Virtio-GPU Venus

Venus is a Virtio-GPU protocol for Vulkan command serialization. The protocol definition and codegen are hosted at venus-protocol. The renderer is hosted at virglrenderer.

The protocol is still under development. This driver and the renderer are both considered experimental.


The Venus renderer requires

  • Vulkan 1.1

  • VK_EXT_external_memory_dma_buf

  • VK_EXT_image_drm_format_modifier

  • VK_EXT_queue_family_foreign

from the host driver. However, it violates the spec in some places currently and also relies on implementation-defined behaviors in others. It is not expected to work on all drivers meeting the requirements. It has only been tested with

  • ANV 21.1 or later

  • RADV 21.1 or later (the host kernel must have CONFIG_TRANSPARENT_HUGEPAGE disabled because of this KVM issue)

The Venus driver requires supports for





from the virtio-gpu kernel driver, unless vtest is used. Currently, this means the context-init kernel branch paired with crosvm.


The simplest way to test Venus is to use virglrenderer’s vtest server. To build virglrenderer with Venus support and to start the vtest server,

$ git clone
$ cd virglrenderer
$ meson out -Dvenus-experimental=true
$ ninja -C out
$ ./out/vtest/virgl_test_server --venus

In another shell,

$ export VK_ICD_FILENAMES=<path-to-virtio_icd.x86_64.json>
$ export VN_DEBUG=vtest
$ vulkaninfo
$ vkcube

If the host driver of the system is not new enough, it is a good idea to build the host driver as well when building the Venus driver. Just remember to set VK_ICD_FILENAMES when starting the vtest server so that the vtest server finds the locally built host driver.


Because the driver requires VIRTGPU_PARAM_CONTEXT_INIT from the virtio-gpu kernel driver, one must make sure the guest kernel includes the changes from the context-init branch.

To build crosvm,

$ mkdir crosvm
$ cd crosvm
$ wget
$ chmod +x repo
$ ./repo init -g crosvm -u
$ ./repo sync
$ cd src/platform/crosvm
$ RUSTFLAGS="-L<path-to-virglrenderer>/out/src" cargo build \
      --features "x virgl_renderer virgl_renderer_next default-no-sandbox"

Note that crosvm must be built with default-no-sandbox or started with --disable-sandbox in this setup.

This is how one might want to start crosvm

$ sudo LD_LIBRARY_PATH=<...> VK_ICD_FILENAMES=<...> ./target/debug/crosvm run \
      --gpu vulkan=true \
      --display-window-keyboard \
      --display-window-mouse \
      --host_ip \
      --netmask \
      --mac 12:34:56:78:9a:bc \
      --rwdisk disk.qcow2 \
      -p root=/dev/vda1 \

assuming a working system is installed to partition 1 of disk.qcow2. sudo or CAP_NET_ADMIN is needed to set up the TAP network device.

Virtio-GPU and Virtio-WL

In this setup, the guest userspace uses Xwayland and a special Wayland compositor to connect guest X11/Wayland clients to the host Wayland compositor, using Virtio-WL as the transport. This setup is more tedious, but that should hopefully change over time.

For now, the guest kernel must be built from the chromeos-5.10 branch of the Chrome OS kernel. crosvm should also be built with wl-dmabuf feature rather than x feature.

To build minigbm and to enable minigbm support in virglrenderer,

$ git clone
$ cd minigbm
$ CFLAGS=-DDRV_<I915-or-your-driver> OUT=out DESTDIR=out/install make install
$ cd ../virglrenderer
$ meson configure out -Dminigbm_allocation=true
$ ninja -C out

Make sure a host Wayland compositor is running. Replace --display-window-keyboard --display-window-mouse by --wayland-sock=<path-to-wayland-socket> when starting crosvm.

In the guest, build and start sommelier, the special Wayland compositor,

$ git clone
$ cd platform2/vm_tools/sommelier
$ meson out -Dxwayland_path=/usr/bin/Xwayland -Dxwayland_gl_driver_path=/usr/lib/dri
$ ninja -C out
$ sudo chmod 777 /dev/wl0
$ ./out/sommelier -X --glamor
      --xwayland-gl-driver-path=<path-to-locally-built-gl-driver> \
      sleep infinity

sommelier requires xdg-shell-unstable-v6 rather than the stable xdg-shell from the host compositor. One must make sure the host compositor still supports the older extension.

Optional Requirements

When virglrenderer is built with -Dminigbm_allocation=true, the Venus renderer might need to import GBM BOs. The imports will fail unless the host driver supports the formats, especially multi-planar ones, and the DRM format modifiers of the GBM BOs.

In the future, if virglrenderer’s virgl_renderer_export_fence is supported, the Venus renderer will require VK_KHR_external_fence_fd with VK_EXTERNAL_FENCE_HANDLE_TYPE_SYNC_FD_BIT from the host driver.

A WSI image of the Venus driver is an external image to the host driver. When the WSI image is transitioned from VK_IMAGE_LAYOUT_UNDEFINED after image acquisition, the Venus driver does not request the Venus renderer to perform an ownership transfer on the external image. It is unclear if the ownership transfer is required or not. A specification issue has been filed for clarifications. See the comment before vn_cmd_fix_image_memory_barrier for more details.


The Venus renderer makes assumptions about VkDeviceMemory that has VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT. The assumptions are illegal and rely on the current behaviors of the host drivers. It should be possible to remove some of the assumptions and incrementally improve compatibilities with more host drivers by imposing platform-specific requirements. But the long-term plan is to create a new Vulkan extension for the host drivers to address this specific use case.

The Venus renderer assumes a device memory that has VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT can be exported as a mmapable dma-buf (in the future, the plan is to export the device memory as an opaque fd). It chains VkExportMemoryAllocateInfo to VkMemoryAllocateInfo without checking if the host driver can export the device memory.

The dma-buf is mapped (in the future, the plan is to import the opaque fd and call vkMapMemory) but the mapping is not accessed. Instead, the mapping is passed to KVM_SET_USER_MEMORY_REGION. The hypervisor, host KVM, and the guest kernel work together to set up a write-back or write-combined guest mapping (see virtio_gpu_vram_mmap of the virtio-gpu kernel driver). CPU accesses to the device memory are via the guest mapping, and are assumed to be coherent when the device memory also has VK_MEMORY_PROPERTY_HOST_COHERENT_BIT.

When a VkImage or a VkBuffer is created, the Venus renderer does not know if the image or the buffer will be bound to such a device memory or not. As a result, the Venus renderer unconditionally chains VkExternalMemoryImageCreateInfo to VkImageCreateInfo and chains VkExternalMemoryBufferCreateInfo to VkBufferCreateInfo without checking for the host driver support.