Linux kernel: Difference between revisions

From NixOS Wiki
Sdht0 (talk | contribs)
Sdht0 (talk | contribs)
→‎Configuration: Add info about SysRq
Line 86: Line 86:
}
}
</nowiki>}}
</nowiki>}}
=== Enable SysRq ===
The Linux kernel is known<ref>https://github.com/hakavlad/nohang?tab=readme-ov-file#what-is-the-problem</ref> to not handle out-of-memory situation properly and can freeze for a long time, often leaving no option but doing a hard reboot. The [https://en.wikipedia.org/wiki/Magic_SysRq_key SysRq shortcuts] can be used to trigger a more graceful reboot. However, most keys are disabled by default on NixOS. To enable:
{{file|/etc/nixos/configuration.nix|nix|<nowiki>
{
  boot.kernel.sysctl."kernel.sysrq" = 1;
}
</nowiki>}}
Useful shortcuts, triggered using {{ic|Alt+SysRq+<key>}}:
* {{ic|h}}: Print help to the system log.
* {{ic|f}}: Trigger the kernel oom killer.
* {{ic|s}}: Sync data to disk.
* {{ic|e}}: SIGTERM all processes except PID 0.
* {{ic|i}}: SIGKILL all processes except PID 0.
Check {{journalclt}} to see if you are triggering the shortcuts correctly, which might be different for your keyboard, as noted in the Wikipedia page.
Also see {{nixos:config|services.earlyoom.enable}} and {{nixos:config|systemd.oomd.enable}}.


=== Custom configuration ===
=== Custom configuration ===

Revision as of 17:17, 13 April 2024

By default, the latest LTS linux kernel is installed (Linux Kernel Version History).

Configuration

You can change the installed kernel using boot.kernelPackages. To use the latest release:

/etc/nixos/configuration.nix
{
  boot.kernelPackages = pkgs.linuxPackages_latest;
}

Then rebuild your system and reboot to use your new kernel:

$ sudo nixos-rebuild boot
$ sudo reboot

List available kernels

You can list available kernels using nix repl (previously nix-repl) by typing the package name and using the tab completion:

$ nix repl
nix-repl> :l <nixpkgs>
Added 12607 variables.

nix-repl> pkgs.linuxPackages
pkgs.linuxPackages                           pkgs.linuxPackages_latest
pkgs.linuxPackages-libre                     pkgs.linuxPackages_latest-libre
pkgs.linuxPackages-rt                        pkgs.linuxPackages_latest_hardened
pkgs.linuxPackages-rt_5_4                    pkgs.linuxPackages_latest_xen_dom0
pkgs.linuxPackages-rt_5_6                    pkgs.linuxPackages_latest_xen_dom0_hardened
pkgs.linuxPackages-rt_latest                 pkgs.linuxPackages_mptcp
pkgs.linuxPackagesFor                        pkgs.linuxPackages_rpi0
pkgs.linuxPackages_4_14                      pkgs.linuxPackages_rpi1
pkgs.linuxPackages_4_19                      pkgs.linuxPackages_rpi2
pkgs.linuxPackages_4_4                       pkgs.linuxPackages_rpi3
pkgs.linuxPackages_4_9                       pkgs.linuxPackages_rpi4
pkgs.linuxPackages_5_4                       pkgs.linuxPackages_testing
pkgs.linuxPackages_5_8                       pkgs.linuxPackages_testing_bcachefs
pkgs.linuxPackages_5_9                       pkgs.linuxPackages_testing_hardened
pkgs.linuxPackages_custom                    pkgs.linuxPackages_xen_dom0
pkgs.linuxPackages_custom_tinyconfig_kernel  pkgs.linuxPackages_xen_dom0_hardened
pkgs.linuxPackages_hardened                  pkgs.linuxPackages_zen
pkgs.linuxPackages_hardkernel_4_14
pkgs.linuxPackages_hardkernel_latest

Custom kernel modules

Note that if you deviate from the default kernel version, you should also take extra care that extra kernel modules must match the same version. The safest way to do this is to use config.boot.kernelPackages to select the correct module set:

/etc/nixos/configuration.nix
{ config, ... }:
{
  boot.extraModulePackages = with config.boot.kernelPackages; [ bbswitch ];
}

Customizing kernel module parameters

You can use boot.extraModprobeConfig, which is analogous to creating modprobe.conf files in /etc/modprobe.d/ in regular Linux distributions. This can be used for both built-in and loadable kernel modules.

