抱歉,此页面暂无中文版。

Use Git to Manage Kernel Sources on Gentoo

Gentoo offers its users a wide array of Linux kernel packages in the sys-kernel/* category, each of which may install the kernel in a different way. In particular, the sys-kernel/*-sources packages (such as sys-kernel/gentoo-sources, sys-kernel/vanilla-sources) install only the kernel’s source code and do nothing else. This is suitable for users who prefer to compile and install the kernel on their own but still would like to let the system package manager update the kernel sources automatically.

However, having a sys-kernel/* package is not required to install and manage the Linux kernel properly on Gentoo: even the kernel sources can be downloaded and updated of the user’s own accord without any package manager intervention. This article introduces one such way, which is to use Git to manage kernel sources (optionally with extra kernel patches as well) that are available from a Git repository.

What This Method Does and Does Not Help with

Before adopting this method of managing the kernel sources, it is better to know the benefits of this method and situations where it might fail to meet the anticipations.

This method is particularly effective than other methods to install the kernel sources in the following use cases:

  • When the Git commit history of the Linux project is being bisected to locate the exact commit that introduces a regression or another kind of bug. Reporters of helpful kernel bug reports that get resolved quickly and more effectively often use git bisect to pinpoint the problematic commit that triggers the bug being reported and include the commit’s SHA-1 hash in the report.

  • When one is trying to make and test small modifications to the kernel sources. The fact that the kernel sources are downloaded with Git means that those modifications can be tracked and managed with Git more systematically than if no version control system was used. Git commits can be created for the modifications, and each modification’s purpose can be documented in the commit message. The modifications can be reverted easily and be shared with others quickly by using git format-patch to export them as patches.

Despite these benefits, this method usually fails to do the following things:

  • Substantially decrease kernel build time. Under the impression that the reusable, intermediate object files (*.o files) generated by the compiler in a kernel build are kept in the same directory when the Git working tree in the directory is updated to a newer kernel release, it might seem that a lot of those object files could be reused, thus an updated kernel could take less time to build. The reality is, the size of changes in a bugfix kernel release (for example, changes from 5.16.10 to 5.16.11) is typically large enough to cause the bulk of those object files to be rebuilt, and not very much time is saved as a result.

    Nevertheless, in the occasional case where a bugfix kernel release contains a very small patch set (e.g. 5.16.9), using this method to manage the kernel sources still helps make the build significantly faster.

Download the Kernel Sources’ Git Repository

The instructions in this article will download a stable kernel or a longterm maintenance kernel from Git servers hosted on kernel.org for example. The clone URLs for the repository hosting these kernels are listed in the Clone section at the bottom of the repository’s homepage. This is the repository recommended to most users.

Advanced users may use sources of an alternative kernel too, as long as they understand the purpose of the alternative kernel and how it relates to the stable kernel. Normal users are advised against using any of them because they require more expert knowledge to use.

Some kernel repositories have multiple branches for different kernel versions. For instance, the repository for the stable and longterm kernels have branches like linux-5.16.y, linux-5.15.y, linux-5.10.y, and so on. In this case, a specific kernel version may be picked by selecting the corresponding branch with the --branch option for git clone.

It is suggested that the Git repository is cloned under /usr/src because the sys-kernel/*-sources packages also install the kernel sources to this directory. To allow eselect kernel to work with the kernel sources being cloned, the name of the directory that will hold the repository’s working tree should start with the linux- prefix. In particular, if the repository’s clone URL ends with linux or linux.git, then the directory’s name should be explicitly specified to be something different from linux. Otherwise, the repository would be cloned into /usr/src/linux in this case, and on Gentoo, /usr/src/linux is supposed to be a symbolic link pointing to the directory containing the running kernel’s sources.

To minimize the download size, adding --depth 1 to the options for git clone is recommended. This causes only the files necessary for the latest kernel release in the selected branch of the Git repository to be downloaded.

Accounting for these points, the commands for downloading the kernel sources’ Git repository may be invoked now. For example, these commands download the sources for the latest release in the Linux 5.15.y branch, which is a longterm kernel branch, to /usr/src/linux-5.15.y.

