Reverse Engineering DVFS Mechanisms

Dynamic Voltage and Frequency Scaling (DVFS) provides superlinear power savings to computer systems by lowering supply voltage and clock frequency when processors wait for other systems components such as memory and I/O, in addition to enforcing thermal chip constraints. More recent DVFS mechanisms are managed fully in hardware to reduce response times, but existing analysis specific to hardware-managed DVFS is lacking. In this study, we aim to reconstruct the DVFS mechanism implemented in modern hardware by constructing a custom high frequency sampling mechanism using a Linux kernel module that enables flexible and low overhead sampling of Model-Specific Registers (MSRs) derived from the power management mechanism. We study an extensive reverse engineering of Intel Core 3100 U's DVFS mechanism. Our study reveals two novel findings: (1) the mapping from abstract firmware standards to an Intel-specific implementation of fixed voltage-frequency states and (2) the function of undocumented bits in an MSR. We then demonstrate that we are able to perform a model stealing attack through a variety of machine-learned proxy models based on access to relevant hardware with <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$99 \% R^{2}$</tex> coefficient for regression models and up to 92 % F1 score for a classifier.

Authors

Institutions

Publication Details

Published
2025-05-05
DOI
https://doi.org/10.1109/host64725.2025.11050037
Primary Topic
Gear and Bearing Dynamics Analysis
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
article

Reverse Engineering DVFS Mechanisms

Ryan Piersma, Simha Sethumadhavan, Tanvir Ahmed Khan, Tawhid Bhuiyan
Gear and Bearing Dynamics Analysis
article

Reverse Engineering DVFS Mechanisms

Ryan Piersma, Simha Sethumadhavan, Tanvir Ahmed Khan, Tawhid Bhuiyan
article en

Abstract

Dynamic Voltage and Frequency Scaling (DVFS) provides superlinear power savings to computer systems by lowering supply voltage and clock frequency when processors wait for other systems components such as memory and I/O, in addition to enforcing thermal chip constraints. More recent DVFS mechanisms are managed fully in hardware to reduce response times, but existing analysis specific to hardware-managed DVFS is lacking. In this study, we aim to reconstruct the DVFS mechanism implemented in modern hardware by constructing a custom high frequency sampling mechanism using a Linux kernel module that enables flexible and low overhead sampling of Model-Specific Registers (MSRs) derived from the power management mechanism. We study an extensive reverse engineering of Intel Core 3100 U's DVFS mechanism. Our study reveals two novel findings: (1) the mapping from abstract firmware standards to an Intel-specific implementation of fixed voltage-frequency states and (2) the function of undocumented bits in an MSR. We then demonstrate that we are able to perform a model stealing attack through a variety of machine-learned proxy models based on access to relevant hardware with <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$99 \% R^{2}$</tex> coefficient for regression models and up to 92 % F1 score for a classifier.

Columbia University (US)
Openalex Percentile: Top 84%
Gear and Bearing Dynamics Analysis
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.