Remote desktop on Gnome hangs randomly

ovicz

Hello. I'm running a Rocky Linux 10 vm on xcp-ng latest version with gpu passthrough enabled. The host is a TOPTON mini-pc with intel Alder Lake N100 processor.

The problem is that at random times the RDP session freezes and to make it work again I have to restart the vm. The VM is running with no issues when this occurs but only the gui is not working. If I run with software rendering there are no problems at all, but that's not ideal.

The passthrough drivers are loaded correctly but something is not right.

lspci
00:00.0 Host bridge: Intel Corporation 440FX - 82441FX PMC [Natoma] (rev 02)
00:01.0 ISA bridge: Intel Corporation 82371SB PIIX3 ISA [Natoma/Triton II]
00:01.1 IDE interface: Intel Corporation 82371SB PIIX3 IDE [Natoma/Triton II]
00:01.2 USB controller: Intel Corporation 82371SB PIIX3 USB [Natoma/Triton II] (rev 01)
00:01.3 Bridge: Intel Corporation 82371AB/EB/MB PIIX4 ACPI (rev 03)
00:02.0 VGA compatible controller: Device 1234:1111 (rev 02)
00:03.0 Unassigned class [ff80]: XenSource, Inc. Xen Platform Device (rev 01)
00:08.0 Audio device: Intel Corporation Alder Lake-N PCH High Definition Audio Controller
00:09.0 VGA compatible controller: Intel Corporation Alder Lake-N [UHD Graphics]

lsmod | grep i915
i915 4677632 28
cec 69632 2 xe,i915
intel_gtt 24576 1 i915
drm_buddy 28672 2 xe,i915
i2c_algo_bit 20480 2 xe,i915
drm_display_helper 286720 2 xe,i915
video 81920 2 xe,i915
ttm 110592 4 drm_vram_helper,drm_ttm_helper,xe,i915

I've tried even with a newer distro as Debian Testing but it's the same result.

ovicz

If I pass the option GRUB_CMDLINE_LINUX_DEFAULT="i915.enable_guc=0 quiet" the gui doesn't freeze totally but still lags.

ph7

@ovicz
Perhaps upgrade bios ???

ovicz

I'm afraid there aren't any bios updates...the device is a chinese mini pc from aliexpress. I've spoken with the seller and he told there aren't any, at least not official ones. I don't think it's a bios issue, but something regarding the passthrough on the newer devices . maybe the kernel from xcp is too old, although it's heavily patched.

ovicz

Setting pci=nomsi seems to do the trick for now. Probably not optimal but it's just a VM and I didn't notice any performance issues.

olivierlambert

Adding @Team-OS-Platform-Release in case that rings a bell

yann

Since GPU passthrough is involved I'd ping @Team-Hypervisor-Kernel

ovicz

If i remove the pci=nomsi from grub..this pops up in xl dmesg.

(XEN) [15627.724045] CPU3: Running in modulated clock mode
(XEN) [15628.457337] d[IDLE]v0: Unsupported MSI delivery mode 7 for Dom5
(XEN) [15628.457339] d[IDLE]v2: Unsupported MSI delivery mode 7 for Dom5

That's when the gnome desktop hangs. Is this fixable ?

dinhngtu

@ovicz Are you up for a test package with a fix? Note that the fix is not yet tested nor supported at the moment.

Add user repos using this command:

yum-config-manager --add-repo https://koji.xcp-ng.org/repos/user/8/8.3/xcpng-users.repo

To know what you need to upgrade, run this command (as xcp-ng-ndinh2 contains some changes that you don't need):

rpm -qa --qf '%{NAME}\t%{SOURCERPM}\n' | grep $'\txapi'

Then upgrade just the XAPI-related packages:

yum update --enablerepo=xcp-ng-ndinh2 <list of packages from above>

Remember to reboot the host after installing.

Finally, disable hvm-pirq on your affected VM and reboot it (parameter name subject to change):

xe vm-param-set uuid=... platform:hvm-pirq=false

To go back, list what you installed and downgrade them:

yum list installed | grep ndinh
yum downgrade ...

ovicz

Hello. I did what you said and it seems to work. Thank you.

dinhngtu

@ovicz Thanks for testing. Is everything working without the pci=nomsi or GRUB command line tweaks? We'll work to integrate this fix into a future update.

ovicz

Yes. I removed pci=nomsi entry from grub. I've tested on a Debian Forky and CentOS Stream 10 VMs with gnome and an intel alderlake igpu passthrough no issues at all.

dinhngtu

@ovicz xapi-26.1.16-1.1.0.pirq.2.xcpng8.3 is now available on xcp-ng-ndinh2, which disables HVM PIRQ for all existing VMs by default on reboot. Please try it out.

I've prepared an upstream PR for this feature. If you'd like to be listed as a tester, please respond with your name and email address (or your fake github email if you prefer) and I'll update my commit with your credit.

ovicz

@dinhngtu Hello. I've updated to the latest commit. There is no need to give me credit. I just want to help so others and myself included can benefit from this. Anyway thank you for your work.

ovicz

IF I update to the latest batch of updates released recently, I suppose I will lose the changes made in that repo that you provided. ?

dinhngtu

@ovicz I've built xapi-26.1.16-1.2.0.pirq.1.xcpng8.3 on xcp-ng-ndinh2. You can update as usual, then install my packages on top.