Binary Searching to Find the Culprit Change
It is usually useful to know when a bug was introduced into the kernel source. If you know that a bug occurred in version 2.6.10, but not 2.4.9, then you have a clear picture of the changes that occurred to cause the bug. The bug fix is often as simple as reverting or otherwise fixing the bad change.
Many times, however, you do not know what kernel version introduced the bug. You know that the bug is in the current kernel, but it seemed to have always have been in the current kernel! It takes some investigative work, but with a little effort, you can find the offending change. With the change in hand, the bug fix is usually near.
To start, you need a reliably reproducible problempreferably, a bug that you can verify immediately after boot. Next, you need a known-good kernel. You might already know this. For example, if you know a couple months back the kernel worked, grab a kernel from that period. If you are wrong, try an earlier release. It shouldn't be too hardunless the bug has existed foreverto find a kernel without the bug.
Next, you need a known-bad kernel. To make things easier, start with the earliest kernel you know to have the bug.
Now, you begin a binary search from the known-bad kernel down to the known-good kernel. Let's look at an example. Assume the latest known-good kernel is 2.4.11 and the earliest known-bad is 2.4.20. Start by picking a kernel in the middle, such as 2.4.15. Test 2.4.15 for the bug. If 2.4.15 works, then you know the problem began in a later kernel, so try a kernel in between 2.4.15 and 2.4.20say, 2.4.17. On the other hand, if 2.4.15 does not work, you know the problem is in an earlier kernel, so you might try 2.4.13. Rinse and repeat.
Eventually you should narrow the problem down to two kernelsone of which has the bug and one of which does not. You then have a clear picture of the changes that caused the bug.
This approach beats looking at every kernel!
|