An Improved Memory Integrity Protection Scheme

Memory integrity protection is crucial for many applications that store critical data in the memory. A secure real-time protection scheme can eliminate many potential attacks and protect such critical data effectively. However, existing schemes are not fast enough to provide on-line protection. The main difficulty in enrolling real-time memory integrity protection schemes is that the cryptographic operations typically cost a few hundred cycles per memory access. In this paper, we make an effort to reduce this overhead by proposing an improved memory integrity protection scheme. In addition, we for the first time propose a provably secure scheme that takes advantage of the error inheritance property, which can eliminate the costly check process that is normally required. More specifically, we propose the parallelization of the scheme, randomization of tagged data to eliminate potential attacks on error inheritance, and hashing of seeds (instead of the data) which enables delayed AES mask computations. The security of the proposed scheme is rigorously analyzed and the performance is measured. When these techniques are implemented simultaneously, the peak performance can be improved by up to a factor of 5 over a previously proposed scheme based on Merkle Trees.