Research On Static Binary Translation For The MIPS Architecture

Security and privacy is considered as a significant requirement in embedded systems,especially considering that most of them are provided to system networks,private networks,or the Internet.However,the specialization of embedded system often comes with one or more inherent characteristics,which make security analysis and testing on embedded systems significantly stricter than on traditional commodity systems.One way to employ security analysis is to leverage hardware debugging interface.This way requires dedicated hardwares and extreme resource requirements,therefore it can not be widely used in off-the-shelf embedded devices.To address,another way is leveraging dynamic binary translation(e.g.,QEMU,Avatar)to simulate and execute the binary program on PC system.The main drawback of dynamic binary translation is the large overhead of runtime translation and runtime optimization.Furthermore,for short-running programs,especially for interactive applications that are common on mobile devices,start-up time and response time are critical to their performances.This thesis presents an framework,BabelFish,to implement static bianry tranlstion technology.We leverage a translator to lift MIPS binary to LLVM-IRs,and re-use LLVM backend to give more complex analysis and optimizations.Meantimes,we give approaches to indirect branch/calling,data relocation and library function calling.To provide a fine-translated code,a better strategy for translating the register operation is implemented.Finally,the prototype is evaluated from two aspects,translating efficiency and code quality.The results illustrate that without any source code information,we can get a translated x86 binary using our BabelFish,and such compiled binary also have the same performance and functionality with that compiled from source code.