1 files changed, 344 insertions, 120 deletions

diff --git a/kernel/events/core.c b/kernel/events/core.c
index 19efcf13375a..7f2fbb8b5069 100644
--- a/kernel/events/core.c
+++ b/kernel/events/core.c
@@ -872,22 +872,32 @@ void perf_pmu_enable(struct pmu *pmu)
 		pmu->pmu_enable(pmu);
 }
 
-static DEFINE_PER_CPU(struct list_head, rotation_list);
+static DEFINE_PER_CPU(struct list_head, active_ctx_list);
 
 /*
- * perf_pmu_rotate_start() and perf_rotate_context() are fully serialized
- * because they're strictly cpu affine and rotate_start is called with IRQs
- * disabled, while rotate_context is called from IRQ context.
+ * perf_event_ctx_activate(), perf_event_ctx_deactivate(), and
+ * perf_event_task_tick() are fully serialized because they're strictly cpu
+ * affine and perf_event_ctx{activate,deactivate} are called with IRQs
+ * disabled, while perf_event_task_tick is called from IRQ context.
  */
-static void perf_pmu_rotate_start(struct pmu *pmu)
+static void perf_event_ctx_activate(struct perf_event_context *ctx)
 {
-	struct perf_cpu_context *cpuctx = this_cpu_ptr(pmu->pmu_cpu_context);
-	struct list_head *head = this_cpu_ptr(&rotation_list);
+	struct list_head *head = this_cpu_ptr(&active_ctx_list);
 
 	WARN_ON(!irqs_disabled());
 
-	if (list_empty(&cpuctx->rotation_list))
-		list_add(&cpuctx->rotation_list, head);
+	WARN_ON(!list_empty(&ctx->active_ctx_list));
+
+	list_add(&ctx->active_ctx_list, head);
+}
+
+static void perf_event_ctx_deactivate(struct perf_event_context *ctx)
+{
+	WARN_ON(!irqs_disabled());
+
+	WARN_ON(list_empty(&ctx->active_ctx_list));
+
+	list_del_init(&ctx->active_ctx_list);
 }
 
 static void get_ctx(struct perf_event_context *ctx)
@@ -907,6 +917,84 @@ static void put_ctx(struct perf_event_context *ctx)
 }
 
 /*
+ * Because of perf_event::ctx migration in sys_perf_event_open::move_group and
+ * perf_pmu_migrate_context() we need some magic.
+ *
+ * Those places that change perf_event::ctx will hold both
+ * perf_event_ctx::mutex of the 'old' and 'new' ctx value.
+ *
+ * Lock ordering is by mutex address. There is one other site where
+ * perf_event_context::mutex nests and that is put_event(). But remember that
+ * that is a parent<->child context relation, and migration does not affect
+ * children, therefore these two orderings should not interact.
+ *
+ * The change in perf_event::ctx does not affect children (as claimed above)
+ * because the sys_perf_event_open() case will install a new event and break
+ * the ctx parent<->child relation, and perf_pmu_migrate_context() is only
+ * concerned with cpuctx and that doesn't have children.
+ *
+ * The places that change perf_event::ctx will issue:
+ *
+ *   perf_remove_from_context();
+ *   synchronize_rcu();
+ *   perf_install_in_context();
+ *
+ * to affect the change. The remove_from_context() + synchronize_rcu() should
+ * quiesce the event, after which we can install it in the new location. This
+ * means that only external vectors (perf_fops, prctl) can perturb the event
+ * while in transit. Therefore all such accessors should also acquire
+ * perf_event_context::mutex to serialize against this.
+ *
+ * However; because event->ctx can change while we're waiting to acquire
+ * ctx->mutex we must be careful and use the below perf_event_ctx_lock()
+ * function.
+ *
+ * Lock order:
+ *	task_struct::perf_event_mutex
+ *	  perf_event_context::mutex
+ *	    perf_event_context::lock
+ *	    perf_event::child_mutex;
+ *	    perf_event::mmap_mutex
+ *	    mmap_sem
+ */
+static struct perf_event_context *
+perf_event_ctx_lock_nested(struct perf_event *event, int nesting)
+{
+	struct perf_event_context *ctx;
+
+again:
+	rcu_read_lock();
+	ctx = ACCESS_ONCE(event->ctx);
+	if (!atomic_inc_not_zero(&ctx->refcount)) {
+		rcu_read_unlock();
+		goto again;
+	}
+	rcu_read_unlock();
+
+	mutex_lock_nested(&ctx->mutex, nesting);
+	if (event->ctx != ctx) {
+		mutex_unlock(&ctx->mutex);
+		put_ctx(ctx);
+		goto again;
+	}
+
+	return ctx;
+}
+
+static inline struct perf_event_context *
+perf_event_ctx_lock(struct perf_event *event)
+{
+	return perf_event_ctx_lock_nested(event, 0);
+}
+
+static void perf_event_ctx_unlock(struct perf_event *event,
+				  struct perf_event_context *ctx)
+{
+	mutex_unlock(&ctx->mutex);
+	put_ctx(ctx);
+}
+
+/*
  * This must be done under the ctx->lock, such as to serialize against
  * context_equiv(), therefore we cannot call put_ctx() since that might end up
  * calling scheduler related locks and ctx->lock nests inside those.
@@ -1155,8 +1243,6 @@ list_add_event(struct perf_event *event, struct perf_event_context *ctx)
 		ctx->nr_branch_stack++;
 
