockp(busiest))
 		return 0;
 
 	if (likely(raw_spin_rq_trylock(busiest)))
@@ -2370,9 +2387,9 @@ static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest)
 static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
 	__releases(busiest->lock)
 {
-	if (rq_lockp(this_rq) != rq_lockp(busiest))
+	if (__rq_lockp(this_rq) != __rq_lockp(busiest))
 		raw_spin_rq_unlock(busiest);
-	lock_set_subclass(&rq_lockp(this_rq)->dep_map, 0, _RET_IP_);
+	lock_set_subclass(&__rq_lockp(this_rq)->dep_map, 0, _RET_IP_);
 }
 
 static inline void double_lock(spinlock_t *l1, spinlock_t *l2)
@@ -2412,7 +2429,7 @@ static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2)
 	__releases(rq1->lock)
 	__releases(rq2->lock)
 {
-	if (rq_lockp(rq1) != rq_lockp(rq2))
+	if (__rq_lockp(rq1) != __rq_lockp(rq2))
 		raw_spin_rq_unlock(rq2);
 	else
 		__release(rq2->lock);