Hello, Sergey.

On Thu, May 14, 2015 at 10:09:13AM +0900, Sergey Senozhatsky wrote:

quoted

+static inline void threadgroup_change_begin(struct task_struct *tsk)
 {
-	down_write(&tsk->signal->group_rwsem);
+	might_sleep();

I think cgroup_threadgroup_change_begin()->down_read() already does
might_sleep() check.

Sure but it's a layering thing.  threadgroup_change_begin() should be
called from a blockable context whether the hook users actually make
use of it or not.  e.g. We want might_sleep() even when
!CONFIG_CGROUP.

Thanks.

-- 
tejun