Security Properties for Open-Source Hardware Designs

Jayden Rogers, Niyaz Shakeel, Divya Mankani, Samantha Espinosa, Cade Chabra, Kaki Ryan and Cynthia Sturton · 2024

Read Security Properties for Open-Source Hardware Designs by Jayden Rogers, Niyaz Shakeel, Divya Mankani, Samantha Espinosa, Cade Chabra, Kaki Ryan and Cynthia Sturton (2024) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2412-08769.txt.

Introduction

Jayden Rogers and six co-authors published security properties in twenty twenty-four for four open-source hardware designs. Hardware security researchers verify designs against known flaw patterns, yet while buggy designs and verification tools circulate openly, the formal properties that catch the flaws mostly do not. That gap makes prior results hard to reproduce and slows the field.

The team wrote SystemVerilog Assertions for the OR1200 processor, and for the intentionally buggy competition systems PULPissimo, CVA6, and OpenPiton from three years of Hack@DAC. Properties are organized by design and tagged with the flaw details and the relevant weakness category. A methodology section explains how they craft such properties, encouraging others to publish theirs and build a shared library.

Catalog record

CREATORJayden Rogers, Niyaz Shakeel, Divya Mankani, Samantha Espinosa, Cade Chabra, Kaki Ryan and Cynthia Sturton
YEAR2024
COLLECTIONPAPERS
FORMATTXT
FILE SIZE36 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2412-08769

