Hardware-Software Codesign for Mitigating Spectre
Spectre attacks exploit control- and data-flow (mis)prediction on modern processors to transiently leak program secrets. Comprehensively mitigating Spectre leakage is hard, and doing so while preserving the program’s performance is even harder: no existing Spectre mitigations are widely deployed due to their high overhead or high complexity. We claim that a comprehensive, efficient, and low-complexity mitigation for Spectre attacks requires engaging in software-compiler-hardware co-design. In our talk, we will pitch such a co-designed Spectre mitigation that will be widely deployable at a low cost in security-critical applications. As a first step towards this goal, we have developed Serberus, a comprehensive and proven-correct Spectre mitigation for constant-time code that targets existing hardware. We are currently exploring lightweight hardware support to improve Serberus’ performance in other application domains.
|Position paper (plarch23-final16.pdf)