You are not logged in.

#1 2022-01-12 14:47:41

Bitals
Member
Registered: 2021-08-07
Posts: 9

Perfomance problems with kernels newer than 5.10

Hello everyone,

I have actually had this problem since the 5.11 release, but I just kept using the LTS kernel because I didn't have time to investigate deep enough (or I was just lazy, but it couldn't possibly be true). It started on OpenSUSE Tumbleweed, but I've been on Arch since August, symptoms are exactly the same. Today the LTS kernel got updated to 5.15, I tried it out and the problem persists, so I guess it's too stupid to rollback and stop updating my kernel now.

Laptop: MouseComputer BC-MB1485UD11A-191
CPU: Intel Core i7-8565U
Audio server: Pulseaudio
Kernel: 5.15.14-1-lts

So, the problem. While running the 5.11+ kernel (5.15.14-1-lts at the moment), my system randomly gets very slow. Sometimes it happened right after boot, sometimes after wake from suspend, sometimes while I was working. But I think this actually got better with new updates, at least to the point when it no longer makes my laptop literally unusable for about 5-15 minutes. Before the touchpad was lagging, Wi-Fi was not connecting and so on.
What didn't get better is sound. My main examples are the YouTube videos and the Zoom calls, because they are usually long enough for this to occur. Sound suddenly starts lagging, freezing, "quacking" and such. With 5.11 there were errors like

CPU Bug - soft lockup...

while lagging and almost always on shutdown, but I don't see them now while my sound is chopping like crazy. In fact I see no errors in dmesg, except for the intel-undervolt triggered ones, and disabling it made no difference. It's just an informative message anyway.

[Wed Jan 12 17:16:43 2022] msr: Write to unrecognized MSR 0x610 by intel-undervolt (pid: 5280).
[Wed Jan 12 17:16:43 2022] msr: See https://git.kernel.org/pub/scm/linux/kernel/git/tip/tip.git/about for details.

So where should I look? My desktop PC with Intel Core i7-3770k runs the latest non-LTS kernel just fine, but my laptop can't work reliably on anything newer than 5.10. Please help me fix this.

Edit:
On second thought, I did find some segfaults in dmesg. Not sure they would tell anything, I didn't for example explicitly launch the gnome-documents at that time, but better than nothing I guess.
dmesg

Last edited by Bitals (2022-01-12 15:17:35)

Offline

#2 2022-01-12 15:20:25

seth
Member
Registered: 2012-09-03
Posts: 26,521

Re: Perfomance problems with kernels newer than 5.10

1. why are you https://wiki.archlinux.org/title/Undervolting_CPU ?
What if you don't?
2. Do you https://wiki.archlinux.org/title/Microc … ly_loading ?
3. Does something charge CPU, RAM or IO (top/iotop)?
4. What cpu governor do you use and where's the frequency when it's fast/slow?
https://wiki.archlinux.org/title/CPU_fr … _governors
5. Please post a complete system journal

Edit: the dmesg is an excerpt only, but intel-undervolt is *very* suspicious here.

Last edited by seth (2022-01-12 15:22:03)

Offline

#3 2022-01-12 15:33:41

Bitals
Member
Registered: 2021-08-07
Posts: 9

Re: Perfomance problems with kernels newer than 5.10

seth wrote:

1. why are you https://wiki.archlinux.org/title/Undervolting_CPU ?
What if you don't?

To actually have some battery life. Without it my laptop lived for about 2 hours under regular load (just multitasking, not rendering or compiling or something like that), which was unacceptable.

Yes.

seth wrote:

3. Does something charge CPU, RAM or IO (top/iotop)?

By charge do you mean utilizes to 100%? No, nothing does, first thing I checked.

seth wrote:

4. What cpu governor do you use and where's the frequency when it's fast/slow?
https://wiki.archlinux.org/title/CPU_fr … _governors

Powersave most of the time. 900-1900MHz usually, same on 5.10 with no problems. I am not talking about less than 60fps in Gnome animations, they were pretty slow on powersave forever, but sound was always ok on older kernels.

seth wrote:

5. Please post a complete system journal
Edit: the dmesg is an excerpt only, but intel-undervolt is *very* suspicious here.

journalct

?
Trying to disable intel-undervolt was the second thing I tried after looking at the CPU load, and it only changed the chassis temperature.

Offline

#4 2022-01-12 15:36:05

seth
Member
Registered: 2012-09-03
Posts: 26,521

Re: Perfomance problems with kernels newer than 5.10

sudo journalctl -b | curl -F 'f:1=<-' ix.io

if this has happened during the present boot, otherwise … "-b -1" for the previous boot etc.

