1 files changed, 258 insertions, 42 deletions

diff --git a/kernel/events/core.c b/kernel/events/core.c
index acdc087..466213f 100644
--- a/kernel/events/core.c
+++ b/kernel/events/core.c
@@ -604,6 +604,77 @@ 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(struct perf_event *event)
+{
+	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(&ctx->mutex);
+	if (event->ctx != ctx) {
+		mutex_unlock(&ctx->mutex);
+		put_ctx(ctx);
+		goto again;
+	}
+
+	return ctx;
+}
+
+static void perf_event_ctx_unlock(struct perf_event *event,
+				  struct perf_event_context *ctx)
+{
+	mutex_unlock(&ctx->mutex);
+	put_ctx(ctx);
+}
+
 static void unclone_ctx(struct perf_event_context *ctx)
 {
 	if (ctx->parent_ctx) {
@@ -1245,7 +1316,7 @@ static 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;
@@ -1287,6 +1358,19 @@ 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);
+}
+
 static void perf_set_shadow_time(struct perf_event *event,
 				 struct perf_event_context *ctx,
 				 u64 tstamp)
@@ -1580,6 +1664,8 @@ perf_install_in_context(struct perf_event_context *ctx,
 	lockdep_assert_held(&ctx->mutex);
 
 	event->ctx = ctx;
+	if (event->cpu != -1)
+		event->cpu = cpu;
 
 	if (!task) {
 		/*
@@ -1720,7 +1806,7 @@ 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;
@@ -1777,7 +1863,19 @@ out:
 	raw_spin_unlock_irq(&ctx->lock);
 }
 
-static int perf_event_refresh(struct perf_event *event, int refresh)
+/*
+ * 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);
+}
+
+static int _perf_event_refresh(struct perf_event *event, int refresh)
 {
 	/*
 	 * not supported on inherited events
@@ -1786,11 +1884,26 @@ static 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;
+}
+
 static void ctx_sched_out(struct perf_event_context *ctx,
 			  struct perf_cpu_context *cpuctx,
 			  enum event_type_t event_type)
@@ -3034,7 +3147,16 @@ static void put_event(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
@@ -3086,12 +3208,13 @@ 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;
@@ -3106,7 +3229,7 @@ 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;
 
@@ -3120,14 +3243,12 @@ 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;
 }
@@ -3186,8 +3307,14 @@ 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;
+
+	ctx = perf_event_ctx_lock(event);
+	ret = perf_read_hw(event, buf, count);
+	perf_event_ctx_unlock(event, ctx);
 
-	return perf_read_hw(event, buf, count);
+	return ret;
 }
 
 static unsigned int perf_poll(struct file *file, poll_table *wait)
@@ -3207,7 +3334,7 @@ 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);
@@ -3306,25 +3433,24 @@ 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);
@@ -3365,13 +3491,30 @@ 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;
+}
+
 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;
@@ -3379,11 +3522,15 @@ 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;
@@ -5416,7 +5563,6 @@ static int swevent_hlist_get_cpu(struct perf_event *event, int cpu)
 	int err = 0;
 
 	mutex_lock(&swhash->hlist_mutex);
-
 	if (!swevent_hlist_deref(swhash) && cpu_online(cpu)) {
 		struct swevent_hlist *hlist;
 
@@ -6216,6 +6362,7 @@ perf_event_alloc(struct perf_event_attr *attr, int cpu,
 	if (!group_leader)
 		group_leader = event;
 
+	mutex_init(&event->group_leader_mutex);
 	mutex_init(&event->child_mutex);
 	INIT_LIST_HEAD(&event->child_list);
 
@@ -6444,6 +6591,15 @@ 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
  *
@@ -6459,7 +6615,7 @@ 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 file *group_file = NULL;
 	struct task_struct *task = NULL;
@@ -6513,6 +6669,16 @@ SYSCALL_DEFINE5(perf_event_open,
 			group_leader = NULL;
 	}
 
