7.4. Building a New Kernel
Rebuilding the kernel sounds like a pastime for hackers, but it is an important skill for any system administrator. First, you should rebuild the kernel on your system to eliminate the device drivers you don't need. This reduces the amount of memory used by the kernel itself, as described in Section 6.2. The kernel is always present in memory, and the memory it uses cannot be reclaimed for use by programs if necessary.
You also need to occasionally upgrade your kernel to a newer version. As with any piece of your system, if you know of important bug fixes or new features in a kernel release, you may want to upgrade to pick them up. Those people who are actively developing kernel code will also need to keep their kernel up-to-date in case changes are made to the code they are working on. Sometimes, it is necessary to upgrade your kernel to use a new version of the compiler or libraries. Some applications (such as the X Window System) require a certain kernel version to run.
rutabaga% uname -a Linux owl 2.4.19-64GB-SMP #2 SMP Fri Aug 9 21:46:03 CEST 2002 i686 unknown
Here, we see a machine running Version 2.4.19 of the kernel (configured for a machine with more than one processor [SMP] and a maximum of 64 GB RAM), which was last compiled on August 9. We see other information as well, such as the hostname of the machine, the number of times this kernel has been compiled (two), and the fact that the machine is a Pentium Pro or better (as denoted by i686). The manual page for uname(1) can tell you more.
The Linux kernel is a many-tentacled beast. Many groups of people work on different pieces of it, and some parts of the code are a patchwork of ideas meeting different design goals. Overall, however, the kernel code is clean and uniform, and those interested in exploring its innards should have little trouble doing so. However, because of the great amount of development going on with the kernel, new releases are made very rapidly — sometimes daily! The chief reason for this is that nearly all device drivers are contained within the kernel code, and every time someone updates a driver, a new release is necessary. As the Linux community moves toward loadable device drivers, the maintainers of those drivers can release them independently of the main kernel, alleviating the necessity of such rapid updates.
Currently, Linus Torvalds maintains the "official" kernel release. Although the GPL allows anyone to modify and rerelease the kernel under the same copyright, Linus's maintenance of an "official" kernel is a helpful convention that keeps version numbers uniform and allows everyone to be on equal footing when talking about kernel revisions. In order for a bug fix or new feature to be included in the kernel, all one must do is send it to Linus (or whoever is in charge for the kernel series in question, Linus himself always maintains the most current kernel), who will usually incorporate the change as long as it doesn't break anything.
major is the major version number, which rarely changes, minor is the minor version number, which indicates the current "strain" of the kernel release, and patchlevel is the number of the patch to the current kernel version. Some examples of kernel versions are 2.4.4, (patch level 4 of kernel Version 2.4), and 2.5.1 (patch level 1 of kernel Version 2.5).
By convention, even-numbered kernel versions (2.2, 2.4, and so on) are "stable" releases, patches that contain only bug fixes and no new features. Odd-numbered kernel versions (2.3, 2.5, and so on) are "development" releases, patches that contain whatever new code developers wish to add and bug fixes for that code. When a development kernel matures to the point where it is stable enough for wide use, it is renamed with the next highest (even) minor version number, and the development cycle begins again.
For example, kernel Versions 2.2 and 2.3 were worked on concurrently. Patches made to 2.2 were bug fixes — meant only to correct problems in the existing code. Patches to 2.3 included bug fixes as well as a great deal of new code — new device drivers, new features, and so on. When kernel Version 2.3 was stable enough, it was renamed to 2.4; a copy was made and named Version 2.5. Development continued with Versions 2.4 and 2.5. 2.4 is the new "stable" kernel, while 2.5 is a development kernel for new features.
Actually, the first versions of the 2.4 kernel series were not as stable as the number implies, which is why many users stayed with the 2.2 series for a while. By now, the current 2.4 kernel can be considered very stable, though, if you don't use any 2.4 kernels before 2.4.16.
Note that this version-numbering convention applies only to Linus's official kernel release and only to kernel versions after 1.0. Prior to 1.0 (this is now ancient history), there was only one "current" kernel version and patches were consistently made to it. The kernel development community has found that having two concurrent kernel versions allows those who want to experiment to use the development kernel, and those who need a reliable platform to stick with the stable kernel. In this way, if the development kernel is seriously broken by new code, it shouldn't affect those who are running the newest stable kernel. The general rule is that you should use development kernels if you want to be on the leading edge of new features and are willing to risk problems with your system. Use the development kernels at your own risk.
