[PATCH tip/core/rcu 0/4] RCU-sync updates for v5.3

[PATCH tip/core/rcu 0/4] RCU-sync updates for v5.3

[Paul E. McKenney]

Hello!

This series contains flavor-consolidation updates to RCU-sync:

1.	Kill rcu_sync_type/gp_type, courtesy of Oleg Nesterov.

2.	Use DEFINE_STATIC_PERCPU_RWSEM() to initialize dup_mmap_sem in
	uprobes, courtesy of Oleg Nesterov.

3.	Add DEFINE_PERCPU_RWSEM(), use it to initialize
	cgroup_threadgroup_rwsem in percpu-rwsem, courtesy of Oleg Nesterov.

4.	Simplify the state machine, courtesy of Oleg Nesterov.

							Thanx, Paul

------------------------------------------------------------------------

 include/linux/percpu-rwsem.h  |   10 +
 include/linux/rcu_sync.h      |   40 ++-----
 kernel/cgroup/cgroup.c        |    3 
 kernel/events/uprobes.c       |    4 
 kernel/locking/percpu-rwsem.c |    2 
 kernel/rcu/sync.c             |  220 +++++++++++++++++++-----------------------
 6 files changed, 121 insertions(+), 158 deletions(-)

[PATCH tip/core/rcu 1/4] rcu/sync: Kill rcu_sync_type/gp_type

[Paul E. McKenney]

From: Oleg Nesterov <oleg@redhat.com>

Now that the RCU flavors have been consolidated, rcu_sync_type makes no
sense because none of internal update functions aside from .held() depend
on gp_type.  This commit therefore removes this field and consolidates
the relevant code.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
[ paulmck: Added RCU and RCU-bh checks to rcu_sync_is_idle(). ]
[ paulmck: And applied subsequent feedback from Oleg Nesterov. ]
Signed-off-by: Paul E. McKenney <paulmck@linux.ibm.com>
---
 include/linux/percpu-rwsem.h  |  2 +-
 include/linux/rcu_sync.h      | 36 +++++++----------------
 kernel/locking/percpu-rwsem.c |  2 +-
 kernel/rcu/sync.c             | 55 ++++-------------------------------
 4 files changed, 17 insertions(+), 78 deletions(-)

diff --git a/include/linux/percpu-rwsem.h b/include/linux/percpu-rwsem.h
index 03cb4b6f842e..6887636ea169 100644
--- a/include/linux/percpu-rwsem.h
+++ b/include/linux/percpu-rwsem.h
@@ -20,7 +20,7 @@ struct percpu_rw_semaphore {
 #define DEFINE_STATIC_PERCPU_RWSEM(name)				\
 static DEFINE_PER_CPU(unsigned int, __percpu_rwsem_rc_##name);		\
 static struct percpu_rw_semaphore name = {				\
-	.rss = __RCU_SYNC_INITIALIZER(name.rss, RCU_SCHED_SYNC),	\
+	.rss = __RCU_SYNC_INITIALIZER(name.rss),			\
 	.read_count = &__percpu_rwsem_rc_##name,			\
 	.rw_sem = __RWSEM_INITIALIZER(name.rw_sem),			\
 	.writer = __RCUWAIT_INITIALIZER(name.writer),			\
diff --git a/include/linux/rcu_sync.h b/include/linux/rcu_sync.h
index 6fc53a1345b3..87971e85519c 100644
--- a/include/linux/rcu_sync.h
+++ b/include/linux/rcu_sync.h
@@ -13,8 +13,6 @@
 #include <linux/wait.h>
 #include <linux/rcupdate.h>
 
-enum rcu_sync_type { RCU_SYNC, RCU_SCHED_SYNC, RCU_BH_SYNC };
-
 /* Structure to mediate between updaters and fastpath-using readers.  */
 struct rcu_sync {
 	int			gp_state;
@@ -23,52 +21,38 @@ struct rcu_sync {
 
 	int			cb_state;
 	struct rcu_head		cb_head;
-
-	enum rcu_sync_type	gp_type;
 };
 