Offline

#5 2022-01-12 15:42:18

Bitals
Member
Registered: 2021-08-07
Posts: 9

Re: Perfomance problems with kernels newer than 5.10

seth wrote:
sudo journalctl -b | curl -F 'f:1=<-' ix.io

if this has happened during the present boot, otherwise … "-b -1" for the previous boot etc.

https://mega.nz/file/Uw8C0DaC#anEaxCVj89zAIfesvAOd9hLZeXkcts_ksoxrmYesI74
curl gave no output, so mega again.

Offline

#6 2022-01-12 15:53:59

seth
Member
Registered: 2012-09-03
Posts: 26,521

Re: Perfomance problems with kernels newer than 5.10

Quota, it's full of zeitgeist/tracker crashes and some game (debug?) logs.
Speaking of which - what if you remove/disable zeitgeist and tracker. It constantly crashing and restarting could very much cause cpu or io stalls.

Offline

#7 2022-01-12 16:01:53

Bitals
Member
Registered: 2021-08-07
Posts: 9

Re: Perfomance problems with kernels newer than 5.10

seth wrote:

Quota, it's full of zeitgeist/tracker crashes and some game (debug?) logs.
Speaking of which - what if you remove/disable zeitgeist and tracker. It constantly crashing and restarting could very much cause cpu or io stalls.

Killed both zeitgeist and tracker, even uninstalled them. No changes.

Offline

#8 2022-01-12 16:30:04

Bitals
Member
Registered: 2021-08-07
Posts: 9

Re: Perfomance problems with kernels newer than 5.10

Bitals wrote:

With 5.11 there were errors like

CPU Bug - soft lockup...

while lagging and almost always on shutdown, but I don't see them now while my sound is chopping like crazy.

Yay, I got it right now while rebooting.
soft lockup
Edit:
iptables are not to blame for sure, it's a different app each time. It's only consistent with audio server, everything else is at mercy of random.
Also it will just hang there until I hard reset the laptop.

Last edited by Bitals (2022-01-12 16:43:48)

Offline

#9 2022-01-12 16:55:33

seth
Member
Registered: 2012-09-03
Posts: 26,521

Re: Perfomance problems with kernels newer than 5.10

What if you limit the c-state, https://wiki.archlinux.org/title/Intel_ … ete_freeze (ignore the context)

Offline

#10 2022-01-12 17:09:09

Bitals
Member
Registered: 2021-08-07
Posts: 9

Re: Perfomance problems with kernels newer than 5.10

seth wrote:

What if you limit the c-state, https://wiki.archlinux.org/title/Intel_ … ete_freeze (ignore the context)

Did that through grub config file and rebooted. Right now I need to charge my laptop, will test for some time.

Last edited by Bitals (2022-01-12 17:26:18)

Offline

#11 2022-01-12 17:42:03

Bitals
Member
Registered: 2021-08-07
Posts: 9

Re: Perfomance problems with kernels newer than 5.10

seth wrote:

What if you limit the c-state, https://wiki.archlinux.org/title/Intel_ … ete_freeze (ignore the context)

Same thing, just a hotter laptop.

Offline

#12 2022-01-13 12:02:20

Bitals
Member
Registered: 2021-08-07
Posts: 9

Re: Perfomance problems with kernels newer than 5.10

seth wrote:

Quota, it's full of zeitgeist/tracker crashes and some game (debug?) logs.
Speaking of which - what if you remove/disable zeitgeist and tracker. It constantly crashing and restarting could very much cause cpu or io stalls.

Hello,
Just found out there are separate packages tracker and tracker3. Tracker was harmless to remove, while half of my system depends on tracker 3. Stopped it for now, will look how it works out. But I'm pretty sure something will restart it sooner or later.

Offline

#13 2022-01-13 13:55:42

Bitals
Member
Registered: 2021-08-07
Posts: 9

Re: Perfomance problems with kernels newer than 5.10

Bitals wrote:
seth wrote:

What if you limit the c-state, https://wiki.archlinux.org/title/Intel_ … ete_freeze (ignore the context)

Same thing, just a hotter laptop.

But battery life didn't seem to decrease significantly, so maybe I was just seeing things. I think that Baytrail might've used a different module to control CPU p-states, so maybe this parameter didn't do anything. I will check this when my workday is over.

Offline

Board footer

Powered by FluxBB