[PATCH 1/2] locking: Use spin primitives for busy loops

[Nicholas Piggin]

Commit fd851a3cdc ("spin loop primitives for busy waiting") introduced
a begin/relax/end sequence for busy loops, to improve behaviour with
some architectures.

Convert most of the generic locking primitives over to use these spin
primitives.

Signed-off-by: Nicholas Piggin <npiggin@xxxxxxxxx>
---
 include/linux/bit_spinlock.h        |  5 ++---
 include/linux/seqlock.h             |  9 ++++-----
 kernel/locking/mcs_spinlock.h       |  6 ++----
 kernel/locking/mutex.c              | 10 ++++++++--
 kernel/locking/osq_lock.c           | 17 +++++++++++++----
 kernel/locking/qrwlock.c            | 11 ++++++++---
 kernel/locking/qspinlock.c          | 14 ++++++++++----
 kernel/locking/qspinlock_paravirt.h | 16 ++++++++++++----
 kernel/locking/rwsem-xadd.c         |  9 +++++++--
 9 files changed, 66 insertions(+), 31 deletions(-)

diff --git a/include/linux/bit_spinlock.h b/include/linux/bit_spinlock.h
index 3b5bafce4337..4cec87d9cde8 100644
--- a/include/linux/bit_spinlock.h
+++ b/include/linux/bit_spinlock.h
@@ -3,6 +3,7 @@
 
 #include <linux/kernel.h>
 #include <linux/preempt.h>
+#include <linux/processor.h>
 #include <linux/atomic.h>
 #include <linux/bug.h>
 
@@ -25,9 +26,7 @@ static inline void bit_spin_lock(int bitnum, unsigned long *addr)
 #if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK)
 	while (unlikely(test_and_set_bit_lock(bitnum, addr))) {
 		preempt_enable();
-		do {
-			cpu_relax();
-		} while (test_bit(bitnum, addr));
+		spin_until_cond(!test_bit(bitnum, addr));
 		preempt_disable();
 	}
 #endif
diff --git a/include/linux/seqlock.h b/include/linux/seqlock.h
index ead97654c4e9..f4bd4a6c89d9 100644
--- a/include/linux/seqlock.h
+++ b/include/linux/seqlock.h
@@ -32,6 +32,7 @@
  * by Keith Owens and Andrea Arcangeli
  */
 
+#include <linux/processor.h>
 #include <linux/spinlock.h>
 #include <linux/preempt.h>
 #include <linux/lockdep.h>
@@ -108,12 +109,10 @@ static inline unsigned __read_seqcount_begin(const seqcount_t *s)
 {
 	unsigned ret;
 
-repeat:
 	ret = READ_ONCE(s->sequence);
-	if (unlikely(ret & 1)) {
-		cpu_relax();
-		goto repeat;
-	}
+	if (unlikely(ret & 1))
+		spin_until_cond( !((ret = READ_ONCE(s->sequence)) & 1) );
+
 	return ret;
 }
 
diff --git a/kernel/locking/mcs_spinlock.h b/kernel/locking/mcs_spinlock.h
index 6a385aabcce7..a91a0cc46a4c 100644
--- a/kernel/locking/mcs_spinlock.h
+++ b/kernel/locking/mcs_spinlock.h
@@ -27,8 +27,7 @@ struct mcs_spinlock {
  */
 #define arch_mcs_spin_lock_contended(l)					\
 do {									\
-	while (!(smp_load_acquire(l)))					\
-		cpu_relax();						\
+	spin_until_cond(smp_load_acquire(l));				\
 } while (0)
 #endif
 
@@ -107,8 +106,7 @@ void mcs_spin_unlock(struct mcs_spinlock **lock, struct mcs_spinlock *node)
 		if (likely(cmpxchg_release(lock, node, NULL) == node))
 			return;
 		/* Wait until the next pointer is set */
-		while (!(next = READ_ONCE(node->next)))
-			cpu_relax();
+		spin_until_cond((next = READ_ONCE(node->next)) != 0);
 	}
 
 	/* Pass lock to next waiter. */
diff --git a/kernel/locking/mutex.c b/kernel/locking/mutex.c
index 858a07590e39..0ffa1cd7f12b 100644
--- a/kernel/locking/mutex.c
+++ b/kernel/locking/mutex.c
@@ -427,6 +427,7 @@ bool mutex_spin_on_owner(struct mutex *lock, struct task_struct *owner,
 	bool ret = true;
 
 	rcu_read_lock();
+	spin_begin();
 	while (__mutex_owner(lock) == owner) {
 		/*
 		 * Ensure we emit the owner->on_cpu, dereference _after_
@@ -450,8 +451,9 @@ bool mutex_spin_on_owner(struct mutex *lock, struct task_struct *owner,
 			break;
 		}
 
-		cpu_relax();
+		spin_cpu_relax();
 	}
+	spin_end();
 	rcu_read_unlock();
 
 	return ret;
@@ -532,6 +534,7 @@ mutex_optimistic_spin(struct mutex *lock, struct ww_acquire_ctx *ww_ctx,
 			goto fail;
 	}
 