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Security Properties for Open-Source Hardware Designs Jayden Rogers, Niyaz Shakeel, Divya Mankani, Samantha Espinosa, Cade Chabra, Kaki Ryan and Cynthia Sturton Abstract . The hardware security community relies on databases of known vulnerabilities and open-source designs to develop formal verification methods for identifying hardware security flaws. While there are plenty of open-source designs and verification tools, there is a gap in open-source properties addressing these flaws, making it difficult to reproduce prior work and slowing research. This paper aims to bridge that gap. We provide SystemVerilog Assertions for four common designs: OR1200, Hack@DAC 2018’s buggy PULPissimo SoC, Hack@DAC 2019’s CVA6, and Hack@DAC 2021’s buggy OpenPiton SoCs. The properties are organized by design and tagged with details about the security flaws and the implicated CWE. To encourage more property reporting, we describe the methodology we use when crafting properties. 1. Introduction “For better or worse, benchmarks shape a field”–David Patterson ( Patterson 2012 ) . The use of formal verification methods for the security validation of hardware designs is an active area of research. The research has led to the development of new tools ( Ryan and Sturton 2023 ; Meng et al. 2021 ; Antón et al. 2023 ; Müller et al. 2021 ) and uncovered new security bugs and vulnerabilities ( Zhang et al. 2018 ; Meng et al. 2021 ; Antón et al. 2023 ) . In support of these efforts, the hardware security community has developed and published a large number of open-source hardware designs for tool development (e.g., ( OR1200: OpenRISC 1200 Implementation [n. d.] ; Schiavone et al. 2018 ; Zaruba and Benini 2019 ) ), along with flawed designs for use as benchmarks in the evaluation of any new methodology (e.g., ( Hack@DAC [n. d.] ; Trust-Hub [n. d.] ) ). Unfortunately, the security properties needed for verification are not similarly available. Many of the new methods being developed are property-based, where the verification engine takes a (possibly buggy) hardware design and a set of properties as input and outputs any found counterexamples (CEX) to the properties. There is general consensus that developing properties is a laborious task ( Tyagi et al. 2022 ; Kastner et al. 2022 ) , with ongoing research into generating properties automatically ( Zhang et al. 2017 ; Deutschbein and Sturton 2020 ; Zhang and Sturton 2020 ) , most recently using large language models ( Paria et al. 2023 ; Dipu et al. 2023 ; Kande et al. 2023 ) . Still, there remains a lack of well-vetted, publicly available, formally written properties tied to a particular snapshot of a design (see Figure 1 ). This work is a step toward providing that resource. Our contribution is largely pragmatic. We write four sets of security properties for four buggy designs commonly used in the literature: the OR1200 OpenRISC CPU core ( OR1200: OpenRISC 1200 Implementation [n. d.] ) , the PULPissimo SoC with RISC-V core ( Schiavone et al. 2018 ) , and two versions of the OpenPiton SoC with the CVA6 RISC-V core ( Zaruba and Benini 2019 ; Balkind et al. 2016 ) . The properties are written to find known security bugs in each design. For OR1200, we create a snapshot of the design with a collection of known bugs taken from the literature inserted ( Zhang et al. 2018 ; Hicks et al. 2015 ; Ryan and Sturton 2023 ) ; for the PULPissimo and OpenPiton SoCs, we use the buggy designs used in the Hack@DAC competitions in 2018, 2019, and 2021 ( Hack@DAC 2018 Phase 2 Buggy SoC [n. d.] ; Hack@DAC 2019 Alpha Stage SoC [n. d.] ; Hack@DAC 2019 [n. d.] ; Hack@DAC 2021 SoC [n. d.] ) . The properties are written as SystemVerilog Assertions (SVA), the industry standard for trace property specification ( sys 2018 ) , and the security bugs they cover are categorized according to the CWE taxonomy ( MITRE [n. d.] ) . We provide this categorization for OR1200 and supplement the existing categorization for the Hack@DAC designs. We write 71 properties for the OR1200, successfully detecting all known bugs and offering additional security assurance based on requirements from prior work. For PULPissimo, we create 20 properties to address 31 known bugs, an improvement over prior efforts that reported (but did not publish) 15 properties. Finally, we design 11 and 20 properties for the 66 and 99 known bugs in the 2019 and 2021 OpenPiton SoC designs, respectively. We use JasperGold, the industry-standard formal verification tool from Cadence, to confirm that the properties find the known bugs. The properties, along with a snapshotted version of the buggy open-source design they target, are available as benchmarks for future evaluation of verification tools. Figure 1 . In the hardware verification community there is no shortage of open-source designs or verification tools, but the verification properties still remain scarce, hard to find, and hard to write. This work aims to bridge that gap through both the contribution of an open-source database of properties, a methodology for property writing , and a call for authors to start providing the properties they use in their work. To further motivate the need for open-source properties, we present a case study illustrating the challenges to reproducibility that arise when properties are missing from the public record. To encourage more property reporting, we describe the methodology we use to craft properties so that others can follow the same steps. Our hope is that together the properties and methodology will accelerate and improve future evaluation of formal verification tools. This paper makes three contributions. • Provide well-vetted properties written for four buggy designs commonly used in the literature along with their associated snapshotted designs: https://github.com/HWSec-UNC/verification-benchmarks • Demonstrate the importance of properties for reproducibility through a case study of the community-standard Hack@DAC 2018 buggy PULPissimo design. • Describe the methodology we use to craft properties to encourage others to contribute new properties. 2. Background We discuss related work and provide context for our work. 2.1. Scope Our focus is on research using formal verification methods for the security validation of hardware designs; functional validation is out of scope. The line between security properties and functional properties can be blurry, as security flaws often stem from functional bugs in the design. However, the resources we consider are designed specifically for security validation activities aimed at identifying vulnerabilities exploitable by an attacker. These resources may or may not be appropriate for functional validation activities. 2.2. Open-Source Resources We are contributing formally stated security properties used during the formal verification process. Here, we discuss the resources currently available. Trust-Hub . TrustHub is an NSF-funded website designed to support research and education in hardware security ( Trust-Hub [n. d.] ) . The website serves as a directory for various hardware security resources and also offers resources directly, including Trojan-infected design components, taxonomies, and datasets related to physical hardware security. Two resources found on the Trust-Hub site are the Security Property/Rule Database and the SOC Vulnerability Database. We discuss each here. The Security Property/Rule Database ( Farzana et al. 2019 ) contains Verilog and SystemVerilog modules implemented at the register transfer level (RTL) or gate level, along with a set of

Read the complete text