*=*=*=* Project to study faults in Linux *=*=*=*
This project study Linux version based on linux-* tags available
in the git repository mentioned in the Herodotos configuration files
(e.g. study.hc.base).
* 1. Repository organisation *
SHORT explanation:
- To retrieve our results, checkout the 'new.hybrid' branch
- To run your own experiment and produce your own results,
checkout the 'master' branch.
LONG explanation:
The branch are incrementally defined. At each step, new results
are added to the tree.
------------------------------------- master
\ \ \
+-------+-------+------------------ collect
\ \ \
+-------+-------+-------------- orig.org
\ \ \
+-------+-------+--------- new.hybrid
The 'master' branch keeps track of the project definitions.
These definitions include which coccinelle files are applied and
which options are used.
The 'collect' branch only concerns the rec-faults directory.
It tracks the list of functions that will be used to look for faults.
The 'orig.org' branch tracks the results of Coccinelle for each
spatch file and each Linux version. The *.orig.org files are stored
under results/linux/linux-*/.
The 'new.hybrid' branch is almost identical. Each new.org file is a copy
of the corresponding edit.org file. Once inspected, the reports should no
longer have TODO remaining.
* 2. Structure of the project *
The project contains information to search for faults in Linux, import the
fault reports in a database, and produce some graphs.
The following directories contains Herodotos projects to search for the faults:
- faults: Coccinelle scripts used to directly search for faults in the Linux kernel
- notes: Coccinelle scripts used to record code site where a fault may appear. It is used to compute
the ratio of faults per code side.
- rec-faults/1st-collect: Coccinelle and shell scripts to collect function name of interest
- rec-faults/2nd-find-faults: Faults related to a wrong/bad usage of a function recorded in the 1st phase
The sql directory contains SQL and shell scripts to set up a database.
The shell scripts allow to fill the database with information about the versions,
the files, the functions.
The scripts allow to create the views needed to extract interpretable results.
* 3. Structure of the correlated files *
A correlated file lists (i.e. edit.org and new.org files) the unique
reports with a status annotation
as a Org headline. The status are the following:
- TODO: Not manually reviewed
- UNKNOWN: Not able to determine
- FP: False Positive report from coccinelle
- BUG: Fault report.
For each report, the list of Linux version affected is given
as subitems of the headline.
* 4. How to clean up some report entries *
psql# delete from reports using full_reports
where reports.report_id = full_reports.report_id
and full_reports.correlation_last_date >= '2011-10-24';
* 5. How to debug/trace the process *
============ debug.config =============
logger "" {
level = trace;
filter = all;
layout = default;
mode = direct;
output = file;
name = "full-trace.log";
}
======================================
Run the tool with BOLT_CONFIG=debug.config
set as an environment variable.
--
Nicolas Palix