/etc/nixos/configuration.nix
{
  boot.extraModprobeConfig = ''
    # example settings
    options yourmodulename optionA=valueA optionB=valueB # syntax
    options thinkpad_acpi  fan_control=1                 # example #1 kernel module parameter
    options usbcore        blinkenlights=1               # example #2 kernel module parameter
  '';
}

boot.kernelParams can be used to set additional kernel command line arguments at boot time. It can only be used for built-in modules. For example:

/etc/nixos/configuration.nix
{
  boot.kernelParams = [
    "quiet"
    "splash"
    "usbcore.blinkenlights=1"
  ];
}

Enable SysRq

The Linux kernel is known[1] to not handle out-of-memory situation properly and can freeze for a long time, often leaving no option but doing a hard reboot. The SysRq shortcuts can be used to trigger a more graceful reboot. However, most keys are disabled by default on NixOS. To enable:

/etc/nixos/configuration.nix
{
  boot.kernel.sysctl."kernel.sysrq" = 1;
}

Useful shortcuts, triggered using Alt+SysRq+<key>:

  • h: Print help to the system log.
  • f: Trigger the kernel oom killer.
  • s: Sync data to disk.
  • e: SIGTERM all processes except PID 0.
  • i: SIGKILL all processes except PID 0.

Check Template:Journalclt to see if you are triggering the shortcuts correctly, which might be different for your keyboard, as noted in the Wikipedia page.

Also see Template:Nixos:config and Template:Nixos:config.

Custom configuration

It is sometimes desirable to change the configuration of your kernel, while keeping the kernel version itself managed through Nixpkgs.

One way is to use boot.kernelPatches[2][3]. For example, boot.crashDump.enable is configured as shown below. Note that the CONFIG_ prefix is not used in the configuration names.

{
      boot.kernelPatches = [ {
        name = "crashdump-config";
        patch = null;
        extraConfig = ''
                CRASH_DUMP y
                DEBUG_INFO y
                PROC_VMCORE y
                LOCKUP_DETECTOR y
                HARDLOCKUP_DETECTOR y
              '';
        } ];
}

Another way is to create a overlay to configure either extraConfig or structuredExtraConfig. This allows for more flexibility, since you can basically override any available kernel option[4]. Using structuredExtraConfig is recommended as it checks if the configured option are correct. You may also want to set ignoreConfigErrors to avoid error: unused option during builds.

/etc/nixos/configuration.nix
{
  nixpkgs = {
    overlays = [
      (self: super: {
        linuxZenWMuQSS = pkgs.linuxPackagesFor (pkgs.linux_zen.kernel.override {
          structuredExtraConfig = with lib.kernel; {
            SCHED_MUQSS = yes;
          };
          ignoreConfigErrors = true;
        });
      })
    ];
  };
}

Pinning a kernel version

/etc/nixos/configuration.nix
{
  boot.kernelPackages = pkgs.linuxPackagesFor (pkgs.linux_4_19.override {
    argsOverride = rec {
      src = pkgs.fetchurl {
            url = "mirror://kernel/linux/kernel/v4.x/linux-${version}.tar.xz";
            sha256 = "0ibayrvrnw2lw7si78vdqnr20mm1d3z0g6a0ykndvgn5vdax5x9a";
      };
      version = "4.19.60";
      modDirVersion = "4.19.60";
      };
  });
}

Debugging a failed configuration

As dependencies between kernel configurations need to be addressed manually, use --keep-failed to inspect the intermediate nix config file after any build failures:

note: keeping build directory '/tmp/nix-build-linux-config-4.19.0-mptcp_v0.94.1.drv-0'

