Skip to content
master
Go to file
Code

Latest commit

 

Git stats

Files

Permalink
Failed to load latest commit information.
Type
Name
Latest commit message
Commit time
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
qlz
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

README.md

Open source case studies for Frama-C

This repository is a collection of open source C codes to be used with Frama-C, in particular with the Eva (Evolved Value Analysis) plug-in.

Each directory contains open-source code that constitutes a "case study".

Requirements

  • GNU Make >= 4.0;
  • Frama-C 20 (Calcium);
  • The frama-c binary must be in the PATH.

Usage

  • cd to one of the case studies;

  • Run make to parse and run Eva on the predefined targets (you can run make help to get the list of base targets). Note that the default file used by GNU Make is GNUmakefile if it exists. We use this to avoid renaming the original Makefile, if any. It also means that, if you want to compile the code using its Makefile, you'll have to explicitly name it (make -f Makefile).

  • For each base target t, the following targets are generated:

    • t.parse: parse the sources;
    • t.eva: run Eva;
    • t.eva.gui: open the GUI.

    Each target depends on the previous one; note that t.parse.gui is also available (e.g. for inspecting the AST before the analysis).

Optional targets

  • For each base target t, the following optional targets are generated:

    • t.stats: print time/memory usage;
    • t.parse.loop and t.eva.loop: use the Loop Analysis plug-in to produce a file with slevel heuristics (running Eva may improve the result of Loop Analysis, so t.eva.loop should be more precise than t.parse.loop). After obtaining this initial set of parameters, consider saving it to a .slevel file and including it in the GNUmakefile. This way, you can improve the parameters for specific functions as you refine the analysis.

Remarks

  • The output of t.eva is verbose, but you can ignore it; the important information (warnings and alarms) can be inspected via the GUI;

  • The result of each analysis is stored in a directory containing the full logs and Frama-C save files; successive runs are copied into timestamped directories, to allow comparing them (e.g. via meld);

  • To try other parametrizations, simply edit variables CPPFLAGS/FCFLAGS/EVAFLAGS in GNUmakefile and re-run make.

Notes

Source code modifications

Only minor modifications were performed on each of these case studies:

  • File GNUmakefile is added to each case study, with Frama-C/Eva-specific rules for parsing and running the analysis;
  • Some case studies contain a .slevel file which is derived from the result obtained by the Loop Analysis plug-in;
  • When necessary, syntactic modifications were performed to ensure better C99-compliance and/or the inclusion of stubs to allow Frama-C to parse the files;
  • In some cases, an eva_main function was added to provide a better initial context for the analysis;
  • When recursive calls are present, the functions containing them need to be replaced with specifications;
  • Some ACSL annotations may have been added to the sources (to illustrate their usage, or to improve the analysis).

Rationale

  • The main objectives of these files are:

    1. Non-regression tests;

    2. Benchmarking;

    Therefore, some of the code bases are voluntarily parametrized with suboptimal parameters, for non-regression testing; alternatively, some code bases may be present several times, with different versions and/or parametrizations;

  • These case studies constitute work in progress and do not represent "finalized" case studies.

Contributions

If you know of other open source code bases where Frama-C/Eva produces interesting results, please contribute with pull requests including the sources and the GNUmakefile that you have devised to run Frama-C.

On the other hand, if you have some interesting open-source C software (ideally, C99-compatible) that you are unable to parse and/or run with Frama-C/Eva, consider creating an issue with the description of the problem you are facing (e.g. missing/incompatible declarations in the Frama-C libc, problems when preprocessing/parsing the software, constructs unsupported by Eva, etc). Ideally, create a (WIP) pull request with the sources in a new directory, ready to be prepared for the case study.

License

License files are kept in each directory where they were originally found, when available. We also summarize the license of each directory below.

  • 2048: MIT
  • chrony: GPL
  • debie1: distribution and use authorized by Patria Aviation Oy, Space Systems Finland Ltd. and Tidorum Ltd, see README.txt and terms_of_use-2014-05.pdf
  • gzip124: GPL
  • hiredis: Redis license (BSD-style), see COPYING
  • icpc: Unlicense
  • itc-benchmarks: BSD 2-clause, see COPYING
  • jsmn: MIT
  • khash: MIT
  • kilo: BSD 2-clause "Simplified" (see https://github.com/antirez/kilo/blob/master/LICENSE)
  • libmodbus: LGPL
  • libspng: BSD 2-clause, see LICENSE
  • mini-gmp: LGPL or GPL
  • monocypher: see README
  • papabench: GPL
  • polarssl: GPL
  • qlz: GPL
  • semver: MIT
  • solitaire: public domain (see solitaire.c)
  • tweetnacl-usable: public domain (see LICENSE.txt)
  • x509-parser : GPLv2 / BSD (see LICENSE)

About

Open-source case studies for Frama-C (mainly EVA/Value Analysis plug-in)

Resources

Releases

No releases published
You can’t perform that action at this time.