Intel GVT-g random kernel panics

BobMccapherey

Member
Apr 25, 2020
33
0
6
42
I'm running into the same issues as the poster here https://github.com/intel/gvt-linux/issues/153

Running newest version of Proxmox with all latest non-subscription updates with Windows VMs with Intel GVT-g active. VMs seem to crash at random times with crash frequency increasing on heavy GPU loads or larger number of VMs active. My stack trace looks similar to the above github issue. I'm also on a Coffee Lake - generation Xeon processor with integrated Intel GPU.

Boot parameters are:
GRUB_CMDLINE_LINUX_DEFAULT="quiet intel_iommu=on i915.enable_dc=0 i915.enable_guc=0 i915.enable_gvt=1 kvm.ignore_msrs=1 kvm.report_ignored_msrs=0"

How quickly are kernel patches rolled into the mainline Proxmox kernel? Is there a way to run a 5.7 or 5.8 branch kernel for testing?
 
Unfortunately these types of bugs have been what's preventing me from paying for support as I need stable GVT support for Coffee Lake. It really sucks since I'm a fan of Proxmox's Ceph hyper-converged concept.
 
Unfortunately these types of bugs have been what's preventing me from paying for support as I need stable GVT support for Coffee Lake.

So you do not pay for support when you need it and you wait till all your issues are fixed and you do not need support anymore?

Re-think about your approach ...
 
Is there a way to see which version of the kernel the subscription version is using? If it lags behind stuff a bit but doesn't constantly get blown up that would be a reason to subscribe.
 
GVT-g is really new, niche, and not exactly a common thing (at least not right now) in the server space. Is it really that shocking this doesn't "100% work"? I've not read many virtualization platforms supporting it at all. Not even bleeding edge linux distro's.
 
BTW, I'm not blaming Proxmox for taking the kernel put together by a lot of other folks. Just curious how long it might take to get to 5.7 where there are some fixes.
If it is a long way I can stop doing GVT I suppose and just pass the entire thing into the VM. Would probably be more stable. Such a cool concept to not waste the entire GPU when the VM just needs a little bit of it.
 
So you do not pay for support when you need it and you wait till all your issues are fixed and you do not need support anymore?

Re-think about your approach ...

I'm not blaming Proxmox but what probably would have been a more helpful response is actually clarifying whether support would include offering a 5.7 kernel to a customer or if this is out of scope.
 
GVT-g is really new, niche, and not exactly a common thing (at least not right now) in the server space. Is it really that shocking this doesn't "100% work"? I've not read many virtualization platforms supporting it at all. Not even bleeding edge linux distro's.

xcp-ng (Basically the open source equivalent of Citrix XenServer) supports GVT-g via the xengt kernel modules (similar to how Proxmox uses the kvmgt modules). This would imply that Citrix XenServer has support as well. Maybe not for newer processor architectures but definitely for Broadwell/Haswell/etc.
 

About

The Proxmox community has been around for many years and offers help and support for Proxmox VE, Proxmox Backup Server, and Proxmox Mail Gateway.
We think our community is one of the best thanks to people like you!

Get your subscription!

The Proxmox team works very hard to make sure you are running the best software and getting stable updates and security enhancements, as well as quick enterprise support. Tens of thousands of happy customers have a Proxmox subscription. Get yours easily in our online shop.

Buy now!