The massively distributed antenna system (MDAS) can offer a significant macro-diversity gain in comparison with traditional co-located massive MIMO. Thus, it is a promising candidate for future network densification. Coordinated antenna selection (CAS), by harmoniously activating a subset of the antennas, can efficiently exploit the benefit of MDAS while reducing the number of radio frequency chains. However, perfect CAS usually requires global channel state information (CSI), which consequently leads to a tremendous amount of system overhead. In order to control the cost of CAS, we in this paper propose the use of visible antennas (VAs) for each mobile terminal (MT). Assuming that only the CSI between a given MT and its VAs is acquired, we use the number of VAs to quantitatively characterize a general partial CSI condition. Then we formulate the corresponding CAS problem as a non-convex integer programming problem. By adopting variable relaxation and successive approximation, we derive a suboptimal solution to the problem based on Geometric Programming (GP). Simulation results illustrate that the proposed CAS scheme can offer a nearoptimal performance gain in terms of achievable sum rate, for any randomly-chosen number of VAs.