# cd /usr/src
# git clone --depth 1 https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git --branch linux-5.15.y linux-5.15.y

After the kernel sources have been downloaded, the /usr/src/linux symbolic link should be set to point to the directory containing them. There are several ways of doing this, and perhaps the easiest way is to use eselect kernel:

# eselect kernel list
Available kernel symlink targets:
  [1]   linux-5.15.y
# eselect kernel set linux-5.15.y
# ls -l /usr/src
total 4
lrwxrwxrwx 1 root root  12 Nov 14 18:00 linux -> linux-5.15.y
drwxr-xr-x 1 root root 504 Nov 14 17:59 linux-5.15.y

Now, the kernel sources are ready to be built in the same way as the sys-kernel/*-sources packages. Patches can also be applied, though it is strongly recommended that a Git commit is created for each patch applied, as this will make kernel sources updates easier later.

Update the Downloaded Git Repository

No matter which kernel is chosen, as long as it is still supported, it is usually subject to frequent updates, i.e. bugfix releases, that all users are expected to install in time, as sometimes they might contain fixes for critical security vulnerabilities or serious regressions. Users of a non-longterm stable kernel are also expected to update to a new mainline kernel in a few weeks after the latter’s release, which happens every 9-10 weeks, because the older kernel will reach end-of-life soon after that. Therefore, the kernel sources in the downloaded Git repository should be updated on a reasonable frequency. This section covers the update instructions for each type of new kernel release.

New Bugfix Release in the Same Branch

Changes in a new bugfix release can be downloaded and applied easily using just some Git commands in the Git repository’s working tree. Before any Git operations, the working directory should be changed to the working tree, which can be accessed via the /usr/src/linux symbolic link assuming that it is set up correctly:

# cd /usr/src/linux

First, use git fetch to check for any updates:

# git fetch
remote: Enumerating objects: 7766, done.
remote: Counting objects: 100% (7766/7766), done.
remote: Compressing objects: 100% (1032/1032), done.
remote: Total 6204 (delta 5216), reused 6153 (delta 5165), pack-reused 0
Receiving objects: 100% (6204/6204), 1.10 MiB | 6.38 MiB/s, done.
Resolving deltas: 100% (5216/5216), completed with 1534 local objects.
From https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux
   7cc36c3e1..3b17187f5  linux-5.15.y -> origin/linux-5.15.y
 * [new tag]             v5.15.3      -> v5.15.3

The output of git fetch in this example shows that Linux 5.15.3 is now available, and the origin/linux-5.15.y remote branch is also updated accordingly.

To apply the updates, rebase the local branch against the remote branch:

# git rebase origin/linux-5.15.y
Successfully rebased and updated refs/heads/linux-5.15.y.

A git rebase operation instead of git pull is strongly recommended because a rebase can cleanly reapply any kernel patches applied locally before in the Git commit history.

Finally, to verify that the kernel sources have been updated successfully, git show may be used to check if the latest commit is for the new bugfix release. Note, however, that if there are any locally-applied kernel patches, then git show might output the last commit for those patches instead of the commit bearing the latest bugfix release’s tag.

# git show
commit 3b17187f5ca1f5d0c641fdc90a6a7e38afdf8fae (HEAD -> linux-5.15.y, tag: v5.15.3, origin/linux-5.15.y)
Author: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Date:   Thu Nov 18 19:17:21 2021 +0100

    Linux 5.15.3

...

Once the kernel sources have been updated, a new kernel is ready to be built from it. Note that some steps are unnecessary in an upgrade to a new bugfix release of the same kernel:

  • The /usr/src/linux symbolic link does not need to be updated since the updated kernel sources are still stored in the same directory as before.

  • Usually, if there is already a configuration file for the kernel at /usr/src/linux/.config, then there is no need to make any modifications to it or replace it because it can be reused directly.

New Mainline Kernel in a Different Branch

When a new mainline kernel is to be installed, it is strongly recommended to download its sources to a new directory in /usr/src instead of keep reusing the existing copy of the Git repository. This can be done by executing the steps to download the kernel sources’ Git repository again but changing the branch name and directory name for the new kernel. For example, the following commands download the linux-5.16.y branch and sets /usr/src/linux to point to it:

# cd /usr/src
# git clone --depth 1 https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git --branch linux-5.16.y linux-5.16.y
# eselect kernel set linux-5.16.y

Do not forget to adjust the configuration file for the new kernel in this case. Also, any kernel patches applied locally before might need to be reapplied manually to the new kernel sources.

Downloading the sources for the new kernel to a different directory provides the following benefits:

  • If the new mainline kernel does not work as optimally as the previous one, it is possible to temporarily roll back to the previous kernel quickly.

  • The odds of accidentally applying a configuration file for a different kernel are smaller.