If you are interested in how the existing kernel versions have evolved, check out http://www.kernel.org.
On your system, the kernel sources most probably live in /usr/src/linux (unless you use the Debian distribution, where you can find the kernel sources in /usr/src/kernel-source-versionsnumber). If you are going to rebuild your kernel only from the current sources, you don't need to obtain any files or apply any patches. If you wish to upgrade your kernel to a new version, you need to follow the instructions in the following section.
7.4.1. Obtaining Kernel Sources
The official kernel is released as a gzipped tar file, containing the sources along with a series of patch files — one per patch level. The tar file contains the source for the unpatched revision; for example, there is a tar file containing the sources for kernel Version 2.4 with no patches applied. Each subsequent patch level is released as a patch file (produced using diff), which can be applied using the patch program. In Section 14.2.8 in Chapter 14, we describe the use of patch in detail.
Let's say you're upgrading to kernel Version 2.4 patch level 4. You'll need the sources for 2.4 (the file might be named v2.4.0.tar.gz) and the patches for patch levels 1 through 4. These files would be named patch1, patch2, and so forth. (You need all the patch files up to the version to which you're upgrading. Usually, these patch files are rather small, and are gzipped on the archive sites.) All these files can be found in the kernel directory of the Linux FTP archive sites; for example, on ftp://ftp.kernel.org, the directory containing the 2.4 sources and patches is /pub/linux/kernel/v2.4. You will find the kernel sources here as tar archives, compressed with both gzip and bzip2.
If you are already at some patch level of the kernel (such as 2.4 patch level 2) and want to upgrade to a newer patch level, you can simply apply the patches from the version you have up to the version to which you'd like to upgrade. If you're upgrading from, say, 2.4 patch level 2 to 2.4 patch level 4, you need the patch files for 2.4.3 and 2.4.4.
22.214.171.124. Unpacking the sources
First, you need to unpack the source tar file from /usr/src. You do this with commands such as:
rutabaga# cd /usr/src rutabaga# mv linux linux.old rutabaga# tar xzf v2.4.0.tar.gz
This saves your old kernel source tree as /usr/src/linux.old and creates /usr/src/linux containing the new sources. Note that the tar file containing the sources includes the linux subdirectory.
You should keep your current kernel sources in the directory /usr/src/linux because there are two symbolic links — /usr/include/linux and /usr/include/asm — that point into the current kernel source tree to provide certain header files when compiling programs. (You should always have your kernel sources available so that programs using these include files can be compiled.) If you want to keep several kernel source trees around, be sure that /usr/src/linux points to the most recent one.
126.96.36.199. Applying patches
If you are applying any patch files, you use the patch program. Let's say that you have the files patch1.gz through patch4.gz, which are gzipped. These patches should be applied from the kernel sources main directory. That doesn't mean the patch files themselves should be located there, but rather that patch should be executed from e.g. /usr/src/linux. For each patch file, use the command:
gunzip -c patchfile | patch -p1
from /usr/src. The -p1 option tells patch it shouldn't strip any part of the filenames contained within the patch file except for the first one.
You must apply each patch in numerical order by patch level. This is very important. Note that using a wildcard such as patch* will not work because the * wildcard uses ASCII order, not numeric order. (Otherwise, if you are applying a larger number of patches, patch1 might be followed by patch10 and patch11, as opposed to patch2 and patch3.) It is best to run the previous command for each patch file in succession, by hand. This way you can ensure you're doing things in the right order.
You shouldn't encounter problems when patching your source tree in this way unless you try to apply patches out of order or apply a patch more than once. Check the patch manual page if you do encounter trouble. If all else fails, remove the new kernel source tree and start over from the original tar file.
To double-check that the patches were applied successfully, use the commands:
find /usr/src/linux -follow -name "*.rej" -print find /usr/src/linux -follow -name "*#" -print
This lists any files that are "rejected" portions of the patch process. If any such files exist, they contain sections of the patch file that could not be applied for some reason. Look into these, and if there's any doubt, start over from scratch. You cannot expect your kernel to compile or work correctly if the patch process did not complete successfully and without rejections.
A handy script for patching the kernel is available and can be found in scripts/patch-kernel. But as always, you should know what you are doing before using automated tools, even more so when it comes to the very core of the operating system, the kernel.
7.4.2. Building the Kernel
All these commands are executed from /usr/src/linux, except for Step 5, which you can do anywhere.
