KB4480963 breaks Hyper-V Live Migration on AMD Opteron 6376

Topic

New Reply

We’re quite aggressive in our monthly patching campaign: 8 days after Patch Tuesday we’re fully patched (3000 Windows Servers). I’m quite proud of that actually 🙂

This month however we’ve hit a hiccup with KB4480963 breaking the Live Migration feature on our 2 oldest Win2012R2 Hyper-V clusters. These are 2x 4 Dell R715 hosts with AMD Opteron 6376 CPU’s (yep, they’re old…).
The VMs themselves are still running fine though. The problem’s limited to Live Migration which is unusable, thus no more servicing of the hosts without bringing the VMs down…
Thought I’d share the knowledge and spare others the trouble, cause Microsoft hasn’t come up with a fix for it yet.

I had contacted our HPE support guys 3 weeks ago, who in turn quite quickly escalated it to MS. At first the MS engineer didn’t even acknowledge it as a potential issue with KB4480963 as this hadn’t been reported before, and just referred to the security advisory for enabling the AMD specific Spectre/Meltdown mitigations, which are new since this KB.
Well, somebody has to be the first to report a bug, right?

Our HPE support guy also confirmed me today he was able to repro the issue on an HPE server with an AMD CPU from the same generation. So, definitely a bug in KB4480963 I’d say, it’s not strictly in our environment.

I suppose there won’t be too many people running a combination of Win2012 R2 Hyper-V on AMD CPU’s with the Piledriver architecture. But for what it’s worth…

2 users thanked author for this post.

Elly, woody

Reply | Quote

Replies

Have you also reported this to patchmanagement.org?

Reply | Quote

no, I haven’t . I come on this site quite regularly, but not patchmanagment.org (although I think in the past I’ve seen it being referenced here, I think)

Reply | Quote

the product team has now officially published it as a known issue in KB4480963:

After installing this update, virtual machines may fail to restore after being “Saved and Restored” the first time. The error message is, “Failed to restore the virtual machine state: Cannot restore this virtual machine because the saved state data cannot be read. Delete the saved state data and then try to start the virtual machine. (0xC0370027).”

This affects AMD Bulldozer Family 15h, AMD Jaguar Family 16h, and AMD Puma Family 16h (2nd-gen) microarchitectures.

There should be a fix mid-Feb (so I hope they mean this evening in the Feb rollup)

2 users thanked author for this post.

Elly, woody

Reply | Quote

MS delivered as promised: Preview Rollup KB4487016 now mentions (amongst other things)
Addresses an issue that may cause virtual machines (VM) to fail to restore successfully if the VM has been saved and restored once before. The error message is, “Failed to restore the virtual machine state: Cannot restore this virtual machine because the saved state data cannot be read. Delete the saved state data and then try to start the virtual machine. (0xC0370027).” This affects AMD Bulldozer Family 15h, AMD Jaguar Family 16h, and AMD Puma Family 16h (second generation) microarchitectures.

However, for my support case I was first asked to try KB4490512, which should tackle the same problem.
I tried it on 2 of the involved hosts, and Live Migrations now work as expected again.

I assume the next step will be to install the Preview Rollup, but I’ve asked for their confirmation first.

1 user thanked author for this post.

woody

Reply | Quote

Thanks for staying on top of this one!

Reply | Quote