Manuel:X86/Installation/Système d'amorçage

From Gentoo Wiki
Jump to:navigation Jump to:search
This page is a translated version of the page Handbook:X86/Installation/Bootloader and the translation is 100% complete.
Sommaire du manuel
Installation
‎À propos de l'installation
Choix du support
Configurer le réseau
Préparer les disques
Installer l'archive stage3
Installer le système de base
Configurer le noyau
Configurer le système
Installer les outils
Configurer le système d'amorçage
Finaliser
Utiliser Gentoo
Introduction à Portage
Les options de la variable USE
Les fonctionnalités de Portage
Scripts d'initialisation systèmes
Variables d'environnement
Utiliser Portage
Fichiers et répertoires
Les variables
Mélanger plusieurs branches logicielles
Outils supplémentaires
Dépôt personnalisé
Fonctionnalités avancées
Configuration du réseau
Bien démarrer
Configuration avancée
Les modules réseau
Sans fil
Ajouter des fonctionnalités
Gestion dynamique


Although installing for a 32-bit CPU, almost all x86 motherboards (starting from around 2006-2007 until the present) that were produced with support for UEFI have 64-bit UEFI firmware. Some users may notice "64" in the name of configuration settings and files in the coming sections below. This is expected in nearly every case.

There are a few very small exceptions to this 64-bit UEFI firmware rule, namely few early Apple Macs and some Intel Atom powered Dell tablet PCs had support for 32-bit UEFI firmware. The vast majority of readers will never encounter 32-bit UEFI firmware in the wild. For this reason 32-bit UEFI firmware is not covered in the x86 Handbook.


Selecting a boot loader

With the Linux kernel configured, system tools installed and configuration files edited, it is time to install the last important piece of a Linux installation: the boot loader.

The boot loader is responsible for firing up the Linux kernel upon boot - without it, the system would not know how to proceed when the power button has been pressed.

For x86, we document how to configure either GRUB or LILO for DOS/Legacy BIOS based systems, and GRUB or efibootmgr for UEFI systems.

In this section of the Handbook a delineation has been made between emerging the boot loader's package and installing a boot loader to a system disk. Here the term emerge will be used to ask Portage to make the software package available to the system. The term install will signify the boot loader copying files or physically modifying appropriate sections of the system's disk drive in order to render the boot loader activated and ready to operate on the next power cycle.

Default: GRUB

By default, the majority of Gentoo systems now rely upon GRUB (found in the sys-boot/grub package), which is the direct successor to GRUB Legacy. With no additional configuration, GRUB gladly supports older BIOS ("pc") systems. With a small amount of configuration, necessary before build time, GRUB can support more than a half a dozen additional platforms. For more information, consult the Prerequisites section of the GRUB article.

Emerge

When using an older BIOS system supporting only MBR partition tables, no additional configuration is needed in order to emerge GRUB:

root #emerge --ask --verbose sys-boot/grub

A note for UEFI users: running the above command will output the enabled GRUB_PLATFORMS values before emerging. When using UEFI capable systems, users will need to ensure GRUB_PLATFORMS="efi-64" is enabled (as it is the case by default). If that is not the case for the setup, GRUB_PLATFORMS="efi-64" will need to be added to the /etc/portage/make.conf file before emerging GRUB so that the package will be built with EFI functionality:

root #echo 'GRUB_PLATFORMS="efi-64"' >> /etc/portage/make.conf
root #emerge --ask sys-boot/grub

If GRUB was somehow emerged without enabling GRUB_PLATFORMS="efi-64", the line (as shown above) can be added to make.conf and then dependencies for the world package set can be re-calculated by passing the --update --newuse options to emerge:

root #emerge --ask --update --newuse --verbose sys-boot/grub

The GRUB software has now been merged onto the system, but it has not yet been installed as a secondary bootloader.

Install

Next, install the necessary GRUB files to the /boot/grub/ directory via the grub-install command. Presuming the first disk (the one where the system boots from) is /dev/sda, one of the following commands will do:

DOS/Legacy BIOS systems

For DOS/Legacy BIOS systems:

root #grub-install /dev/sda

UEFI systems

Important
Make sure the EFI system partition has been mounted before running grub-install. It is possible for grub-install to install the GRUB EFI file (grubx64.efi) into the wrong directory without providing any indication the wrong directory was used.

