Efficiënte mechanismes voor het behoud van control flow confidentialiteit

In 2018, a group of leading scientists convened at Schloss Dagstuhl for a seminar on secure compilation, emphasizing the need for efficient hardware enforcement mechanisms to "protect a source language's abstractions in compiled code, even against low-level attacks". Simultaneously, Gernot Heiser called for a new hardware-software contract, arguing that the current ISA abstraction is insufficient for ensuring system security. This dissertation addresses both of these needs. We revisit a traditional side-channel problem, often considered resolved by the community: control-flow leakage attacks. In these attacks, a low-level adversary seeks to infer a program's control flow through side-channel observations. These threats are inherently absent in high-level programming languages, which exhibit control-flow confidentiality (CFC), a security property ensuring that programs secrets do not leak through control flow. Our research explores countermeasures against control-flow leakage attacks from a novel perspective. The key contributions include: (1) innovative hardware- software co-designed enforcement primitives that establish an explicit ISA-level security contract for control-flow confidentiality, and (2) low-level compiler algorithms. These contributions lay the groundwork for secure compilers that efficiently preserve control-flow confidentiality of high-level language constructs in low-level machine code. Additionally, we critically reassess the constant-time programming discipline, traditionally linked closely to source code. We argue that preserving confidentiality should be the primary objective, rather than rigidly adhering to constant-time rules in source code. Recognizing that side-channel vulnerabilities are inherently tied to the underlying hardware, we propose that concrete side- channel protection principles should be articulated at the ISA level, as part of a comprehensive hardware-software security contract. By decoupling target- dependent security policies from target-independent source code, we demonstrate