MigGPT: Harnessing Large Language Models for Automated Migration of Out-of-Tree Linux Kernel Patches Across Versions

TL;DR

MigGPT introduces a two-stage framework for automating the migration of out-of-tree Linux kernel patches across kernel versions, anchored by a Code Fingerprint (CFP) representation and three specialized modules that enhance code retrieval, boundary alignment, and migration-point localization. A robust benchmark based on real-world patches demonstrates that CFP-guided MigGPT significantly outperforms vanilla LLMs, achieving an average migration completion of 74.07% and substantial gains in semantic and error-resilience metrics, while remaining time-efficient. The work provides strong empirical support for integrating deterministic code-structure signals with LLMs to tackle complex software maintenance tasks and outlines a scalable path for broader downstream patch backporting and driver-migration challenges. Overall, MigGPT contributes a practical, benchmark-backed framework that reduces manual effort and accelerates the evolution of patched kernel code across versions, with potential implications for automated maintenance in other large-scale codebases.

Abstract

Out-of-tree kernel patches are essential for adapting the Linux kernel to new hardware or enabling specific functionalities. Maintaining and updating these patches across different kernel versions demands significant effort from experienced engineers. Large language models (LLMs) have shown remarkable progress across various domains, suggesting their potential for automating out-of-tree kernel patch migration. However, our findings reveal that LLMs, while promising, struggle with incomplete code context understanding and inaccurate migration point identification. In this work, we propose MigGPT, a framework that employs a novel code fingerprint structure to retain code snippet information and incorporates three meticulously designed modules to improve the migration accuracy and efficiency of out-of-tree kernel patches. Furthermore, we establish a robust benchmark using real-world out-of-tree kernel patch projects to evaluate LLM capabilities. Evaluations show that MigGPT significantly outperforms the direct application of vanilla LLMs, achieving an average completion rate of 74.07 for migration tasks.

Figures (17)

• Figure 1: MigGPT can assist in automating the version migration and maintenance of out-of-tree kernel patches of the Linux kernel. This saves on expert labor costs and reduces the development cycle.
• Figure 2: Overview of MigGPT. MigGPT employs a code fingerprint (CFP) structure to retain code snippet information, enhanced by three modules to improve migration accuracy and efficiency. The migration process involves two steps: 1) locating the migration position in $\text{file}_\text{new}$ to find $s_{\text{new}}$, and 2) completing the migration to obtain $s^{\prime}_{\text{new}}$.
• Figure 3: A code snippet containing inline assembly statements and comment annotations.
• Figure 4: Compared to AST, CFP extracts key code structures, and its linear representation enables clearer localization of code modification points.
• Figure 5: The semantic match accuracy of the target code snippets retrieval task and the target code snippets migration task across various LLMs. "One-step" indicates the direct utilization of an LLM to complete the migration task in a single step.