For UEFI systems:

root #grub-install --efi-directory=/efi
Installing for x86_64-efi platform.
Installation finished. No error reported.

Upon successful installation, the output should match the output of the previous command. If the output does not match exactly, then proceed to Debugging GRUB, otherwise jump to the Configure step.

Optional: Secure Boot
The information in this section has been added or modified recently and may not be as fully-tested as the rest of the documentation. If encountering documentation-related issues, please help get the issue fixed in the wiki if possible. Support is available for technical issues, if required. The wiki team is also able to help with fixing documentation.

The sys-boot/grub package does not recognize the secureboot USE flag, this is because the GRUB EFI executable is not installed during package installation, but is instead built and installed by the grub-install command via post-package installation. GRUB must therefore be manually signed after installation to the boot partition. Additionally, GRUB is a modular bootloader but loading modules is prohibited when Secure Boot is enabled. Therefore all necessary modules must be compiled into the GRUB EFI executable, below an example is shown including some basic modules, this may have to be adjusted for more advanced configurations:

root #emerge --noreplace sbsigntools
root #export GRUB_MODULES="all_video boot btrfs cat chain configfile echo efifwsetup efinet ext2 fat font gettext gfxmenu gfxterm gfxterm_background gzio halt help hfsplus iso9660 jpeg keystatus loadenv loopback linux ls lsefi lsefimmap lsefisystab lssal memdisk minicmd normal ntfs part_apple part_msdos part_gpt password_pbkdf2 png probe reboot regexp search search_fs_uuid search_fs_file search_label sleep smbios squash4 test true video xfs zfs zfscrypt zfsinfo"
root #grub-install --target=x86_64-efi --efi-directory=/efi --modules="${GRUB_MODULES}" --sbat /usr/share/grub/sbat.csv
root #sbsign /efi/EFI/gentoo/grubx64.efi --key /path/to/kernel_key.pem --cert /path/to/kernel_key.pem --out /efi/EFI/gentoo/grubx64.efi

To successfully boot with secure boot enabled the used certificate must either be accepted by the UEFI firmware, or shim must be used as a pre-loader. Shim is pre-signed with the third-party Microsoft Certificate, accepted by default by most UEFI motherboards.

How to configure the UEFI firmware to accept custom keys depends on the firmware vendor, which is beyond the scope of the handbook. Below is shown how to setup shim instead:

root #emerge sys-boot/shim sys-boot/mokutil sys-boot/efibootmgr
root #cp /usr/share/shim/BOOTX64.EFI /efi/EFI/gentoo/shimx64.efi
root #cp /usr/share/shim/mmx64.efi /efi/EFI/gentoo/mmx64.efi

Shims MOKlist requires keys in the DER format, since the OpenSSL key generated in the example here is in the PEM format, the key must be converted first:

root #openssl x509 -in /path/to/kernel_key.pem -inform PEM -out /path/to/kernel_key.der -outform DER
Remarque
The path used here must be the path to the pem file containing the certificate belonging to the generated key. In this example both key and certificate are in the same pem file.

Then the converted certificate can be imported into Shims MOKlist:

root #mokutil --import /path/to/kernel_key.der

And finally we register Shim with the UEFI firmware. In the following command, boot-disk and boot-partition-id must be replaced with the disk and partition identifier of the EFI system partition:

root #efibootmgr --create --disk /dev/boot-disk --part boot-partition-id --loader '\EFI\GRUB\shimx64.efi' --label 'shim' --unicode
Debugging GRUB

When debugging GRUB, there are a couple of quick fixes that may result in a bootable installation without having to reboot to a new live image environment.

In the event that "EFI variables are not supported on this system" is displayed somewhere in the output, it is likely the live image was not booted in EFI mode and is presently in Legacy BIOS boot mode. The solution is to try the removable GRUB step mentioned below. This will overwrite the executable EFI file located at /EFI/BOOT/BOOTX64.EFI. Upon rebooting in EFI mode, the motherboard firmware may execute this default boot entry and execute GRUB.

