Automatic Recovery of Fine-grained Compiler Artifacts at the Binary Level
summary
Tool reconstructs per-function compiler passes from binaries using a modified LLVM to generate ground truth, a lightgbm multi-label classifier with seven engineered static features and optional binary emulation for dynamic features, and evaluates on 552 binaries / 149,814 functions. Results show an average F1 of 92.1% across 83 passes and 89.8% for security-impacting passes; it finds real bugs such as dead-store elimination removing secret erasures in BusyBox and Apache utilities.
AI-written summary · about
chapters (13)
- 0:00Start
- 0:51Introduction
- 1:29Secure Coding Practice
- 2:43The Correctness-Security Gap
- 6:16Background: Optimization Passes
- 7:06Detecting Compiler Optimizations
- 8:07Compiler Configuration Identification
- 8:57Features
- 11:42Evaluating Compiler Config Identification
- 12:20Generating Pass Dataset
- 13:30Evaluating Compiler Pass Identification
- 18:14Case Studies
- 20:24Next Step
related talks
57mCode Genome - Fingerprinting Code to Build Trustworthy SBOMs
21mDemystify Today's Binary Disassembling and How Modern ABI Makes it Easier
23mAn In-Depth Analysis of Disassembly on Full-Scale x86/x64 Binaries
23mPrecise and Accurate Patch Presence Test for Binaries
15mNot So Fast: Understanding and Mitigating Negative Impacts of Compiler Optimizations on Code Reuse G
14mNeural Reverse Engineering of Stripped Binaries using Augmented Control Flow Graphs
45mUsing Machines to exploit Machines
35mRepurposed Purpose: Using git's DAG for Supply Chain Artifact Resolution