This entry is purely a guideline on howto start this beautiful journey of Linux kernel programming. The learnig curve is pretty steep and choosing the right direction might be somewhat difficult.
When I finished the Eudyptula Challenge I got involved into a discussion with little who is the penguin that runs the challenge. He asked me if I would like to contribute and when I said yes he asked me if I have any idea what I would like to do in the kernel. I answered that given my current level of knowledge the best for me would be to work with one of the mantainers who could tell me what shall be done and later review my work so that I can learn and do something usefull at th esame time. Can you guess the answer? It was:
“No mantainer has that time, sorry”
In this particular moment I understood what should be my attitude like. That I should be proactive because noone will do any kind of work for me. This reminds me the quote from one of the movie about hackers; it fits:
This business is all about bits. It is up to us if we are one or zero.
I decided to show the “1” attitude and find a couple ways of contributing.
Improve the code quality
This is the easiest thing and it has a really good ratio of easy to learning. In general it boils down to either making sure that the code follows the coding style or eliminating the static code checker errors and warnings. Most of the code is in a pretty good shape due to the policy of not allowing patches which contain such a flaws. However there are a few weak spots where things should be improved. The biggest advantage of doing this kind of work is that it allows to learn a lot. First the coding style, second the various areas of the kernel code and last but not least it makes you better programmer.
Apply the coding style
There is a tool called checkpatch.pl which resides in the scripts directory of the kernel repository. This very clever script checks either the patches or files for issues in the coding style. Additionaly, if the input file is a patch, it verifies if it conforms to the patch format.
The usage is very simple as it checks the input files which by default are understood to be patches (the product of git format-patch). There is -f option which tells the script that the input is a regular file so it will not check the format of the patch.
% scripts/checkpatch.pl ../patches/*patch % scripts/checkpatch.pl -f drivers/tty/serial/jsm/*c
This is the sample output for one of the patches that I prepared. As you can see I was careless enough to sneak one coding style issue.
konrad in linux-mainline on jsm-work % scripts/checkpatch.pl ../patches/0001*patch WARNING: Unnecessary space before function pointer arguments #28: FILE: drivers/tty/serial/jsm/jsm.h:123: + void (*clear_break) (struct jsm_channel *ch); total: 0 errors, 1 warnings, 32 lines checked ../patches/0001-serial-jsm-Remove-unnecessary-parameter-from-clear_b.patch has style problems, please review. If any of these errors are false positives, please report them to the maintainer, see CHECKPATCH in MAINTAINERS. konrad in linux-mainline on jsm-work %
Static check the code
The next step, after improving the style, is to acctually fix the broken code. The Linux kernel build system uses the sparse static code analyzer which, when enabled, runs over every file that is compiled and if there is anything wrong with it reports it. This is simple as that.
The prerequisite however is to install it which shouldn’t be an issue for a modern Linux distribution. On Ubuntu it is enough to type in the terminal:
% sudo apt-get install sparse
moreover the releases can be downloaded from kernel.org and installed using the make & make install combo.
The usage is very simple as there is make option for it.
% make C=1
which will run static checker over evey file that is being compiled in.
This is the example output of running sparse over the dgap driver from the staging area. The code is in a pretty good state as there is only one warning reported :)
konrad in linux-mainline (jsm-work) % make C=1 M=drivers/staging/dgap LD drivers/staging/dgap/built-in.o CHECK drivers/staging/dgap/dgap.c drivers/staging/dgap/dgap.c:365:25: warning: too long initializer-string for array of char CC [M] drivers/staging/dgap/dgap.o Building modules, stage 2. MODPOST 1 modules CC drivers/staging/dgap/dgap.mod.o LD [M] drivers/staging/dgap/dgap.ko konrad in linux-mainline (jsm-work) %
Read the contents of the TODOs
Naturally, after a while of improving the code quality it is good to move on and seek for different kind of assignments. The source tree contains the number of TODO files which, to some extent, contain description of work that shall be done. This is a starting point and a source of inspiration for anyone willing to do something in the kernel without a real idea on where to start. This was my way of getting involved into the developent of staging/dgnc and later the jsm driver.
At the time of writing there are 53 TODO files in the kernel source tree. I figured out that the best way to browse them would be to merge the contents of each into one big text file for easier reading. I also wanted to know how long time ago the TODO file has been updated. This is check how accurate (or innacurate) the information might be. As a result I developed a terrible bash one liner which gave me what I wanted, here it is:
% echo "" > /tmp/todolist-kernel.txt; count=0; for entry in `find . -name "*TODO*"`; do echo $count". "$entry`git log --pretty=format:" Last edited %ar" $entry | head -1` >> /tmp/todolist-kernel.txt; echo "" >> /tmp/todolist-kernel.txt; sed 's/^/ /' $entry >> /tmp/todolist-kernel.txt; echo "" >> /tmp/todolist-kernel.txt; ((count=$count+1)); done
It takes a while to execute but at the end all the information is ready for checking out in the /tmp/todolist-kernel.txt file. It might be hard to browse as it contains around 1.2k lines however it gets better with Vim and the foldmethod set to indent. Nevertheless choosing the area of interest is a time consuming process so do not rush, read it carefully and make your choice.
I have been told that the information in the TODOs, at least in the staging area, shall be accurate. However for example for the dgnc driver it was not the case as some of the bullet points have already been addressed. I assume that this might be the case for a few other TODOs as well. In my case I found this to be good for me because I had a chance (been forced to) to learn the code and understand how it works to be able to sort out what is done and what is not. So in general this is a good experience unless the TODO is painfully old. In this case everything is possible including the driver being obsolete and/or abandoned for good.
As for me I focused on the driver for which I could understand the bullet points from the TODO. I did not really want to be stuck in some kind of really difficult development. I believe that this is very good approach for newcomers into the Linux kernel development and in general because having a chance to acctually accomplish something is very important for self confidence. So if you do not know much about Linux kernel programming I recommend my approach.
The drivers/staging area
I mentioned the staging area a few times so far. Now it is high time to elaborate more on what it is.
So basically the drivers/staging area is a home for not yet officially supported drivers. The whole area is supported by Greg KH, the Linux Driver Project is behind it and people involved communicate on the driverdev-devel mailing list.
The code in the staging does not meet the quality standards and the job is to make it good enough to be promoted as the ‘real’ kernel driver. This makes it a perfect place to start from especially when compibed with the information from TODOs.
It is important to know that there is a significant amount of people working in this area nowadays. So it is good to follow the mailing list just to get the gist on what people are working on. Would be unfortunate to learn, after sending the patch, that someone has made this change like a week ago or something.
Fix kernel bug
As any other software project the kernel has bugs. It can be either a direct crash or just a glitch reported in the bug tracking system. Regardless what it is fixing it is a great, challenging adventure as fixing bugs is more advanced than improving the code quality.
The kernel OOPS is a crash and it usually do not happen nevertheless once a while it can be seen. Debugging such an issue is an advanced stuff however it is a great learning experience. My best kernel patch so far is the one liner I implemented for the crash that I had a few months ago.
Kernel bugs are tracked using the Bugzilla. This is a good source of inspiration for the brave :)
As you see there is many ways of contributing and still a lot of work to be done. In short you can:
- start in the drivers/staging area
- improve the code quality
- find the inspiration in the TODOs
- fix an acctuall bug in the kernel
plus many more inspirations that you can find when you acctually start.