 	list_add_rcu(&event->event_entry, &ctx->event_list);
-	if (!ctx->nr_events)
-		perf_pmu_rotate_start(ctx->pmu);
 	ctx->nr_events++;
 	if (event->attr.inherit_stat)
 		ctx->nr_stat++;
@@ -1275,6 +1361,8 @@ static void perf_group_attach(struct perf_event *event)
 	if (group_leader == event)
 		return;
 
+	WARN_ON_ONCE(group_leader->ctx != event->ctx);
+
 	if (group_leader->group_flags & PERF_GROUP_SOFTWARE &&
 			!is_software_event(event))
 		group_leader->group_flags &= ~PERF_GROUP_SOFTWARE;
@@ -1296,6 +1384,10 @@ static void
 list_del_event(struct perf_event *event, struct perf_event_context *ctx)
 {
 	struct perf_cpu_context *cpuctx;
+
+	WARN_ON_ONCE(event->ctx != ctx);
+	lockdep_assert_held(&ctx->lock);
+
 	/*
 	 * We can have double detach due to exit/hot-unplug + close.
 	 */
@@ -1380,6 +1472,8 @@ static void perf_group_detach(struct perf_event *event)
 
 		/* Inherit group flags from the previous leader */
 		sibling->group_flags = event->group_flags;
+
+		WARN_ON_ONCE(sibling->ctx != event->ctx);
 	}
 
 out:
@@ -1442,6 +1536,10 @@ event_sched_out(struct perf_event *event,
 {
 	u64 tstamp = perf_event_time(event);
 	u64 delta;
+
+	WARN_ON_ONCE(event->ctx != ctx);
+	lockdep_assert_held(&ctx->lock);
+
 	/*
 	 * An event which could not be activated because of
 	 * filter mismatch still needs to have its timings
@@ -1471,7 +1569,8 @@ event_sched_out(struct perf_event *event,
 
 	if (!is_software_event(event))
 		cpuctx->active_oncpu--;
-	ctx->nr_active--;
+	if (!--ctx->nr_active)
+		perf_event_ctx_deactivate(ctx);
 	if (event->attr.freq && event->attr.sample_freq)
 		ctx->nr_freq--;
 	if (event->attr.exclusive || !cpuctx->active_oncpu)
@@ -1654,7 +1753,7 @@ int __perf_event_disable(void *info)
  * is the current context on this CPU and preemption is disabled,
  * hence we can't get into perf_event_task_sched_out for this context.
  */
-void perf_event_disable(struct perf_event *event)
+static void _perf_event_disable(struct perf_event *event)
 {
 	struct perf_event_context *ctx = event->ctx;
 	struct task_struct *task = ctx->task;
@@ -1695,6 +1794,19 @@ retry:
 	}
 	raw_spin_unlock_irq(&ctx->lock);
 }
+
+/*
+ * Strictly speaking kernel users cannot create groups and therefore this
+ * interface does not need the perf_event_ctx_lock() magic.
+ */
+void perf_event_disable(struct perf_event *event)
+{
+	struct perf_event_context *ctx;
+
+	ctx = perf_event_ctx_lock(event);
+	_perf_event_disable(event);
+	perf_event_ctx_unlock(event, ctx);
+}
 EXPORT_SYMBOL_GPL(perf_event_disable);
 
 static void perf_set_shadow_time(struct perf_event *event,
@@ -1782,7 +1894,8 @@ event_sched_in(struct perf_event *event,
 
 	if (!is_software_event(event))
 		cpuctx->active_oncpu++;
-	ctx->nr_active++;
+	if (!ctx->nr_active++)
+		perf_event_ctx_activate(ctx);
 	if (event->attr.freq && event->attr.sample_freq)
 		ctx->nr_freq++;
 
