mirror of
https://github.com/PhyllomeOS/wiki.git
synced 2024-12-22 21:55:21 +00:00
173 lines
4.7 KiB
Markdown
173 lines
4.7 KiB
Markdown
---
|
|
title: Virtual Function I/O Mediated devices (vfio-mdev)
|
|
description: Create and Configure Virtual Function I/O Mediated devices (vfio-mdev)
|
|
published: true
|
|
date: 2022-07-27T23:53:40.933Z
|
|
tags:
|
|
editor: markdown
|
|
dateCreated: 2022-07-21T21:10:41.046Z
|
|
---
|
|
|
|
# Configure Virtual Function I/O Mediated devices
|
|
|
|
Virtual Function I/O Mediated devices (vfio-mdev) allows you to split a compatible GPU into multiple virtual GPUs (vGPUs). These vGPUs can then be assigned to a virtual machine, just as real GPUs.
|
|
|
|
> These instructions only cover Intel GPUs that are compatible with vfio-mdev (5th to 10th generation). Since generation 11th, Intel graphics do not support vfio-mdev but SR-IOV.
|
|
{.is-info}
|
|
|
|
## Preparation
|
|
|
|
### Update the GRUB
|
|
|
|
* On a freshly deployed edition of Phyllome OS optimized for Intel Graphics such as [Phyllome OS Desktop II](https://wiki.phyllo.me/deploy/rightforyou), make sure that the GRUB has been updated.
|
|
|
|
```
|
|
sudo grub2-mkconfig -o /boot/grub2/grub.cfg
|
|
```
|
|
|
|
* Then reboot your computer
|
|
|
|
```
|
|
sudo reboot
|
|
```
|
|
|
|
### Modify GPU Memory in BIOS/UEFI
|
|
|
|
> Only available for Intel Graphics cards
|
|
{.is-info}
|
|
|
|
Some computers allow you to modify the GPU memory allocated your desktop-based computers. It will reserve system memory for the GPUs.
|
|
|
|
To do so, you need to enter the BIOS/UEFI and to look for a setting called GPU Aperture size, or GPU memory.
|
|
|
|
Use the highest value possible, but make sure you have enough system memory to accomodate both the GPU and your operating system.
|
|
|
|
## Create a virtual GPU
|
|
|
|
Upon reboot, you should then be able to list available GPUs using the `mdevctl` command.
|
|
|
|
* List available virtual GPUs
|
|
|
|
```
|
|
mdevctl types
|
|
```
|
|
|
|
```
|
|
0000:00:02.0
|
|
i915-GVTg_V5_1
|
|
Available instances: 1
|
|
Device API: vfio-pci
|
|
Description: low_gm_size: 512MB, high_gm_size: 2048MB, fence: 4, resolution: 1920x1200, weight: 16
|
|
i915-GVTg_V5_2
|
|
Available instances: 2
|
|
Device API: vfio-pci
|
|
Description: low_gm_size: 256MB, high_gm_size: 1024MB, fence: 4, resolution: 1920x1200, weight: 8
|
|
i915-GVTg_V5_4
|
|
Available instances: 5
|
|
Device API: vfio-pci
|
|
Description: low_gm_size: 128MB, high_gm_size: 512MB, fence: 4, resolution: 1920x1200, weight: 4
|
|
i915-GVTg_V5_8
|
|
Available instances: 7
|
|
Device API: vfio-pci
|
|
Description: low_gm_size: 64MB, high_gm_size: 384MB, fence: 4, resolution: 1024x768, weight: 2
|
|
```
|
|
|
|
* You may need to increase GPU aperture size if there is no available instance.
|
|
|
|
In this case, the `i915-GVTg_V5_4` kind seems to offer the best trade-offs between the available resolution and the number of available instances.
|
|
|
|
* Generate a universally unique identifier (UUID) with the following command:
|
|
|
|
```
|
|
uuidgen
|
|
```
|
|
|
|
```
|
|
7686131b-b229-4768-a02c-35d1dbed7c66
|
|
```
|
|
|
|
* Start a vGPU based on the kind `i915-GVTg_V5_4` using the previously generated UUID
|
|
|
|
```
|
|
sudo mdevctl start -u 7686131b-b229-4768-a02c-35d1dbed7c66 -p 0000:00:02.0 --type i915-GVTg_V5_4
|
|
```
|
|
|
|
* Define, or make this vGPU permanent.
|
|
|
|
```
|
|
sudo mdevctl define -u 7686131b-b229-4768-a02c-35d1dbed7c66
|
|
```
|
|
|
|
* Set the vGPU to auto-start after the host boots up, so that it is available to guest virtual machines without further action
|
|
|
|
```
|
|
sudo mdevctl modify -u 7686131b-b229-4768-a02c-35d1dbed7c66 --auto
|
|
```
|
|
|
|
* Finally, verify that the vGPU has successfully been created and is set to auto-start:
|
|
|
|
```
|
|
mdevctl list -d
|
|
```
|
|
|
|
```
|
|
7686131b-b229-4768-a02c-35d1dbed7c66 0000:00:02.0 i915-GVTg_V5_4 auto (active)
|
|
```
|
|
|
|
Great you have
|
|
|
|
## Assign a virtual GPU to a virtual machine
|
|
|
|
* Add that segment to a virtual machine's definition. Make sure the provided ```uuid``` matches the previously generated UUID.
|
|
|
|
```
|
|
<domain>
|
|
[...]
|
|
<device>
|
|
[...]
|
|
<hostdev mode="subsystem" type="mdev" managed="no" model="vfio-pci" display="on" ramfb="on">
|
|
<source>
|
|
<address uuid="7686131b-b229-4768-a02c-35d1dbed7c66"/>
|
|
</source>
|
|
<address type="pci" domain="0x0000" bus="0x09" slot="0x00" function="0x0"/>
|
|
</hostdev>
|
|
[...]
|
|
</device>
|
|
[...]
|
|
</domain>
|
|
```
|
|
|
|
> Notice that the RAMFB is set to on, which activates Drect Memory Access Buffers (DMA-BUFs), making available the output of a virtual monitor before the guest operating system takes over
|
|
{.is-info}
|
|
|
|
## Configure Spice / SDL
|
|
|
|
*To-do*
|
|
|
|
## Remove any video device
|
|
|
|
* Remove any video device such as `virtio-gpu` and set the last one to the `none`.
|
|
|
|
|
|
```
|
|
<domain>
|
|
[...]
|
|
<device>
|
|
[...]
|
|
<video>
|
|
<model type="none"/>
|
|
</video>
|
|
[...]
|
|
</device>
|
|
[...]
|
|
</domain>
|
|
```
|
|
|
|
* Then starts the domain
|
|
|
|
## Resources
|
|
|
|
* Official page for vfio-mdev: https://www.kernel.org/doc/html/latest/driver-api/vfio-mediated-device.html
|
|
* Archlinux's entry on Intel GVT-g: https://wiki.archlinux.org/title/Intel_GVT-g
|
|
* DMA-BUF Linux documentation: https://www.kernel.org/doc/html/latest/driver-api/dma-buf.html
|