What You Need to Start

So, you are ready to start bug hunting? It might be a long and frustrating journey. Some bugs have perplexed the entire kernel development community for months. Fortunately, for every one of these evil bugs, there are many simple bugs with an equally simple fix. With luck, all your bugs will remain simple and trivial. You will not know that, however, until you start investigating. For that, you need

• A bug. It might sound silly, but you need a well-defined and specific bug. It helps if it is reliably reproducible (at least for someone), but unfortunately bugs are not always well behaved or well defined.

• A kernel version on which the bug exists (presumably in the latest kernel, or who cares?). Even better is if you know the kernel version where the bug first appeared. You will learn how to figure that out if you do not know it.

• Some good luck, kernel hacking skill, or a combination thereof.

If you cannot reproduce the bug, many of the following approaches become worthless. It is rather crucial that you be able to duplicate the problem. If you cannot, fixing the bug is limited to conceptualizing the problem and finding a flaw in the code. This does often happen (yep, the Linux kernel developers are that good), but chances of success are obviously much more favorable if you can reproduce the problem.

It might seem strange that there would be a bug that someone cannot reproduce. In user-space programs, bugs are quite often a lot more straightforward. For example, doing foo makes me core dump. The kernel is an entirely different beast. The interactions between the kernel, user-space, and hardware can be quite delicate. Race conditions might rear their ugly heads only once in a million iterations of an algorithm. Poorly designed or even miscompiled code can result in acceptable performances on some systems, but terrible performances on others. It is very common for some specific configuration, on some random machine, under some odd workload, to trigger a bug otherwise unseen. The more information you have when tackling a bug, the better. Many times, if you can reliably reproduce the bug, you are more than halfway home.