@@ -2158,7 +2271,7 @@ unlock:
  * perf_event_for_each_child or perf_event_for_each as described
  * for perf_event_disable.
  */
-void perf_event_enable(struct perf_event *event)
+static void _perf_event_enable(struct perf_event *event)
 {
 	struct perf_event_context *ctx = event->ctx;
 	struct task_struct *task = ctx->task;
@@ -2214,9 +2327,21 @@ retry:
 out:
 	raw_spin_unlock_irq(&ctx->lock);
 }
+
+/*
+ * See perf_event_disable();
+ */
+void perf_event_enable(struct perf_event *event)
+{
+	struct perf_event_context *ctx;
+
+	ctx = perf_event_ctx_lock(event);
+	_perf_event_enable(event);
+	perf_event_ctx_unlock(event, ctx);
+}
 EXPORT_SYMBOL_GPL(perf_event_enable);
 
-int perf_event_refresh(struct perf_event *event, int refresh)
+static int _perf_event_refresh(struct perf_event *event, int refresh)
 {
 	/*
 	 * not supported on inherited events
@@ -2225,10 +2350,25 @@ int perf_event_refresh(struct perf_event *event, int refresh)
 		return -EINVAL;
 
 	atomic_add(refresh, &event->event_limit);
-	perf_event_enable(event);
+	_perf_event_enable(event);
 
 	return 0;
 }
+
+/*
+ * See perf_event_disable()
+ */
+int perf_event_refresh(struct perf_event *event, int refresh)
+{
+	struct perf_event_context *ctx;
+	int ret;
+
+	ctx = perf_event_ctx_lock(event);
+	ret = _perf_event_refresh(event, refresh);
+	perf_event_ctx_unlock(event, ctx);
+
+	return ret;
+}
 EXPORT_SYMBOL_GPL(perf_event_refresh);
 
 static void ctx_sched_out(struct perf_event_context *ctx,
@@ -2612,12 +2752,6 @@ static void perf_event_context_sched_in(struct perf_event_context *ctx,
 
 	perf_pmu_enable(ctx->pmu);
 	perf_ctx_unlock(cpuctx, ctx);
-
-	/*
-	 * Since these rotations are per-cpu, we need to ensure the
-	 * cpu-context we got scheduled on is actually rotating.
-	 */
-	perf_pmu_rotate_start(ctx->pmu);
 }
 
 /*
@@ -2905,25 +3039,18 @@ static void rotate_ctx(struct perf_event_context *ctx)
 		list_rotate_left(&ctx->flexible_groups);
 }
 
-/*
- * perf_pmu_rotate_start() and perf_rotate_context() are fully serialized
- * because they're strictly cpu affine and rotate_start is called with IRQs
- * disabled, while rotate_context is called from IRQ context.
- */
 static int perf_rotate_context(struct perf_cpu_context *cpuctx)
 {
 	struct perf_event_context *ctx = NULL;
-	int rotate = 0, remove = 1;
+	int rotate = 0;
 
 	if (cpuctx->ctx.nr_events) {
-		remove = 0;
 		if (cpuctx->ctx.nr_events != cpuctx->ctx.nr_active)
 			rotate = 1;
 	}
 
 	ctx = cpuctx->task_ctx;
 	if (ctx && ctx->nr_events) {
-		remove = 0;
 		if (ctx->nr_events != ctx->nr_active)
 			rotate = 1;
 	}
@@ -2947,8 +3074,6 @@ static int perf_rotate_context(struct perf_cpu_context *cpuctx)
 	perf_pmu_enable(cpuctx->ctx.pmu);
 	perf_ctx_unlock(cpuctx, cpuctx->task_ctx);
 done:
-	if (remove)
-		list_del_init(&cpuctx->rotation_list);
 
 	return rotate;
 }
@@ -2966,9 +3091,8 @@ bool perf_event_can_stop_tick(void)
 
 void perf_event_task_tick(void)
 {
-	struct list_head *head = this_cpu_ptr(&rotation_list);
-	struct perf_cpu_context *cpuctx, *tmp;
-	struct perf_event_context *ctx;
+	struct list_head *head = this_cpu_ptr(&active_ctx_list);
+	struct perf_event_context *ctx, *tmp;
 	int throttled;
 