Embedded Linux Cross-compile

To configure and cross-compile Linux kernels for embedded development, often distributed separately instead of using the stock kernel, you can setup a development environment as shown below:

shell.nix
let
  pkgs = import &lt;nixpkgs&gt; { };
in
(pkgs.buildFHSUserEnv {
  name = "kernel-build-env";
  targetPkgs = pkgs: (with pkgs;
    [
      pkgconfig
      ncurses
      qt5.qtbase
      # new gcc usually causes issues with building kernel so use an old one
      pkgsCross.aarch64-multiplatform.gcc8Stdenv.cc
      (hiPrio gcc8)
    ]
    ++ pkgs.linux.nativeBuildInputs);
  runScript = pkgs.writeScript "init.sh" ''
    export ARCH=arm64
    export hardeningDisable=all
    export CROSS_COMPILE=aarch64-unknown-linux-gnu-
    export PKG_CONFIG_PATH="${pkgs.ncurses.dev}/lib/pkgconfig:${pkgs.qt5.qtbase.dev}/lib/pkgconfig"
    export QT_QPA_PLATFORM_PLUGIN_PATH="${pkgs.qt5.qtbase.bin}/lib/qt-${pkgs.qt5.qtbase.version}/plugins"
    export QT_QPA_PLATFORMTHEME=qt5ct
    exec bash
  '';
}).env

Clone the kernel sources, enter the environment using nix-shell, and then do development normally.

make menuconfig

It is (currently) not possible to run make menuconfig in the checked out linux kernel sources. This is because ncurses is not part of your working environment when you start it with nix-shell '<nixpkgs>' -A linuxPackages.kernel.

This nix-shell hack adds ncurses as a build dependency to the kernel:

$ nix-shell -E 'with import <nixpkgs> {}; linux.overrideAttrs (o: {nativeBuildInputs=o.nativeBuildInputs ++ [ pkg-config ncurses ];})'
[nix-shell] $ unpackPhase && cd linux-*
[nix-shell] $ patchPhase
[nix-shell] $ make menuconfig

(thanks to sphalerite)

make xconfig

Similarly to make menuconfig, you need to import qt in the environment:

$ nix-shell -E 'with import <nixpkgs> {}; linux.overrideAttrs (o: {nativeBuildInputs=o.nativeBuildInputs ++ [ pkg-config qt5.qtbase ];})'

If the source was unpacked and an initial config exists, you can run make xconfig KCONFIG_CONFIG=build/.config

Requesting a change in the default nixos kernel configuration

Please provide a comparison with other distributions' kernel: - arch: https://github.com/archlinux/svntogit-packages/blob/packages/linux/trunk/config - debian: https://salsa.debian.org/kernel-team/linux/blob/master/debian/config/config and the ARCH specific ones

Booting a kernel from a custom source

The following example shows how to configure NixOS to compile and boot a kernel from a custom source, and with custom configuration options.

{ pkgs, ... }:

{
  boot.kernelPackages = let
      linux_sgx_pkg = { fetchurl, buildLinux, ... } @ args:

        buildLinux (args // rec {
          version = "5.4.0-rc3";
          modDirVersion = version;

          src = fetchurl {
            url = "https://github.com/jsakkine-intel/linux-sgx/archive/v23.tar.gz";
            # After the first build attempt, look for "hash mismatch" and then 2 lines below at the "got:" line.
            # Use "sha256-....." value here.
            hash = "";
          };
          kernelPatches = [];

          extraConfig = ''
            INTEL_SGX y
          '';

          extraMeta.branch = "5.4";
        } // (args.argsOverride or {}));
      linux_sgx = pkgs.callPackage linux_sgx_pkg{};
    in 
      pkgs.recurseIntoAttrs (pkgs.linuxPackagesFor linux_sgx);
}

Out-of-tree kernel modules

Packaging out-of-tree kernel modules

There are a couple of steps that you will most likely need to do a couple of things. Here is an annotated example:

{ stdenv, lib, fetchFromGitHub, kernel, kmod }:

stdenv.mkDerivation rec {
  name = "v4l2loopback-dc-${version}-${kernel.version}";
  version = "1.6";

  src = fetchFromGitHub {
    owner = "aramg";
    repo = "droidcam";
    rev = "v${version}";
    sha256 = "1d9qpnmqa3pfwsrpjnxdz76ipk4w37bbxyrazchh4vslnfc886fx";
  };

  sourceRoot = "source/linux/v4l2loopback";
  hardeningDisable = [ "pic" "format" ];                                             # 1
  nativeBuildInputs = kernel.moduleBuildDependencies;                       # 2

  makeFlags = [
    "KERNELRELEASE=${kernel.modDirVersion}"                                 # 3
    "KERNEL_DIR=${kernel.dev}/lib/modules/${kernel.modDirVersion}/build"    # 4
    "INSTALL_MOD_PATH=$(out)"                                               # 5
  ];

  meta = with lib; {
    description = "A kernel module to create V4L2 loopback devices";
    homepage = "https://github.com/aramg/droidcam";
    license = licenses.gpl2;
    maintainers = [ maintainers.makefu ];
    platforms = platforms.linux;
  };
}

1. For kernel modules it is necessary to disable pic in compiler hardenings as the kernel need different compiler flags.

2. In addition to other dependencies in nativeBuildInputs you should include kernel.moduleBuildDependencies as this propagates additional libraries required during the build.

