[pve-devel] [RFC kernel] revert problematic TSC multiplier commit

[Fiona Ebner]

Am 14.09.22 um 11:38 schrieb Eneko Lacunza:
> Hi,
> 
> El 14/9/22 a las 10:40, Eneko Lacunza escribió:
>>
>> El 14/9/22 a las 9:50, Fiona Ebner escribió:
>>> Am 05.09.22 um 10:25 schrieb Eneko Lacunza:
>>>> I just confirmed that in addition to issue reported in
>>>> https://bugzilla.proxmox.com/show_bug.cgi?id=4073 (live migrated VM hung
>>>> using 100% CPU), we also reproduce issue reported in
>>>>
>>>> https://forum.proxmox.com/threads/zeitspr%C3%BCnge-in-vms-seit-pve-7-2.112756/ 
>>>>
>>>> At least what I understand google translating it :)
>>>>
>>>> - From 5.15.39-4-pve (Ryzen 2600X) to 5.15.39-4-pve (Ryzen 1700): Problems
>>>>   * Migration of a VM hung 4 other VMs in destination host (no 100% CPU use, consoles show time related issues, watchdog, etc.).
>>>>
>>>> We can test the patch if you can provide a kernel .deb package.
>>>>
>>> Hi,
>>> a kernel package for 5.19 is available in all repositories now [0]. The
>>> fix for the TSC multiplier issue is included there.
>>>
>>> [0]: https://forum.proxmox.com/threads/115090/
>>
>> Thanks, I'll try to test today and will report back.
> 
> I just made a kick test and 5.19.7-1 kernel continues to have issues for
> us (tried live migration of 5 VMs from 5.13 to 5.19, 4 VMs froze using
> 100% CPU).

IIUC, the migrated VMs themselves hung? Is the other issue with causing
time jumps and CPU stalls in *other* VMs now gone?

I'm not sure if a fix for the FPU-related issue is present in this
kernel or still outstanding, likely @Thomas knows?