Photovoltaic (PV)-assisted electric vehicle (EV) charging stations require real-time energy management that coordinates uncertain renewable generation, stochastic charging demand, and time-varying electricity prices while considering user incentives and fairness. This paper proposes a bi-level online optimization framework for PV-assisted EV charging stations. At the upper level, a Lyapunov-based online algorithm determines the aggregate EV charging command using only real-time information, with time-coupled EV energy requirements transformed into per-slot feasible charging ranges. The resulting time-average operating cost is theoretically shown to remain within a bounded gap of the offline optimum. A real-time pricing mechanism is designed to transfer part of the economic benefit of PV generation to EV users while ensuring that the charging price does not exceed the grid electricity price. At the lower level, a fairness-aware allocation strategy decomposes the aggregate command into individual EV charging powers by minimizing the variance of users" relative deviations from their preferred charging profiles, while ensuring exact tracking of the aggregate command. Numerical results demonstrate reduced station operating cost, lower average user charging price, improved allocation fairness, and millisecond-level computation time.