WHYjun / rt-kernel-test-analytics

Modify the Linux kernel to be able to read a new ELF section

PREEMPT-RT-LINUX

Kernel Modifications

We have modified the kernel to read section headers of ELF files and look for a specific header. If found, the kernel will attempt to run a module that alters program behavior based on the contents of the section header found in the elf file.

Please check kernel/README.md to see how our modification works with a new section and a new module.

Configuration

Please give permissions to run the bash script chmod u+x rt_kernel_analysis.sh.

Bash Script

Usage

Usage: ./rt_kernel_analysis.sh [-h (help)] [-n program name] [-t type (rt | non-rt)]

• h flag will echo the Usage string
• n flag will assign the program name to run the rt-kernel analysis
• t flag will assign the type of the kernel to be either rt or non-rt

What it will run

The following executions will be made:

• Error handle user input and validate if program is in rt-kernel
• ftrace graph
• ftrace functions
• strace table
• strace timestamp
• perf context switching
• perf cache profiling
• python data_reduction.py

The output of each trace function will be output to either non-rt/logfiles or rt/logfiles. The python script will read from those directories and store the created csv files into non-rt/python-output or rt/python-output.

TODO

1. Get kernel module working with modded kernel.
2. Possibly read data from the python-output file.
3. Incorporate assembly code analysis using perf annotate --stdio > perf_assembly.txt

Python Script

Adding log files

Please consider if you're kernel is running real-time or non-real-time and push into the respected folder called logfiles

Creating csv files

After inserting log files into the respected logfiles folder, please run this command:

for rt

python data_reduction.py -i rt/logfiles -s pi_stress -o rt/python-output

for non-rt

python data_reduction.py -i non-rt/logfiles -s pi_stress -o non-rt/python-output

The output csv files will be created in the folder python-output

Future work

• Statistical analysis on data frames created
• Refactor into Object-Orientated Design

Possible issues

If these issues arise, please notify in github issues or on discord.

• Our strace output file is missing process id
• The timestamp log file may contain errors that the data_reduction.py does not recognize