+	spin_begin();
 	for (;;) {
 		struct task_struct *owner;
 
@@ -553,8 +556,9 @@ mutex_optimistic_spin(struct mutex *lock, struct ww_acquire_ctx *ww_ctx,
 		 * memory barriers as we'll eventually observe the right
 		 * values at the cost of a few extra spins.
 		 */
-		cpu_relax();
+		spin_cpu_relax();
 	}
+	spin_end();
 
 	if (!waiter)
 		osq_unlock(&lock->osq);
@@ -563,6 +567,8 @@ mutex_optimistic_spin(struct mutex *lock, struct ww_acquire_ctx *ww_ctx,
 
 
 fail_unlock:
+	spin_end();
+
 	if (!waiter)
 		osq_unlock(&lock->osq);
 
diff --git a/kernel/locking/osq_lock.c b/kernel/locking/osq_lock.c
index a3167941093b..9dd58bbe60b7 100644
--- a/kernel/locking/osq_lock.c
+++ b/kernel/locking/osq_lock.c
@@ -53,6 +53,7 @@ osq_wait_next(struct optimistic_spin_queue *lock,
 	 */
 	old = prev ? prev->cpu : OSQ_UNLOCKED_VAL;
 
+	spin_begin();
 	for (;;) {
 		if (atomic_read(&lock->tail) == curr &&
 		    atomic_cmpxchg_acquire(&lock->tail, curr, old) == curr) {
@@ -80,8 +81,9 @@ osq_wait_next(struct optimistic_spin_queue *lock,
 				break;
 		}
 
-		cpu_relax();
+		spin_cpu_relax();
 	}
+	spin_end();
 
 	return next;
 }
@@ -107,6 +109,8 @@ bool osq_lock(struct optimistic_spin_queue *lock)
 	if (old == OSQ_UNLOCKED_VAL)
 		return true;
 
+	spin_begin();
+
 	prev = decode_cpu(old);
 	node->prev = prev;
 	WRITE_ONCE(prev->next, node);
@@ -129,8 +133,9 @@ bool osq_lock(struct optimistic_spin_queue *lock)
 		if (need_resched() || vcpu_is_preempted(node_cpu(node->prev)))
 			goto unqueue;
 
-		cpu_relax();
+		spin_cpu_relax();
 	}
+	spin_end();
 	return true;
 
 unqueue:
@@ -152,10 +157,12 @@ bool osq_lock(struct optimistic_spin_queue *lock)
 		 * in which case we should observe @node->locked becomming
 		 * true.
 		 */
-		if (smp_load_acquire(&node->locked))
+		if (smp_load_acquire(&node->locked)) {
+			spin_end();
 			return true;
+		}
 
-		cpu_relax();
+		spin_cpu_relax();
 
 		/*
 		 * Or we race against a concurrent unqueue()'s step-B, in which
@@ -164,6 +171,8 @@ bool osq_lock(struct optimistic_spin_queue *lock)
 		prev = READ_ONCE(node->prev);
 	}
 
+	spin_end();
+
 	/*
 	 * Step - B -- stabilize @next
 	 *
diff --git a/kernel/locking/qrwlock.c b/kernel/locking/qrwlock.c
index 2655f26ec882..186ff495097d 100644
--- a/kernel/locking/qrwlock.c
+++ b/kernel/locking/qrwlock.c
@@ -54,10 +54,12 @@ struct __qrwlock {
 static __always_inline void
 rspin_until_writer_unlock(struct qrwlock *lock, u32 cnts)
 {
+	spin_begin();
 	while ((cnts & _QW_WMASK) == _QW_LOCKED) {
-		cpu_relax();
+		spin_cpu_relax();
 		cnts = atomic_read_acquire(&lock->cnts);
 	}
+	spin_end();
 }
 
 /**
@@ -124,6 +126,7 @@ void queued_write_lock_slowpath(struct qrwlock *lock)
 	 * Set the waiting flag to notify readers that a writer is pending,
 	 * or wait for a previous writer to go away.
 	 */
+	spin_begin();
 	for (;;) {
 		struct __qrwlock *l = (struct __qrwlock *)lock;
 
@@ -131,7 +134,7 @@ void queued_write_lock_slowpath(struct qrwlock *lock)
 		   (cmpxchg_relaxed(&l->wmode, 0, _QW_WAITING) == 0))
 			break;
 
-		cpu_relax();
+		spin_cpu_relax();
 	}
 
 	/* When no more readers, set the locked flag */
@@ -142,8 +145,10 @@ void queued_write_lock_slowpath(struct qrwlock *lock)
 					    _QW_LOCKED) == _QW_WAITING))
 			break;
 
-		cpu_relax();
+		spin_cpu_relax();
 	}
+	spin_end();
+
 unlock:
 	arch_spin_unlock(&lock->wait_lock);
 }
diff --git a/kernel/locking/qspinlock.c b/kernel/locking/qspinlock.c
index fd24153e8a48..52ebcebf6fa8 100644
--- a/kernel/locking/qspinlock.c
+++ b/kernel/locking/qspinlock.c
@@ -362,6 +362,7 @@ void queued_spin_unlock_wait(struct qspinlock *lock)
 {
 	u32 val;
 
