Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net

4 files changed, 76 insertions, 14 deletions

diff --git a/kernel/cgroup.c b/kernel/cgroup.c
index d1c51b7f5221fb..5e8dab5bf9adb9 100644
--- a/kernel/cgroup.c
+++ b/kernel/cgroup.c
@@ -6270,6 +6270,12 @@ void cgroup_sk_alloc(struct sock_cgroup_data *skcd)
 	if (cgroup_sk_alloc_disabled)
 		return;
 
+	/* Socket clone path */
+	if (skcd->val) {
+		cgroup_get(sock_cgroup_ptr(skcd));
+		return;
+	}
+
 	rcu_read_lock();
 
 	while (true) {
diff --git a/kernel/events/core.c b/kernel/events/core.c
index a7b8c1c75fa71c..9fc3be0c7d01df 100644
--- a/kernel/events/core.c
+++ b/kernel/events/core.c
@@ -2496,11 +2496,11 @@ static int __perf_event_stop(void *info)
 	return 0;
 }
 
-static int perf_event_restart(struct perf_event *event)
+static int perf_event_stop(struct perf_event *event, int restart)
 {
 	struct stop_event_data sd = {
 		.event		= event,
-		.restart	= 1,
+		.restart	= restart,
 	};
 	int ret = 0;
 
@@ -3549,10 +3549,18 @@ static int perf_event_read(struct perf_event *event, bool group)
 			.group = group,
 			.ret = 0,
 		};
-		ret = smp_call_function_single(event->oncpu, __perf_event_read, &data, 1);
-		/* The event must have been read from an online CPU: */
-		WARN_ON_ONCE(ret);
-		ret = ret ? : data.ret;
+		/*
+		 * Purposely ignore the smp_call_function_single() return
+		 * value.
+		 *
+		 * If event->oncpu isn't a valid CPU it means the event got
+		 * scheduled out and that will have updated the event count.
+		 *
+		 * Therefore, either way, we'll have an up-to-date event count
+		 * after this.
+		 */
+		(void)smp_call_function_single(event->oncpu, __perf_event_read, &data, 1);
+		ret = data.ret;
 	} else if (event->state == PERF_EVENT_STATE_INACTIVE) {
 		struct perf_event_context *ctx = event->ctx;
 		unsigned long flags;
@@ -4837,6 +4845,19 @@ static void ring_buffer_attach(struct perf_event *event,
 		spin_unlock_irqrestore(&rb->event_lock, flags);
 	}
 
+	/*
+	 * Avoid racing with perf_mmap_close(AUX): stop the event
+	 * before swizzling the event::rb pointer; if it's getting
+	 * unmapped, its aux_mmap_count will be 0 and it won't
+	 * restart. See the comment in __perf_pmu_output_stop().
+	 *
+	 * Data will inevitably be lost when set_output is done in
+	 * mid-air, but then again, whoever does it like this is
+	 * not in for the data anyway.
+	 */
+	if (has_aux(event))
+		perf_event_stop(event, 0);
+
 	rcu_assign_pointer(event->rb, rb);
 
 	if (old_rb) {
@@ -6112,7 +6133,7 @@ static void perf_event_addr_filters_exec(struct perf_event *event, void *data)
 	raw_spin_unlock_irqrestore(&ifh->lock, flags);
 
 	if (restart)
-		perf_event_restart(event);
+		perf_event_stop(event, 1);
 }
 
 void perf_event_exec(void)
@@ -6156,7 +6177,13 @@ static void __perf_event_output_stop(struct perf_event *event, void *data)
 
 	/*
 	 * In case of inheritance, it will be the parent that links to the
-	 * ring-buffer, but it will be the child that's actually using it:
+	 * ring-buffer, but it will be the child that's actually using it.
+	 *
+	 * We are using event::rb to determine if the event should be stopped,
+	 * however this may race with ring_buffer_attach() (through set_output),
+	 * which will make us skip the event that actually needs to be stopped.
+	 * So ring_buffer_attach() has to stop an aux event before re-assigning
+	 * its rb pointer.
 	 */
 	if (rcu_dereference(parent->rb) == rb)
 		ro->err = __perf_event_stop(&sd);
@@ -6670,7 +6697,7 @@ static void __perf_addr_filters_adjust(struct perf_event *event, void *data)
 	raw_spin_unlock_irqrestore(&ifh->lock, flags);
 
 	if (restart)
-		perf_event_restart(event);
+		perf_event_stop(event, 1);
 }
 
 /*
@@ -7933,7 +7960,7 @@ static void perf_event_addr_filters_apply(struct perf_event *event)
 	mmput(mm);
 
 restart:
-	perf_event_restart(event);
+	perf_event_stop(event, 1);
 }
 
 /*
diff --git a/kernel/events/ring_buffer.c b/kernel/events/ring_buffer.c
index ae9b90dc9a5a66..257fa460b84603 100644
--- a/kernel/events/ring_buffer.c
+++ b/kernel/events/ring_buffer.c
@@ -330,15 +330,22 @@ void *perf_aux_output_begin(struct perf_output_handle *handle,
 	if (!rb)
 		return NULL;
 
-	if (!rb_has_aux(rb) || !atomic_inc_not_zero(&rb->aux_refcount))
+	if (!rb_has_aux(rb))
 		goto err;
 
 	/*
-	 * If rb::aux_mmap_count is zero (and rb_has_aux() above went through),
-	 * the aux buffer is in perf_mmap_close(), about to get freed.
+	 * If aux_mmap_count is zero, the aux buffer is in perf_mmap_close(),
+	 * about to get freed, so we leave immediately.
+	 *
+	 * Checking rb::aux_mmap_count and rb::refcount has to be done in
+	 * the same order, see perf_mmap_close. Otherwise we end up freeing
+	 * aux pages in this path, which is a bug, because in_atomic().
 	 */
 	if (!atomic_read(&rb->aux_mmap_count))
-		goto err_put;
+		goto err;
+
+	if (!atomic_inc_not_zero(&rb->aux_refcount))
+		goto err;
 
 	/*
 	 * Nesting is not supported for AUX area, make sure nested
diff --git a/kernel/sched/core.c b/kernel/sched/core.c
index 2a906f20fba7c4..44817c640e99b1 100644
--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -2016,6 +2016,28 @@ try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags)
 	success = 1; /* we're going to change ->state */
 	cpu = task_cpu(p);
 
+	/*
+	 * Ensure we load p->on_rq _after_ p->state, otherwise it would
+	 * be possible to, falsely, observe p->on_rq == 0 and get stuck
+	 * in smp_cond_load_acquire() below.
+	 *
+	 * sched_ttwu_pending()                 try_to_wake_up()
+	 *   [S] p->on_rq = 1;                  [L] P->state
+	 *       UNLOCK rq->lock  -----.
+	 *                              \
+	 *				 +---   RMB
+	 * schedule()                   /
+	 *       LOCK rq->lock    -----'
+	 *       UNLOCK rq->lock
+	 *
+	 * [task p]
+	 *   [S] p->state = UNINTERRUPTIBLE     [L] p->on_rq
+	 *
+	 * Pairs with the UNLOCK+LOCK on rq->lock from the
+	 * last wakeup of our task and the schedule that got our task
+	 * current.
+	 */
+	smp_rmb();
 	if (p->on_rq && ttwu_remote(p, wake_flags))
 		goto stat;