Flexible Hardware-based Security-aware Mechanisms and Architectures

For decades, software security has been the primary focus in securing our computing platforms. Hardware was always assumed trusted, and inherently served as the foundation, and thus the root of trust, of our systems. This has been further leveraged in developing hardware-based dedicated security extensions and architectures to protect software from attacks exploiting software vulnerabilities such as memory corruption. However, the recent outbreak of microarchitectural attacks has shaken these long-established trust assumptions in hardware entirely, thereby threatening the security of all of our computing platforms and bringing hardware and microarchitectural security under scrutiny. These attacks have undeniably revealed the grave consequences of hardware/microarchitecture security flaws to the entire platform security, and how they can even subvert the security guarantees promised by dedicated security architectures. Furthermore, they shed light on the sophisticated challenges particular to hardware/microarchitectural security; it is more critical (and more challenging) to extensively analyze the hardware for security flaws prior to production, since hardware, unlike software, cannot be patched/updated once fabricated. Hardware cannot reliably serve as the root of trust anymore, unless we develop and adopt new design paradigms where security is proactively addressed and scrutinized across the full stack of our computing platforms, at all hardware design and implementation layers. Furthermore, novel flexible security-aware design mechanisms are required to be incorporated in processor microarchitecture and hardware-assisted security architectures, that can practically address the inherent conflict between performance and security by allowing that the trade-off is configured to adapt to the desired requirements. In this thesis, we investigate the prospects and implications at the intersection of hardware and security that emerge across the full stack of our computing platforms and System-on-Chips (SoCs). On one front, we investigate how we can leverage hardware and its advantages, in contrast to software, to build more efficient and effective security extensions that serve security architectures, e.g., by providing execution attestation and enforcement, to protect the software from attacks exploiting software vulnerabilities. We further propose that they are microarchitecturally configured at runtime to provide different types of security services, thus adapting flexibly to different deployment requirements. On another front, we investigate how we can protect these hardware-assisted security architectures and extensions themselves from microarchitectural and software attacks that exploit design flaws that originate in the hardware, e.g., insecure resource sharing in SoCs. More particularly, we focus in this thesis on cache-based side-channel attacks, where we propose sophisticated cache designs, that fundamentally mitigate these attacks, while still preserving performance by enabling that the performance security trade-off is configured by design. We also investigate how these can be incorporated into flexible and customizable security architectures, thus complementing them to further support a wide spectrum of emerging applications with different performance/security requirements. Lastly, we inspect our computing platforms further beneath the design layer, by scrutinizing how the actual implementation of these mechanisms is yet another potential attack surface. We explore how the security of hardware designs and implementations is currently analyzed prior to fabrication, while shedding light on how state-of-the-art hardware security analysis techniques are fundamentally limited, and the potential for improved and scalable approaches