检索增强生成在嵌入式固件系统仿真中的应用方法  

Application of Retrieval-Augmented Generation in System Emulation of Embedded Firmware

作　　者：孙浩楠 井靖[1] 吕鑫 SUN Haonan;JING Jing;LYU Xin(Information Engineering University,Zhengzhou 450001,China)

机构地区：[1]信息工程大学,河南郑州450001

出　　处：《信息工程大学学报》2025年第1期113-119,共7页Journal of Information Engineering University

摘　　要：嵌入式固件的系统仿真环境构建是面向嵌入式设备固件动态分析的基础,针对当前固件系统仿真环境构建中外设模型代码生成人工成本高且保真度低等问题,提出一种基于检索增强生成的嵌入式固件系统仿真技术。首先,通过数据手册预处理提取仿真环境构建所需的内存映射表和外设寄存器字段文本描述等信息;其次,利用大语言模型的归纳总结能力生成文本描述概要,并将生成的概要文本嵌入后构建向量知识库;最后,通过多向量检索器查找知识库中与目标外设相关的上下文并自动生成外设模型代码,实现目标固件的仿真环境构建。实验结果表明,该方法能成功提取并生成内存映射表、外设寄存器名和外设行为等描述,并能够正确引导系统初始化、响应固件中的外设访问等。与基于符号执行的固件托管技术相比,生成的仿真环境能使固件通过更多的状态检查代码,具有更高的代码覆盖率。The system emulation environment for embedded firmware serves as the foundation for dynamic analysis of embedded devices.However,there are issues like high labor costs and low fidelity in peripheral model code generation within current system emulation environments.Therefore,an embedded firmware system emulation technology based on retrieval-augmented generation is proposed in this study.Through preprocessing of the data manual,the memory map table and text descriptions of peripheral register fields are extracted.Subsequently,a large language model is employed to generate summaries from raw data and construct a vector knowledge base using embedded texts.Utilizing a multi-vector retriever,the contexts related to the target peripheral are identified and peripheral model codes are automatically generated to establish the desired firmware emulation environment.Experimental results demonstrate the memory maps,peripheral register names,and descriptions of peripheral behavior from the data manual can be successfully extraced and generate.Using the method the emulation environment effectively guides.And the system initialization can be effutively and peripheral accesses in the firmware with accuracy by the emulation euvirenment.Compared to firmware rehosting based on symbolic execution,the generated emulation environment enables more state-checking code execution with higher code coverage.

关 键 词：固件仿真 全系统仿真 嵌入式安全 动态分析 检索增强生成 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]