Closed rnd-ash closed 2 years ago
Try disabling full-screen unredirection. If that fixes the issue, then something's wrong with the unredirect code.
Tried that. It didn't help unfortunately...
Hmm.... May I see your kwinrc? Let's check if perhaps block compositing is on...
I completely forgot that 5.19.5 was being released today, but I am not releasing it until I fix this bug...
What I guess is that something is making the compositor do nothing for some time, which causes the wait for VBlank function to fail and wastes a core.
I completely forgot that 5.19.5 was being released today, but I am not releasing it until I fix this bug...
For what it's worth, I upgraded to 5.19.5 with the 5.19.4 lowlatency patch applied and it seems to work perfectly fine.
5.19.5 released...... without any fix to this problem due to rush from Gentoo side :<
This could also be an issue in the VBlank waiting code... because we use the "horrible hack" on Intel...
Still a problem in 5.19.5...
[$Version]
update_info=kwin.upd:replace-scalein-with-scale,kwin.upd:port-minimizeanimation-effect-to-js,kwin.upd:port-scale-effect-to-js,kwin.upd:port-dimscreen-effect-to-js,kwin.upd:auto-bordersize,kwin.upd:animation-speed
[Compositing]
LatencyControl=2
OpenGLIsUnsafe=false
WindowsBlockCompositing=false
[Desktops]
Id_1=a2a40f0a-7857-4aaf-9043-e3d6de46cc55
Id_2=c32d011e-1ff3-40e6-80f1-d8a24ed40a0b
Id_3=cffbea09-b132-4e8c-815f-5bbbb7370559
Id_4=003561ab-9768-4638-82d5-5eb5490b7892
Name_2=New Desktop
Name_3=New Desktop
Name_4=New Desktop
Number=4
Rows=1
[Effect-PresentWindows]
Accuracy=5
BorderActivate=7
BorderActivateAll=9
LayoutMode=1
[ElectricBorders]
TopLeft=None
[Plugins]
ParachuteEnabled=true
lattewindowcolorsEnabled=false
minimizeallEnabled=false
synchronizeskipswitcherEnabled=false
toggleparachuteEnabled=true
videowallEnabled=false
[TabBox]
BorderActivate=9
DesktopLayout=org.kde.breeze.desktop
DesktopListLayout=org.kde.breeze.desktop
LayoutName=org.kde.breeze.desktop
[org.kde.kdecoration2]
library=org.kde.breeze
theme=Breeze
Issue reproduces when VSync mechanism is set to "hope for the best" or "SGI video sync with horrible hack".
Now to see where is the program getting stuck at...
Issue does not reproduce on latest Git version. I am not sure what did I do, but it does not reproduce anymore.
@rnd-ash Are you able to test using the latest version from Git branch Plasma/5.19?
@tildearrow unfortunately not, I'm on 5.20 now. Will happily roll back to test though
can confirm same issue with 5.20 branch using NVIDIA or intel GPU
If anyone is bothered by the problem (like me having a laptop), I created a dirty workaround. Lockscreen might lag or even freeze (type in password and press ENTER, it will work anyway), but the workaround does the trick quite well.
First, install required dependencies: sudo apt install cpulimit
(same package name in many other distros, too)
This script should added to user autostart:
#!/bin/sh
sleep 10
dbus-monitor --session "type='signal',interface='org.freedesktop.ScreenSaver'" |
while read x; do
case "$x" in
*"boolean true"*) sh /home/lukas/locked.sh ;;
*"boolean false"*) sh /home/lukas/unlocked.sh ;;
esac
done
locked.sh :
#!/bin/sh
echo SCREEN_LOCKED; sleep 0.2
cpulimit -b -e kwin_x11 -l 6 &
unlocked.sh :
#/bin/sh
echo SCREEN_UNLOCKED
killall -u USERNAME cpulimit # change USERNAME to your user
killall will kill all cpulimit instances, so beware if you are using cpulimit for something else, too.
You may want to increase the number specified in cpulimit's -l option if your experience is suboptimal.
EDIT: small syntax change that for some reason is required in Manjaro, but not in *buntu.
This bug has been confirmed again, on Plasma 5.20. Now what happened that broke the wait code again? :<
Hi there. I've just installed 5.19.5 and have similar problem. If I play Firefox/Youtube or VLC in fullscreen mode (windowed mode is ok) - kwin eats 1 core with 90-100% of usage.
UP. It seems unchecked "Suspend compositor for full-screen windows" helps.
The bug also exist in vanilla kwin or is this what you talking of anyway?
The bug also exist in vanilla kwin or is this what you talking of anyway?
It's (or was) a bug exclusive to KWin-lowlatency.
Doesn't happen anymore on KWin-lowlatency 5.22/5.23 with the unredirection code rewrite.
Regardless on GPU (Intel/NVIDIA), when I lock my laptop 1 CPU core gets pegged at 100% by kwin_x11 process (Checked htop in another tty to verify)
Any debugging steps i can do?