In this work, we consider waveform design for a MIMO dual-function radar communication (DFRC) system, which enables sensing and communication for multiple users. Waveform design with perfect radar waveform constraint can guarantee the performance of radar, but it greatly loses the performance of communication. Therefore, we consider waveform design with radar waveform similarity constraint (SC) and constant mode constraint (CMC). Through minimizing downlink multi-user interference (MUI), the optimal problem can be formulated with the constraint of a tolerable mismatch between the actual radar beam pattern and the expectation, and the waveform envelope is guaranteed to be constant. Since the CMC is non-convex with quadratic equation constraint, a concave-convex penalty (CCP) algorithm is proposed. The proposed CCP can be decomposed and then the non-convex sub-constraint can be relaxed into a convex one by successive convex approximation (SCA). And finally it can be solved with a penalty factor that regulates the feasible region. The convergence of the proposed algorithm is verified by simulation, and the performance the designed waveform of DFRC are analyzed.