A shot-noise limited single-user single-input single-output (SISO) Poisson fading channel with partial channel state information (CSI) at the transmitter and perfect CSI at the receiver is considered. We address an optimal transmit power allocation problem that maximizes the ergodic capacity of the SISO Poisson fading channel subject to peak and average power constraints with only quantized CSI available at the transmitter, acquired via a no-delay and error-free feedback link with finite-rate from the receiver to the transmitter. Due to the non-convexity of the proposed optimization problem, a globally optimum solution is difficult to obtain. However, we manage to obtain a locally optimal quantized power allocation (QPA) scheme by solving its dual Lagrangian optimization problem. We develop two efficient optimal QPA algorithms for solving the dual optimization problem and show that both of these algorithms converge to the globally optimal solution of the dual problem. A low-complexity near-optimal QPA algorithm is also derived for the case of large number of feedback bits. The results are then extended to the high peak signal-to-noise ratio (SNR) regime and an explicit expression for the approximate asymptotic ergodic capacity behavior in the high SNR regime with high rate quantization (as the number of feedback bits goes to infinity) is also provided. It is seen via numerical simulations that this asymptotic capacity expression correctly approaches the capacity of the corresponding full CSI case as the number of feedback bits becomes large. Finally, the effectiveness of the derived algorithms is examined through numerical simulations.