We consider the optimization of the downlink transmission for simultaneous wireless information and power transfer (SWIPT) in multi-cell massive multiple-input multipleoutput (MIMO) systems. The system comprises a two-antenna active energy harvester (EH) which is capable of legitimately harvesting energy via one antenna, and illegitimately and actively eavesdropping the signal intended for certain information user(s) (IU(s)) via the other antenna for the purpose of information decoding or energy harvesting. Thereby, the considered problems are: 1) when the EH eavesdrops for the purpose of information decoding, i.e., the EH is information-untrusted by the base station (BS), we propose to maximize the worst-case secrecy rate under a constraint on a worst-case average harvested energy (AHE); 2) when the EH eavesdrops one or multiple IUs for energy harvesting, i.e., the EH is information-trusted by the BS, we propose to maximize the sum-rate of the IUs under a constraint on a worst-case AHE by the EH. We derive asymptotic expressions for a lower bound on ergodic secrecy rate (ESR) and AHE in large system limit. Then, we use these results to optimize the power allocation for downlink SWIPT transmissions which include: information signals, artificial noise (AN) and energy signal towards the IUs, legitimate and illegitimate antennas of the EH, respectively. Simulation results show the accuracy of our asymptotic analysis. We show that there is a performance tradeoff between the worst-case ESR and the worst-case AHE. In addition, the impact of the combined legitimate/illegitimate operation of the EH on the SWIPT performance is evaluated.