 	WARN_ON(!irqs_disabled());
@@ -2976,14 +3100,8 @@ void perf_event_task_tick(void)
 	__this_cpu_inc(perf_throttled_seq);
 	throttled = __this_cpu_xchg(perf_throttled_count, 0);
 
-	list_for_each_entry_safe(cpuctx, tmp, head, rotation_list) {
-		ctx = &cpuctx->ctx;
+	list_for_each_entry_safe(ctx, tmp, head, active_ctx_list)
 		perf_adjust_freq_unthr_context(ctx, throttled);
-
-		ctx = cpuctx->task_ctx;
-		if (ctx)
-			perf_adjust_freq_unthr_context(ctx, throttled);
-	}
 }
 
 static int event_enable_on_exec(struct perf_event *event,
@@ -3142,6 +3260,7 @@ static void __perf_event_init_context(struct perf_event_context *ctx)
 {
 	raw_spin_lock_init(&ctx->lock);
 	mutex_init(&ctx->mutex);
+	INIT_LIST_HEAD(&ctx->active_ctx_list);
 	INIT_LIST_HEAD(&ctx->pinned_groups);
 	INIT_LIST_HEAD(&ctx->flexible_groups);
 	INIT_LIST_HEAD(&ctx->event_list);
@@ -3421,7 +3540,16 @@ static void perf_remove_from_owner(struct perf_event *event)
 	rcu_read_unlock();
 
 	if (owner) {
-		mutex_lock(&owner->perf_event_mutex);
+		/*
+		 * If we're here through perf_event_exit_task() we're already
+		 * holding ctx->mutex which would be an inversion wrt. the
+		 * normal lock order.
+		 *
+		 * However we can safely take this lock because its the child
+		 * ctx->mutex.
+		 */
+		mutex_lock_nested(&owner->perf_event_mutex, SINGLE_DEPTH_NESTING);
+
 		/*
 		 * We have to re-check the event->owner field, if it is cleared
 		 * we raced with perf_event_exit_task(), acquiring the mutex
@@ -3440,7 +3568,7 @@ static void perf_remove_from_owner(struct perf_event *event)
  */
 static void put_event(struct perf_event *event)
 {
-	struct perf_event_context *ctx = event->ctx;
+	struct perf_event_context *ctx;
 
 	if (!atomic_long_dec_and_test(&event->refcount))
 		return;
@@ -3448,7 +3576,6 @@ static void put_event(struct perf_event *event)
 	if (!is_kernel_event(event))
 		perf_remove_from_owner(event);
 
-	WARN_ON_ONCE(ctx->parent_ctx);
 	/*
 	 * There are two ways this annotation is useful:
 	 *
@@ -3461,7 +3588,8 @@ static void put_event(struct perf_event *event)
 	 *     the last filedesc died, so there is no possibility
 	 *     to trigger the AB-BA case.
 	 */
-	mutex_lock_nested(&ctx->mutex, SINGLE_DEPTH_NESTING);
+	ctx = perf_event_ctx_lock_nested(event, SINGLE_DEPTH_NESTING);
+	WARN_ON_ONCE(ctx->parent_ctx);
 	perf_remove_from_context(event, true);
 	mutex_unlock(&ctx->mutex);
 
@@ -3547,12 +3675,13 @@ static int perf_event_read_group(struct perf_event *event,
 				   u64 read_format, char __user *buf)
 {
 	struct perf_event *leader = event->group_leader, *sub;
-	int n = 0, size = 0, ret = -EFAULT;
 	struct perf_event_context *ctx = leader->ctx;
-	u64 values[5];
+	int n = 0, size = 0, ret;
 	u64 count, enabled, running;
+	u64 values[5];
+
+	lockdep_assert_held(&ctx->mutex);
 
-	mutex_lock(&ctx->mutex);
 	count = perf_event_read_value(leader, &enabled, &running);
 
 	values[n++] = 1 + leader->nr_siblings;
@@ -3567,7 +3696,7 @@ static int perf_event_read_group(struct perf_event *event,
 	size = n * sizeof(u64);
 
 	if (copy_to_user(buf, values, size))
-		goto unlock;
+		return -EFAULT;
 
 	ret = size;
 
