c11547a952
The "mysql" extension is gone, using "mysqli" now. Plus tiny syntax/operator precendence changes. Change-Id: Icad4329521a2a42d7cd73408588d210431ec04c6
VisualFAIL*
Guest-system setup
For analysis with VisualFAIL*, the guest-system ELF binary must be compiled with debugging information (gcc/g++/clang/clang++ compiler flag "-g"). Note that debugging information quality/accuracy usually decreases with higher optimization levels.
Database preparation
- import-trace -t yourtrace.tc -i MemoryImporter
- import-trace -t yourtrace.tc -i FullTraceImporter
- import-trace -t yourtrace.tc -i ElfImporter -e yourbinary.elf --objdump $(which objdump) --sources
Step 1 is the prerequisite to run the fault-injection campaign (you may use other importers as well, e.g., the RegisterImporter). Steps 2 and 3 are required for VisualFAIL* to work.
Setup
Copy CONFIGURATION.php.dist to CONFIGURATION.php, edit it, and add your MySQL database credentials and the result-table name (usually starting with "result...", "echo SHOW TABLES | mysql yourdatabase" on the command line should give you the correct table name).
(In a later version of VisualFAIL*, we will probably add automatic loading of ~/.my.cnf if available.)
Running and using VisualFAIL*
./StartVF.sh requires PHP 5.4.0 or newer and uses its simple built-in web server. You can connect to it by using http://localhost:1234 in a web browser on the same machine. (If you need to connect from another machine, manually run "php -S 0.0.0.0:1234 -t .")
Alternatively, VisualFAIL* can be set up using a "real" web server with a recent PHP version (5.x should suffice).
Once the web server runs, you can use VisualFAIL*:
-
Pick a coloring (currently, only "Right margin" really makes sense), a benchmark and a variant from the drop-down menus. Click "Analyze". Wait.
-
Enable the experiment result types you want to highlight by clicking them in the top row.
-
Scroll down the left column with the disassembled machine code. Instructions that activated faults in the FI experiments, and led to one of the enabled result types, are highlighted with red color. The stronger the coloring, the more experiments led to one of these failure modes. (Actually, not the raw number of actual experiments is used, but is weighted with the size of the corresponding def/use equivalence class.)
-
Click one of the highlighted instructions to display a popup with the actual result numbers, and the distribution among the different failure modes.
-
Pick a source-code file from the right-hand side pulldown to look into the instructions generated from that file. The right column now contains the source code from that file, interspersed with the disassembled machine code generated from it. These disassembled machine-code lines are, just like in the disassembly column, clickable to display details. (The source-code lines are currently not clickable, although the mouse cursor indicates otherwise.)