I've been daily driving a Win 10 Vm setup on Unraid for about 5 years now, and while occasionally something weird will pop up that I need to tinker with, it's mostly been a great experience. Recently, though, any other VM I try to run aside from my main passthrough one has become unusable. Both Windows and Linux VMs that have worked previously no longer work, and creating totally new VMs from templates and new vdisk installations doesn't help. 

 

Whenever I run a VM, the assigned CPU cores for that VM go to 100% and the Unraid webUI becomes super slow even through I don't have my first two cores pinned for VM usage and they show less than 100%. Eventually I have to force close the VM, but while the broken VM is running my passthrough VM starts to become slow, with the mouse visually lagging behind where I'm moving it and Windows becoming unresponsive. If I open task manager in the passthrough VM, once it loads I don't see high usage on any of the hardware while this is happening. This laggy behavior has started to happen at other times when using the passthrough VM, usually when I'm browsing the web with FIrefox and interacting with something like Claude or ChatGPT. I can't quite pin this down either though, as again both FF and Windows task manager do not show high hardware usage. I will say that that lag on the passthrough VM happens while another VM is running even if I have Firefox closed.

 

Running the other VMs with my passthrough vm off does not change the broken behavior. All of my appdata/system/vm data is on a Samsung 990 nvme that works great otherwise. The passthrough VM has 10 cores isolated and USB/NVME/GPU passed through at the IOMMU level. CPU governor is set to performance and I've tweaked many things over the years to minimize VM latency. Like I said, mostly things have been great. I do have a bunch of Docker containers running normally but the behavior persists even if I have the Docker service disabled.

 

Attached is a console screenshot from a previously working Mint installation, saying that there's a firmware bug and a register is already in use. This makes me think it's a memory problem or something but Memtest didn't give me any errors and this is the only place I'm experiencing issues really. Windows 10 just hangs and I don't get an error message, and a Windows XP vm that I had previously working has started to blue-screen. I have a 16 core 5950x and 128 GB of RAM with usage hovering around 65-70% so it's not a matter of resources.

 

It's hard to say when this issue started exactly, I can't quite pinpoint a hardware or settings change that I can undo. I'm on 7.0.1 now but t was also happening on the most recent 6 versions. I was hoping 7.0 and a Libvert or QEMU version bump may resolve it but so far that has not happened. I tried updating my passthrough vm virtio drivers as well just in case, but I also get an installation error there so I'm not sure what's up. It may be related, but probably not since the other VMs still have issues when this VM is not running.

 

If anyone has any ideas or can tell me where else in my settings to start poking around, I'd appreciate it!
 

bishop-diagnostics-20250310-0901.zip

I'm having similar issues with VMs recently on 7.0. No passthrough, just a load of VMs with a 3960X thradripper on ASRock TRX40 Creator. Windows VMs will start to freeze and start to stutter and not be as smooth as usual, and often ultimately lock up.  My fresh W11 VM is showing weird behavior as well. Remote desktop will connect to an existing logged in session, but load the pre-login screen - but not respond. But connecting via the VNC console and logging into an existing session that way seems to kick it back into life momentarily and you can login just fine, and then it freezes again. 

 

Previous W10 VM essentially broke when I increased it's RAM. It would drop to EFI shell and refuse to boot. After switching machine types around, nothing would recusitate it. I eventually copied the qcow2 disk and created a new VM pointing to the copied disk, and restored the VM, but the poor performance, stutter, and etc remained.

 

 

No memory over commitment, cpu cores are pinned and isolated (not shared). Just upgraded to 7.0.1 about 12 hours ago, we'll see how stable it is. Seems to initially be better on my machine.

Hi bitcore, thanks for the report. I'm really not sure what's up here.

 

I will say that I noticed kcompactd0 was consuming 100% of one of my cores pretty much permanently, I'm not sure if that has anything to do with this issue though. I'm trying to look into it but I can't find much info online. But yeah, same as you, no memory overcommitment and my iowait is low at around 0.1%.