Towards Remote Attestation of Microarchitectural Attacks: The Case of Rowhammer

Martin Herrmann, Oussama Draissi, Christian Niesler, Ahmad-Reza Sadeghi, Lucas Davi · 2026

Read Towards Remote Attestation of Microarchitectural Attacks: The Case of Rowhammer by Martin Herrmann, Oussama Draissi, Christian Niesler, Ahmad-Reza Sadeghi, Lucas Davi (2026) 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-2603-24172.txt.

Introduction

Herrmann and co-authors propose spotting Rowhammer attacks through remote attestation. Rowhammer is a hardware fault where repeated memory access disturbs nearby rows. Prevention tools often lag behind new attack tricks. Their protocol, HammerWatch, lets an outside verifier judge whether a system shows such disturbance. It uses error records from ECC memory and per-row activation counts. TPM-anchored hash chains protect the evidence from kernel-level attackers. On 20,000 test patterns, the verifier splits attack-like behavior from benign use.

Catalog record

CREATORMartin Herrmann, Oussama Draissi, Christian Niesler, Ahmad-Reza Sadeghi, Lucas Davi
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE52066
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2603.24172

Browse the full library · About the Tower of Babel

Full text

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

Towards Remote Attestation of Microarchitectural Attacks: The Case of Rowhammer Martin Herrmann, Oussama Draissi, Christian Niesler, Ahmad-Reza Sadeghi, Lucas Davi Abstract Microarchitectural vulnerabilities increasingly undermine the assumption that hardware can be treated as a reliable root of trust. Prevention mechanisms often lag behind evolving attack techniques, leaving deployed systems unable to assume continued trustworthiness. We propose a shift from prevention to detection through microarchitectural-aware remote attestation. As a first instantiation of this idea, we present HammerWatch , a Rowhammer-aware remote attestation protocol that enables an external verifier to assess whether a system exhibits hardware-induced disturbance behavior. HammerWatch leverages memory-level evidence available on commodity platforms, specifically Machine-Check Exceptions (MCEs) from ECC DRAM and counter-based indicators from Per-Row Activation Counting (PRAC), and protects these measurements against kernel-level adversaries using TPM-anchored hash chains. We implement HammerWatch on commodity hardware and evaluate it on 20 000 20\,000 simulated benign and malicious access patterns. Our results show that the verifier reliably distinguishes Rowhammer-like behavior from benign operation under conservative heuristics, demonstrating that detection-oriented attestation is feasible and can complement incomplete prevention mechanisms. 1 Introduction System security rests on hardware as a root of trust. Software-based protections, from operating system isolation to integrity measurement, assume that the underlying hardware reliably preserves memory integrity [ 4 ] . This assumption has guided decades of system design. Yet modern performance optimizations increasingly introduce microarchitectural vulnerabilities that undermine this foundation [ 19 , 41 , 50 ] . Features designed to improve latency, throughput, or density can unintentionally expose fault behavior or side effects that are exploitable below the software layer. Prominent examples include cache timing channels [ 69 ] and speculative execution attacks [ 35 , 40 ] , which exploit access times and speculative execution to leak sensitive information. Similarly, increasing DRAM density has introduced disturbance errors, called Rowhammer, that can be triggered through carefully crafted memory access patterns, to induce bit-flips enabling privilege escalation from unprivileged code to kernel-level execution [ 28 , 33 , 36 ] . These attacks do not merely leak information; they can directly corrupt memory, violating isolation guarantees assumed by all higher software layers. Because the corruption originates below the operating system, traditional integrity checks fail to detect or attribute the compromise and leave no reliable software-visible trace [ 27 ] . A decade of defensive research has shown that eliminating such vulnerabilities in deployed hardware is difficult. Mitigations are often reactive, vendor-specific, and incomplete. In the case of Rowhammer, countermeasures such as Target Row Refresh (TRR) and ECC memory have been overwhelmed by many-sided and Half-Double attacks [ 25 , 30 , 36 ] or through multi-bit faults [ 16 , 37 ] . Even DDR5’s Per-Row Activation Counting (PRAC) remains vulnerable [ 56 , 70 ] . Hardware-level fixes are costly and cannot protect already-deployed systems [ 26 , 47 ] . Attack techniques continue to outpace defenses [ 9 , 44 , 46 , 68 , 71 ] , leaving systems unable to assume continued trustworthiness