[patch V2 24/24] cpu/hotplug: Convert hotplug locking to percpu rwsem

[Thomas Gleixner]

There are no more (known) nested calls to get_online_cpus() so it's
possible to remove the nested call magic and convert the mutex to a
percpu-rwsem, which speeds up get/put_online_cpus() significantly for the
uncontended case.

The contended case (write locked for hotplug operations) is slow anyway, so
the slightly more expensive down_write of the percpu rwsem does not matter.

[ peterz: Add lockdep assertions ]

Signed-off-by: Thomas Gleixner <tglx@xxxxxxxxxxxxx>
---
 include/linux/cpu.h   |    2 
 kernel/cpu.c          |  110 +++++++-------------------------------------------
 kernel/jump_label.c   |    2 
 kernel/padata.c       |    1 
 kernel/stop_machine.c |    2 
 5 files changed, 23 insertions(+), 94 deletions(-)

--- a/include/linux/cpu.h
+++ b/include/linux/cpu.h
@@ -105,6 +105,7 @@ extern void cpu_hotplug_begin(void);
 extern void cpu_hotplug_done(void);
 extern void get_online_cpus(void);
 extern void put_online_cpus(void);
+extern void lockdep_assert_hotplug_held(void);
 extern void cpu_hotplug_disable(void);
 extern void cpu_hotplug_enable(void);
 void clear_tasks_mm_cpumask(int cpu);
@@ -118,6 +119,7 @@ static inline void cpu_hotplug_done(void
 #define put_online_cpus()	do { } while (0)
 #define cpu_hotplug_disable()	do { } while (0)
 #define cpu_hotplug_enable()	do { } while (0)
+static inline void lockdep_assert_hotplug_held(void) {}
 #endif		/* CONFIG_HOTPLUG_CPU */
 
 #ifdef CONFIG_PM_SLEEP_SMP
--- a/kernel/cpu.c
+++ b/kernel/cpu.c
@@ -27,6 +27,7 @@
 #include <linux/smpboot.h>
 #include <linux/relay.h>
 #include <linux/slab.h>
+#include <linux/percpu-rwsem.h>
 
 #include <trace/events/power.h>
 #define CREATE_TRACE_POINTS
@@ -196,121 +197,41 @@ void cpu_maps_update_done(void)
 	mutex_unlock(&cpu_add_remove_lock);
 }
 
-/* If set, cpu_up and cpu_down will return -EBUSY and do nothing.
+/*
+ * If set, cpu_up and cpu_down will return -EBUSY and do nothing.
  * Should always be manipulated under cpu_add_remove_lock
  */
 static int cpu_hotplug_disabled;
 
 #ifdef CONFIG_HOTPLUG_CPU
 
-static struct {
-	struct task_struct *active_writer;
-	/* wait queue to wake up the active_writer */
-	wait_queue_head_t wq;
-	/* verifies that no writer will get active while readers are active */
-	struct mutex lock;
-	/*
-	 * Also blocks the new readers during
-	 * an ongoing cpu hotplug operation.
-	 */
-	atomic_t refcount;
-
-#ifdef CONFIG_DEBUG_LOCK_ALLOC
-	struct lockdep_map dep_map;
-#endif
-} cpu_hotplug = {
-	.active_writer = NULL,
-	.wq = __WAIT_QUEUE_HEAD_INITIALIZER(cpu_hotplug.wq),
-	.lock = __MUTEX_INITIALIZER(cpu_hotplug.lock),
-#ifdef CONFIG_DEBUG_LOCK_ALLOC
-	.dep_map = STATIC_LOCKDEP_MAP_INIT("cpu_hotplug.dep_map", &cpu_hotplug.dep_map),
-#endif
-};
-
-/* Lockdep annotations for get/put_online_cpus() and cpu_hotplug_begin/end() */
-#define cpuhp_lock_acquire_read() lock_map_acquire_read(&cpu_hotplug.dep_map)
-#define cpuhp_lock_acquire_tryread() \
-				  lock_map_acquire_tryread(&cpu_hotplug.dep_map)
-#define cpuhp_lock_acquire()      lock_map_acquire(&cpu_hotplug.dep_map)
-#define cpuhp_lock_release()      lock_map_release(&cpu_hotplug.dep_map)
-
+DEFINE_STATIC_PERCPU_RWSEM(cpu_hotplug_lock);
 
 void get_online_cpus(void)
 {
-	might_sleep();
-	if (cpu_hotplug.active_writer == current)
-		return;
-	cpuhp_lock_acquire_read();
-	mutex_lock(&cpu_hotplug.lock);
-	atomic_inc(&cpu_hotplug.refcount);
-	mutex_unlock(&cpu_hotplug.lock);
+	percpu_down_read(&cpu_hotplug_lock);
 }
 EXPORT_SYMBOL_GPL(get_online_cpus);
 
 void put_online_cpus(void)
 {
-	int refcount;
-
-	if (cpu_hotplug.active_writer == current)
-		return;
-
-	refcount = atomic_dec_return(&cpu_hotplug.refcount);
-	if (WARN_ON(refcount < 0)) /* try to fix things up */
-		atomic_inc(&cpu_hotplug.refcount);
-
-	if (refcount <= 0 && waitqueue_active(&cpu_hotplug.wq))
-		wake_up(&cpu_hotplug.wq);
-
-	cpuhp_lock_release();
-
+	percpu_up_read(&cpu_hotplug_lock);
 }
 EXPORT_SYMBOL_GPL(put_online_cpus);
 