If grub-install returns an error that says "Could not prepare Boot variable: Read-only file system", and the live environment was correctly booted in UEFI mode, then it should be possible to remount the efivars special mount as read-write and then re-run the aforementioned grub-install command:

root #mount -o remount,rw,nosuid,nodev,noexec --types efivarfs efivarfs /sys/firmware/efi/efivars

This is caused by certain non-official Gentoo environments not mounting the special EFI filesystem by default. If the previous command does not run, then reboot using an official Gentoo live image environment in EFI mode.

Some motherboard manufacturers with poor UEFI implementations seem to only support the /EFI/BOOT directory location for the .EFI file in the EFI System Partition (ESP). The GRUB installer can create the .EFI file in this location automatically by appending the --removable option to the install command. Ensure the ESP has been mounted before running the following command; presuming it is mounted at /efi (as defined earlier), run:

root #grub-install --target=x86_64-efi --efi-directory=/efi --removable

This creates the 'default' directory defined by the UEFI specification, and then creates a file with the default name: bootx64.efi.

Configure

Next, generate the GRUB configuration based on the user configuration specified in the /etc/default/grub file and /etc/grub.d scripts. In most cases, no configuration is needed by users as GRUB will automatically detect which kernel to boot (the highest one available in /boot/) and what the root file system is. It is also possible to append kernel parameters in /etc/default/grub using the GRUB_CMDLINE_LINUX variable.

To generate the final GRUB configuration, run the grub-mkconfig command:

root #grub-mkconfig -o /boot/grub/grub.cfg
Generating grub.cfg ...
Found linux image: /boot/vmlinuz-5.15.52-gentoo
Found initrd image: /boot/initramfs-genkernel-x86-5.15.52-gentoo
done

The output of the command must mention that at least one Linux image is found, as those are needed to boot the system. If an initramfs is used or genkernel was used to build the kernel, the correct initrd image should be detected as well. If this is not the case, go to /boot/ and check the contents using the ls command. If the files are indeed missing, go back to the kernel configuration and installation instructions.

Conseil
The os-prober utility can be used in conjunction with GRUB to detect other operating systems from attached drives. Windows 7, 8.1, 10, and other distributions of Linux are detectable. Those desiring dual boot systems should emerge the sys-boot/os-prober package then re-run the grub-mkconfig command (as seen above). If detection problems are encountered be sure to read the GRUB article in its entirety before asking the Gentoo community for support.

Alternative 1: LILO

Emerge

LILO, the LInuxLOader, is the tried and true workhorse of Linux boot loaders. However, it lacks features when compared to GRUB. LILO is still used because, on some systems, GRUB does not work and LILO does. Of course, it is also used because some people know LILO and want to stick with it. Either way, Gentoo supports both bootloaders.

Installing LILO is a breeze; just use emerge.

root #emerge --ask sys-boot/lilo

Configure

To configure LILO, first create /etc/lilo.conf:

root #nano -w /etc/lilo.conf

In the configuration file, sections are used to refer to the bootable kernel. Make sure that the kernel files (with kernel version) and initramfs files are known, as they need to be referred to in this configuration file.

Remarque
If the root filesystem is JFS, add an append="ro" line after each boot item since JFS needs to replay its log before it allows read-write mounting.
FILE /etc/lilo.confExample LILO configuration
boot=/dev/sda             # Install LILO in the MBR
prompt                    # Give the user the chance to select another section
timeout=50                # Wait 5 (five) seconds before booting the default section
default=gentoo            # When the timeout has passed, boot the "gentoo" section
compact                   # This drastically reduces load time and keeps the map file smaller; may fail on some systems
  
image=/boot/vmlinuz-5.15.52-gentoo
  label=gentoo            # Name we give to this section
  read-only               # Start with a read-only root. Do not alter!
  root=/dev/sda3          # Location of the root filesystem
  
image=/boot/vmlinuz-5.15.52-gentoo
  label=gentoo.rescue     # Name we give to this section
  read-only               # Start with a read-only root. Do not alter!
  root=/dev/sda3         # Location of the root filesystem
  append="init=/bin/bb"   # Launch the Gentoo static rescue shell
  
# The next two lines are for dual booting with a Windows system.
# In this example, Windows is hosted on /dev/sda6.
other=/dev/sda6
  label=windows
Remarque
If a different partitioning scheme and/or kernel image is used, adjust accordingly.

