.. SPDX-License-Identifier: GPL-2.0-or-later

=================================== Testing VFIO display with mdev mdpy

.. contents:: Table of Contents

The kernel provides a sample mediated device driver, mdpy (samples/vfio-mdev/mdpy.c), that exposes a fake framebuffer through the VFIO display region interface. It can be used to test VFIO display support, including hotplug, without any real GPU hardware.

The kernel modules

The mdpy driver depends on the mdev subsystem. Enable, build and load the modules.

The minimal set is::

CONFIG_SAMPLE_VFIO_MDEV_MDPY=m
CONFIG_SAMPLE_VFIO_MDEV_MDPY_FB=m   # guest framebuffer driver

CONFIG_VFIO_MDEV is selected automatically.

Verify that the driver registered successfully:

.. code-block:: bash

ls /sys/devices/virtual/mdpy/mdpy/mdev_supported_types/

Creating an mdev instance

Available types correspond to different resolutions (e.g. mdpy-vga for 640x480, mdpy-xga for 1024x768, mdpy-hd for 1920x1080).

Each mdev instance is identified by a UUID:

.. code-block:: bash

uuid=$(uuidgen)
echo "$uuid" > /sys/devices/virtual/mdpy/mdpy/mdev_supported_types/mdpy-xga/create

To remove the instance later:

.. code-block:: bash

echo 1 > /sys/bus/mdev/devices/$uuid/remove

Make sure your user has the necessary permissions to access the vfio group. (ex: chmod 666 /dev/vfio/16)

Starting QEMU

Boot-time attachment

.. code-block:: bash

qemu-system-x86_64 -machine q35 -m 1G \
    -device vfio-pci,sysfsdev=/sys/bus/mdev/devices/$uuid,display=on \
    -display gtk,gl=on

Hotplug via HMP

Start QEMU with a PCIe root port (required for PCIe hotplug) and a monitor:

.. code-block:: bash

qemu-system-x86_64 -machine q35 -m 1G \
    -device pcie-root-port,id=rp0,slot=1 \
    -display gtk,gl=on \
    -monitor stdio

Then at the (qemu) prompt:

.. code-block:: none

device_add vfio-pci,sysfsdev=/sys/bus/mdev/devices/<uuid>,display=on,bus=rp0,id=mdpy0

To hot-unplug:

.. code-block:: none

device_del mdpy0