-extern void rcu_sync_lockdep_assert(struct rcu_sync *);
-
 /**
  * rcu_sync_is_idle() - Are readers permitted to use their fastpaths?
  * @rsp: Pointer to rcu_sync structure to use for synchronization
  *
- * Returns true if readers are permitted to use their fastpaths.
- * Must be invoked within an RCU read-side critical section whose
- * flavor matches that of the rcu_sync struture.
+ * Returns true if readers are permitted to use their fastpaths.  Must be
+ * invoked within some flavor of RCU read-side critical section.
  */
 static inline bool rcu_sync_is_idle(struct rcu_sync *rsp)
 {
-#ifdef CONFIG_PROVE_RCU
-	rcu_sync_lockdep_assert(rsp);
-#endif
+	RCU_LOCKDEP_WARN(!rcu_read_lock_held() &&
+			 !rcu_read_lock_bh_held() &&
+			 !rcu_read_lock_sched_held(),
+			 "suspicious rcu_sync_is_idle() usage");
 	return !rsp->gp_state; /* GP_IDLE */
 }
 
-extern void rcu_sync_init(struct rcu_sync *, enum rcu_sync_type);
+extern void rcu_sync_init(struct rcu_sync *);
 extern void rcu_sync_enter_start(struct rcu_sync *);
 extern void rcu_sync_enter(struct rcu_sync *);
 extern void rcu_sync_exit(struct rcu_sync *);
 extern void rcu_sync_dtor(struct rcu_sync *);
 
-#define __RCU_SYNC_INITIALIZER(name, type) {				\
+#define __RCU_SYNC_INITIALIZER(name) {					\
 		.gp_state = 0,						\
 		.gp_count = 0,						\
 		.gp_wait = __WAIT_QUEUE_HEAD_INITIALIZER(name.gp_wait),	\
 		.cb_state = 0,						\
-		.gp_type = type,					\
 	}
 
-#define	__DEFINE_RCU_SYNC(name, type)	\
-	struct rcu_sync_struct name = __RCU_SYNC_INITIALIZER(name, type)
-
-#define DEFINE_RCU_SYNC(name)		\
-	__DEFINE_RCU_SYNC(name, RCU_SYNC)
-
-#define DEFINE_RCU_SCHED_SYNC(name)	\
-	__DEFINE_RCU_SYNC(name, RCU_SCHED_SYNC)
-
-#define DEFINE_RCU_BH_SYNC(name)	\
-	__DEFINE_RCU_SYNC(name, RCU_BH_SYNC)
+#define	DEFINE_RCU_SYNC(name)	\
+	struct rcu_sync name = __RCU_SYNC_INITIALIZER(name)
 
 #endif /* _LINUX_RCU_SYNC_H_ */
diff --git a/kernel/locking/percpu-rwsem.c b/kernel/locking/percpu-rwsem.c
index f17dad99eec8..48cab93a47fd 100644
--- a/kernel/locking/percpu-rwsem.c
+++ b/kernel/locking/percpu-rwsem.c
@@ -17,7 +17,7 @@ int __percpu_init_rwsem(struct percpu_rw_semaphore *sem,
 		return -ENOMEM;
 
 	/* ->rw_sem represents the whole percpu_rw_semaphore for lockdep */
-	rcu_sync_init(&sem->rss, RCU_SCHED_SYNC);
+	rcu_sync_init(&sem->rss);
 	__init_rwsem(&sem->rw_sem, name, rwsem_key);
 	rcuwait_init(&sem->writer);
 	sem->readers_block = 0;
diff --git a/kernel/rcu/sync.c b/kernel/rcu/sync.c
index a8304d90573f..ee427e138dad 100644
--- a/kernel/rcu/sync.c
+++ b/kernel/rcu/sync.c
@@ -10,65 +10,20 @@
 #include <linux/rcu_sync.h>
 #include <linux/sched.h>
 