  • The names of directories under /usr/src that contain the kernel sources accurately and precisely indicate the kernel releases they are for. Otherwise, for example, if the linux-5.16.y branch were downloaded through the pre-existing Git repository stored at /usr/src/linux-5.15.y, then this single directory might contain sources for either of those branches, which could be confusing when the linux-5.16.y branch was checked out in /usr/src/linux-5.15.y. Renaming /usr/src/linux-5.15.y to a more general name like /usr/src/linux-stable could resolve the accuracy issue but would compromise the precision of information given by the directory name.

Though not recommended, it is still possible to download and check out the new mainline kernel’s branch in the existing Git repository. For example, the following commands check out the linux-5.16.y branch and attempt to minimize the download size for the operation:

# cd /usr/src/linux
# git fetch --depth 1 origin linux-5.16.y:linux-5.16.y
# git checkout linux-5.16.y

In this case, do not forget to update kernel configuration and reapply local kernel patches for the new kernel too.

General Kernel Management Tips

The tips in this section are applicable to both users who follow this article to use Git to manage kernel sources and those who use a sys-kernel/*-sources package.

Compile and Install a Kernel Without genkernel

Using genkernel is quite a popular way to build a custom kernel among users who are new to the process; however, it is introduced as an alternative way to build the kernel in the Gentoo Handbook. Building a kernel without genkernel is definitely possible, is not very hard, and is the primary method presented by the Handbook.

Before starting to compile and install a kernel, it is strongly recommended to install a package that provides the /sbin/installkernel utility, such as sys-kernel/installkernel-gentoo. This is because, if the installation scripts in the kernel sources cannot locate /sbin/installkernel, they will perform their own kernel installation logic, which is to install the kernel’s executable image to /boot/vmlinuz if it is compressed or /boot/vmlinux if it is not; however, some bootloaders might not support these installation paths. For example, GRUB 2 searches the kernel’s image using /boot/vmlinuz-* and /boot/vmlinux-* patterns; the extra hyphen causes GRUB 2 to disregard /boot/vmlinuz and /boot/vmlinux. Once /sbin/installkernel is installed, the kernel sources’ Makefiles will call it to handle kernel installation instead, and the /sbin/installkernel provided by sys-kernel/installkernel-gentoo installs the kernel’s image to /boot/vmlinuz-* or /boot/vmlinux-* (with the hyphen), so bootloaders like GRUB 2 can properly detect it.

# emerge --ask --noreplace sys-kernel/installkernel-gentoo

Assuming that a kernel configuration file has been created, compiling and installing the kernel without an initramfs is as simple as just running these commands as described in the Handbook:

# cd /usr/src/linux
# make -j "$(nproc)"
# make modules_install
# make install

In case an initramfs is needed, the Handbook recommends sys-kernel/dracut for building it. With Dracut, an initramfs can be built and installed without very much effort:

# cd /usr/src/linux
# dracut --force "" "$(cat include/config/kernel.release)"

If it is uncertain whether an initramfs is necessary, it is recommended to just build and install one to avoid potential boot failure. Whether or not the initramfs is redundant can be determined later after more knowledge is gained and more experiment is done. In the short term, getting a bootable system is usually more important than getting a minimal system.

After the kernel (and the initramfs, if it is needed) is installed, the configuration of any bootloader used to boot the system should be updated, so the new kernel can show up in the bootloader’s menu and thus be booted. The exact steps to update the configuration vary depending on which bootloader is used. For GRUB 2, the following command updates the bootloader configuration:

# grub-mkconfig -o /boot/grub/grub.cfg

A script can even be created to eliminate the need to type these commands every time when a new kernel is to be built:

#!/usr/bin/env bash

set -e

cd /usr/src/linux

# Compile and install kernel
make -j "$(nproc)"
make modules_install
make install

# Build and install initramfs
KERNEL_RELEASE="$(cat include/config/kernel.release)"
dracut --force "" "${KERNEL_RELEASE}"

# Update bootloader configuration
grub-mkconfig -o /boot/grub/grub.cfg

This leaves kernel configuration as the only challenging task in building a kernel manually.

Clean Up Files for an Old Kernel

app-admin/eclean-kernel is a tool that automates file clean-up for old kernels. Although an old kernel’s files can be removed manually, using a tool like eclean-kernel makes the process less error-prone.

To preserve only the newest kernel and let eclean-kernel clean up the rest kernels, the following command may be used:

# eclean-kernel -n 1