3. Some kernel modules try guess the kernel version based on the running kernel via uname. Usually they save this information in a makefile variable like KERNELRELEASE. If this is the case you can override the kernel version via makeFlags. The right kernel version string can be found in kernel.modDirVersion.

4. You need to find out how the build environment (Makefile in general) finds the kernel tree. This is sometimes KDIR and sometimes KERNEL_DIR.

5. Lastly it is required to give the kernel build system the right location where to install the kernel module. This is done by setting INSTALL_MOD_PATH to $out Otherwise an error like mkdir: cannot create directory '/lib': Permission denied is generated.

You can then call your program using something like let yourprogram = config.boot.kernelPackages.callPackage ./your-derivation.nix {}; in … (or if you want to compile it for the default kernel used by nix you can use pkgs.linuxPackages.callPackage but be aware that if you install it on a system running another kernel it will not work) and install the module in the kernel using boot.extraModulePackages = [ yourprogram ];.

Developing out-of-tree kernel modules

See also: NixOS Manual, 12.2. Developing kernel modules

If you work on an out-of-tree kernel module the workflow could look as follow:

#include <linux/module.h>
#define MODULE_NAME "hello"
static int __init hello_init(void)
{
    printk(KERN_INFO "hello world!\n");
    return 0;
}
static void __exit hello_cleanup(void) {
    printk(KERN_INFO "bye world!\n");
}
module_init(hello_init);
module_exit(hello_cleanup);
obj-m += hello.o
$ nix-shell '<nixpkgs>' -A linux.dev
$ make -C $(nix-build -E '(import <nixpkgs> {}).linux.dev' --no-out-link)/lib/modules/*/build M=$(pwd) modules
$ insmod ./hello.ko
$ dmesg | grep hello
[   82.027229] hello world!

If wishing to develop out-of-tree kernel modules for kernels aside from the default variant shipped under pkgs.linuxPackages, you can replace linux.dev with (for instance) linuxPackages_latest.kernel.dev.

If you want a local development environment and are only interested in Module.symvers, then you can do instead:

$ cp $(nix-build -E '(import <nixpkgs> {}).linux.dev' --no-out-link)/lib/modules/*/build/Module.symvers .

Loading out-of-tree kernel modules

As far as I understand, if you developed a kernel module, you should end up with having some .ko files inside a subfolder inside $out/lib/modules/${kernel.modDirVersion}. Now, if you want to make your module loadable inside the kernel by modprobe, you should do:

boot.extraModulePackages = [ yourmodulename ];

Then, the user can load it using:

$ sudo modprobe yourmodulename

or unload it using

$ sudo modprobe -r yourmodulename

However, if you want to autoload your module at startup in stage 2, you need to do:

boot.kernelModules = [ "yourmodulename" ];

and the module will be automatically loaded after a reboot. If you want instead to load it at stage 1 (before the root is even mounted), you need to add it to boot.initrd.availableKernelModules and boot.initrd.kernelModules.

Note that if you don't reboot, you can still load manually the module using modprobe yourmodulename>, and to automatically enable a module during configuration switch/reboot, you can put modprobe yourmodulename || true inside the script of a systemctl service (it is for example what does wireguard).

