Unrelyzer _{A static analyzer for C-like unreliable programs}

The tool parses any language defined by the Mini-C grammar. Each arithmetic and boolean operation in the program is probabilistic/unreliable. This tool statically analyze the program using Abstract Interpretation. As byproduct it also generates Parse-Tree, Abstract-Syntax-Tree and Control-Flow-Graph for the program.

The type of programs it can analyse are similar to that of C. A specimen input file is shipped that states all the necessary constraints that any program parsable by this tool must follow. It also contains few sample code blocks.

Moreover, one file named hardware_specification.h is also there, that specifies correctness probabilities of the Arithmetic, Logical and Memory operations used in the language. The hardware manufacturer sets these probabilities.

Usage

Build

1. ./configure

• This step will check for existing flex and bison installations in the system. Otherwise compile from the source that is shipped with the project and install in a directory called binary.

2. make

• Builds the project inside src itself.

Run

1. cd src
2. ./unrelyzer --help

• Shows all the options and required arguments to run this tool.

Some of the primary options are listed below :

-a, --abstract                                  Analyse the program to produce abstract (range of) values

-c, --concrete                                  Analyse the program to produce concrete/discrete values

-p, --parse-tree                              Generate the Parse Tree in a file called parse.dot

-s, --ast                                            Generate the Abstract Syntax Tree in a file called ast.dot

-f, --cfg                                            Generate the Control Flow Graph in a file called cfg.dot

-i, --iteration[=COUNT]                Iterate through the Data Flow Equations maximum COUNT (default 20) times

-l, --column[=COUNT]                      Display output in COUNT (default 1) column(s) of variables

-o, --output=FILE                            Output to FILE instead of standard output

-v, -d, --verbose, --debug          Produce verbose output

The following option can only be used along with option -a (or --abstract):

-w, --widening                                  Use widening technique to accelerate convergence (reduced number of iterations through the Data Flow Equations) of the abstract result

For example, suppose the following C code is stored in a file named input:

int limit = 10;

int foo(int num) {
  while (num < limit)
    num = num + 3;
  return 0;
}

To analysis the function foo in concrete domain, the command would be ./unrelyzer -cf -l2 input foo num=0. It generates the following Control Flow Graph followed by the analysis result:

G₁ =  ⊥
G₃ =  sp(G₁, limit=10)
G₄ =  sp(G₃, num=0) ⊔  sp(G₅, num=(num+3))
G₅ =  sp(G₄, num<limit)
G₇ =  sp(G₄, num>=limit)

In the following result   m = -32768  & M = 32767

---------------------- Result of Concrete Analysis (65 clock ticks) ----------------------
Reached fixed point after 6 iterations

G₁  ::
    limit=<{a∈ ℤ | m ≤ a ≤ M}, 1>    num=<{a∈ ℤ | m ≤ a ≤ M}, 1>
G₃  ::
    limit=<{10}, 0.999999899998>     num=<{a∈ ℤ | m ≤ a ≤ M}, 1>
G₄  ::
    limit=<{10}, 0.999999899998>     num=<{0 3 6 9 12}, 0.999998499987>
G₅  ::
    limit=<{10}, 0.999999749998>     num=<{0 3 6 9}, 0.99999869999>
G₇  ::
    limit=<{10}, 0.999999749998>     num=<{12}, 0.999998349988>

Visualization of graphs

1. dot -Tpng [FILE].dot -o [FILE].png

• Converts the graphs from .dot format to .png image format.
• Graphviz is required for this step.

2. alternatively, an online service like http://graphviz.herokuapp.com, can also be used.

Clean

1. make clean

• To clean up everything.

Licensing

This code is released under GNU General Public License (Version 3), a copy of which is also shipped with the software.