+	/*
+	 * Take the group_leader's group_leader_mutex before observing
+	 * anything in the group leader that leads to changes in ctx,
+	 * many of which may be changing on another thread.
+	 * In particular, we want to take this lock before deciding
+	 * whether we need to move_group.
+	 */
+	if (group_leader)
+		mutex_lock(&group_leader->group_leader_mutex);
+
 	if (pid != -1 && !(flags & PERF_FLAG_PID_CGROUP)) {
 		task = find_lively_task_by_vpid(pid);
 		if (IS_ERR(task)) {
@@ -6572,7 +6738,7 @@ SYSCALL_DEFINE5(perf_event_open,
 	/*
 	 * Get the target context (task or percpu):
 	 */
-	ctx = find_get_context(pmu, task, cpu);
+	ctx = find_get_context(pmu, task, event->cpu);
 	if (IS_ERR(ctx)) {
 		err = PTR_ERR(ctx);
 		goto err_alloc;
@@ -6627,9 +6793,14 @@ SYSCALL_DEFINE5(perf_event_open,
 	}
 
 	if (move_group) {
-		struct perf_event_context *gctx = group_leader->ctx;
+		gctx = group_leader->ctx;
+
+		/*
+		 * See perf_event_ctx_lock() for comments on the details
+		 * of swizzling perf_event::ctx.
+		 */
+		mutex_lock_double(&gctx->mutex, &ctx->mutex);
 
-		mutex_lock(&gctx->mutex);
 		perf_remove_from_context(group_leader);
 
 		/*
@@ -6644,28 +6815,41 @@ SYSCALL_DEFINE5(perf_event_open,
 			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) {
-		perf_install_in_context(ctx, group_leader, cpu);
+		/*
+		 * 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, event->cpu);
 		get_ctx(ctx);
 		list_for_each_entry(sibling, &group_leader->sibling_list,
 				    group_entry) {
-			perf_install_in_context(ctx, sibling, cpu);
+			perf_install_in_context(ctx, sibling, event->cpu);
 			get_ctx(ctx);
 		}
 	}
 
-	perf_install_in_context(ctx, event, cpu);
+	perf_install_in_context(ctx, event, event->cpu);
 	++ctx->generation;
 	perf_unpin_context(ctx);
+
+	if (move_group) {
+		mutex_unlock(&gctx->mutex);
+		put_ctx(gctx);
+	}
 	mutex_unlock(&ctx->mutex);
 
+	if (group_leader)
+		mutex_unlock(&group_leader->group_leader_mutex);
+
 	event->owner = current;
 
 	mutex_lock(&current->perf_event_mutex);
@@ -6697,6 +6881,8 @@ err_task:
 	if (task)
 		put_task_struct(task);
 err_group_fd:
+	if (group_leader)
+		mutex_unlock(&group_leader->group_leader_mutex);
 	fput_light(group_file, fput_needed);
 err_fd:
 	put_unused_fd(event_fd);
@@ -6751,6 +6937,42 @@ err:
 }
 EXPORT_SYMBOL_GPL(perf_event_create_kernel_counter);
 
+void perf_pmu_migrate_context(struct pmu *pmu, int src_cpu, int dst_cpu)
+{
+	struct perf_event_context *src_ctx;
+	struct perf_event_context *dst_ctx;
+	struct perf_event *event, *tmp;
+	LIST_HEAD(events);
+
+	src_ctx = &per_cpu_ptr(pmu->pmu_cpu_context, src_cpu)->ctx;
+	dst_ctx = &per_cpu_ptr(pmu->pmu_cpu_context, dst_cpu)->ctx;
+
+	/*
+	 * 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);
+		put_ctx(src_ctx);
+		list_add(&event->event_entry, &events);
+	}
+
+	synchronize_rcu();
+
+	list_for_each_entry_safe(event, tmp, &events, event_entry) {
+		list_del(&event->event_entry);
+		if (event->state >= PERF_EVENT_STATE_OFF)
+			event->state = PERF_EVENT_STATE_INACTIVE;
+		perf_install_in_context(dst_ctx, event, dst_cpu);
+		get_ctx(dst_ctx);
+	}
+	mutex_unlock(&dst_ctx->mutex);
+	mutex_unlock(&src_ctx->mutex);
+}
+EXPORT_SYMBOL_GPL(perf_pmu_migrate_context);
+
 static void sync_child_event(struct perf_event *child_event,
 			       struct task_struct *child)
 {
@@ -7309,12 +7531,6 @@ static void perf_event_exit_cpu_context(int cpu)
 
 static void perf_event_exit_cpu(int cpu)
 {
-	struct swevent_htable *swhash = &per_cpu(swevent_htable, cpu);
-
-	mutex_lock(&swhash->hlist_mutex);
-	swevent_hlist_release(swhash);
-	mutex_unlock(&swhash->hlist_mutex);
-
 	perf_event_exit_cpu_context(cpu);
 }
 #else