Finally, if you want to define some options by default (used when you load manually a module using modprobe, or when the system boots), you can specify them in:

boot.extraModprobeConfig = ''
  options yourmodulename optionA=valueA optionB=valueB
'';

If you have developed a Nix package for your module, and want to only add the module to your configuration.nix instead of complete rebuilding the system based on your local nixpkgs, you need to update boot.kernelPackages as well, so kernel and modules can match each other:

{
  boot = {
    kernelPackages = pkgs.wip.linuxPackages;
    extraModulePackages = with config.boot.kernelPackages; [ yourmodulename ];
  }
  nixpkgs.config = {
    # Allow unfree modules
    allowUnfree = true;
    packageOverrides = pkgs: {
      wip = import (fetchGit { url = "/home/user/nixpkgs"; shallow = true;}) {
        config = config.nixpkgs.config;
      };
    };
  };
};

Cross-compiling Linux from source

Save the following as shell.nix for your nix-shell:

{ pkgs ? import <nixpkgs> {} }:
let
  # more platforms are defined here: https://github.com/NixOS/nixpkgs/blob/master/lib/systems/examples.nix
  aarch64 = pkgs.pkgsCross.aarch64-multiplatform;
in
aarch64.linux.overrideAttrs (old: {
  # the override is optional if you need for example more build dependencies
  nativeBuildInputs = old.nativeBuildInputs ++ [ pkgs.gllvm ];
}

Then you can run the following commands (in the nix-shell $makeFlags will contain the necessary cross-compiling arguments)

$ nix-shell
# to configure
$ nix-shell> make $makeFlags defconfig -j $(nproc)
# to build
$ nix-shell> make $makeFlags -j $(nproc)

Compiling Linux with clang

Save this as shell.nix

{ pkgs ? import <nixpkgs> {} }:
pkgs.linux.override {
   # needs to be clang11Stdenv or newer
   stdenv = clang11Stdenv;
}

Then you can use:

$ nix-shell
# to configure
$ nix-shell> make $makeFlags defconfig -j $(nproc)
# to build
$ nix-shell> make $makeFlags -j $(nproc)

Overriding kernel packages

In order to override linuxPackages, use the extend attribute. Example:

linuxPackages.extend (self: super: {
  xpadneo = super.xpadneo.overrideAttrs (o: rec {
    src = pkgs.fetchFromGitHub {
      # Custom override goes here.
    };
  });
});

Here is a fully worked example to enable debugging for the zfsUnstable module:

 nixpkgs.overlays = [
    (self: super: {
      linuxPackages = super.linuxPackages.extend (lpself: lpsuper: {
        zfsUnstable = super.linuxPackages.zfsUnstable.overrideAttrs (oldAttrs: {
          configureFlags = oldAttrs.configureFlags ++ [ "--enable-debug" ];
        });
      });
    })
  ];

Patching a single In-tree kernel module

If you wish to patch a single kernel module, you can avoid rebuilding the entire linux kernel by packaging that module applying patches to it.

Packaging a single kernel module

Here is an example for how to package the amdgpu kernel module.

{ pkgs
, lib
, kernel ? pkgs.linuxPackages_latest.kernel
}:

pkgs.stdenv.mkDerivation {
  pname = "amdgpu-kernel-module";
  inherit (kernel) src version postPatch nativeBuildInputs;

  kernel_dev = kernel.dev;
  kernelVersion = kernel.modDirVersion;

  modulePath = "drivers/gpu/drm/amd/amdgpu";

  buildPhase = ''
    BUILT_KERNEL=$kernel_dev/lib/modules/$kernelVersion/build

    cp $BUILT_KERNEL/Module.symvers .
    cp $BUILT_KERNEL/.config        .
    cp $kernel_dev/vmlinux          .

    make "-j$NIX_BUILD_CORES" modules_prepare
    make "-j$NIX_BUILD_CORES" M=$modulePath modules
  '';

  installPhase = ''
    make \
      INSTALL_MOD_PATH="$out" \
      XZ="xz -T$NIX_BUILD_CORES" \
      M="$modulePath" \
      modules_install
  '';

  meta = {
    description = "AMD GPU kernel module";
    license = lib.licenses.gpl3;
  };
}

Replacing default kernel modules

After packaging the kernel module you can add it to your system just like an out-of-tree kernel module, it will replace the default module provided by the linux package because it has the same name:

{pkgs, config, ...}:
let
  amdgpu-kernel-module = pkgs.callPackage ./amdgpu-kernel-module.nix {
    # Make sure the module targets the same kernel as your system is using.
    kernel = config.boot.kernelPackages.kernel;
  };
in
{
  boot.extraModulePackages = [
    (amdgpu-kernel-module.overrideAttrs (_: {
      patches = [ ./patches/amdgpu-foo-bar.patch ];
    })
  ];
}

Troubleshooting

Build fails

Too high ram usage

turn off `DEBUG_INFO_BTF`

See also

Kernel Debugging with QEMU

References