-#ifdef CONFIG_PROVE_RCU
-#define __INIT_HELD(func)	.held = func,
-#else
-#define __INIT_HELD(func)
-#endif
-
-static const struct {
-	void (*sync)(void);
-	void (*call)(struct rcu_head *, void (*)(struct rcu_head *));
-	void (*wait)(void);
-#ifdef CONFIG_PROVE_RCU
-	int  (*held)(void);
-#endif
-} gp_ops[] = {
-	[RCU_SYNC] = {
-		.sync = synchronize_rcu,
-		.call = call_rcu,
-		.wait = rcu_barrier,
-		__INIT_HELD(rcu_read_lock_held)
-	},
-	[RCU_SCHED_SYNC] = {
-		.sync = synchronize_rcu,
-		.call = call_rcu,
-		.wait = rcu_barrier,
-		__INIT_HELD(rcu_read_lock_sched_held)
-	},
-	[RCU_BH_SYNC] = {
-		.sync = synchronize_rcu,
-		.call = call_rcu,
-		.wait = rcu_barrier,
-		__INIT_HELD(rcu_read_lock_bh_held)
-	},
-};
-
 enum { GP_IDLE = 0, GP_PENDING, GP_PASSED };
 enum { CB_IDLE = 0, CB_PENDING, CB_REPLAY };
 
 #define	rss_lock	gp_wait.lock
 
-#ifdef CONFIG_PROVE_RCU
-void rcu_sync_lockdep_assert(struct rcu_sync *rsp)
-{
-	RCU_LOCKDEP_WARN(!gp_ops[rsp->gp_type].held(),
-			 "suspicious rcu_sync_is_idle() usage");
-}
-
-EXPORT_SYMBOL_GPL(rcu_sync_lockdep_assert);
-#endif
-
 /**
  * rcu_sync_init() - Initialize an rcu_sync structure
  * @rsp: Pointer to rcu_sync structure to be initialized
  * @type: Flavor of RCU with which to synchronize rcu_sync structure
  */
-void rcu_sync_init(struct rcu_sync *rsp, enum rcu_sync_type type)
+void rcu_sync_init(struct rcu_sync *rsp)
 {
 	memset(rsp, 0, sizeof(*rsp));
 	init_waitqueue_head(&rsp->gp_wait);
-	rsp->gp_type = type;
 }
 
 /**
@@ -114,7 +69,7 @@ void rcu_sync_enter(struct rcu_sync *rsp)
 
 	WARN_ON_ONCE(need_wait && need_sync);
 	if (need_sync) {
-		gp_ops[rsp->gp_type].sync();
+		synchronize_rcu();
 		rsp->gp_state = GP_PASSED;
 		wake_up_all(&rsp->gp_wait);
 	} else if (need_wait) {
@@ -167,7 +122,7 @@ static void rcu_sync_func(struct rcu_head *rhp)
 		 * to catch a later GP.
 		 */
 		rsp->cb_state = CB_PENDING;
-		gp_ops[rsp->gp_type].call(&rsp->cb_head, rcu_sync_func);
+		call_rcu(&rsp->cb_head, rcu_sync_func);
 	} else {
 		/*
 		 * We're at least a GP after rcu_sync_exit(); eveybody will now
@@ -195,7 +150,7 @@ void rcu_sync_exit(struct rcu_sync *rsp)
 	if (!--rsp->gp_count) {
 		if (rsp->cb_state == CB_IDLE) {
 			rsp->cb_state = CB_PENDING;
-			gp_ops[rsp->gp_type].call(&rsp->cb_head, rcu_sync_func);
+			call_rcu(&rsp->cb_head, rcu_sync_func);
 		} else if (rsp->cb_state == CB_PENDING) {
 			rsp->cb_state = CB_REPLAY;
 		}
@@ -220,7 +175,7 @@ void rcu_sync_dtor(struct rcu_sync *rsp)
 	spin_unlock_irq(&rsp->rss_lock);
 
 	if (cb_state != CB_IDLE) {
-		gp_ops[rsp->gp_type].wait();
+		rcu_barrier();
 		WARN_ON_ONCE(rsp->cb_state != CB_IDLE);
 	}
 }
-- 
2.17.1

