TSL: A System for Generating Abstract Interpreters and its Application to Machine-Code Analysis

Abstract

This paper describes the design and implementation of a system, called TSL (for "Transformer Specification Language"), that provides a systematic solution to the problem of creating retargetable tools for analyzing machine code. TSL is a tool generator--i.e., a meta-tool--that automatically creates different abstract interpreters for machine-code instruction sets. The most challenging technical issue that we faced in designing TSL was how to automate the generation of the set of abstract transformers for a given abstract interpretaton of a given instruction set. From a description of the concrete operational semantics of an instruction set, together with the datatypes and operations that define an abstract domain, TSL automatically creates the set of abstract transformers for the instructions of the instruction set. TSL advances the state of the art in program analysis because it provides two dimensions of parameterizability: (i) a given analysis component can be retargeted to different instruction sets; (ii) multiple analysis components can be created automatically from a single specification of the concrete operational semantics of the language to be analyzed. TSL is an abstract-transformer-generator generator. The paper describes the principles behind TSL, and discusses how one uses TSL to develop different abstract interpreters.