In this paper, we consider high-rate low-density parity-check (LDPC) codes for magnetic recording systems. We propose an efficient approach to reduce the computational complexity of the sum-product algorithm (SPA) for LDPC decoding by restricting the check operation to a small subset of input messages. Furthermore, the first-order MacLaurin Series is used to simplify the core operation and an attenuation factor is introduced to improve bit-error-rate (BER) performance by scaling down log-likelihood ratio (LLR) reliabilities in the decoding process. Simulation results show that the proposed decoder can obtain a noticeable performance gain over the conventional SPA at high signal-to-noise ratios (SNR) when used in turbo equalization schemes; while achieving a significantly lower computational complexity.