[PATCH tip/core/rcu 2/4] uprobes: Use DEFINE_STATIC_PERCPU_RWSEM() to initialize dup_mmap_sem

[Paul E. McKenney]

From: Oleg Nesterov <oleg@redhat.com>

Use DEFINE_STATIC_PERCPU_RWSEM() to initialize dup_mmap_sem.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Reviewed-by: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Paul E. McKenney <paulmck@linux.ibm.com>
---
 kernel/events/uprobes.c | 4 +---
 1 file changed, 1 insertion(+), 3 deletions(-)

diff --git a/kernel/events/uprobes.c b/kernel/events/uprobes.c
index 78f61bfc6b79..97c367f0a9aa 100644
--- a/kernel/events/uprobes.c
+++ b/kernel/events/uprobes.c
@@ -46,7 +46,7 @@ static DEFINE_SPINLOCK(uprobes_treelock);	/* serialize rbtree access */
 static struct mutex uprobes_mmap_mutex[UPROBES_HASH_SZ];
 #define uprobes_mmap_hash(v)	(&uprobes_mmap_mutex[((unsigned long)(v)) % UPROBES_HASH_SZ])
 
-static struct percpu_rw_semaphore dup_mmap_sem;
+DEFINE_STATIC_PERCPU_RWSEM(dup_mmap_sem);
 
 /* Have a copy of original instruction */
 #define UPROBE_COPY_INSN	0
@@ -2302,7 +2302,5 @@ void __init uprobes_init(void)
 	for (i = 0; i < UPROBES_HASH_SZ; i++)
 		mutex_init(&uprobes_mmap_mutex[i]);
 
-	BUG_ON(percpu_init_rwsem(&dup_mmap_sem));
-
 	BUG_ON(register_die_notifier(&uprobe_exception_nb));
 }
-- 
2.17.1

[PATCH tip/core/rcu 3/4] locking/percpu-rwsem: Add DEFINE_PERCPU_RWSEM(), use it to initialize cgroup_threadgroup_rwsem

[Paul E. McKenney]

From: Oleg Nesterov <oleg@redhat.com>

Turn DEFINE_STATIC_PERCPU_RWSEM() into __DEFINE_PERCPU_RWSEM() with the
additional "is_static" argument to introduce DEFINE_PERCPU_RWSEM().

Change cgroup.c to use DEFINE_PERCPU_RWSEM(cgroup_threadgroup_rwsem).

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Reviewed-by: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Paul E. McKenney <paulmck@linux.ibm.com>
---
 include/linux/percpu-rwsem.h | 8 ++++++--
 kernel/cgroup/cgroup.c       | 3 +--
 2 files changed, 7 insertions(+), 4 deletions(-)

diff --git a/include/linux/percpu-rwsem.h b/include/linux/percpu-rwsem.h
index 6887636ea169..2809b44cbbee 100644
--- a/include/linux/percpu-rwsem.h
+++ b/include/linux/percpu-rwsem.h
@@ -17,14 +17,18 @@ struct percpu_rw_semaphore {
 	int			readers_block;
 };
 
-#define DEFINE_STATIC_PERCPU_RWSEM(name)				\
+#define __DEFINE_PERCPU_RWSEM(name, is_static)				\
 static DEFINE_PER_CPU(unsigned int, __percpu_rwsem_rc_##name);		\
-static struct percpu_rw_semaphore name = {				\
+is_static struct percpu_rw_semaphore name = {				\
 	.rss = __RCU_SYNC_INITIALIZER(name.rss),			\
 	.read_count = &__percpu_rwsem_rc_##name,			\
 	.rw_sem = __RWSEM_INITIALIZER(name.rw_sem),			\
 	.writer = __RCUWAIT_INITIALIZER(name.writer),			\
 }
+#define DEFINE_PERCPU_RWSEM(name)		\
+	__DEFINE_PERCPU_RWSEM(name, /* not static */)
+#define DEFINE_STATIC_PERCPU_RWSEM(name)	\
+	__DEFINE_PERCPU_RWSEM(name, static)
 
