Verified Fault Handling for Modern Board Management Controllers

Abstract

Fault handling is the timely and crash-free response to critical changes in a system's operating characteristics, such as rapid temperature increases, or electrical shorts. In a typical computer system, it is the board management controller's job to correctly respond to such anomalous situations.

We develop an Isabelle/HOL model of a state machine for fault handling and define semantics for correctness of this procedure. Additionally, we formalize a notion of refinement that allows us to prove the correctness of implementations of this state machine. We also provide the first verified implementation of a C-based fault handler for board management controllers. Our implementation and the accompanying proofs are open-sourced and available online.

Furthermore, we successfully deploy our verified fault handler on top of the seL4 microkernel and alongside a production-grade, open source software stack widely deployed today, applying the cyber-retrofit approach to securing board management controllers in practice. The implementation and proof effort required is moderate, and our efforts indicate that already a small team of a handful of people can significantly raise the level of assurance of a modern, highly privileged software system.