ll_ra_count_get() is designed to allow over reservation for performance reasons, this could happen that multiple threads consuming a small readahead window limit, it will fallback to small 4K read which hurts performances.