 extern int __percpu_down_read(struct percpu_rw_semaphore *, int);
 extern void __percpu_up_read(struct percpu_rw_semaphore *);
diff --git a/kernel/cgroup/cgroup.c b/kernel/cgroup/cgroup.c
index 217cec4e22c6..b112e93388dc 100644
--- a/kernel/cgroup/cgroup.c
+++ b/kernel/cgroup/cgroup.c
@@ -101,7 +101,7 @@ static DEFINE_SPINLOCK(cgroup_idr_lock);
  */
 static DEFINE_SPINLOCK(cgroup_file_kn_lock);
 
-struct percpu_rw_semaphore cgroup_threadgroup_rwsem;
+DEFINE_PERCPU_RWSEM(cgroup_threadgroup_rwsem);
 
 #define cgroup_assert_mutex_or_rcu_locked()				\
 	RCU_LOCKDEP_WARN(!rcu_read_lock_held() &&			\
@@ -5616,7 +5616,6 @@ int __init cgroup_init(void)
 	int ssid;
 
 	BUILD_BUG_ON(CGROUP_SUBSYS_COUNT > 16);
-	BUG_ON(percpu_init_rwsem(&cgroup_threadgroup_rwsem));
 	BUG_ON(cgroup_init_cftypes(NULL, cgroup_base_files));
 	BUG_ON(cgroup_init_cftypes(NULL, cgroup1_base_files));
 
-- 
2.17.1

[PATCH tip/core/rcu 4/4] rcu/sync: Simplify the state machine

[Paul E. McKenney]

From: Oleg Nesterov <oleg@redhat.com>

With this patch rcu_sync has a single state variable and the transition rules
become really simple:

	GP_IDLE   - owned by the first rcu_sync_enter() which moves it to

	GP_ENTER  - owned by rcu-callback which moves it to

	GP_PASSED - owned by the last rcu_sync_exit() which moves it to

	GP_EXIT   - and this is the only "nontrivial" state.

		rcu-callback moves it back to GP_IDLE unless another enter()
		comes before a GP pass.

		If rcu-callback is invoked before the next rcu_sync_exit() it
		must see gp_count incremented by that enter() and set GP_PASSED.

		Otherwise, if the next rcu_sync_exit() wins the race, it will
		move it to

	GP_REPLAY - owned by rcu-callback which moves it to GP_EXIT

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
[ paulmck: While here, apply READ_ONCE() and WRITE_ONCE() to ->gp_state. ]
[ paulmck: Tweaks to make htmldocs happy. (Reported by kbuild test robot.) ]
Signed-off-by: Paul E. McKenney <paulmck@linux.ibm.com>
---
 include/linux/rcu_sync.h |   4 +-
 kernel/rcu/sync.c        | 165 ++++++++++++++++++++++-----------------
 2 files changed, 96 insertions(+), 73 deletions(-)

diff --git a/include/linux/rcu_sync.h b/include/linux/rcu_sync.h
index 87971e85519c..9b83865d24f9 100644
--- a/include/linux/rcu_sync.h
+++ b/include/linux/rcu_sync.h
@@ -19,7 +19,6 @@ struct rcu_sync {
 	int			gp_count;
 	wait_queue_head_t	gp_wait;
 
-	int			cb_state;
 	struct rcu_head		cb_head;
 };
 