If an initramfs is necessary, then change the configuration by referring to this initramfs file and telling the initramfs where the root device is located:

FILE /etc/lilo.confAdding initramfs information to a boot entry
image=/boot/vmlinuz-5.15.52-gentoo
  label=gentoo
  read-only
  append="root=/dev/sda3"
  initrd=/boot/initramfs-genkernel-x86-5.15.52-gentoo

If additional options need to be passed to the kernel, use an append statement. For instance, to add the video statement to enable framebuffer:

FILE /etc/lilo.confAdding video parameter to the boot options
image=/boot/vmlinuz-5.15.52-gentoo
  label=gentoo
  read-only
  root=/dev/sda3
  append="video=uvesafb:mtrr,ywrap,1024x768-32@85"

Users that used genkernel should know that their kernels use the same boot options as is used for the installation CD. For instance, if SCSI device support needs to be enabled, add doscsi as kernel option.

Now save the file and exit.

Install

To finish up, run the /sbin/lilo executable so LILO can apply the /etc/lilo.conf settings to the system (i.e. install itself on the disk). Keep in mind that /sbin/lilo must be executed each time a new kernel is installed or a change has been made to the lilo.conf file in order for the system to boot if the filename of the kernel has changed.

root #/sbin/lilo

Alternative 2: efibootmgr

Computer systems with UEFI-based firmware technically do not need secondary bootloaders (e.g. GRUB) in order to boot kernels. Secondary bootloaders exist to extend the functionality of UEFI firmware during the boot process. Using GRUB (see the prior section) is typically easier and more robust because it offers a more flexible approach for quickly modifying kernel parameters at boot time.

System administrators who desire to take a minimalist, although more rigid, approach to booting the system can avoid secondary bootloaders and boot the Linux kernel as an EFI stub.

The sys-boot/efibootmgr application is a tool to used interact with UEFI firmware - the system's primary bootloader. Normally this looks like adding or removing boot entries to the firmware's list of bootable entries. It can also update firmware settings so that the Linux kernels that were previously added as bootable entries can be executed with additional options. These interactions are performed through special data structures called EFI variables (hence the need for kernel support of EFI vars).

Ensure the EFI stub kernel article has been reviewed before continuing. The kernel must have specific options enabled to be directly bootable by the UEFI firmware. It may be necessary to recompile the kernel in order to build-in this support.

It is also a good idea to take a look at the efibootmgr article for additional information.

Remarque
To reiterate, efibootmgr is not a requirement to boot an UEFI system; it is merely necessary to add an entry for an EFI-stub kernel into the UEFI firmware. When built appropriately with EFI stub support, the Linux kernel itself can be booted directly. Additional kernel command-line options can be built-in to the Linux kernel (there is a kernel configuration option called CONFIG_CMDLINE. Similarly, support for initramfs can be 'built-in' to the kernel as well. These decisions must be made prior to kernel compilation, resulting in a more static boot configuration.

Install the efibootmgr software:

root #emerge --ask sys-boot/efibootmgr

Create the /efi location, and copy the kernel into this location, calling it bootx64.efi:

root #mkdir -p /efi
root #cp /boot/vmlinuz-* /efi/bootx64.efi
Remarque
The use of a backslash (\) as directory path separator is mandatory when using UEFI definitions.

Create boot entry called "gentoo" for the freshly compiled EFI stub kernel within the UEFI firmware:

root #efibootmgr --create --disk /dev/sda --part 1 --label "gentoo" --loader "\bootx64.efi"

If an initial RAM file system (initramfs) is used, then add the proper boot option to it:

root #efibootmgr --create --disk /dev/sda --part 1 --label "gentoo" --loader "\bootx64.efi" --unicode "initrd=\efi\initramfs-genkernel-x86-5.15.52-gentoo"

Note that the above command presumes an initramfs file was copied into the ESP inside the same directory as the bootx64.efi file.

With these changes done, when the system reboots, a boot entry called "gentoo" will be available.

Unified Kernel Image

If installkernel was configured to build and install unified kernel images. The unified kernel image should already be installed to the EFI/Linux directory on the EFI system partition, if this is not the case ensure the directory exists and then run the kernel installation again as described earlier in the handbook.

