Closed racerxdl closed 11 months ago
So it looks its not related to LG. I did some tunning (since it was correlated to CPU/GPU usage) and skipped the VM from using core 0 (and its logic thread) and removed the RR scheduler:
(...)
<cputune>
<vcpupin vcpu="0" cpuset="4"/>
<vcpupin vcpu="1" cpuset="5"/>
<vcpupin vcpu="2" cpuset="6"/>
<vcpupin vcpu="3" cpuset="7"/>
<vcpupin vcpu="4" cpuset="8"/>
<vcpupin vcpu="5" cpuset="9"/>
<vcpupin vcpu="6" cpuset="10"/>
<vcpupin vcpu="7" cpuset="11"/>
<emulatorpin cpuset="0-1"/>
</cputune>
(...)
This seens to solved the issue, but not sure why (the host machine doesnt have anything running)
Your CPU only has 6 performance cores, you're over allocating your system. Efficiency cores are useless for KVM usage and should be treated like they do not exist when configuring and pinning.
As this is not a LG bug but rather a support issue I am closing this. Please join the LG discord if you need further assistance.
Just for completeness if someone reads this in the future:
Your CPU only has 6 performance cores, you're over allocating your system. Efficiency cores are useless for KVM usage and should be treated like they do not exist when configuring and pinning.
As this is not a LG bug but rather a support issue I am closing this. Please join the LG discord if you need further assistance.
It is 6 performance cores with 2 threads each, which I correctly mapped/pinned:
CPU NODE SOCKET CORE L1d:L1i:L2:L3 ONLINE MAXMHZ MINMHZ MHZ
0 0 0 0 0:0:0:0 sim 4600,0000 400,0000 2700.000
1 0 0 0 0:0:0:0 sim 4600,0000 400,0000 2700.000
2 0 0 1 4:4:1:0 sim 4600,0000 400,0000 3912.706
3 0 0 1 4:4:1:0 sim 4600,0000 400,0000 2700.000
4 0 0 2 8:8:2:0 sim 4700,0000 400,0000 2700.000
5 0 0 2 8:8:2:0 sim 4700,0000 400,0000 2700.000
6 0 0 3 12:12:3:0 sim 4700,0000 400,0000 2700.000
7 0 0 3 12:12:3:0 sim 4700,0000 400,0000 2700.000
8 0 0 4 16:16:4:0 sim 4600,0000 400,0000 3680.392
9 0 0 4 16:16:4:0 sim 4600,0000 400,0000 1899.743
10 0 0 5 20:20:5:0 sim 4600,0000 400,0000 2763.059
11 0 0 5 20:20:5:0 sim 4600,0000 400,0000 2700.000
And in libvirt settings I was only passing 5 cores (which would give 10 threads).
<cpu mode="host-passthrough" check="none" migratable="on">
<topology sockets="1" dies="1" cores="5" threads="2"/>
</cpu>
So I was not over allocating cores. I specifically only pinned at P-Cores.
Still, the issue has been solved by skipping the first machine core and removing the vcpusched.
Also sorry for putting a support ticket, it looked like a bug when it happened.
Read the minimum requirements for LG, you ARE over-allocating your system https://looking-glass.io/docs/B6/requirements/#minimum
I'm facing some weird issues in my laptop with looking glass. Whenever there is a high cpu usage peak or gpu usage peak, the looking glass restarts the capture. Not sure if that is happening inside the vm or outside (I would guess inside). The issue appears a bit random but highly correlated to CPU/GPU usage. Sometimes I can play for hours without any issue, but it always happen when I start a game or windows is starting up.
I can easily simulating by opening space engineers. This is what happens:
Gravação de tela de 11-08-2023 12:57:09.webm
In my desktop (with a 3070TI as well) this issue does not happen.
Extra information which might be needed:
Virtual Machine libvirt config:
Bug Report Required Information
The log dont have anything special, just telling the capture restarted (no crash or anything). No related info in syslog, dmesg or vm log.
Client Log:
Host Log
Sometimes I get a random client crash: