Envisioned as one of the most promising technologies, holographic multiple-input multiple-output (H-MIMO) recently attracts notable research interests for its great potential in expanding wireless possibilities and achieving fundamental wireless limits. Empowered by the nearly continuous, large and energy-efficient surfaces with powerful electromagnetic (EM) wave control capabilities, H-MIMO opens up the opportunity for signal processing in a more fundamental EM-domain, paving the way for realizing holographic imaging level communications in supporting the extremely high spectral efficiency and energy efficiency in future networks. In this article, we try to implement a generalized EM-domain near-field channel modeling and study its capacity limit of point-to-point H-MIMO systems that equips arbitrarily placed surfaces in a line-of-sight (LoS) environment. Two effective and computational-efficient channel models are established from their integral counterpart, where one is with a sophisticated formula but showcases more accurate, and another is concise with a slight precision sacrifice. Furthermore, we unveil the capacity limit using our channel model, and derive a tight upper bound based upon an elaborately built analytical framework. Our result reveals that the capacity limit grows logarithmically with the product of transmit element area, receive element area, and the combined effects of $1/{{d}_{mn}^2}$, $1/{{d}_{mn}^4}$, and $1/{{d}_{mn}^6}$ over all transmit and receive antenna elements, where $d_{mn}$ indicates the distance between each transmit and receive elements. Numerical evaluations validate the effectiveness of our channel models, and showcase the slight disparity between the upper bound and the exact capacity, which is beneficial for predicting practical system performance.