QEMU and KVM

Roles

Component	Role in the lecture’s Linux stack
QEMU	Userspace VM setup, machine/device models, lifecycle and migration support
KVM	Linux kernel virtualization interface and hardware-assisted guest execution
Linux kernel	Host scheduling, memory management, drivers, and supporting infrastructure
Guest OS	Kernel and applications running inside the virtual machine

QEMU can also emulate a whole system across different ISAs. Its user-mode emulation runs an individual foreign-ISA program; system mode supplies a complete machine. KVM acceleration in the lecture requires a compatible host/guest ISA and virtualization hardware.

VM process and vCPU threads

The illustrated VM is backed by a QEMU process. Guest RAM belongs to that process’s address space; each guest vCPU executes through a host thread. A main loop handles events and other threads can perform expensive I/O.

The host scheduler sees host threads, not every guest application thread as a separate schedulable host entity. The guest OS schedules its own work onto its vCPUs.

Lifecycle through /dev/kvm

The slide’s conceptual flow is:

1. Open /dev/kvm.
2. Create a VM and register guest-memory backing.
3. Create vCPUs and map their kvm_run communication structures.
4. Initialize processor state.
5. Call KVM_RUN to execute.
6. Handle returned exit reasons where userspace emulation is needed, then resume.

QEMU (root ring 3) -> host syscall -> KVM (root ring 0)
KVM -> VM entry -> guest (non-root kernel/apps)
Guest -> VM exit -> KVM -> handle in kernel or return to QEMU

Not every VM exit makes a userspace round trip.

Correction to the schematic code

PDF page 80 / printed slide 85 is not a complete executable example. In the x86 API, KVM_SET_REGS sets general registers including RIP, while KVM_SET_SREGS sets special registers. The slide mixes these categories. Linux KVM API.

Lecture demonstrations

Pages 76-77 compare x64 emulation with -enable-kvm and an ARM64 guest emulated on x86. The supplied Ubuntu 20.04/21.04 downloads and firmware commands are recorded as historical demonstrations, not tested setup instructions. No VM was launched for these notes.

Common mistakes

• Assuming KVM replaces all QEMU responsibilities.
• Assuming hardware acceleration translates arbitrary foreign ISAs.
• Assuming a guest’s ten application threads become ten host threads.
• Copying the schematic API example as working C.

Related: Hypervisor Architectures, Hardware-Assisted Virtualization and VMCS, Libvirt virsh and virt-manager, Virtio Virtqueues and vhost-net.

Source

Lecture 03 PDF pages 71-83 (printed slides 74-88, with gaps). Context: Lecture 03 - Computing Virtualization Technologies and Tools. The diagram and cautionary distinctions are study explanations; the API correction is separately sourced above.