One clocksource available in (VM) PVE 9.1

[marigo]

Hi Guys,

I upgraded from 8 to 9 and notice a high(er) CPU load on one of my VM's.
After some investigation I saw three clocksources available on PVE8.x

# PVE8.x (inside KVM)

cat /sys/devices/system/clocksource/clocksource0/available_clocksource
tsc hpet acpi_pm

# PVE9.1 (inside KVM)

cat /sys/devices/system/clocksource/clocksource0/available_clocksource
acpi_pm

The 'acpi_pm' show a much higher CPU load than before.
I have already edited the GRUB boot parameters with: "clocksource=tsc tsc=reliable"
But it didn't do the trick.

I am missing something here?
Hopefully can someone help me.

 

[leesteken]

If you load the ptp_kvm module then you also get kvm-clock: https://forum.proxmox.com/threads/h...etween-qemu-guest-and-host.173396/post-806460
I don't know why tsc and hpet are no longer emulated. Maybe switching to an older machine type might bring them back. But I assume there are downsides to both if they have been removed.

EDIT: Your command is slightly wrong, it should be: cat /sys/devices/system/clocksource/clocksource0/available_clocksource

 

[marigo]

Thx for reply!
My PM host and VM both support the "constant_tsc" flag. I have no problem with time mgt as refered to NTP.

I will dig deeper into the problem. Maybe I have overlooked some setting.

EDIT:
I saw that my other VM's use tsc as the clocksource.
Switching to a newer kernel did the trick!