@@ -3581,14 +3710,11 @@ static int perf_event_read_group(struct perf_event *event,
 		size = n * sizeof(u64);
 
 		if (copy_to_user(buf + ret, values, size)) {
-			ret = -EFAULT;
-			goto unlock;
+			return -EFAULT;
 		}
 
 		ret += size;
 	}
-unlock:
-	mutex_unlock(&ctx->mutex);
 
 	return ret;
 }
@@ -3660,8 +3786,14 @@ static ssize_t
 perf_read(struct file *file, char __user *buf, size_t count, loff_t *ppos)
 {
 	struct perf_event *event = file->private_data;
+	struct perf_event_context *ctx;
+	int ret;
 
-	return perf_read_hw(event, buf, count);
+	ctx = perf_event_ctx_lock(event);
+	ret = perf_read_hw(event, buf, count);
+	perf_event_ctx_unlock(event, ctx);
+
+	return ret;
 }
 
 static unsigned int perf_poll(struct file *file, poll_table *wait)
@@ -3687,7 +3819,7 @@ static unsigned int perf_poll(struct file *file, poll_table *wait)
 	return events;
 }
 
-static void perf_event_reset(struct perf_event *event)
+static void _perf_event_reset(struct perf_event *event)
 {
 	(void)perf_event_read(event);
 	local64_set(&event->count, 0);
@@ -3706,6 +3838,7 @@ static void perf_event_for_each_child(struct perf_event *event,
 	struct perf_event *child;
 
 	WARN_ON_ONCE(event->ctx->parent_ctx);
+
 	mutex_lock(&event->child_mutex);
 	func(event);
 	list_for_each_entry(child, &event->child_list, child_list)
@@ -3719,14 +3852,13 @@ static void perf_event_for_each(struct perf_event *event,
 	struct perf_event_context *ctx = event->ctx;
 	struct perf_event *sibling;
 
-	WARN_ON_ONCE(ctx->parent_ctx);
-	mutex_lock(&ctx->mutex);
+	lockdep_assert_held(&ctx->mutex);
+
 	event = event->group_leader;
 
 	perf_event_for_each_child(event, func);
 	list_for_each_entry(sibling, &event->sibling_list, group_entry)
 		perf_event_for_each_child(sibling, func);
-	mutex_unlock(&ctx->mutex);
 }
 
 static int perf_event_period(struct perf_event *event, u64 __user *arg)
@@ -3796,25 +3928,24 @@ static int perf_event_set_output(struct perf_event *event,
 				 struct perf_event *output_event);
 static int perf_event_set_filter(struct perf_event *event, void __user *arg);
 
-static long perf_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
+static long _perf_ioctl(struct perf_event *event, unsigned int cmd, unsigned long arg)
 {
-	struct perf_event *event = file->private_data;
 	void (*func)(struct perf_event *);
 	u32 flags = arg;
 
 	switch (cmd) {
 	case PERF_EVENT_IOC_ENABLE:
-		func = perf_event_enable;
+		func = _perf_event_enable;
 		break;
 	case PERF_EVENT_IOC_DISABLE:
-		func = perf_event_disable;
+		func = _perf_event_disable;
 		break;
 	case PERF_EVENT_IOC_RESET:
-		func = perf_event_reset;
+		func = _perf_event_reset;
 		break;
 
 	case PERF_EVENT_IOC_REFRESH:
-		return perf_event_refresh(event, arg);
+		return _perf_event_refresh(event, arg);
 
 	case PERF_EVENT_IOC_PERIOD:
 		return perf_event_period(event, (u64 __user *)arg);
@@ -3861,6 +3992,19 @@ static long perf_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
 	return 0;
 }
 
+static long perf_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
+{
+	struct perf_event *event = file->private_data;
+	struct perf_event_context *ctx;
+	long ret;
+
+	ctx = perf_event_ctx_lock(event);
+	ret = _perf_ioctl(event, cmd, arg);
+	perf_event_ctx_unlock(event, ctx);
+
+	return ret;
+}
+
 #ifdef CONFIG_COMPAT
 static long perf_compat_ioctl(struct file *file, unsigned int cmd,
 				unsigned long arg)
@@ -3883,11 +4027,15 @@ static long perf_compat_ioctl(struct file *file, unsigned int cmd,
 
 int perf_event_task_enable(void)
 {
+	struct perf_event_context *ctx;
 	struct perf_event *event;
 
