In this article, we consider the problem of secure state estimation in the presence of a passive eavesdropper for linear cyber-physical systems. By exploiting the system dynamic characteristic, a time-varying coding algorithm with low computational complexity is proposed to defend eavesdropping attacks. To derive the minimum coded dimension and the upper bound of update period in this algorithm, a convex estimator and a non-convex optimization bilinear programming model are established, respectively. Furthermore, a novel privacy-preserving mechanism is designed by combining coding and noise against an eavesdropper, where the estimation error of eavesdropper is monotonically increasing with the covariance of added noise. It is shown that both schemes can precisely recover the encoded message for defender and impair the estimation performance of eavesdropper. Finally, the theoretical results are validated by simulation examples.