Revert "sched/core: Prevent race condition between cpuset and __sched_setscheduler()"

This reverts commit 710da3c8ea.

When changing a thread's scheduling priority, binder calls
sched_setscheduler_nocheck() while holding the node lock and
proc inner lock. This was safe until v5.3 when this change was
introduced where cpuset_read_lock() is called in this path
which can sleep.

This patch was introduced to fix a possible accounting error
in sched deadline (potential oversell of CPU bandwidth) due
to a race condition between cpusets and __sched_setscheduler().
This is not an issue for Android.

This should be fixed in the binder driver, but that may take
some time.

Bug: 151861772
Change-Id: Ica1ef71b3cdcdc509b341ea1b57c41f8ee73794a
Signed-off-by: Todd Kjos <tkjos@google.com>

kernel/cgroup/cpuset.c

@@ -334,17 +334,6 @@ static struct cpuset top_cpuset = {
  */
 
 DEFINE_STATIC_PERCPU_RWSEM(cpuset_rwsem);
-
-void cpuset_read_lock(void)
-{
-	percpu_down_read(&cpuset_rwsem);
-}
-
-void cpuset_read_unlock(void)
-{
-	percpu_up_read(&cpuset_rwsem);
-}
-
 static DEFINE_SPINLOCK(callback_lock);
 
 static struct workqueue_struct *cpuset_migrate_mm_wq;