Due to the effects of channel aging and estimation errors, perfect instantaneous channel state information (CSI) is unavailable at a base station. Therefore, robust precoding under imperfect CSI is important for practical communications. In this letter, we propose an efficient robust precoding design for massive multiple-input multiple-output systems. Based on the fractional programming technique, we transform the original non-convex optimization problem into a much more tractable equivalent problem, which can be iteratively solved by alternating optimization. Since all variables are updated via closed-form optimal solutions, the proposed algorithm is guaranteed to converge to a locally optimal point. Simulation results reveal that the proposed robust precoding algorithm has fast convergence and achieves significant performance improvement over the conventional ones.