Day 0: The Training Ground
Goal: Two lab VMs that can run VMs themselves, an SSH key and the companion code. · Duration: approx. 30 min · Prerequisite: none
Briefing
- KVM (Kernel-based Virtual Machine) is a module in the Linux kernel. It uses the virtualization features of the CPU (Intel VT-x, AMD-V) and turns Linux itself into a hypervisor. KVM alone does not start a VM, though.
- QEMU is the program that builds the VM: virtual
disks, network cards, BIOS, console. Every VM is an ordinary
qemu-system-x86_64process, and KVM runs its CPU instructions directly on the real CPU. - libvirt manages many such QEMU processes: with
configuration in XML, networks, storage pools and the
virshtool. Proxmox, OpenNebula and OpenStack build on exactly this stack.
Laptop ──ssh──► kvm1 (Ubuntu 24.04, 4 vCPU, 8 GB) kvm2 (2 vCPU, 4 GB)
├─ libvirtd ├─ libvirtd
├─ qemu-system-x86_64 VM ◄── live migration (Day 12) ──►
└─ qemu-system-x86_64 VM notes
Kernel: kvm, kvm_intel ── /dev/kvm
Drill 1: The lab VMs
You need two Linux VMs with Ubuntu 24.04 in which KVM works:
| Name | vCPU | RAM | Disk |
|---|---|---|---|
kvm1 |
4 | 8 GB | 80 GB |
kvm2 |
2 | 4 GB | 60 GB |
Both are on the same network and reachable via SSH. You only need kvm2 from Day 7 on. For VMs to run inside a VM, the hypervisor underneath must allow nested virtualization:
| Underneath | What you set |
|---|---|
| Proxmox | CPU type host in the VM’s hardware settings |
| libvirt/virt-manager | CPU model host-passthrough |
| VirtualBox | “Enable Nested VT-x/AMD-V” |
| Mac with Apple Silicon | does not work, use an x86 machine or a server |
| Physical hardware | nothing, KVM runs directly there |
On a Proxmox host, this file shows whether nested virtualization is
enabled (on AMD the module is called kvm_amd):
# on the Proxmox host
cat /sys/module/kvm_intel/parameters/nestedExpected:
Y
On current Proxmox versions this is the default.
Save your progress? Sign up for free
Drill 2: Does the VM see the CPU features?
Log in to kvm1. From here on, everything runs there unless a chapter says otherwise.
# on kvm1
lscpu | grep -E 'Model name|Virtualization|Hypervisor'Expected (your processor will differ):
Model name: Intel(R) Xeon(R) CPU E5-2620 v4 @ 2.10GHz
Virtualization: VT-x
Hypervisor vendor: KVM
Virtualization type: full
Virtualization: VT-x(on AMDAMD-V): The VM sees the feature. If the line is missing, the CPU type is set incorrectly.Hypervisor vendor: KVM: kvm1 itself runs under KVM, namely under Proxmox.
grep -c vmx /proc/cpuinfoExpected: a number greater than 0 (here 8:
the flags and vmx flags lines for each CPU).
On AMD you search for svm.
Save your progress? Sign up for free
Drill 3: The kernel module and /dev/kvm
lsmod | grep '^kvm'
ls -l /dev/kvmExpected:
kvm_intel 487424 0
kvm 1404928 1 kvm_intel
crw-rw---- 1 root kvm 10, 232 Nov 13 12:50 /dev/kvm
kvmis the generic part,kvm_intel(orkvm_amd) the part for your processor. Ubuntu loads both automatically./dev/kvmis the door to KVM. Whoever may open it may start accelerated VMs:rootand thekvmgroup.
A small check tool sums this up:
sudo apt-get install -y cpu-checker
kvm-okExpected:
INFO: /dev/kvm exists
KVM acceleration can be used
Save your progress? Sign up for free
Drill 4: You in the kvm group
On Day 1 you start QEMU as a regular user. For that you need access
to /dev/kvm:
sudo usermod -aG kvm $USERThe new group only applies after a new login. Log out and back in:
exit# logged in again on kvm1
id -nGExpected: kvm is in the list.
ubuntu adm cdrom sudo dip lxd kvm
Save your progress? Sign up for free
Drill 5: An SSH key
Your VMs receive your public key on first boot. You then only need a password for the console:
ssh-keygen -q -t ed25519 -N "" -f ~/.ssh/id_ed25519
cat ~/.ssh/id_ed25519.pubExpected (your key will differ):
ssh-ed25519 AAAAC3NzaC1lZDI1NTE5AAAAIKBb5JErN+8f… ubuntu@kvm1
Save your progress? Sign up for free
Drill 6: Working directory and companion code
mkdir -p ~/kvm-lab/images
git clone -q https://github.com/thomaszachmann/kvm-zum-anfassen.git ~/kvm-lab/code
ls ~/kvm-lab/codeExpected:
LICENSE loesung README.md wettkampf
loesung/tagNN contains the files you write each day. You
only need wettkampf/ on Day 15. The downloaded images end
up in ~/kvm-lab/images.
Save your progress? Sign up for free
Checkpoint
kvm-okExpected: the line KVM acceleration can be used.
id -nGExpected: kvm is in the list. If the group is missing,
you did not log in again after Drill 4.
ls ~/.ssh/id_ed25519.pub ~/kvm-lab/code/README.mdExpected: both files are listed without errors.
Challenge
Find out how many CPU cores and how much RAM kvm1 has left for its own VMs if Ubuntu itself needs about 1 GB.
Hint: nproc and free -m
are enough.
Solution
nproc
free -mExpected (roughly):
4
total used free shared buff/cache available
Mem: 7941 528 3373 1 4347 7412
Swap: 0 0 0
Four cores and a good 6.5 GB of RAM. You can overcommit cores: three VMs with 2 vCPUs each, that is 6 vCPUs, share the four cores. With RAM this is more dangerous, because when all VMs actually use their memory, the host’s OOM killer steps in and terminates a QEMU process.
Save your progress? Sign up for free
Enterprise Note
- No nesting in production: KVM hosts run on physical hardware. Nested virtualization costs performance and is not supported by many vendors.
- Check the BIOS/UEFI: On new servers, VT-x/AMD-V is
sometimes disabled in the BIOS.
kvm-oktells you before you start troubleshooting. - Who uses KVM: Proxmox VE, OpenNebula, OpenStack Nova, oVirt and the big clouds are built on KVM and QEMU. What you learn here helps when debugging all of these platforms.