Intel iGPU passthough

xerxist

Yes I'm running 8.3 Beta

hawkpro

Is this just the mediated (gvt-g?) device passthrough, so the XCP side/server maintains video but a VM can make use of the resources as well?

I am very interested in this (plex, frigate type use) as a stepping stone away from Proxmox.

Thanks

flakpyro

In my testing of this, iGPU passthru works fine in Linux but in Windows the device will show an error in the device manager, disable/enabling the drive in the device manager will allow it to work, until next reboot.

CJ

@bullerwins @xerxist @flakpyro

What are you using for display output on the host since you're passing the iGPU to the VM?

flakpyro

@CJ Im running server grade hardware that has remote lights out management with iKVM support. Otherwise yeah you would lose access to the display output.

xerxist

@CJ

No output just need the Intel quick sync.

CJ

@xerxist Yes, QS for the VM but there are times when having a local display of the console comes in handy. That's why I haven't done mine.

Mefosheez

This post is deleted!

Mefosheez

@bullerwins said in Intel iGPU passthough:

@olivierlambert I tried but getting this error when turning on the VM

INTERNAL_ERROR(xenopsd internal error: (Failure
"Error from xenguesthelper: Populate on Demand and PCI Passthrough are mutually exclusive"))

Not sure what it means

EDIT: after googleing it seems that static and dynamic memory has to be the same:

@bullerwins thanks for this. I had to restore a VM from a backup and while it was down changed the dynamic value. This certainly saved me some time. I was positive it had to do with remapping the UID on the VM or something on the host side of it. anyways, thanks.

scawtydont

I followed this post and it worked for me.
One thing to note in my testing, if I had a monitor / HDMI dummy plug hooked up the transcoding worked reliably.
When I unplugged it I found activity would be present on the iGPU but the video in plex would play/pause constantly then eventually stop.

olivierlambert

@scawtydont Thanks for keeping us posted. Dummy plug is still a thing it seems

vhaelan

I'm also trying to pass through my Intel iGPU to a VM, but running into issues. It does work on Debian, however not on Fedora Server which I'd like to use. I have tried changing to BIOS mode, even plugged in a dummy HDMI plug, but nothing helped. SELinux is disabled on Fedora to rule out any issues related to that. I even tried using XCP-NG Center which was suggested in an older post, but that didn't help either: https://xcp-ng.org/forum/topic/6211/intel-gpu-passthrough-question/3

I also posted this on the Fedora forum with further details: https://discussion.fedoraproject.org/t/intel-igpu-in-containers/176029

Are there any other things I could try to investigate this further?

ovicz

@vhaelan Hello. What's the output of rpm -qa | grep intel-gpu-firmware ?

vhaelan

@ovicz said in Intel iGPU passthough:

rpm -qa | grep intel-gpu-firmware

Hello @ovicz, I see the following: intel-gpu-firmware-20251021-1.fc43.noarch

ovicz

@vhaelan Is the intel graphics module loaded ?

lsmod | grep i915
i915 5373952 24
drm_buddy 32768 2 xe,i915
ttm 135168 3 drm_ttm_helper,xe,i915
i2c_algo_bit 20480 2 xe,i915
drm_display_helper 331776 2 xe,i915
cec 106496 3 drm_display_helper,xe,i915
video 81920 2 xe,i91

sudo dmesg | grep i915
[ 5.855595] i915 0000:00:08.0: [drm] Found alderlake_p/alderlake_n (device ID 46d1) integrated display version 13.00 stepping D0
[ 5.858477] i915 0000:00:08.0: [drm] VT-d active for gfx access
[ 5.858739] i915 0000:00:08.0: [drm] Using Transparent Hugepages
[ 5.859307] i915 0000:00:08.0: Invalid PCI ROM header signature: expecting 0xaa55, got 0x4556
[ 5.859312] i915 0000:00:08.0: [drm] Failed to find VBIOS tables (VBT)
[ 5.860012] i915 0000:00:08.0: vgaarb: VGA decodes changed: olddecodes=io+mem,decodes=none:owns=io+mem
[ 5.878556] i915 0000:00:08.0: [drm] Finished loading DMC firmware i915/adlp_dmc.bin (v2.20)
[ 7.806109] i915 0000:00:08.0: [drm] [ENCODER:204:DDI A/PHY A] failed to retrieve link info, disabling eDP
[ 7.843768] i915 0000:00:08.0: [drm] GT0: GuC firmware i915/tgl_guc_70.bin version 70.49.4
[ 7.843776] i915 0000:00:08.0: [drm] GT0: HuC firmware i915/tgl_huc.bin version 7.9.3
[ 7.849282] i915 0000:00:08.0: [drm] GT0: HuC: authenticated for all workloads
[ 7.850868] i915 0000:00:08.0: [drm] GT0: GUC: submission enabled
[ 7.850874] i915 0000:00:08.0: [drm] GT0: GUC: SLPC enabled
[ 7.851225] i915 0000:00:08.0: [drm] GT0: GUC: RC enabled
[ 7.852797] i915 0000:00:08.0: [drm] Protected Xe Path (PXP) protected content support initialized
[ 7.854314] i915 0000:00:08.0: [drm] Registered 3 planes with drm panic
[ 7.854318] [drm] Initialized i915 1.6.0 for 0000:00:08.0 on minor 1

