Mind the Gap: Studying the Insecurity of Provably Secure Embedded Trusted Execution Architectures

The security claims of a system can be supported or refuted by different kinds of evidence. On the one hand, attack research uses empirical, experimental, inductive methods to refute security claims. If motivated and competent attackers do not succeed in breaking a specific security property, this provides some support (but no definite proof) that the system is secure.

On the other hand, formal methods use mathematical, deductive methods that can prove the security of a model of the system. The process of constructing a proof can uncover vulnerabilities that can then be fixed. The use of formal methods can be very powerful and is attractive because it seems to provide irrefutable evidence of security. However, that evidence applies only to the mathematical model, not to any actual system, and, hence, it is important to understand the gap between the model and the real-world system.

In this paper, we present a case study that examines this gap for two embedded security architectures that use formal methods to prove their security properties. Despite strong formal evidence for security, we discover numerous attacks against the implementations, all of which falsify proven security properties. These attacks range from exploiting simple programming errors to a novel DMA-based side-channel attack. The simple attacks demonstrate that the construction of systems and proofs is errorprone, while some of the more sophisticated attacks serve as examples to show that formal methods alone can never guarantee the security of a real-world system.

From our case study, we also distill actionable guidelines on how to provide stronger evidence for the security of a system.