list_lru: per-node list infrastructure fix

list_lru: per-node list infrastructure fix

After a while investigating, it seems to us that the imbalance we are
seeing are due to a multi-node race already in tree (our guess).  Although
the WARN is useful to show us the race, BUG_ON is too much, since it seems
the kernel should be fine going on after that.

Signed-off-by: Glauber Costa <glommer@openvz.org>
Cc: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>