When developing new features for the kernel, chances are that you will screw up. Often. Depending on the severity, the kernel may or may not recover. So to avoid restarting your PC over and over, it's better to work in a minimal virtualized environment. As such, we will first bootstrap a loopback disk image with a basic Ubuntu system, but without the kernel. Eventually, we will boot a virtual macine with qemu-system-x86_64 from this disk image, with a custom kernel that we will build ourselves.

For the bootstrapping process, we will use debootstrap. This tool will download a Debian-based ecosystem and install it in whatever directory we tell it to. Incidentally, that directory will be the mount point of the disk image that we are going to create.

# create a 5GB empty file -- this will be our disk image
[student@host]$ qemu-img create images/ubuntu.raw 5G
 
# build an ext4 filesystem onto the disk image -- now we can mount it
[student@host]$ mkfs.ext4 images/ubuntu.raw
 
# mount the ext4 filesystem -- now we can copy files onto it
[student@host]$ sudo mount images/ubuntu.raw /mnt
 
# bootstrat the Ubuntu system
[student@host]$ sudo debootstrap --arch amd64 focal /mnt https://mirrors.kernel.org/ubuntu

Almost there… if you list the contents of /mnt/, you will see most of the usual entries from your root directory. At this point, we should be able to boot into this machine (if we had a kernel image), but we wouldn't be able to log in. The only thing that's left for us to do is set a password for the root user. For this, we need to trick the passwd tool to thing that /mnt/ is in fact the root of our filesystem. So we use chroot:

# pretend that /mnt/ is our new / and start a bash instance inside
[student@host]$ sudo chroot /mnt /bin/bash
 
# change password for current user (root)
[   root@jail]$ passwd
New password: root
Retype new password: root
passwd: password updated successfully
 
# exit from this bash instance and escape from the chroot jail
[   root@jail]$ exit
 
# finally, unmount our disk -- we're done with it for now
[student@host]$ sudo umount /mnt

Next step is to get the kernel source code and compile it. By separating the kernel from the disk image, we are able to checkout to other branches / commits and test out different versions without installing them anywhere. Normally, you would have to select what options you want included in the compilation process (e.g.: memory allocators, cryptographic systems, etc.) by running make menuconfig. After finishing your selection and saving the configuration, a .config file would be created. Because we haven't the faintest idea what most of the things enumerated in that menu even are, we will rely on default configurations.

# clone the linux kernel locally
[student@host]$ git clone --depth=1 https://github.com/torvalds/linux.git
 
# go into the repo directory
[student@host]$ pushd linux
 
# create a default configuration file (.config)
[student@host]$ make x86_64_defconfig kvm_guest.config
 
# optional: check out the generated .config file
[student@host]$ less .config
 
# compile the kernel using all cores
[student@host]$ make -j $(nproc) bzImage
 
# required for building out-of-tree modules -- see warning below
[student@host]$ make modules_prepare 
 
# return to the previous direcotry
[student@host]$ popd

make[3]: *** No rule to make target 'scripts/module.lds', needed by '...'.   Stop.
make[2]: *** [scripts/Makefile.modpost:140: __modpost] Error 2
make[1]: *** [Makefile:1761: modules] Error 2

We are finally here. Let's boot up the VM from our bootstrapped disk image, with our personally compiled Linux kernel.

$ sudo qemu-system-x86_64                            \
    -m 4G                                            \
    -smp 1                                           \
    -enable-kvm                                      \
    -kernel linux/arch/x86/boot/bzImage              \
    -drive file=images/ubuntu.raw,format=raw,index=0 \
    -append 'root=/dev/sda rw console=ttyS0'         \
    -nographic

Let us have a look at this command, line by line:

1. -m 4G: allocate 4GB of memory (change this as you wish)
2. -smp 1: use only 1 vCPU; this is recommended for debugging purposes
3. -enable-kvm: KVM is a Linux kernel module that transforms your operating system intro a bare-metal hypervisor. This is what allows you to run actual virtual machines on Linux. Without it, qemu would try to emulate the system, resulting in worse performance.
4. -kernel …/bzImage: this specifies the compiled & compressed kernel image to use when booting the virtual machine
5. -drive … : specifies the disk image to load; note that index=0 will make the VM consider this to be /dev/sda. Adding another drive with index=1 will cause it to be regarded as /dev/sdb.
6. -append …: these are command line arguments for the kernel (yes, even it has those). root=/dev/sda rw marks /dev/sda (i.e.: our Ubuntu.raw disk image) as the root device to be mounted onto the root directory (i.e.: /) in read-write mode. console=ttyS0 exposes an UART serial interface to the VM and tells Linux to use it for I/O.
7. -nographic: tells qemu: not to open a separate window for the GUI. In stead, it will take the virtual serial device (which the VM will recognize as ttyS0) and link it to the terminal. So whatever the VM sends via the serial to be printed will end out in your stdout. Whatever you type into stdin will be forwarded to the VM as input.

# list available interfaces (in a colorful fashion)
[root@guest]$ ip -c addr show
 
# send a DHCP request on your eth interface
[root@guest]$ dhclient eth0

After starting the VM and logging in as root (with the password set earlier), try finding out the kernel version in both the host and guest operating systems:

# host has the latest Arch Linux kernel (you may have Ubuntu, etc.)
[student@host]$ uname -r
5.15.2-arch1-1
 
# guest has the newest Linux kernel release candidate
[  root@guest]$ uname -r
5.16.0-rc2+