Skip to content

Usage

Build

  • Build: scons
  • Clean: scons -c

Running

Here we explain how to run the injector manually for each phase. Running the error injector using the script we provide:

$PIN_INJECTOR_DIR/bin/injector.sh <args> -- <application>

In general, running the error injector is broken into two steps: profiling and injection.

Profiling

First, a program is profiled to get information on the dynamic instruction stream.

$PIN_INJECTOR_DIR/bin/injector.sh -c <op_class> -- <application>

This generates a logfile (err_prof_inj.out) which provides information useful for injection. Mainly you should look at the global/local (image) filtered Op Count to select (possibly randomly) a dynamic instruction for injection. op_class is the class filter (see below).

Injection

To run an injection:

$PIN_INJECTOR_DIR/bin/injector.sh -c <op_class> -i <instruction_count> -e <error_model> -- <application>

This will possibly inject an error into the application specified. If there is no logfile at the end then the target instruction was not found/reached. Otherwise, if successful, there will be a logfile which describes information about the error.

Batch running

We provide a launcher script for launching error injectors in a templated manner. The launcher is located at $HAMARTIA_DIR/src/python/launcher. It allows user to launch experiments in an automatic and parallel manner. Injection result analysis can also be pipelined if needed.

$LAUNCHER_DIR/launcher.py -c <config_file> -d <output_dir> -t <timeout_in_sec>

See examples of config_file within $LAUNCHER_DIR/configs directory.

Command-Line Options

  • Injection Site Filter
    • Class filter (-c): filters the profiled/injected operation by:
      • Operation class (e.g. OPS_INT, OPS_FP), XED ICLASS, or OPCODE
    • Dynamic instruction (-i): dynamic instruction index to inject at (0 performs profiling)
    • Ignore images (-n): image names to ignore (while profiling and injecting)
    • Include images (-l): images to include in injection by index (0 = all, 1 = target program)
    • Regions (-r): only inject into designated code regions
    • Width (-w): width of operation to inject at
    • Injection point (-p): operand to inject (post = output operand (default); pre = input operand; any = both)
  • Error model (-e): specifies the error model to use for injection
  • Error config (-q): specifies the error configuration file
  • Detector model (-d): specifies the detector model to use for injection
  • Detector config (-a): specifies the detector configuration file
  • Output file (-o): name of the output file (default = err_prof_inj.out)
  • MPI Injection
    • MPI injection switch (-m): set to 1 to enable MPI injection
    • Target rank (-t): MPI process rank id to inject
  • CD Injection
    • CD injection switch (-cd): set to 1 to enable CD injection
    • CD injection mode (-cdmode): 0 = use dynamic instruction count, 1 = use CD name
    • CD to be injected (-cdname): valid only if cdmode is 1
  • PIN inherent options
    • Unique logfile (-unique_logfile): insert the PID into the log filename
    • Pin logfile (-logfile): filename of pin logfile
    • Help (-help): print out help