Frequency domain (FD) multi-user detection (MUD) has been shown to be an effective means of approaching the theoretical per-user capacity for single carrier modulation (SCM) schemes in massive MIMO scenarios with highly dispersive channels. When a cyclic prefix is added to the SCM waveform, the circulant structures of the resulting convolutional channel matrix allows for relatively simple expressions for the FD detection. In this paper, we develop a computationally efficient minimum mean squared error (MMSE) FD-MUD technique in a time-division duplexing (TDD) massive MIMO setup and show how processing steps can be shared between uplink and downlink.