Research And Implementation Of GPU Infinite Loop Malicious Program Detection Technology Based On PTX Instructions

At present,GPGPU programming technology has been widely used in artificial intelligence,deep learning and other fields.The GPGPU programming framework represented by the CUDA framework launched by NVIDIA can apply the powerful parallel computing capabilities of GPUs to non-graphic tasks,and GPUs are gradually opening up.The application of computing power in many fields has also brought related security risks.However,the industry is still mainly concerned with how to tap the potential security issues of GPU,and there is less research on the protection of known issues.This paper proposes a detection and evaluation method for the infinite loop structure of the PTX instruction codes on the GPU side,and programmed the method.prototype tool to detect whether the GPU codes written based on the CUDA framework has an infinite loop structure,which will cause the risk of the operating system to lose response.In addition,in the research process,the problem of residual memory data on the GPU side is also concerned.This paper designs a detection method for API calling features to evaluate whether there are such problems in the codes.It is introduced as a preliminary research work for the detection of infinite loop structures and good results have been achieved.The tool prototypes of the two detection methods are implemented using the Pass module development capabilities provided by the LLVM compiler project.The method in this paper has been tested to have good accuracy and effectiveness,which can provide a basis for subsequent GPU code security research.