 	mutex_lock(&current->perf_event_mutex);
-	list_for_each_entry(event, &current->perf_event_list, owner_entry)
-		perf_event_for_each_child(event, perf_event_enable);
+	list_for_each_entry(event, &current->perf_event_list, owner_entry) {
+		ctx = perf_event_ctx_lock(event);
+		perf_event_for_each_child(event, _perf_event_enable);
+		perf_event_ctx_unlock(event, ctx);
+	}
 	mutex_unlock(&current->perf_event_mutex);
 
 	return 0;
@@ -3895,11 +4043,15 @@ int perf_event_task_enable(void)
 
 int perf_event_task_disable(void)
 {
+	struct perf_event_context *ctx;
 	struct perf_event *event;
 
 	mutex_lock(&current->perf_event_mutex);
-	list_for_each_entry(event, &current->perf_event_list, owner_entry)
-		perf_event_for_each_child(event, perf_event_disable);
+	list_for_each_entry(event, &current->perf_event_list, owner_entry) {
+		ctx = perf_event_ctx_lock(event);
+		perf_event_for_each_child(event, _perf_event_disable);
+		perf_event_ctx_unlock(event, ctx);
+	}
 	mutex_unlock(&current->perf_event_mutex);
 
 	return 0;
@@ -5889,6 +6041,8 @@ end:
 	rcu_read_unlock();
 }
 
+DEFINE_PER_CPU(struct pt_regs, __perf_regs[4]);
+
 int perf_swevent_get_recursion_context(void)
 {
 	struct swevent_htable *swhash = this_cpu_ptr(&swevent_htable);
@@ -5904,21 +6058,30 @@ inline void perf_swevent_put_recursion_context(int rctx)
 	put_recursion_context(swhash->recursion, rctx);
 }
 
-void __perf_sw_event(u32 event_id, u64 nr, struct pt_regs *regs, u64 addr)
+void ___perf_sw_event(u32 event_id, u64 nr, struct pt_regs *regs, u64 addr)
 {
 	struct perf_sample_data data;
-	int rctx;
 
-	preempt_disable_notrace();
-	rctx = perf_swevent_get_recursion_context();
-	if (rctx < 0)
+	if (WARN_ON_ONCE(!regs))
 		return;
 
 	perf_sample_data_init(&data, addr, 0);
-
 	do_perf_sw_event(PERF_TYPE_SOFTWARE, event_id, nr, &data, regs);
+}
+
+void __perf_sw_event(u32 event_id, u64 nr, struct pt_regs *regs, u64 addr)
+{
+	int rctx;
+
+	preempt_disable_notrace();
+	rctx = perf_swevent_get_recursion_context();
+	if (unlikely(rctx < 0))
+		goto fail;
+
+	___perf_sw_event(event_id, nr, regs, addr);
 
 	perf_swevent_put_recursion_context(rctx);
+fail:
 	preempt_enable_notrace();
 }
 
@@ -6780,7 +6943,6 @@ skip_type:
 
 		__perf_cpu_hrtimer_init(cpuctx, cpu);
 
-		INIT_LIST_HEAD(&cpuctx->rotation_list);
 		cpuctx->unique_pmu = pmu;
 	}
 
@@ -6853,6 +7015,20 @@ void perf_pmu_unregister(struct pmu *pmu)
 }
 EXPORT_SYMBOL_GPL(perf_pmu_unregister);
 
+static int perf_try_init_event(struct pmu *pmu, struct perf_event *event)
+{
+	int ret;
+
+	if (!try_module_get(pmu->module))
+		return -ENODEV;
+	event->pmu = pmu;
+	ret = pmu->event_init(event);
+	if (ret)
+		module_put(pmu->module);
+
+	return ret;
+}
+
 struct pmu *perf_init_event(struct perf_event *event)
 {
 	struct pmu *pmu = NULL;
@@ -6865,24 +7041,14 @@ struct pmu *perf_init_event(struct perf_event *event)
 	pmu = idr_find(&pmu_idr, event->attr.type);
 	rcu_read_unlock();
 	if (pmu) {
-		if (!try_module_get(pmu->module)) {
-			pmu = ERR_PTR(-ENODEV);
-			goto unlock;
-		}
-		event->pmu = pmu;
-		ret = pmu->event_init(event);
+		ret = perf_try_init_event(pmu, event);
 		if (ret)
 			pmu = ERR_PTR(ret);
 		goto unlock;
 	}
 
 	list_for_each_entry_rcu(pmu, &pmus, entry) {
-		if (!try_module_get(pmu->module)) {
-			pmu = ERR_PTR(-ENODEV);
-			goto unlock;
-		}
-		event->pmu = pmu;
-		ret = pmu->event_init(event);
+		ret = perf_try_init_event(pmu, event);
 		if (!ret)
 			goto unlock;
 
