When Side Channel Meets Row Hammer: Cache-Memory Attacks in Clouds and Mobile Devices

Dec 07, 2016

Download: MP4 Video Size: 168.8MB  
Watch on YouTube

Abstract

Processor caches and memory chips are hardware components used by all software programs on a computer system. They are designed, and thereafter fine-tuned over the years, for better performance and power efficiency, but not for strong isolation between mutually distrustful software programs. However, modern computing paradigm has been shifting towards resource sharing without full trust: In multi-tenant public clouds, virtual machines controlled by different customers are scheduled to run on the same cloud servers; in mobile devices, untrusted third-party apps, though isolated using sandboxes, share the same devices with sensitive apps. Our research question is whether sharing of memory resources will introduce new security threats to these systems.

In this talk, we highlight a type of security threats that we call cache-memory attacks. These attacks are possible due to insufficient isolation in hardware memory resources (e.g., various levels of caches, memory controllers, buses and chips, etc.) that are shared between malevolent and sensitive software programs. We coin cache-memory attacks as the umbrella terms of side-channel attacks (i.e., confidentiality attacks), row-hammer attacks (i.e., integrity attacks) and resource contention attacks (i.e., availability attacks). We will discuss the root vulnerabilities of these attacks and their exploitation in the context of clouds and mobile devices. We will also cover some defense techniques against these attacks that we have developed over the past few years.