Cracking Randomized Coalescing Techniques with An Efficient Profiling-Based Side-Channel Attack to GPU
Xin Wang, Wei Zhang · 2019
GPUs have been increasingly used to accelerate a wide range of general purpose applications, including the encryption/decryption algorithms. However, recent research has shown that GPUs can be vulnerable to side-channel attacks (SCAs). Aiming at protecting GPUs against these SCAs, the Randomized Coalescing (RCoal) techniques are proposed with proven effectiveness on security improvement. In this paper, we propose to leverage a Profiling-based Side-Channel Attack (pSCA) to suppress the effectiveness of the RCoal techniques on security improvement or force the most secure choice which inevitably results in substantial performance degradation. Without the detailed information of the RCoal configuration, the pSCA is still able to rebuild the secure key of a GPU-supported AES algorithm under the RCoal protection in a reasonable duration.