@@ -36,7 +35,7 @@ static inline bool rcu_sync_is_idle(struct rcu_sync *rsp)
 			 !rcu_read_lock_bh_held() &&
 			 !rcu_read_lock_sched_held(),
 			 "suspicious rcu_sync_is_idle() usage");
-	return !rsp->gp_state; /* GP_IDLE */
+	return !READ_ONCE(rsp->gp_state); /* GP_IDLE */
 }
 
 extern void rcu_sync_init(struct rcu_sync *);
@@ -49,7 +48,6 @@ extern void rcu_sync_dtor(struct rcu_sync *);
 		.gp_state = 0,						\
 		.gp_count = 0,						\
 		.gp_wait = __WAIT_QUEUE_HEAD_INITIALIZER(name.gp_wait),	\
-		.cb_state = 0,						\
 	}
 
 #define	DEFINE_RCU_SYNC(name)	\
diff --git a/kernel/rcu/sync.c b/kernel/rcu/sync.c
index ee427e138dad..d4558ab7a07d 100644
--- a/kernel/rcu/sync.c
+++ b/kernel/rcu/sync.c
@@ -10,15 +10,13 @@
 #include <linux/rcu_sync.h>
 #include <linux/sched.h>
 
-enum { GP_IDLE = 0, GP_PENDING, GP_PASSED };
-enum { CB_IDLE = 0, CB_PENDING, CB_REPLAY };
+enum { GP_IDLE = 0, GP_ENTER, GP_PASSED, GP_EXIT, GP_REPLAY };
 
 #define	rss_lock	gp_wait.lock
 
 /**
  * rcu_sync_init() - Initialize an rcu_sync structure
  * @rsp: Pointer to rcu_sync structure to be initialized
- * @type: Flavor of RCU with which to synchronize rcu_sync structure
  */
 void rcu_sync_init(struct rcu_sync *rsp)
 {
@@ -41,56 +39,26 @@ void rcu_sync_enter_start(struct rcu_sync *rsp)
 	rsp->gp_state = GP_PASSED;
 }
 
-/**
- * rcu_sync_enter() - Force readers onto slowpath
- * @rsp: Pointer to rcu_sync structure to use for synchronization
- *
- * This function is used by updaters who need readers to make use of
- * a slowpath during the update.  After this function returns, all
- * subsequent calls to rcu_sync_is_idle() will return false, which
- * tells readers to stay off their fastpaths.  A later call to
- * rcu_sync_exit() re-enables reader slowpaths.
- *
- * When called in isolation, rcu_sync_enter() must wait for a grace
- * period, however, closely spaced calls to rcu_sync_enter() can
- * optimize away the grace-period wait via a state machine implemented
- * by rcu_sync_enter(), rcu_sync_exit(), and rcu_sync_func().
- */
-void rcu_sync_enter(struct rcu_sync *rsp)
-{
-	bool need_wait, need_sync;
 
-	spin_lock_irq(&rsp->rss_lock);
-	need_wait = rsp->gp_count++;
-	need_sync = rsp->gp_state == GP_IDLE;
-	if (need_sync)
-		rsp->gp_state = GP_PENDING;
-	spin_unlock_irq(&rsp->rss_lock);
+static void rcu_sync_func(struct rcu_head *rhp);
 
-	WARN_ON_ONCE(need_wait && need_sync);
-	if (need_sync) {
-		synchronize_rcu();
-		rsp->gp_state = GP_PASSED;
-		wake_up_all(&rsp->gp_wait);
-	} else if (need_wait) {
-		wait_event(rsp->gp_wait, rsp->gp_state == GP_PASSED);
-	} else {
-		/*
-		 * Possible when there's a pending CB from a rcu_sync_exit().
-		 * Nobody has yet been allowed the 'fast' path and thus we can
-		 * avoid doing any sync(). The callback will get 'dropped'.
-		 */
-		WARN_ON_ONCE(rsp->gp_state != GP_PASSED);
-	}
+static void rcu_sync_call(struct rcu_sync *rsp)
+{
+	call_rcu(&rsp->cb_head, rcu_sync_func);
 }
 
