Skip to content

Overview

The Hamartia instruction-level Injector is a Pin-based injector which implements the Hamartia Error and Detection API. This allows any model to be run as long as it is implemented with the API. The features which are currently supported are:

  • Hamartia Error/Detector model execution (C++ or Python)
    • Simple error models (e.g., randomly flip bits)
    • Interface to RTL gate-level injection
  • Injection point logging
    • Values before and after injection
    • Context: functions and line number
    • Registers
    • Memory addresses
  • Flexible injection
    • Instruction filters (e.g., by type, width, etc.)
    • Inject intput/output operands of instruction
    • Inject in region(s) annotated in source code (e.g., specific functions, lines)
    • Inject a particular MPI rank
    • Inject into Containment Domain

Error Injection Flow

The injection process consists of two phases: profiling and injection. The profiling phase obtains the bound of possible injection sites and other information required for customized injection experiments. The injection phase injects error(s) into the program based on command-line inputs specified by user. Reported are the injection site context, target instruction operands before and after injection, and other statistics useful for analysis. We also provide a script that automates the flow and integrates these phases seamlessly.

flow