[CentOS] kvm: C8 as guest on C6 host / huge delay while booting

Mon May 18 12:51:03 UTC 2020
Leon Fauster <leonfauster at googlemail.com>

Am 16.05.20 um 17:48 schrieb Kay Diederichs:
> On 5/16/20 5:09 PM, Leon Fauster via CentOS wrote:
>> Am 11.05.20 um 15:59 schrieb Leon Fauster:
>>> Since C8.1  kvm guests have a huge delay while booting on a kvm host based on C6. This delay was not present with C8.0. The "pause" happend
>>> direct after the grub step. The VNC session shows only a "_" character.
>>>
>>
>> and this are the corresponding logs. The guest system continues to boot after ~5 minutes. At that point this appears
>>
>> May 16 16:44:14 ev kernel: kvm: 8458: cpu0 unhandled rdmsr: 0x140
>> May 16 16:44:14 ev kernel: kvm: 8458: cpu0 disabled perfctr wrmsr: 0xc2 data 0xffff
>> May 16 16:44:14 ev kernel: kvm: 8458: cpu1 unhandled rdmsr: 0x140
>> May 16 16:44:14 ev kernel: kvm: 8458: cpu2 unhandled rdmsr: 0x140
>> May 16 16:44:14 ev kernel: kvm: 8458: cpu3 unhandled rdmsr: 0x140
>> May 16 16:44:15 ev kernel: kvm: emulating exchange as write
>>
>> Any ideas?
>>
>> -- 
>> Leon
>> _______________________________________________
>> CentOS mailing list
>> CentOS at centos.org
>> https://lists.centos.org/mailman/listinfo/centos
> 
> looks like a mismatch between the hardware that the KVM guest is told to use, and the actual hardware provided by the host.
> 
> You can change the CPU type that the guest expects.
> 

Hi Kai,

I tested a couple of cpu configurations (/usr/libexec/qemu-kvm -cpu ?).
I also changed a overcommiting configuration but the behavior (delay 
after grub menu) still persists (for EL8 guests).

It seems that this have to do with Intel's "Uncore Performance 
Monitoring functionality" in the kernel and the log entries are 
classified as harmless (stated by RH).

So I think I better migrate the kvm host to EL8 ASAP.

Does anyone has already a kvm host in production? Any pitfalls?

--
Thanks
Leon