 /**
  * rcu_sync_func() - Callback function managing reader access to fastpath
  * @rhp: Pointer to rcu_head in rcu_sync structure to use for synchronization
  *
- * This function is passed to one of the call_rcu() functions by
+ * This function is passed to call_rcu() function by rcu_sync_enter() and
  * rcu_sync_exit(), so that it is invoked after a grace period following the
- * that invocation of rcu_sync_exit().  It takes action based on events that
+ * that invocation of enter/exit.
+ *
+ * If it is called by rcu_sync_enter() it signals that all the readers were
+ * switched onto slow path.
+ *
+ * If it is called by rcu_sync_exit() it takes action based on events that
  * have taken place in the meantime, so that closely spaced rcu_sync_enter()
  * and rcu_sync_exit() pairs need not wait for a grace period.
  *
@@ -107,35 +75,88 @@ static void rcu_sync_func(struct rcu_head *rhp)
 	struct rcu_sync *rsp = container_of(rhp, struct rcu_sync, cb_head);
 	unsigned long flags;
 
-	WARN_ON_ONCE(rsp->gp_state != GP_PASSED);
-	WARN_ON_ONCE(rsp->cb_state == CB_IDLE);
+	WARN_ON_ONCE(READ_ONCE(rsp->gp_state) == GP_IDLE);
+	WARN_ON_ONCE(READ_ONCE(rsp->gp_state) == GP_PASSED);
 
 	spin_lock_irqsave(&rsp->rss_lock, flags);
 	if (rsp->gp_count) {
 		/*
-		 * A new rcu_sync_begin() has happened; drop the callback.
+		 * We're at least a GP after the GP_IDLE->GP_ENTER transition.
 		 */
-		rsp->cb_state = CB_IDLE;
-	} else if (rsp->cb_state == CB_REPLAY) {
+		WRITE_ONCE(rsp->gp_state, GP_PASSED);
+		wake_up_locked(&rsp->gp_wait);
+	} else if (rsp->gp_state == GP_REPLAY) {
 		/*
-		 * A new rcu_sync_exit() has happened; requeue the callback
-		 * to catch a later GP.
+		 * A new rcu_sync_exit() has happened; requeue the callback to
+		 * catch a later GP.
 		 */
-		rsp->cb_state = CB_PENDING;
-		call_rcu(&rsp->cb_head, rcu_sync_func);
+		WRITE_ONCE(rsp->gp_state, GP_EXIT);
+		rcu_sync_call(rsp);
 	} else {
 		/*
-		 * We're at least a GP after rcu_sync_exit(); eveybody will now
-		 * have observed the write side critical section. Let 'em rip!.
+		 * We're at least a GP after the last rcu_sync_exit(); eveybody
+		 * will now have observed the write side critical section.
+		 * Let 'em rip!.
 		 */
-		rsp->cb_state = CB_IDLE;
-		rsp->gp_state = GP_IDLE;
+		WRITE_ONCE(rsp->gp_state, GP_IDLE);
 	}
 	spin_unlock_irqrestore(&rsp->rss_lock, flags);
 }
 