@@ -7246,6 +7412,15 @@ out:
 	return ret;
 }
 
+static void mutex_lock_double(struct mutex *a, struct mutex *b)
+{
+	if (b < a)
+		swap(a, b);
+
+	mutex_lock(a);
+	mutex_lock_nested(b, SINGLE_DEPTH_NESTING);
+}
+
 /**
  * sys_perf_event_open - open a performance event, associate it to a task/cpu
  *
@@ -7261,7 +7436,7 @@ SYSCALL_DEFINE5(perf_event_open,
 	struct perf_event *group_leader = NULL, *output_event = NULL;
 	struct perf_event *event, *sibling;
 	struct perf_event_attr attr;
-	struct perf_event_context *ctx;
+	struct perf_event_context *ctx, *uninitialized_var(gctx);
 	struct file *event_file = NULL;
 	struct fd group = {NULL, 0};
 	struct task_struct *task = NULL;
@@ -7459,43 +7634,68 @@ SYSCALL_DEFINE5(perf_event_open,
 	}
 
 	if (move_group) {
-		struct perf_event_context *gctx = group_leader->ctx;
-
-		mutex_lock(&gctx->mutex);
-		perf_remove_from_context(group_leader, false);
+		gctx = group_leader->ctx;
 
 		/*
-		 * Removing from the context ends up with disabled
-		 * event. What we want here is event in the initial
-		 * startup state, ready to be add into new context.
+		 * See perf_event_ctx_lock() for comments on the details
+		 * of swizzling perf_event::ctx.
 		 */
-		perf_event__state_init(group_leader);
+		mutex_lock_double(&gctx->mutex, &ctx->mutex);
+
+		perf_remove_from_context(group_leader, false);
+
 		list_for_each_entry(sibling, &group_leader->sibling_list,
 				    group_entry) {
 			perf_remove_from_context(sibling, false);
-			perf_event__state_init(sibling);
 			put_ctx(gctx);
 		}
-		mutex_unlock(&gctx->mutex);
-		put_ctx(gctx);
+	} else {
+		mutex_lock(&ctx->mutex);
 	}
 
 	WARN_ON_ONCE(ctx->parent_ctx);
-	mutex_lock(&ctx->mutex);
 
 	if (move_group) {
+		/*
+		 * Wait for everybody to stop referencing the events through
+		 * the old lists, before installing it on new lists.
+		 */
 		synchronize_rcu();
-		perf_install_in_context(ctx, group_leader, group_leader->cpu);
-		get_ctx(ctx);
+
+		/*
+		 * Install the group siblings before the group leader.
+		 *
+		 * Because a group leader will try and install the entire group
+		 * (through the sibling list, which is still in-tact), we can
+		 * end up with siblings installed in the wrong context.
+		 *
+		 * By installing siblings first we NO-OP because they're not
+		 * reachable through the group lists.
+		 */
 		list_for_each_entry(sibling, &group_leader->sibling_list,
 				    group_entry) {
+			perf_event__state_init(sibling);
 			perf_install_in_context(ctx, sibling, sibling->cpu);
 			get_ctx(ctx);
 		}
+
+		/*
+		 * Removing from the context ends up with disabled
+		 * event. What we want here is event in the initial
+		 * startup state, ready to be add into new context.
+		 */
+		perf_event__state_init(group_leader);
+		perf_install_in_context(ctx, group_leader, group_leader->cpu);
+		get_ctx(ctx);
 	}
 
 	perf_install_in_context(ctx, event, event->cpu);
 	perf_unpin_context(ctx);
+
+	if (move_group) {
+		mutex_unlock(&gctx->mutex);
+		put_ctx(gctx);
+	}
 	mutex_unlock(&ctx->mutex);
 
 	put_online_cpus();
@@ -7603,7 +7803,11 @@ void perf_pmu_migrate_context(struct pmu *pmu, int src_cpu, int dst_cpu)
 	src_ctx = &per_cpu_ptr(pmu->pmu_cpu_context, src_cpu)->ctx;
 	dst_ctx = &per_cpu_ptr(pmu->pmu_cpu_context, dst_cpu)->ctx;
 
