This project provides an easy method for deploying RISC-V QEMU instances in Kubernetes. Using Helm, you can quickly configure and deploy a RISC-V VM and access its console through a web interface. The Helm chart is compatible with Knot. Pre-compiled configurations support RVV 1.0 and RVV 0.7.1.
When runnning Kubernetes in Docker Desktop, you can just run the following and visit http://localhost:8080:
helm install myvm ./chart/qemu-riscv64To run a second instance, specify a different port:
helm install myvm ./chart/qemu-riscv64 \
--set service.port=8081If running Kubernetes with minikube, the above commands apply, but you also need to enable service forwarding to localhost with minikube tunnel.
The table below lists all available options (the current version is set in the Makefile):
| Variable | Default | Notes |
|---|---|---|
images.launcher |
carvicsforth/qemu-riscv64-launcher:<version> |
Default launcher, supporting RVV 1.0 |
images.launcherRVV071 |
carvicsforth/qemu-riscv64-launcher-rvv-0.7.1:<version> |
Custom launcher, supporting RVV 0.7.1 |
images.console |
carvicsforth/qemu-riscv64-console:<version> |
|
images.data |
carvicsforth/qemu-riscv64-ubuntu:22.04-1 |
Use for custom files, kernel, and BIOS |
service.type |
LoadBalancer |
|
service.port |
8080 |
|
cpu.features |
Can be overriden by the data image |
|
cpu.count |
2 |
Also sets resource request in deployment |
memory |
2048 |
Also sets resource request in deployment |
append |
Kernel options. Can be overriden by the data image |
Available data images:
| Image | Notes |
|---|---|
carvicsforth/qemu-riscv64-ubuntu:22.04-1 |
Standard Ubuntu image (login: ubuntu/ubuntu) |
carvicsforth/qemu-riscv64-buildroot:00709af-1 |
Custom built image, uses RVV 1.0 (login: root/) |
carvicsforth/qemu-riscv64-buildroot-rvv-0.7.1:954aeb7-1 |
Custom built image, uses RVV 0.7.1 (login: root/) |
To images set the CPU specifications they need, as well as custom kernel options. If the image name contains rvv-0.7.1, the chart automatically switches to the respective launcher.
To run the custom Buildroot image:
helm install myvm ./chart/qemu-riscv64 \
--set images.data="carvicsforth/qemu-riscv64-buildroot:00709af-1"We use two basic containers for each VM:
- The
launchercontains the QEMU binary. A script starts QEMU automatically when it runs, using environmental variables to define the number of CPUs, the memory, and other parameters. QEMU is configured to serve its console via telnet at local port 10023. Networking is handled by passt, so the VM gets the same IP as the pod and can communicate with other VMs in the same Kubernetes environment. - The
consoleis based on gotty. Gotty runs a local command (in this casetelnet), captures input and output streams, exposes them via websockets, and then offers a web-based terminal emulator (xterm.js) for the user to interact with what has been run.
The root filesystem used by QEMU is also packaged up in a container image (which we call data), using a similar method to KubeVirt. Before each VM starts, the data container copies its contents to an ephemeral volume that is shared with the launcher. As data contains only VM-specific files (it can optionally also hold a custom kernel and BIOS), we copy over a BusyBox binary before it starts, so it can run cp to populate the shared volume.
Container images are available for both amd64 and arm64 architectures.
To build your own locally, set the REGISTRY variable and run:
REGISTRY=<Docker Hub username> make containerThe image tag for the launcher and console images is controlled via the VERSION variable.
This project has received funding from the European Union’s Horizon Europe research and innovation programme through project RISER ("RISC-V for Cloud Services", GA-101092993) and from the Key Digital Technologies Joint Undertaking through project REBECCA ("Reconfigurable Heterogeneous Highly Parallel Processing Platform for safe and secure AI", GA-101097224). KDT JU projects are jointly funded by the European Commission and the involved state members (including the Greek General Secretariat for Research and Innovation).