With interfacing various equipment in modern power systems, oscillation stability is the prior requirement for safe operation of power systems. For oscillation analysis, eigenvalue computation by state space model of power systems is a well-developed method, but still worthy of studying due to existing issues in practical application. Especially, for large-scale systems, high dimension of state space model, undesired distribution of eigenvalues and finite computing resources make eigenvalue computation challenging. This paper proposes a method for searching eigenvalues in user-specified spectrum region of interest. The method is designed with two layers, which are search navigator in upper layer and kernel solver in lower layer, respectively. The search navigator divides computation of all target eigenvalues into a sequence of search tasks, and enables flexibility on specifying search region and adaptivity to computing platforms. The kernel solver is implemented by Krylov-Schur algorithm with restarting strategy highly adaptive to various distributions of eigenvalues. The restarting strategy contributes well reliability on finding eigenvalues without omission. Effectiveness and reliability of the proposed method are validated by numerical experiments on three test power systems.