To add a direct boot entry for the installed unified kernel image:

root #efibootmgr --create --disk /dev/sda --part 1 --label "gentoo" --loader /efi/EFI/Linux/gentoo-x.y.z.efi

Alternative 3: Syslinux

Syslinux is yet another bootloader alternative for the x86 architecture. It supports MBR and, as of version 6.00, it supports EFI boot. PXE (network) boot and lesser-known options are also supported. Although Syslinux is a popular bootloader for many it is unsupported by the Handbook. Readers can find information on emerging and then installing this bootloader in the Syslinux article.

Alternative 4: systemd-boot

Another option is systemd-boot, which works on both OpenRC and systemd machines. It is a thin chainloader and works well with secure boot.

To install systemd-boot:

root #bootctl install
Important
Make sure the EFI system partition has been mounted before running bootctl install.

When using this bootloader, before rebooting, verify that a new bootable entry exists using:

root #bootctl list

If no new entry exists, ensure the sys-kernel/installkernel package has been installed with the systemd-boot USE flag enabled, and re-run the kernel installation.

For the distribution kernels:

root #emerge --ask --config sys-kernel/gentoo-kernel

For a manually configured and compiled kernel:

root #make install
Important
When installing kernels for systemd-boot, no root= kernel command line argument is added by default. On systemd systems that are using an initramfs users may rely instead on systemd-gpt-auto-generator to automatically find the root partition at boot. Otherwise users should manually specify the location of the root partition by setting root= in /etc/kernel/cmdline as well as any other kernel command line arguments that should be used. And then reinstalling the kernel as described above.

Optional: Secure Boot

When the secureboot USE flag is enabled, the systemd-boot EFI executable will be signed automatically. bootctl install will automatically install the signed version.

To successfully boot with secure boot enabled the used certificate must either be accepted by the UEFI firmware, or shim must be used as a pre-loader. Shim is pre-signed with the third-party Microsoft Certificate, accepted by default by most UEFI motherboards.

How to configure the UEFI firmware to accept custom keys depends on the firmware vendor, which is beyond the scope of the handbook. A postinst hook to automatically update systemd-boot and set it up with shim instead is provided on the systemd-boot wiki page. However the first time this should be done manually by following the steps below:

root #emerge --ask sys-boot/shim sys-boot/mokutil sys-boot/efibootmgr
root #cp /usr/share/shim/BOOTX64.EFI /efi/EFI/BOOT/BOOTX64.EFI
root #cp /usr/share/shim/mmx64.efi /efi/EFI/BOOT/mmx64.efi
root #cp /efi/EFI/systemd/systemd-bootx64.efi /efi/EFI/BOOT/grubx64.efi
Remarque
Shim is hardcoded to load grubx64.efi. As such the systemd-boot bootloader must be named as if it were GRUB.

Shims MOKlist requires keys in the DER format, since the OpenSSL key generated in the example here is in the PEM format, the key must be converted first:

root #openssl x509 -in /path/to/kernel_key.pem -inform PEM -out /path/to/kernel_key.der -outform DER
Remarque
The path used here must be the path to the pem file containing the certificate belonging to the generated key. In this example both key and certificate are in the same pem file.

Then the converted certificate can be imported into Shims MOKlist:

root #mokutil --import /path/to/kernel_key.der

And finally we register Shim with the UEFI firmware. In the following command, boot-disk and boot-partition-id must be replaced with the disk and partition identifier of the EFI system partition:

root #efibootmgr --create --disk /dev/boot-disk --part boot-partition-id --loader '\EFI\BOOT\BOOTX64.EFI' --label 'shim' --unicode




Redémarrer le système

Quittez l'environnement et démontez toutes les partitions montées. Ensuite, exécutez cette commande magique qui lance le vrai test final : reboot.

root #exit
cdimage ~#cd
cdimage ~#umount -l /mnt/gentoo/dev{/shm,/pts,}
cdimage ~#umount -R /mnt/gentoo
cdimage ~#reboot

N'oubliez pas de retirer le CD d'installation, sinon il pourrait être redémarré à la place du nouveau système Gentoo.

Une fois redémarré dans le nouvel environnement Gentoo, finir avec Finalisation de l’installation de Gentoo.