Quantifying side-channels-based information-leakage for GPGPU programs
Title: Quantifying side-channels-based information-leakage for GPGPU programs
SNIC Project: SNIC 2020/13-62
Project Type: SNIC Small Compute
Principal Investigator: Ahmed Rezine <ahmed.rezine@liu.se>
Affiliation: Linköpings universitet
Duration: 2020-08-10 – 2021-09-01
Classification: 10201
Homepage: http://ida.liu.se/~ahmre43
Keywords:

Abstract

Micro-architecture based timing side-channel attacks (e.g., recent Meltdown, Spectre and Foreshadow attacks) are difficult to uncover and defend against. These threats undermine fundamental constructions in modern computer systems, such as virtual memory and user/kernal spaces separation. We aim to develop a framework that will allow to systematically assess and validate arbitrary programs against realistic classes of timing-channel attacks for entire families of micro-architectural configurations. The vision is a symbolic framework that can reproduce current and future attacks and that formally verifies and rigorously quantifies capacities and weight-distributions of timing-channels attack-observations for entire families of micro-architectural configurations. In this project, we develop techniques to analyze, without executing, GPGPU programs and to quantify information leaked via memory access patterns: e.g., bank conflicts.