Speculative disassembly of binary code
Speculative disassembly of binary code
复制标题
DOI:
10.1145/2968455.2968505
复制
发表时间:
2016-10
期刊:
影响因子:
--
通讯作者:
M. Ammar;Benguettache Khadra;D. Stoffel;W. Kunz
中科院分区:
文献类型:
--
作者:
M. Ammar;Benguettache Khadra;D. Stoffel;W. Kunz
Embedded software is rapidly increasing in complexity. To cope with this, developers rely on third-party IPs to accelerate product delivery. However, IP source code might not be available which limits verifiability. This creates a particular challenge especially in safety-critical applications, e.g., automotive. Static Binary Analysis (SBA) is a promising technique to address such a challenge by providing engineers with the ability to reason about the actual instructions executed for all possible inputs. Disassembly is the fundamental first step for any SBA where assembly instructions are recovered from binary code. Correct disassembly, however, is challenging since data is mixed with code in binaries. Moreover, variable-size ISA, e.g., Thumb and TriCore, allow a single byte sequence to have multiple valid interpretations. We introduce Spedi, an open source SPEculative DIsassembler for Thumb ISA. Spedi is based on a principled approach to disassembly where all possible basic blocks are speculatively recovered. Then, basic blocks are refined using conflict analyses to identify assembly instructions. Experiments using a wide range of benchmarks demonstrate that Spedi is both fast and effective. It outperforms IDA Pro, the de-facto industry standard disassembler, in terms of disassembly correctness. Spedi can also recover the majority of the call graph and switch table targets. It is resilient to obfuscation and doesn't use any symbol information which makes it a suitable front-end for a wide variety of SBA applications including security analysis.