Trust less : shrinking the trusted parts of trusted systems

Abstract

Modern computers, industrial control systems, and other automation are broadly vulnerable as a result of decades of systemic forces that have prioritized cost and performance over security. Computers across the board face a crisis in the form of motivated software adversaries with access to our imperfect and enormously complex software. Considering these weaknesses, trust in modern computing systems is often not well-placed. Looking ahead to a shift in our collective priorities, this thesis is centered around a rigorous discussion of hardware-assisted isolation and enclaves -- authenticated software modules -- as a means to drastically reduce the complexity of trusted systems. By allowing trustworthy enclaved software to co-exist with, but remain strongly isolated from, existing software, we enable a gentle transition toward trustworthy systems. Specifically, this thesis refines formal definitions of enclaved execution and threat model via a series of hardware and software co-designs. These case studies explore enclave processors with small trusted computing bases spanning a gradient from an embedded SoC to a modern high-performance processor. This work is complementary to, and enables more effective application of, many powerful ideas such as information flow control, formal verification, multi-party computation, and other tools for trustworthy computing.