memcg: do not check for mm in __mem_cgroup_count_vm_event

The mm given to __mem_cgroup_count_vm_event() cannot be NULL because the
function is either called from the page fault path or vma->vm_mm is used.
So the check can be dropped.

The check was introduced by commit 456f998 ("memcg: add the
pagefault count into memcg stats") because the originally proposed patch
used current->mm for shmem but this has been changed to vma->vm_mm later
on without the check being removed (thanks to Hugh for this
recollection).

Signed-off-by: Michal Hocko <mhocko@suse.cz>
Cc: Hugh Dickins <hughd@google.com>
Cc: Ying Han <yinghan@google.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/memcontrol.c

@@ -1021,9 +1021,6 @@ void __mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx)
 {
 	struct mem_cgroup *memcg;
 
-	if (!mm)
-		return;
-
 	rcu_read_lock();
 	memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
 	if (unlikely(!memcg))