after deployment. This persistent gap motivates a shift from preventing Rowhammer to detecting it. Once a system is compromised, local detection becomes unreliable because an attacker with kernel-level privileges can suppress or forge any software-based monitoring [ 63 ] . Trust decisions must therefore be made externally. Remote attestation provides this capability by enabling a verifier to assess system integrity through cryptographically protected evidence anchored in tamper-resistant hardware [ 17 ] . A Trusted Platform Module (TPM) can record measurements in a manner that remains unforgeable even to kernel-level adversaries [ 1 ] : Yet existing remote attestation protocols cannot detect such attacks. Software-based schemes such as IMA [ 60 ] and Keylime [ 59 ] measure boot-time and runtime software integrity. Control-flow attestation protocols including C-FLAT [ 2 ] and LO-FAT [ 20 ] verify execution paths but assume correct underlying memory. Hardware-anchored approaches using TPM [ 1 ] or TEEs [ 38 ] protect attestation evidence but still measure only software state. Collectively, these mechanisms target firmware, boot loaders, and OS components, not hardware fault behavior. To our knowledge, no attestation protocol systematically incorporates microarchitectural fault indicators into integrity evaluation. In this paper, we explore how remote attestation can be extended to account for microarchitectural attacks. We instantiate this concept using Rowhammer as a representative and well-studied hardware-level disturbance attack. Rowhammer enables corruption of security-critical data structures such as page tables through DRAM disturbance effects [ 33 , 63 , 36 ] . Its longevity, practical exploitability, and continued evolution make it an ideal case study for attesting to microarchitectural fault behavior. Our prototype implementation, called HammerWatch , constitutes the first Rowhammer-aware remote attestation protocol. HammerWatch integrates hardware-level evidence, specifically Machine-Check Exceptions (MCEs) from ECC memory and Alert Back-Off (ABO) events from PRAC, into a TPM-anchored attestation architecture. Our contributions are: • We design and implement a microarchitectural-aware remote attestation protocol, instantiated for Rowhammer, that captures memory-fault indicators from commodity hardware, protects measurements through TPM-anchored hash chains, and enables external verification of Rowhammer-related behavior. Our design ensures measurement integrity, authenticity, and freshness even against kernel-level privileged adversaries. Our implementation integrates MCE collection from the Linux EDAC subsystem and PRAC-based detection using Ramulator 2.0 [ 45 ] . • We analyze the detection coverage of HammerWatch against state-of-the-art Rowhammer variants, including Half-Double [ 36 ] , many-sided attacks [ 25 ] , and ECCploit-style exploitation [ 16 ] , as well as the related Rowpress vulnerability [ 44 ] . • We evaluate HammerWatch on 20\,000 benign and malicious access patterns and demonstrate that the verifier reliably distinguishes Rowhammer-like behavior from benign operation under conservative detection heuristics. 2 Background Rowhammer. DRAM stores data in capacitor-based cells organized into rows and columns, with cells requiring periodic refresh to retain their charge [ 33 ] . Rowhammer exploits the physical proximity of DRAM rows: repeatedly activating an aggressor row causes charge leakage in adjacent victim rows, inducing bit-flips if the charge drops below the sensing threshold before refresh [ 33 , 63 ] . These bit-flips can corrupt security-critical data structures such as page tables, enabling privilege escalation [ 27 ] . Susceptibility has worsened with increasing DRAM density due to reduced capacitor sizes and tighter row spacing [ 32 , 51 ] . Error Correcting Code (ECC) and Machine-Check Exceptions (MCE). ECC memory detects and corrects bit errors by storing redundant parity information alongside data [ 22 ] . Commodity ECC DRAM typically implements Single Error Correction, Double Error Detection, which can correct single-bit errors and detect two-bit errors per word [ 16 ] . When the memory controller detects an uncorrectable error or observes repeated correctable errors, it raises a MCE [ 16 ] . On Linux, MCEs are logged by the EDAC (Error Detection and Correction) kernel subsystem and include metadata such as error type and memory address. While ECC can be bypassed through carefully templated multi-bit-flips [ 16 ] , the templating process itself triggers detectable errors [ 21 ] . Per-Row Activation Counting (PRAC). PRAC is a Rowhammer mitigation mechanism standardized in DDR5 [ 31 ] . It maintains per-row activation counters within DRAM and triggers an Alert Back-Off (ABO) signal when a row exceeds a predefined activation

Read the complete text