In this paper a hierarchical distributed model predictive control scheme is proposed for heating, ventilation and air-conditioning (HVAC) systems in buildings. The building consists in multiple connected rooms and zones. The control objective is to keep the temperature of each room and zone at a given comfortable level with minimal energy consumption. This control scheme is divided into two levels. The upper level controller collects temperature and predictive information of all rooms and zones to generate reference trajectories while a lower level controller only uses local information to track the reference and optimize energy efficiency and thermal comfort. By using a contraction property of building's dynamics, recursive feasibility of the proposed algorithm is guaranteed. Simulation results are given to show the performance of our proposed control strategy.