Kernel Debugging with QEMU

Setup

Clone the repository

$ git clone https://github.com/torvalds/linux.git

For kernel dependencies, create a shell.nix file in the cloned repo:

{ pkgs ? import <nixpkgs> {} }:

pkgs.stdenv.mkDerivation {
  name = "linux-kernel-build";
  nativeBuildInputs = with pkgs; [
    getopt
    flex
    bison
    gcc
    gnumake
    bc
    pkg-config
    binutils
  ];
  buildInputs = with pkgs; [
    elfutils
    ncurses
    openssl
    zlib
  ];
}

Older kernel versions might buildFHSUserEnv as they have absolute shebangs:

{ pkgs ? import <nixpkgs> {} }:

(pkgs.buildFHSUserEnv {
  name = "linux-kernel-build";
  targetPkgs = pkgs: (with pkgs;  [
    getopt
    flex
    bison
    elfutils
    binutils
    ncurses.dev
    openssl.dev
    zlib.dev
    gcc
    gnumake
    bc
  ]);
  runScript = "bash";
}).env

Generate a config for KVM

If on make you get asked some questions, just press enter till you are done, this will select the default answer.

$ cd linux
$ make mrproper # Clears all artifacts, do this especially if you upgrade from a significant old version
$ nix-shell shell.nix
$ make defconfig kvm_guest.config
$ scripts/config --set-val DEBUG_INFO y # For gdb debug symbols
$ scripts/config --set-val DEBUG y # All pr_debug messages get printed
$ scripts/config --set-val GDB_SCRIPTS y
$ scripts/config --set-val DEBUG_DRIVER y # Enable printk messages in drivers
# everything as one command for copy'n'paste
$ scripts/config --set-val DEBUG_INFO y --set-val DEBUG y  --set-val GDB_SCRIPTS y --set-val DEBUG_DRIVER y
# this might ask for further options, just press enter for every question
$ make -j$(nproc)

Create a bootable NixOS image with no kernel

Save this as nixos-image.nix:

{ pkgs ? import <nixpkgs> {} }:
import (pkgs.path + "/nixos/lib/make-disk-image.nix") {
  config = (import (pkgs.path + "/nixos/lib/eval-config.nix") {
    inherit (pkgs) system;
    modules = [{
      imports = [ ./nixos-config.nix ];
    }];
  }).config;
  inherit pkgs;
  inherit (pkgs) lib;
  diskSize = 1024;
  partitionTableType = "none";
  # for a different format
  format = "qcow2";
}

Than follows the nixos configuration in a file named nixos-config.nix

{ pkgs, lib, modulesPath, ... }:

{
  imports = [
    (modulesPath + "/profiles/qemu-guest.nix")
  ];
  boot.loader.grub.enable = false;
  boot.initrd.enable = false;
  boot.isContainer = true;
  boot.loader.initScript.enable = true;
  ## login with empty password
  users.extraUsers.root.initialHashedPassword = "";

  networking.firewall.enable = false;

  services.getty.helpLine = ''
    Log in as "root" with an empty password.
    If you are connect via serial console:
    Type Ctrl-a c to switch to the qemu console
    and `quit` to stop the VM.
  '';

  services.getty.autologinUser = lib.mkDefault "root";

  documentation.doc.enable = false;
  documentation.man.enable = false;
  documentation.nixos.enable = false;
  documentation.info.enable = false;
  programs.bash.enableCompletion = false;
  programs.command-not-found.enable = false;
}

Than build with the following commands:

$ nix-build
# copy out
$ install -m644 result/nixos.qcow2 qemu-image.img

Than follow with the next step is launching qemu.

Create a bootable Debian image with replaceable kernel

If you want to build a different Linux distro you can use the following instructions to build a debian instead:

 $ nix-shell -p debootstrap qemu
 $ qemu-img create qemu-image.img 5G
 $ mkfs.ext2 qemu-image.img
 $ mkdir mount-point.dir
 $ sudo mount -o loop qemu-image.img mount-point.dir
 $ sudo debootstrap --arch amd64 buster mount-point.dir
 $ sudo chroot mount-point.dir /bin/bash -i
 $ export PATH=$PATH:/bin
 $ passwd # Set root password
 $ exit
 $ sudo umount mount-point.dir

Installing tools to the image

The filesystem is mounted read only so to add tools like lspci. Mount and chroot then use apt to install the needed binaries.

 $ sudo  mount -o loop qemu-image.img mount-point.dir
 $ sudo chroot mount-point.dir /bin/bash -i
 $ export PATH=$PATH:/bin
 $ apt install pciutils tree
 $ sudo umount mount-point.dir

Launch qemu

You can find a slighty stripped version of qemu in a package called qemu_kvm (qemu without emulation support for other cpu architectures). The nokaslr kernel flag is important to be able to set breakpoints in kernel memory. You can also skip the -S to not make qemu break on startup and waiting for gdb.

 $ qemu-system-x86_64 -s -S \
    -kernel arch/x86/boot/bzImage \
    -hda qemu-image.img \
    -append "root=/dev/sda console=ttyS0 nokaslr" \
    -enable-kvm \
    -nographic

Connect with gdb

 $ echo "add-auto-load-safe-path `pwd`/scripts/gdb/vmlinux-gdb.py" >> ~/.gdbinit
 $ gdb -ex "target remote :1234" ./vmlinux
 (gdb) continue

Note that setting breakpoints in early boot might not work for all functions. If a breakpoint is not triggered as expected try to set the breakpoint later when the VM is fully booted.

Language server support

If you want language server support for the kernel code you can generate a compile_commands.json with

$ python ./scripts/clang-tools/gen_compile_commands.py

This can be used for example in combination with clangd, which scales well to size of the linux kernel.

Debugging drivers

Make sure the driver you want to inspect is not compiled into the kernel, look for the option to enable compilation of your driver, to do this execute:

 $ make nconfig

press F8 and search for your driver, and check if it is set to "Module" with <M>. After compilation copy the driver.ko into the mounted qemu-image.img. Unmount start the kernel and break at the load_module function and insmod driver.ko. Happy hacking!

Bugs

1. With the nixos-config provided above, the console does not work properly. boot.isContainer = true; implies console.enable = false; that disables console. The following can be used as a workaround.

console.enable = true;
systemd.services."serial-getty@ttyS0".enable = true;

Using ktest with NixOS

Yellow onion has integrated nixos vms into ktest:

https://github.com/YellowOnion/ktest/commit/73fadcff949236927133141fcba4bfd76df632e7

This integration also allows to use incremental kernel builds for rapid development. Checkout the commit message for details.