 /**
- * rcu_sync_exit() - Allow readers back onto fast patch after grace period
+ * rcu_sync_enter() - Force readers onto slowpath
+ * @rsp: Pointer to rcu_sync structure to use for synchronization
+ *
+ * This function is used by updaters who need readers to make use of
+ * a slowpath during the update.  After this function returns, all
+ * subsequent calls to rcu_sync_is_idle() will return false, which
+ * tells readers to stay off their fastpaths.  A later call to
+ * rcu_sync_exit() re-enables reader slowpaths.
+ *
+ * When called in isolation, rcu_sync_enter() must wait for a grace
+ * period, however, closely spaced calls to rcu_sync_enter() can
+ * optimize away the grace-period wait via a state machine implemented
+ * by rcu_sync_enter(), rcu_sync_exit(), and rcu_sync_func().
+ */
+void rcu_sync_enter(struct rcu_sync *rsp)
+{
+	int gp_state;
+
+	spin_lock_irq(&rsp->rss_lock);
+	gp_state = rsp->gp_state;
+	if (gp_state == GP_IDLE) {
+		WRITE_ONCE(rsp->gp_state, GP_ENTER);
+		WARN_ON_ONCE(rsp->gp_count);
+		/*
+		 * Note that we could simply do rcu_sync_call(rsp) here and
+		 * avoid the "if (gp_state == GP_IDLE)" block below.
+		 *
+		 * However, synchronize_rcu() can be faster if rcu_expedited
+		 * or rcu_blocking_is_gp() is true.
+		 *
+		 * Another reason is that we can't wait for rcu callback if
+		 * we are called at early boot time but this shouldn't happen.
+		 */
+	}
+	rsp->gp_count++;
+	spin_unlock_irq(&rsp->rss_lock);
+
+	if (gp_state == GP_IDLE) {
+		/*
+		 * See the comment above, this simply does the "synchronous"
+		 * call_rcu(rcu_sync_func) which does GP_ENTER -> GP_PASSED.
+		 */
+		synchronize_rcu();
+		rcu_sync_func(&rsp->cb_head);
+		/* Not really needed, wait_event() would see GP_PASSED. */
+		return;
+	}
+
+	wait_event(rsp->gp_wait, READ_ONCE(rsp->gp_state) >= GP_PASSED);
+}
+
+/**
+ * rcu_sync_exit() - Allow readers back onto fast path after grace period
  * @rsp: Pointer to rcu_sync structure to use for synchronization
  *
  * This function is used by updaters who have completed, and can therefore
@@ -146,13 +167,16 @@ static void rcu_sync_func(struct rcu_head *rhp)
  */
 void rcu_sync_exit(struct rcu_sync *rsp)
 {
+	WARN_ON_ONCE(READ_ONCE(rsp->gp_state) == GP_IDLE);
+	WARN_ON_ONCE(READ_ONCE(rsp->gp_count) == 0);
+
 	spin_lock_irq(&rsp->rss_lock);
 	if (!--rsp->gp_count) {
-		if (rsp->cb_state == CB_IDLE) {
-			rsp->cb_state = CB_PENDING;
-			call_rcu(&rsp->cb_head, rcu_sync_func);
-		} else if (rsp->cb_state == CB_PENDING) {
-			rsp->cb_state = CB_REPLAY;
+		if (rsp->gp_state == GP_PASSED) {
+			WRITE_ONCE(rsp->gp_state, GP_EXIT);
+			rcu_sync_call(rsp);
+		} else if (rsp->gp_state == GP_EXIT) {
+			WRITE_ONCE(rsp->gp_state, GP_REPLAY);
 		}
 	}
 	spin_unlock_irq(&rsp->rss_lock);
@@ -164,18 +188,19 @@ void rcu_sync_exit(struct rcu_sync *rsp)
  */
 void rcu_sync_dtor(struct rcu_sync *rsp)
 {
-	int cb_state;
+	int gp_state;
 
-	WARN_ON_ONCE(rsp->gp_count);
+	WARN_ON_ONCE(READ_ONCE(rsp->gp_count));
+	WARN_ON_ONCE(READ_ONCE(rsp->gp_state) == GP_PASSED);
 
 	spin_lock_irq(&rsp->rss_lock);
-	if (rsp->cb_state == CB_REPLAY)
-		rsp->cb_state = CB_PENDING;
-	cb_state = rsp->cb_state;
+	if (rsp->gp_state == GP_REPLAY)
+		WRITE_ONCE(rsp->gp_state, GP_EXIT);
+	gp_state = rsp->gp_state;
 	spin_unlock_irq(&rsp->rss_lock);
 
-	if (cb_state != CB_IDLE) {
+	if (gp_state != GP_IDLE) {
 		rcu_barrier();
-		WARN_ON_ONCE(rsp->cb_state != CB_IDLE);
+		WARN_ON_ONCE(rsp->gp_state != GP_IDLE);
 	}
 }
-- 
2.17.1