Exorcising Spectres with Secure Compilers

Abstract

Attackers can access sensitive information of programs by exploiting the side-effects of speculatively-executed instructions using Spectre attacks. To mitigate these attacks, popular compilers deployed a wide range of countermeasures whose security, however, has not been ascertained: while some are believed to be secure, others are known to be insecure and result in vulnerable programs. This paper develops formal foundations for reasoning about the security of these defenses. For this, it proposes a framework of secure compilation criteria that characterise when compilers produce code resistant against Spectre v1 attacks. With this framework, this paper performs a comprehensive security analysis of countermeasures against Spectre v1 attacks implemented in major compilers, deriving the first security proofs of said countermeasures.