In this letter, we propose a downlink precoder for massive multi-input multi-output (MIMO) system in the presence of I/Q imbalance (IQI) at both base station and user terminals (UTs). We formulate a minimum mean-square error (MMSE) precoding problem with multiple regularization parameters for different channel qualities and channel correlations of UTs with different IQIs. Then we obtain the asymptotically optimal precoding matrix for the massive MIMO system. Moreover, we derive the asymptotic sum rate of the proposed regularized zero-forcing (RZF) precoder. We show that the proposed RZF precoder achieves significantly higher sum rate than conventional RZF precoder. In addition, the proposed precoder shows robust performances to severe IQI conditions.