Improved stochastic control flow model for LLVM-based software reliability analysis
Recently we have proposed a new method for error propagation analysis of the safety-critical software using the transformation of the source code to the Dual-graph Error Propagation Model (DEPM) based on the Low-Level Virtual Machine (LLVM) compiler framework, that allows the automatic analysis of C-code or another LLVM supported front-end. The source code is compiled into the LLVM Intermediate Representation and instrumented in order to analyze control and data flow structures of the software and the control flow transition probabilities between the basic blocks. Based on this information a DEPM for further analysis is generated. The DEPM is a stochastic model that captures system properties relevant to error propagation processes such as control and data flow structures and reliability characteristics of single components, LLVM instructions in this particular case. The DEPM helps to estimate the impact of a fault in a particular instruction on the overall system reliability, e.g. to compute the mean number of erroneous values in a critical system output during given operation time. The feasibility of the method has been proven on several case studies and also reveals several limitations of the current control flow model.
This paper address the improvement of the control flow model using a new customizable heuristic method for the analysis of control flow sequences and their mapping into discrete-time Markov chain models. The method is designed in a way to keep a required tradeoff between the model size and precision.