Modern advanced CPU designs are frequently exposed to transient execution vulnerabilities, which allow attackers to harness microarchitectural side effects for data exfiltration. The leaked data may encompass direct target data, such as RSA keys, or indirect information, like physical page mappings. Thus, transient execution attacks can be divided into data-leakage and address-leakage, depending on the specific targets that are exposed. Existing studies have developed practical defenses and detection mechanisms against the microarchitectural attacks. However, almost all of them focus solely on data leakage and are thus unable to detect and counter address-leakage attacks, like Spoiler, due to their unique mechanisms. This paper introduces AALERT, the first detection mechanism specifically designed for address-leakage transient execution attacks. AALERT integrates a Cuckoo filter module within the CPUs Memory Order Buffer (MOB) to screen buffered addresses on the fly. We further optimize the filtering algorithm to minimize false positives. We discuss and implement several countermeasures to defeat the detected attacks. Finally, we evaluate the effectiveness and performance of AALERT based on prototype implementations, demonstrating a detection rate of 99.99% with negligible performance overhead.

Confidential computing technologies, such as that enabled by Intel SGX (Software Guard eXtensions), have been widely deployed in various commercial cloud platforms. Specifically, SGX uses hardware-isolated compartments named enclaves to shield user applications from Operating Systems (OSes) and hypervisors, thus providing confidentiality and integrity guarantees for code and data. However, some crucial problems are not fully analyzed yet, especially for the compatibility with binary applications. This work first delivers an overview of Intel SGX and reviews its five design constraints that may affect compatibility. Subsequently, we revisit three distinct compatibility solutions from the internals and analyze their impact on security, performance, and flexibility. At last, we lay out some fundamental lessons learned from prior SGX studies.

Spoiler attacks leak physical address information, which is exploited to accelerate reverse engineering of virtual-to-physical address mapping, thus greatly boosting Rowhammer and cache attacks. However, existing approaches that detect data-leakage attacks no longer suit the requirements of identifying Spoiler. This paper proposes Spoiler-alert,the first hardware-level mechanism to detect the address-leakage Spoiler attacks in real time. It leverages a cuckoo filter module embedded into Memory Order Buffer component to screen buffer addresses on-the-fly. We further optimise the filtering algorithm to reduce false positives. We assess the effectiveness and performance based on prototype implementations, which achieve a detection rate of 99.99% and negligible performance loss. Finally, we discuss potential reactions of our detection mechanism after a Spoiler attack was discovered.