-	mutex_lock(&src_ctx->mutex);
+	/*
+	 * See perf_event_ctx_lock() for comments on the details
+	 * of swizzling perf_event::ctx.
+	 */
+	mutex_lock_double(&src_ctx->mutex, &dst_ctx->mutex);
 	list_for_each_entry_safe(event, tmp, &src_ctx->event_list,
 				 event_entry) {
 		perf_remove_from_context(event, false);
@@ -7611,11 +7815,36 @@ void perf_pmu_migrate_context(struct pmu *pmu, int src_cpu, int dst_cpu)
 		put_ctx(src_ctx);
 		list_add(&event->migrate_entry, &events);
 	}
-	mutex_unlock(&src_ctx->mutex);
 
+	/*
+	 * Wait for the events to quiesce before re-instating them.
+	 */
 	synchronize_rcu();
 
-	mutex_lock(&dst_ctx->mutex);
+	/*
+	 * Re-instate events in 2 passes.
+	 *
+	 * Skip over group leaders and only install siblings on this first
+	 * pass, siblings will not get enabled without a leader, however a
+	 * leader will enable its siblings, even if those are still on the old
+	 * context.
+	 */
+	list_for_each_entry_safe(event, tmp, &events, migrate_entry) {
+		if (event->group_leader == event)
+			continue;
+
+		list_del(&event->migrate_entry);
+		if (event->state >= PERF_EVENT_STATE_OFF)
+			event->state = PERF_EVENT_STATE_INACTIVE;
+		account_event_cpu(event, dst_cpu);
+		perf_install_in_context(dst_ctx, event, dst_cpu);
+		get_ctx(dst_ctx);
+	}
+
+	/*
+	 * Once all the siblings are setup properly, install the group leaders
+	 * to make it go.
+	 */
 	list_for_each_entry_safe(event, tmp, &events, migrate_entry) {
 		list_del(&event->migrate_entry);
 		if (event->state >= PERF_EVENT_STATE_OFF)
@@ -7625,6 +7854,7 @@ void perf_pmu_migrate_context(struct pmu *pmu, int src_cpu, int dst_cpu)
 		get_ctx(dst_ctx);
 	}
 	mutex_unlock(&dst_ctx->mutex);
+	mutex_unlock(&src_ctx->mutex);
 }
 EXPORT_SYMBOL_GPL(perf_pmu_migrate_context);
 
@@ -7811,14 +8041,19 @@ static void perf_free_event(struct perf_event *event,
 
 	put_event(parent);
 
+	raw_spin_lock_irq(&ctx->lock);
 	perf_group_detach(event);
 	list_del_event(event, ctx);
+	raw_spin_unlock_irq(&ctx->lock);
 	free_event(event);
 }
 
 /*
- * free an unexposed, unused context as created by inheritance by
+ * Free an unexposed, unused context as created by inheritance by
  * perf_event_init_task below, used by fork() in case of fail.
+ *
+ * Not all locks are strictly required, but take them anyway to be nice and
+ * help out with the lockdep assertions.
  */
 void perf_event_free_task(struct task_struct *task)
 {
@@ -8137,7 +8372,7 @@ static void __init perf_event_init_all_cpus(void)
 	for_each_possible_cpu(cpu) {
 		swhash = &per_cpu(swevent_htable, cpu);
 		mutex_init(&swhash->hlist_mutex);
-		INIT_LIST_HEAD(&per_cpu(rotation_list, cpu));
+		INIT_LIST_HEAD(&per_cpu(active_ctx_list, cpu));
 	}
 }
 
@@ -8158,22 +8393,11 @@ static void perf_event_init_cpu(int cpu)
 }
 
 #if defined CONFIG_HOTPLUG_CPU || defined CONFIG_KEXEC
-static void perf_pmu_rotate_stop(struct pmu *pmu)
-{
-	struct perf_cpu_context *cpuctx = this_cpu_ptr(pmu->pmu_cpu_context);
-
-	WARN_ON(!irqs_disabled());
-
-	list_del_init(&cpuctx->rotation_list);
-}
-
 static void __perf_event_exit_context(void *__info)
 {
 	struct remove_event re = { .detach_group = true };
 	struct perf_event_context *ctx = __info;
 
-	perf_pmu_rotate_stop(ctx->pmu);
-
 	rcu_read_lock();
 	list_for_each_entry_rcu(re.event, &ctx->event_list, event_entry)
 		__perf_remove_from_context(&re);