+	spin_begin();
 	for (;;) {
 		val = atomic_read(&lock->val);
 
@@ -372,14 +373,15 @@ void queued_spin_unlock_wait(struct qspinlock *lock)
 			break;
 
 		/* not locked, but pending, wait until we observe the lock */
-		cpu_relax();
+		spin_cpu_relax();
 	}
 
 	/* any unlock is good */
 	while (atomic_read(&lock->val) & _Q_LOCKED_MASK)
-		cpu_relax();
+		spin_cpu_relax();
 
 done:
+	spin_end();
 	smp_acquire__after_ctrl_dep();
 }
 EXPORT_SYMBOL(queued_spin_unlock_wait);
@@ -428,8 +430,10 @@ void queued_spin_lock_slowpath(struct qspinlock *lock, u32 val)
 	 * 0,1,0 -> 0,0,1
 	 */
 	if (val == _Q_PENDING_VAL) {
+		spin_begin();
 		while ((val = atomic_read(&lock->val)) == _Q_PENDING_VAL)
-			cpu_relax();
+			spin_cpu_relax();
+		spin_end();
 	}
 
 	/*
@@ -609,8 +613,10 @@ void queued_spin_lock_slowpath(struct qspinlock *lock, u32 val)
 	 * contended path; wait for next if not observed yet, release.
 	 */
 	if (!next) {
+		spin_begin();
 		while (!(next = READ_ONCE(node->next)))
-			cpu_relax();
+			spin_cpu_relax();
+		spin_end();
 	}
 
 	arch_mcs_spin_unlock_contended(&next->locked);
diff --git a/kernel/locking/qspinlock_paravirt.h b/kernel/locking/qspinlock_paravirt.h
index 4ccfcaae5b89..88817e41fadf 100644
--- a/kernel/locking/qspinlock_paravirt.h
+++ b/kernel/locking/qspinlock_paravirt.h
@@ -293,15 +293,19 @@ static void pv_wait_node(struct mcs_spinlock *node, struct mcs_spinlock *prev)
 	bool wait_early;
 
 	for (;;) {
+		spin_begin();
 		for (wait_early = false, loop = SPIN_THRESHOLD; loop; loop--) {
-			if (READ_ONCE(node->locked))
+			if (READ_ONCE(node->locked)) {
+				spin_end();
 				return;
+			}
 			if (pv_wait_early(pp, loop)) {
 				wait_early = true;
 				break;
 			}
-			cpu_relax();
+			spin_cpu_relax();
 		}
+		spin_end();
 
 		/*
 		 * Order pn->state vs pn->locked thusly:
@@ -417,11 +421,15 @@ pv_wait_head_or_lock(struct qspinlock *lock, struct mcs_spinlock *node)
 		 * disable lock stealing before attempting to acquire the lock.
 		 */
 		set_pending(lock);
+		spin_begin();
 		for (loop = SPIN_THRESHOLD; loop; loop--) {
-			if (trylock_clear_pending(lock))
+			if (trylock_clear_pending(lock)) {
+				spin_end();
 				goto gotlock;
-			cpu_relax();
+			}
+			spin_cpu_relax();
 		}
+		spin_end();
 		clear_pending(lock);
 
 
diff --git a/kernel/locking/rwsem-xadd.c b/kernel/locking/rwsem-xadd.c
index 34e727f18e49..2d0e539f1a95 100644
--- a/kernel/locking/rwsem-xadd.c
+++ b/kernel/locking/rwsem-xadd.c
@@ -358,6 +358,7 @@ static noinline bool rwsem_spin_on_owner(struct rw_semaphore *sem)
 		goto out;
 
 	rcu_read_lock();
+	spin_begin();
 	while (sem->owner == owner) {
 		/*
 		 * Ensure we emit the owner->on_cpu, dereference _after_
@@ -373,12 +374,14 @@ static noinline bool rwsem_spin_on_owner(struct rw_semaphore *sem)
 		 */
 		if (!owner->on_cpu || need_resched() ||
 				vcpu_is_preempted(task_cpu(owner))) {
+			spin_end();
 			rcu_read_unlock();
 			return false;
 		}
 
-		cpu_relax();
+		spin_cpu_relax();
 	}
+	spin_end();
 	rcu_read_unlock();
 out:
 	/*
@@ -408,6 +411,7 @@ static bool rwsem_optimistic_spin(struct rw_semaphore *sem)
 	 *  2) readers own the lock as we can't determine if they are
 	 *     actively running or not.
 	 */
+	spin_begin();
 	while (rwsem_spin_on_owner(sem)) {
 		/*
 		 * Try to acquire the lock
@@ -432,8 +436,9 @@ static bool rwsem_optimistic_spin(struct rw_semaphore *sem)
 		 * memory barriers as we'll eventually observe the right
 		 * values at the cost of a few extra spins.
 		 */
-		cpu_relax();
+		spin_cpu_relax();
 	}
+	spin_end();
 	osq_unlock(&sem->osq);
 done:
 	preempt_enable();
-- 
2.13.3