QEMU 9.2 available on pvetest and pve-no-subscription as of now

stanthewizzard2025 said:
Hello
Is there a need to reboot ? PVE or vm or none ?
Thanks
Click to expand...
You need to fully restart the VMs (not just a reboot from inside the VM OS) or live-migrate them to another host that eas already updated so that the VM uses the newer QEMU executable that includes the fix.
 
Hi guys,

are you sure it's fixed in `pve-qemu-kvm=9.2.0-3` ?

I have home assistant VM, doing mostly nothing, idling. (2 vCPU)

- On my older `pve-qemu-kvm=9.0.2-5` it was idling on `0.6-1.2%`
- After upgrade to `pve-qemu-kvm=9.2.0-2` idle is `2.5-3%`
- With `pve-qemu-kvm=9.2.0-2` and `qm set {id} --args '-machine hpet=off'` idle back to `0.6-1.2%`
- After upgrade to `pve-qemu-kvm=9.2.0-3` idle is again `2.5-3%` (any hpet=off do not work now)

So i went again back to `pve-qemu-kvm=9.0.2-5` and it's stable `0.6-1.2%`.

After every upgrade/downgrade or set hpet i rebooted the VM in proxmox.

proxmox.jpg
 
Hi,
raz said:
That note should be added to the changelog for 8.4 under Known Issues & Breaking Changes. One of my systems requires the pass-through of an Intel iGPU and broke with the update to 8.4.
Click to expand...
thank you for pointing this out. It was added to the known issues section there too now.
 
  • Like
Reactions: Johannes S
Hi,
jackLevendon said:
I know your frustations and is very hard to keep up with that..
Click to expand...
please elaborate what your referring to. Such statements are not really helpful to anybody otherwise.
 
My AMD GPU, which is passed through, has not been recognized since an update. The GPU is passed through, but the AMD drivers show error 43. I have also tried different versions of QEMU, but unfortunately, the problem still persists. Does anyone have a solution?
 
Hi,
vigorianer said:
My AMD GPU, which is passed through, has not been recognized since an update. The GPU is passed through, but the AMD drivers show error 43. I have also tried different versions of QEMU, but unfortunately, the problem still persists. Does anyone have a solution?
Click to expand...
have you tried booting/installing different kernel versions? Does the issue occur with both kernel 6.8 and the opt-in 6.14? You can check /var/log/apt/history.log to see which packages were upgraded during the problematic upgrade. Are there any messages in the system logs/journal? Please share your VM configuration qm config <ID> and output of pveversion -v
 
Here is the requested data. I don't always have my HDRaid turned on, so that shouldn't be a problem. I did a complete update, but unfortunately it doesn't go back that far. I hope you can help me.
 

Attachments

I'm using PVE 8.4.5 with lastest version of qemu.

My Home Assistant VM is also keeping the host CPU to acheive any C-states.

It was created using the helper script by Tteck, machine i440fx.

I've tried switching to q35, applied --args '-machine hpet=off' but my cpu still can't do any c-states.
 
I’m seeing what looks like the same issue.


Yesterday I upgraded to Proxmox 9 and also did a fresh install of Proxmox 9 on a second host.
As soon as I start a Windows VM (q35; tested with QEMU 9.0 / 9.2 / 10), the host will not enter any package C-states (pc2–pc10 = 0 %). With a Linux VM running, the host reaches pc6/pc8 without problems.


Tested without success:
  • Full VM shutdown and cold start after QEMU upgrade; various machine revisions (pc-q35-9.0 / 9.2+pve1 / 10.x).
  • CPU model host; also tested x86-64-v2/v3; SMT and core count varied; fresh minimal Windows VM (1 vCPU, 2 GB RAM) without extra software.
  • QEMU args: -machine hpet=off, -global kvm-pit.lost_tick_policy=discard.
  • Host KVM halt polling: set to 0 and reverted to defaults (halt_poll_ns/grow/shrink).
  • Windows guest: Balanced power plan with min CPU lowered, ASPM enabled, bcdedit /deletevalue useplatformclock, HPET device disabled, high-res timer apps closed.





    Maybe someone here still knows a trick or workaround I haven’t tried yet

 
You must log in or register to reply here.
Top Bottom