mm/memcontrol.c: fix the margin computation in mem_cgroup_margin()

mem_cgroup_margin() might return (memory.limit - memory_count) when the
memsw.limit is in excess.  This doesn't happen usually because we do not
allow excess on hard limits and (memory.limit <= memsw.limit), but
__GFP_NOFAIL charges can force the charge and cause the excess when no
memory is really swappable (swap is full or no anonymous memory is
left).

[[email protected]: rewrote changelog]
  Link: http://lkml.kernel.org/r/[email protected]
Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: Li RongQing <[email protected]>
Acked-by: Vladimir Davydov <[email protected]>
Acked-by: Michal Hocko <[email protected]>
Cc: Johannes Weiner <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

mm/memcontrol.c

@@ -1108,6 +1108,8 @@ static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg)
 		limit = READ_ONCE(memcg->memsw.limit);
 		if (count <= limit)
 			margin = min(margin, limit - count);
+		else
+			margin = 0;
 	}
 
 	return margin;