A README is included in the kernel sources, which should be located at /usr/src/linux/README on your system. Read it. It contains up-to-date notes on kernel compilation, which may be more current than the information presented here. Be sure to follow the steps described there, using the descriptions given later in this section as a guide.
The first step is to run make config. This executes a script that asks you a set of yes/no questions about which drivers to include in the kernel. There are defaults for each question, but be careful: the defaults probably don't correspond to what you want. (When several options are available, the default will be shown as a capital letter, as in [Y/n].) Your answers to each question will become the default the next time you build the kernel from this source tree.
Simply answer each question, either by pressing Enter for the default, or pressing y or n (followed by Enter). Some questions don't have a yes/no answer; you may be asked to enter a number or some other value. A number of the configuration questions allow an answer of m in addition to y or n. This option allows the corresponding kernel feature to be compiled as a loadable kernel module, as opposed to building it into the kernel image itself. Loadable modules, covered in the following section, Section 7.5, allow portions of the kernel (such as device drivers) to be loaded and unloaded as needed on a running system. If you are unsure about an option, type ? at the prompt; for most options, a message will be shown that tells you more about the option.
Some people say that make config has so many options now that it is hardly feasible to run it by hand any longer, as you have to concentrate for a long time to press the right keys in response to the right questions. Therefore, people are moving to the alternatives described next.
An alternative to running make config is make xconfig, which compiles and runs an X-Window-based kernel configuration program. In order for this to work, you must have the X Window System running, have the appropriate X11 and Tcl/Tk libraries installed, and so forth. Instead of asking a series of questions, the X-based configuration utility allows you to use checkboxes to select which kernel options you want to enable. The system remembers your configuration options each time you run make config, so if you're adding or removing only a few features from your kernel, you need not reenter all the options.
Also available is make menuconfig, which uses the text-based curses library, providing a similar menu-based kernel configuration if you don't have X installed. make menuconfig and make xconfig are much more comfortable than make config, especially because you can go back to an option and change your mind up to the point where you save your configuration.
The following is part of a session with make config. When using make menuconfig or make xconfig, you will encounter the same options, only presented in a more user-friendly fashion (and we actually recommend the use of these tools if at all possible, as it is very easy to get confused by the myriad of configuration options):
rm -f include/asm ( cd include ; ln -sf asm-i386 asm) /bin/sh scripts/Configure arch/i386/config.in # # Using defaults found in .config # * * Code maturity level options * Prompt for development and/or incomplete code/drivers (CONFIG_EXPERIMENTAL) [Y/n/?] * * Loadable module support * Enable loadable module support (CONFIG_MODULES) [Y/n/?] Set version information on all module symbols (CONFIG_MODVERSIONS) [N/y/?] Kernel module loader (CONFIG_KMOD) [Y/n/?] * * Processor type and features * Processor family (386, 486, 586/K5/5x86/6x86/6x86MX, Pentium-Classic, ... defined CONFIG_MPENTIUMIII Toshiba Laptop support (CONFIG_TOSHIBA) [N/y/m/?] /dev/cpu/microcode - Intel IA32 CPU microcode support (CONFIG_MICROCODE) [M/n/y/?] /dev/cpu/*/msr - Model-specific register support (CONFIG_X86_MSR) [M/n/y/?] /dev/cpu/*/cpuid - CPU information support (CONFIG_X86_CPUID) [M/n/y/?] High Memory Support (off, 4GB, 64GB) [4GB] defined CONFIG_HIGHMEM4G Math emulation (CONFIG_MATH_EMULATION) [N/y/?] MTRR (Memory Type Range Register) support (CONFIG_MTRR) [Y/n/?] Symmetric multi-processing support (CONFIG_SMP) [Y/n/?] * * General setup * Networking support (CONFIG_NET) [Y/n/?] ...and so on... *** End of Linux kernel configuration. *** Check the top-level Makefile for additional configuration. *** Next, you may run 'make bzImage', 'make bzdisk', or 'make install'.
If you have gathered the information about your hardware when installing Linux, that information is probably sufficient to answer the configuration questions, most of which should be straightforward. If you don't recognize some feature, it's a specialized feature that you don't need. The following questions are found in the kernel configuration for Version 2.4.4. If you have applied other patches, additional questions might appear. The same is true for later versions of the kernel. Note that in the following list we don't show all the kernel configuration options; there are simply too many of them, and most are self-explanatory. We have highlighted only those that may require further explanation. Remember that if you're not sure how to answer a particular question, the default answer is often the best choice. When in doubt, it is also a good idea to type ? and check the help message.