-/*
- * This ensures that the hotplug operation can begin only when the
- * refcount goes to zero.
- *
- * Note that during a cpu-hotplug operation, the new readers, if any,
- * will be blocked by the cpu_hotplug.lock
- *
- * Since cpu_hotplug_begin() is always called after invoking
- * cpu_maps_update_begin(), we can be sure that only one writer is active.
- *
- * Note that theoretically, there is a possibility of a livelock:
- * - Refcount goes to zero, last reader wakes up the sleeping
- *   writer.
- * - Last reader unlocks the cpu_hotplug.lock.
- * - A new reader arrives at this moment, bumps up the refcount.
- * - The writer acquires the cpu_hotplug.lock finds the refcount
- *   non zero and goes to sleep again.
- *
- * However, this is very difficult to achieve in practice since
- * get_online_cpus() not an api which is called all that often.
- *
- */
 void cpu_hotplug_begin(void)
 {
-	DEFINE_WAIT(wait);
-
-	cpu_hotplug.active_writer = current;
-	cpuhp_lock_acquire();
-
-	for (;;) {
-		mutex_lock(&cpu_hotplug.lock);
-		prepare_to_wait(&cpu_hotplug.wq, &wait, TASK_UNINTERRUPTIBLE);
-		if (likely(!atomic_read(&cpu_hotplug.refcount)))
-				break;
-		mutex_unlock(&cpu_hotplug.lock);
-		schedule();
-	}
-	finish_wait(&cpu_hotplug.wq, &wait);
+	percpu_down_write(&cpu_hotplug_lock);
 }
 
 void cpu_hotplug_done(void)
 {
-	cpu_hotplug.active_writer = NULL;
-	mutex_unlock(&cpu_hotplug.lock);
-	cpuhp_lock_release();
+	percpu_up_write(&cpu_hotplug_lock);
+}
+
+void lockdep_assert_hotplug_held(void)
+{
+	percpu_rwsem_assert_held(&cpu_hotplug_lock);
 }
 
 /*
@@ -344,8 +265,6 @@ void cpu_hotplug_enable(void)
 EXPORT_SYMBOL_GPL(cpu_hotplug_enable);
 #endif	/* CONFIG_HOTPLUG_CPU */
 
-/* Notifier wrappers for transitioning to state machine */
-
 static int bringup_wait_for_ap(unsigned int cpu)
 {
 	struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
@@ -1482,6 +1401,8 @@ int __cpuhp_setup_state_cpuslocked(enum
 	int cpu, ret = 0;
 	bool dynstate;
 
+	lockdep_assert_hotplug_held();
+
 	if (cpuhp_cb_check(state) || !name)
 		return -EINVAL;
 
@@ -1600,6 +1521,7 @@ void __cpuhp_remove_state_cpuslocked(enu
 	int cpu;
 
 	BUG_ON(cpuhp_cb_check(state));
+	lockdep_assert_hotplug_held();
 
 	mutex_lock(&cpuhp_state_mutex);
 	if (sp->multi_instance) {
--- a/kernel/jump_label.c
+++ b/kernel/jump_label.c
@@ -130,6 +130,7 @@ void __static_key_slow_inc(struct static
 	 * the all CPUs, for that to be serialized against CPU hot-plug
 	 * we need to avoid CPUs coming online.
 	 */
+	lockdep_assert_hotplug_held();
 	jump_label_lock();
 	if (atomic_read(&key->enabled) == 0) {
 		atomic_set(&key->enabled, -1);
@@ -158,6 +159,7 @@ EXPORT_SYMBOL_GPL(static_key_slow_inc_cp
 static void __static_key_slow_dec(struct static_key *key,
 		unsigned long rate_limit, struct delayed_work *work)
 {
+	lockdep_assert_hotplug_held();
 	/*
 	 * The negative count check is valid even when a negative
 	 * key->enabled is in use by static_key_slow_inc(); a
--- a/kernel/padata.c
+++ b/kernel/padata.c
@@ -1013,6 +1013,7 @@ static struct padata_instance *padata_al
  */
 struct padata_instance *padata_alloc_possible(struct workqueue_struct *wq)
 {
+	lockdep_assert_hotplug_held();
 	return padata_alloc(wq, cpu_possible_mask, cpu_possible_mask);
 }
 EXPORT_SYMBOL(padata_alloc_possible);
--- a/kernel/stop_machine.c
+++ b/kernel/stop_machine.c
@@ -562,6 +562,8 @@ int stop_machine_cpuslocked(cpu_stop_fn_
 		.active_cpus = cpus,
 	};
 
+	lockdep_assert_hotplug_held();
+
 	if (!stop_machine_initialized) {
 		/*
 		 * Handle the case where stop_machine() is called