AutoLock: Why Cache Attacks on ARM Are Harder Than You Think
summary
Claims an undocumented ARM cache feature called AutoLock exists in inclusive last-level caches, silently preventing cross-core evictions and breaking standard Evict+Time/Prime+Probe/Evict+Reload patterns; authors devise three detection tests and find AutoLock on several Cortex‑A SoCs (ARMv7/ARMv8) but not on all cores. Same‑core evictions and recent flush instructions still work. A redundant cross‑core AES eviction attack is implemented and recovers full keys, proving AutoLock hinders but does not stop real attacks.
AI-written summary · about
chapters (7)
- 0:00Intro
- 0:29The Big Picture
- 1:45Cache Basics
- 7:23AutoLock
- 11:05Implications
- 15:11Countermeasures
- 20:10Conclusion
related talks
29mARMageddon: Cache Attacks on Mobile Devices
41mHand Me Your SECRET, MCU! Microarchitectural Timing Attacks on Microcontrollers are Practical
35mPage Cache Attacks: Microarchitectural Attacks on Flawless Hardware
23mMalicious Management Unit: Why Stopping Cache Attacks in Software is Harder Than You Think
15mI Know What You Said: Unveiling Hardware Cache Side-Channels in Local Large
47mA Security RISC? The State of Microarchitectural Attacks on RISC-V
34mCore Escalation: Unleashing the Power of Cross-Core Attacks on Heterogeneous System
42mAre Microarchitectural Attacks still possible on ...