This paper presents a message passing approach to joint channel estimation, data detection and decoding over time-selective Rayleigh fading channel with residual carrier frequency offset (CFO). The proposed algorithm utilizes the sum product algorithm (SPA) implemented on a factor graph (FG) representing the joint a posteriori probability distribution of the unknown CFO, information bits and channel coefficients vector given the channel output. A combination of particle filtering and Gaussian parameterization is employed to approximate the exact probability density function in message passing for CFO and channel estimation. Computer simulations demonstrate the effectiveness of the proposed algorithm in combating the CFO over unknown Rayleigh fading channels.