Also see in /etc/default/grub if there is NOMODESET in GRUB_CMDLINE_LINUX line.

That's from my Fedora 43 vm with passthrough enabled.

vhaelan

@ovicz Yes, it's loaded

lsmod | grep i915
i915                 5373952  0
i2c_algo_bit           20480  2 xe,i915
drm_buddy              32768  2 xe,i915
video                  81920  2 xe,i915
ttm                   135168  3 drm_ttm_helper,xe,i915
drm_display_helper    331776  2 xe,i915
cec                   106496  3 drm_display_helper,xe,i915

dmesg starts the same as your output, the only difference is that I have GuC disabled.

dmesg | grep i915
[    1.609000] i915 0000:00:08.0: [drm] Found alderlake_s (device ID 4690) integrated display version 12.00 stepping C0
[    1.609986] i915 0000:00:08.0: [drm] VT-d active for gfx access
[    1.610073] i915 0000:00:08.0: [drm] Using Transparent Hugepages
[    1.611342] i915 0000:00:08.0: Invalid PCI ROM header signature: expecting 0xaa55, got 0x4556
[    1.611345] i915 0000:00:08.0: [drm] Failed to find VBIOS tables (VBT)
[    1.612136] i915 0000:00:08.0: vgaarb: VGA decodes changed: olddecodes=io+mem,decodes=none:owns=io+mem
[    1.614455] i915 0000:00:08.0: [drm] Finished loading DMC firmware i915/adls_dmc_ver2_01.bin (v2.1)
[    3.583673] i915 0000:00:08.0: [drm] [ENCODER:261:DDI A/PHY A] failed to retrieve link info, disabling eDP
[    3.605422] i915 0000:00:08.0: [drm] GT0: GuC firmware i915/tgl_guc_70.bin version 70.49.4
[    3.605426] i915 0000:00:08.0: [drm] GT0: HuC firmware i915/tgl_huc.bin version 7.9.3
[    3.615402] i915 0000:00:08.0: [drm] GT0: HuC: authenticated for all workloads
[    3.615410] i915 0000:00:08.0: [drm] GT0: GUC: submission disabled
[    3.615414] i915 0000:00:08.0: [drm] GT0: GUC: SLPC disabled
[    3.616508] i915 0000:00:08.0: [drm] Protected Xe Path (PXP) protected content support initialized
[    3.617330] i915 0000:00:08.0: [drm] Registered 4 planes with drm panic
[    3.617354] [drm] Initialized i915 1.6.0 for 0000:00:08.0 on minor 1
[    3.627390] i915 0000:00:08.0: [drm] Cannot find any crtc or sizes
[    3.637349] i915 0000:00:08.0: [drm] Cannot find any crtc or sizes
[    3.657358] i915 0000:00:08.0: [drm] Cannot find any crtc or sizes

Having edited /etc/modprobe.d/i915.conf by adding options i915 enable_guc=3 enable_huc=1 fixed the GuC part as well, however I still have the same issue. Cannot find any crtc or sizes is the only part which is not in your output.

/etc/default/grub is clean, I don't have NOMODESET:

cat /etc/default/grub
GRUB_TIMEOUT=5
GRUB_DISTRIBUTOR="$(sed 's, release .*$,,g' /etc/system-release)"
GRUB_DEFAULT=saved
GRUB_DISABLE_SUBMENU=true
GRUB_TERMINAL_OUTPUT="console"
GRUB_CMDLINE_LINUX="rd.lvm.lv=fedora/root rhgb quiet"
GRUB_DISABLE_RECOVERY="true"