Now is the time to see how well the feature-frozen kernel works on your hardware, with your load.
A lot of people ask me, “When do you think the 2.6 kernel will be released?” My response to that question usually is, “Well, how well is the 2.5 kernel working for you?” Inevitably, during the resulting conversation where I plead with the person to please at least run the kernel once on their hardware, they respond with one of the following reasons why they have not tried 2.5:
Too many drivers are broken.
Modules are broken.
Alan Cox says it's not ready.
It might eat my filesystem.
Here are my responses to these reasons:
“Too many drivers are broken.”
The reason why a reasonably large number of drivers do not even compile in the kernel tree today is no one has proven the driver is even needed. Usually drivers are broken due to API changes during the development cycle, when it is not a trivial change to fix. Examples of this problem were the large number of block layer changes and the removal of the cli() function.
The only way these drivers will ever get fixed is if people say they need them fixed, as they are still using that kind of hardware. If you run across such a driver and you have the hardware, post to the mailing lists that you really want this driver fixed and that you are willing to test any changes people make. The kernel-janitors' mailing list is a very good place to ask this, as a lot of people there are looking for small tasks like this to accomplish.
“Modules are broken.”
Modules have been working for quite a while now. Make sure you get the latest version of the module-init-tools package. If you are use an RPM-based distribution, get the .src.rpm file in that directory, rebuild it and install it. This file preserves backwards compatibility with 2.4. Debian users can apt-get module-init-tools, and all others can use the module-init-tools source tarball in that directory. (Make sure you read the documentation on how to install it properly.)
“Alan Cox says it's not ready.”
In this message, Alan Cox stated that a number of things were still wrong with the 2.5 kernel, things that needed to be fixed before it could be called 2.6.0-test. His complaints about the IDE and tty layers have basically been addressed since then. (He has been pushing the 2.4 IDE changes to Linus, and the tty layer has had a number of locking problems fixed.) The major objections of people wanting to use the 2.5 kernel are gone. His other points about drivers building and things generally working are going to be addressed only by being tested by a large number of people, running the kernel on a wide range of machines. So what are you waiting for?
“It might eat my filesystem.”
Sure, any kernel might do this; make sure you have backed up any data you really cannot live without. But this is true for any kernel release or operating system. Now, I don't recommend using the 2.5 kernel in a production environment yet. Others are already, read the latest newsletter from OSDL about how they are starting to do this for their datacenter. But, I will state that I run the 2.5 kernel on all of my machines except for my firewall, and it runs particularly better on my laptop than did the 2.4 kernel (due to the better ACPI support and scheduler changes.) I have never had a data loss due to the 2.5 kernel yet. (Knock on wood.)
So, there go all of the arguments. The only way kernel developers will know that the 2.5 kernel is working well enough to declare it a stable 2.6 kernel, is for people to actually run it on their machines. And for that we need you to actually run the kernel.
Remember, any bugs you might find can be filed in the new kernel Bugzilla, where it will be filed to the proper developer. It's a good alternative if you don't want to wade through the mess that the Linux-kernel mailing list can become at times.
Go forth to build and reboot!
Greg Kroah-Hartman is currently the Linux USB and PCI Hot Plug kernel maintainer. He works for IBM, doing various Linux kernel-related things.