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

Dang, Pucheng; Huang, Di; Li, Dong; Chen, Kang; Wen, Yuanbo; Guo, Qi; Hu, Xing

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

Pucheng Dang^1,2,3, Di Huang¹, Dong Li^1,3, Kang Chen⁴, Yuanbo Wen¹, Qi Guo¹, Xing Hu^1,3

¹State Key Lab of Processors, Institute of Computing Technology, Chinese Academy of Sciences
²University of Chinese Academy of Sciences
³Zhongguancun Laboratory
⁴Tsinghua University

NeurIPS 2025 Spotlight

Paper Code arXiv

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.

Challenges

Through analyzing LLM behavior and results, we identify key challenges hindering their success in out-of-tree kernel patch migration:

Challenge 1 of vanilla LLMs.

Challenge 2 of vanilla LLMs.

Challenge 3 of vanilla LLMs.

Challenge 4 of vanilla LLMs.

MigGPT

MigGPT works in two stages: identifying target code snippets in the new version and migrating the out-of-tree patch. MigGPT consists of three core modules: the Retrieval Augmentation Module (addressing Challenges 1 and 3), the Retrieval Alignment Module (addressing Challenge 2), and the Migration Enhancement Module (addressing Challenge 4). Each module uses a Code Fingerprint (CFP) structure to enhance LLM performance and migration accuracy.

Figure 5. Overview of the MigGPT system architecture.

Code Fingerprint (CFP)

The Code Fingerprint (CFP) is a lightweight sequential data structure for analyzing code snippets, offering three key advantages:

Effective processing of incomplete code snippets.
Preservation of critical information such as comments and inline assembly, vital for out-of-tree kernel patch migration.
A streamlined design that focuses on essential statements, **improving migration efficiency** and reducing overhead.

Figure 6. Compared to AST, CFP extracts key code structures, and its linear representation enables clearer localization of code modification points.

Figure 7. An example of step-by-step CFP generation.

Results

BibTeX

@article{dang2025miggpt,
  title={MigGPT: Harnessing Large Language Models for Automated Migration of Out-of-Tree Linux Kernel Patches Across Versions},
  author={Dang, Pucheng and Huang, Di and Li, Dong and Chen, Kang and Wen, Yuanbo and Guo, Qi and Hu, Xing},
  journal={arXiv preprint arXiv:2504.09474},
  year={2025}
}