Optimized trusted execution for hard real-time applications on COTS processors

Abstract

While trusted execution environments (TEE) provide industry standard security and isolation, its implementation through secure monitor calls (SMC) attribute to large time overhead and weakened temporal predictability, potentially prohibiting the use of TEE in hard real-time systems. We propose super-TEEs, where multiple trusted execution sections are fused together to amortize TEE execution overhead and improve predictability through minimized I/O traffic and reduced switching between normal mode and TEE mode of execution. Super-TEEs may, however, violate a task's timing requirement and impact the schedulability of the system. We present a technique to enforce the correct timing requirement of a task, along with a sufficient test for schedulability in uniprocessors. We also, discuss ct-RM, a static task assignment and partitioned scheduling algorithm to schedule super-TEEs, alongside other real-time tasks, on multicore systems. Experimental results on a Raspberry Pi 3B, further confirmed by simulations, show that ct-RM outperforms the state-of-the-art technique in terms of usable utilization by 12% on average and up to 27%.