AEPIC Leak: Architecturally Leaking Uninitialized Data from the Microarchitecture

CPU vulnerabilities undermine the security guarantees provided by software- and hardware-security improvements. While the discovery of transient-execution attacks increased the interest in CPU vulnerabilities on a microarchitectural level, architectural CPU vulnerabilities are still understudied. In this paper, we systematically analyze existing CPU vulnerabilities showing that CPUs suffer from vulnerabilities whose root causes match with those in complex software. We show that transient-execution attacks and architectural vulnerabilities often arise from the same type of bug and identify the blank spots. Investigating the blank spots, we focus on architecturally improperly initialized data locations. We discover AEPIC Leak, the first architectural CPU bug that leaks stale data from the microarchitecture without using a side channel. AEPIC Leak works on all recent Sunny-Cove-based Intel CPUs (i.e., Ice Lake and Alder Lake). It architecturally leaks stale data incorrectly returned by reading undefined APIC-register ranges. AEPIC Leak samples data transferred between the L2 and last-level cache, including SGX enclave data, from the superqueue. We target data in use, e.g., register values and memory loads, as well as data at rest, e.g., SGX-enclave data pages. Our end-to-end attack extracts AES-NI, RSA, and even the Intel SGX attestation keys from enclaves within a few seconds. We discuss mitigations and conclude that the only short-term mitigations for AEPIC Leak are to disable APIC MMIO or not rely on SGX.