It should be noted here that not all Linux device drivers are actually built into the kernel. Instead, some drivers are available only as loadable modules, distributed separately from the kernel sources. (As mentioned earlier, some drivers can be either built into the kernel or compiled as modules.) One notable kernel driver available only as a module is the "floppy tape" driver for QIC-117 tape drives that connect to the floppy controller.
If you can't find support for your favorite hardware device in the list presented by make config, it's quite possible that the driver is available as a module or a separate kernel patch. Scour the FTP sites and archive CD-ROMs if you can't find what you're looking for. In the next section, Section 7.5, kernel modules are covered in detail.
After running make config or its equivalent, you'll be asked to edit "the top-level Makefile," which means /usr/src/linux/Makefile. In most cases, it's not necessary to do this. If you wanted to alter some of the compilation options for the kernel, or change the default root device or SVGA mode, you could edit the makefile to accomplish this. Setting the root device and SVGA mode can easily be done by running rdev on a compiled kernel image, as we saw in Section 5.2.1 in Chapter 5.
If you wish to force a complete recompilation of the kernel, you should issue make clean at this point. This removes from this source tree all object files produced from a previous build. If you have never built the kernel from this tree, you're probably safe skipping this step (although it can't hurt to perform it). If you are tweaking minor parts of the kernel, you might want to avoid this step so that only those files that have changed will be recompiled. At any rate, running make clean simply ensures the entire kernel will be recompiled "from scratch," and if you're in any doubt, use this command to be on the safe side.
Now you're ready to compile the kernel. This is done with the command make bzImage. It is best to build your kernel on a lightly loaded system, with most of your memory free for the compilation. If other users are accessing the system, or if you're trying to run any large applications yourself (such as the X Window System, or another compilation), the build may slow to a crawl. The key here is memory. If a system is low on memory and starts swapping, it will be slow no matter how fast the processor is.
The kernel compilation can take anywhere from a few minutes to many hours, depending on your hardware. There is a great deal of code — well over 10 MB — in the entire kernel, so this should come as no surprise. Slower systems with 4 MB (or less) of RAM can expect to take several hours for a complete rebuild; faster machines with more memory can complete it in less than half an hour. Your mileage will most assuredly vary.
If any errors or warnings occur while compiling, you cannot expect the resulting kernel to work correctly; in most cases, the build will halt if an error occurs. Such errors can be the result of incorrectly applying patches, problems with the make config step, or actual bugs in the code. In the "stock" kernels, this latter case is rare, but is more common if you're working with development code or new drivers under testing. If you have any doubt, remove the kernel source tree altogether and start over.
When the compilation is complete, you will be left with the file bzImage in the directory /usr/src/linux/arch/i386/boot. (Of course, if you're attempting to build Linux on a platform other than the Intel x86, the kernel image will be found in the corresponding subdirectory under arch.) The kernel is so named because it is the executable image of the kernel, and it has been internally compressed using the bzip2 algorithm. When the kernel boots, it uncompresses itself into memory: don't attempt to use bzip2 or bunzip2 on bzImage yourself! The kernel requires much less disk space when compressed in this way, allowing kernel images to fit on a floppy. Earlier kernels supported both the gzip and the bzip2 compression algorithms, the former resulting in a file called zImage. Because bzImage gives better compression results, however, gzip should not be used, as the resulting kernels are usually too big to be installed these days.
If you pick too much kernel functionality, you can get a kernel too big error at the end of the kernel compilation. This happens rarely because you need only a very limited amount of hardware support for one machine, but it can happen. In this case, there is one way out: compile some kernel functionality as modules (see Section 7.5).
You should now run rdev on the new kernel image to verify that the root filesystem device, console SVGA mode, and other parameters have been set correctly. This is described in Section 5.2.1 in Chapter 5.
With your new kernel in hand, you're ready to configure it for booting. This involves either placing the kernel image on a boot floppy, or configuring LILO to boot the kernel from the hard drive. These topics are discussed in Section 5.2 in Chapter 5. To use the new kernel, configure it for booting in one of these ways, and reboot the system.
A warning: you should always keep a known good kernel available for booting. Either keep a previous backup kernel selectable from LILO or test new kernels using a floppy first. This will save you if you make a mistake such as omitting a crucial driver in your new kernel, making your system not bootable.
Copyright © 2003 O'Reilly & Associates. All rights reserved.