Title: Securing Intel SGX against Side-channel Attacks via Load-time Synthesis

Ming-Wei Shih

Ph.D. Student

Information Security

School of Computer Science

Georgia Institute of Technology

Date: Sept 20 (Thursday)

Start Time: 2:00pm

Location: KACB 3126

Committee

———————

Dr. Taesoo Kim (Advisor, School of Computer Science, Georgia Institute of Technology)

Dr. Wenke Lee(School of Computer Science, Georgia Institute of Technology)

Dr. Marcus Pienado (Microsoft Research)

Abstract

———————

One real-world technique that has the potential to ensure

the security of remote execution in an untrusted end-point is

Intel Software Guard Extensions (SGX). SGX allows an application

to run in a hardware-assisted isolated execution environment,

or an enclave, that excludes even privileged software such as

an operating system. Unfortunately, recent studies have shown

that such isolation is vulnerable to various types of side-channel

attacks. Given the strong threat model (i.e., an attacker controls

privileged software), the side-channel attacks are uniquely powerful.

More specifically, the attacker can learn accurate, high-resolution

side-channel information of the protected application, which

undermines the security guarantees that SGX intends to provide.

Although researchers have proposed several countermeasures

against the side-channel attacks on SGX, these countermeasures

focus only on a limited number of side-channel types. To achieve a

complete defense, one idea is to statically compose multiple countermeasures

and to enforce them on an application before deploying it to the end-point.

However, some of the countermeasures may not work as expected because

the statically enforced countermeasures cannot be adaptive to the hardware

specification at the end-point. For example, a TSX-based countermeasure

may provide no protection if the end-point does not support Intel TSX.

To address the limitation of static composition, we propose an alternative

approach, load-time synthesis, which adaptively composes multiple

countermeasures according to the hardware specification at the end-point.

More specifically, an enclave starts by executing a loader program.

The loader program then detects the hardware specification, selectively

enables the countermeasures, and finally produces an application binary

that runs with the possibly highest level of protection at the end-point.