Vinix supports running as a guest under Linux KVM. It also has an experimental native Intel VT-x hypervisor. These use different userspace APIs:
| Use case | Current support |
|---|---|
| Run amd64 Vinix under QEMU/KVM on an x86_64 Linux host | Supported by the desktop launcher and ISO boot checks. |
| Run arm64 Vinix under QEMU/KVM on an AArch64 Linux host | Supported by the ISO boot checks. |
| Create hardware-accelerated VMs inside amd64 Vinix | Intel VT-x through /dev/hypervisor; one vCPU and at most 1 GiB RAM per VM. Larger-memory allocation and mappings are tested separately from physical VM entry. |
Use Linux's /dev/kvm API inside Vinix |
Not implemented on either architecture. |
| Run another arm64 Vinix instance through QEMU inside Vinix | TCG software emulation through vinix-qemu. |
KVM runs on the Linux host, which needs a working KVM kernel backend,
hardware virtualization enabled in firmware, a QEMU build with KVM support,
and read/write access to the host's /dev/kvm. On x86_64 the host can use
Intel VT-x or AMD SVM; the Intel-only limit of Vinix's own hypervisor does not
restrict running Vinix as a guest. Use an image matching the host architecture.
QEMU documents its supported host accelerators in its
system emulator introduction.
Check the host before launching an amd64 guest:
qemu-system-x86_64 -accel help
ls -l /dev/kvm
test -r /dev/kvm && test -w /dev/kvmThe accelerator list should include kvm, and the access check should exit
successfully. Listing an accelerator alone does not prove it can initialize.
If /dev/kvm is missing, check the Linux host's KVM setup and firmware; if
access is denied, use the host distribution's KVM device permissions.
With the released vinix-amd64.iso, create a new disk once and boot:
qemu-img create -f qcow2 vinix-amd64.qcow2 16G
qemu-system-x86_64 -machine q35 -accel kvm -cpu host -m 4096 -smp 2 \
-cdrom vinix-amd64.iso \
-drive file=vinix-amd64.qcow2,format=qcow2 \
-nic user,model=e1000Keep the disk for later launches. The first boot installs onto the empty disk; subsequent boots preserve users, files, and installed apps. Keep the ISO attached as described in the image instructions.
For an amd64 desktop built from this checkout:
./scripts/run-desktop-amd64.shThe launcher selects KVM with -cpu host when /dev/kvm is readable and
writable. Otherwise it selects HVF on an Intel Mac or TCG software emulation.
An explicit VINIX_QEMU_ACCEL overrides this selection; specify the CPU too:
VINIX_QEMU_ACCEL=kvm ./scripts/run-desktop-amd64.sh --no-build -cpu host
VINIX_QEMU_ACCEL=tcg ./scripts/run-desktop-amd64.sh --no-build -cpu maxThe selection checks device permissions; it does not retry with TCG if KVM
initialization fails. Use the explicit TCG command in that case. --no-build
requires an existing vinix-desktop-amd64.iso. The desktop launcher also needs
x86_64 UEFI firmware; set VINIX_OVMF_CODE if it is not found automatically.
For release-image boot checks on a matching Linux host, run either:
./scripts/test-iso.sh --no-virtualbox vinix-amd64.iso
./scripts/test-iso.sh --no-virtualbox vinix-arm64.isoThe ISO runner selects KVM when the host device is writable (and, for arm64,
the host is AArch64). The AArch64 development runner
scripts/run-desktop-aarch64.sh uses HVF by default, or TCG with USE_TCG=1;
it does not automatically select Linux KVM.
The native backend requires Intel VMX, EPT with INVEPT, and unrestricted-guest support
available to the amd64 Vinix kernel. When Vinix itself runs in a VM, the outer
hypervisor must expose nested VT-x for /dev/hypervisor to become available.
Running Vinix as an ordinary KVM guest does not require nested VT-x.
At boot, Vinix checks VMX and runs a real-mode guest self-test before publishing
the root-only /dev/hypervisor device. A successful boot logs
hypervisor: /dev/hypervisor ready. If VMX is unavailable or the self-test
fails, Vinix continues booting without the device.
This backend targets small firmware and device-model experiments. AMD SVM, an ARM hardware virtualization backend, multiple vCPUs, interrupt injection, and a complete device model are not implemented. See the native hypervisor API for its ioctl sequence and C example, and the hypervisor checks for validation and the distinction between host model checks and actual VM-entry coverage.
Linux KVM applications expect /dev/kvm and KVM_* ioctls, as specified in the
Linux KVM API. Vinix's
VINIX_HV_* ioctls are a separate interface, so QEMU -accel kvm cannot use
the native backend. For QEMU running inside arm64 Vinix, follow the
nested QEMU instructions, which use -accel tcg.
The V VM manager imports raw images into separate qcow2 disks, keeps firmware variables per VM, controls guests over QMP, and pairs offline disk snapshots with firmware snapshots. It supports explicit QEMU accelerators, multiple virtual CPUs, user-mode networking, loopback SSH forwarding, read-only virtio-9p shares and a guest-agent channel. Its real QEMU host tests cover lifecycle, locking and snapshot restoration. Installed Vinix uses TCG with its existing authorized JIT policy; native Intel/AMD/ARM hardware-accelerated VM hosting remains incomplete.