2020/07/31 by Elmira Karimi, Karimi, Elmira, Yunsi Fei +3 · 1 citation
Computer Science · Engineering · #Advanced Memory and Neural Computing #Cryptography and Security (cs.CR) #FOS: Computer and information sciences #Hardware Architecture (cs.AR) #Parallel Computing and Optimization Techniques #Security and Verification in Computing
paper · pdf · doi:10.48550/arxiv.2007.16175
openalex publication_date 2020/07/31 · openalex created_date 2022/07/26 · openalex updated_date 2026/07/28
GPUs are increasingly being used in security applications, especially for\naccelerating encryption/decryption. While GPUs are an attractive platform in\nterms of performance, the security of these devices raises a number of\nconcerns. One vulnerability is the data-dependent timing information, which can\nbe exploited by adversary to recover the encryption key. Memory system features\nare frequently exploited since they create detectable timing variations. In\nthis paper, our attack model is a coalescing attack, which leverages a critical\nGPU microarchitectural feature -- the coalescing unit. As multiple concurrent\nGPU memory requests can refer to the same cache block, the coalescing unit\ncollapses them into a single memory transaction. The access time of an\nencryption kernel is dependent on the number of transactions. Correlation\nbetween a guessed key value and the associated timing samples can be exploited\nto recover the secret key. In this paper, a series of hardware/software\ncountermeasures are proposed to obfuscate the memory timing side channel,\nmaking the GPU more resilient without impacting performance. Our hardware-based\napproach attempts to randomize the width of the coalescing unit to lower the\nsignal-to-noise ratio. We present a hierarchical Miss Status Holding Register\n(MSHR) design that can merge transactions across different warps. This feature\nboosts performance, while, at the same time, secures the execution. We also\npresent a software-based approach to permute the organization of critical data\nstructures, significantly changing the coalescing behavior and introducing a\nhigh degree of randomness. Equipped with our new protections, the effort to\nlaunch a successful attack is increased up to 1433X . 178X, while also\nimproving encryption/decryption performance up to 7%.\n