RNA hairpins play a pivotal role in a diverse range of cellular functions, and are integral components of ribozymes, mRNA, and riboswitches. However, the mechanistic and kinetic details of RNA hairpin folding, which are key determinants of most of its biological functions, are poorly understood. In this work, we use the discrete path sampling (DPS) approach to explore the energy landscapes of two RNA tetraloop hairpins, and provide insights into their folding mechanisms and kinetics in atomistic detail. Our results show that the potential energy landscapes have a distinct funnel-like bias toward the folded hairpin state, consistent with efficient structure-seeking properties. Mechanistic and kinetic information is analyzed in terms of kinetic transition networks. We find microsecond folding times, consistent with temperature jump experiments, for hairpin folding initiated from relatively compact unfolded states. This process is essentially driven by an initial collapse, followed by rapid zippering of the helix stem in the final phase. Much lower folding rates are predicted when the folding is initiated from extended chains, which undergo longer excursions on the energy landscape before nucleation events can occur. Our work therefore explains recent experiments and coarse-grained simulations, where the folding kinetics exhibit precisely this dependency on the initial conditions.