High cpu usage with fcitx5 when screen is turned off via `niri msg`

[BRS5672023]

System Information

• niri version: niri 0.1.8 (unknown commit)

• Distro: Arch Linux

• GPU: Radeon 780M Graphics

• CPU: AMD Ryzen 7 8845H

I noticed this issue when I locked the screen and wait until the screen is off, then the cpu fan ramped up after a couple of seconds, but if I then wake up the screen, the cpu usage goes down as normal.. I found something goes wrong with fcitx5 using ssh on my phone..

This issue doesn't happen on hyprland though, and neither when I use the shortcut to turn off the screen Mod+Shift+P { power-off-monitors; }. But if I type niri msg output eDP-1 off within the terminal then the issue appears (and I don't know how to turn on the screen in this case).

One can reproduce this issue with niri msg invoked by swayidle, e.g., the following

timeout 600 'pidof swaylock || swaylock -f'
timeout 630 'niri msg output eDP-1 off' resume 'niri msg output eDP-1 on'
timeout 30 'pidof swaylock && niri msg output eDP-1 off' resume 'niri msg output eDP-1 on'
before-sleep 'pidof swaylock || swaylock -f'

[YaLTeR]

Hm, guess something is busylooping. If you can, please build and run niri with debuginfo and then when the issue occurs, record some seconds with:

perf record -F 99 --call-graph dwarf,65528 -p $(pidof niri)

Then make and upload here a flamegraph using something like:

perf script | inferno-collapse-perf | inferno-flamegraph > flamegraph.svg

(this uses commands from inferno)

It should look similar to this: https://github.com/YaLTeR/niri/assets/1794388/9d81ac17-50a6-4994-abd9-6d1b4debcb34

Sidenote, why are you turning off the monitor in swayidle rather than niri msg action power-off-monitors?

[YaLTeR]

Hm, this looks really weird honestly.

[mexisme]

This seems to be happening to me, as well.

Niri 0.1.9 Distro: NixOS GPU: AMD Radeon 780M CPU: AMD Ryzen 9 7940HS

Just tried it, using the Mod+P option to trigger the power-off-monitors action.

One additional weird thing is Niri can't seem to parse the EDID blob, as I get the following in the journal:

Sep 25 16:47:06 xxx niri[2579]: 2024-09-25T04:47:06.844301Z  WARN niri::backend::tty: error getting EDID info for eDP-1: error parsing EDID
Sep 25 16:47:06 xxx niri[2579]: Caused by:
Sep 25 16:47:06 xxx niri[2579]:     Parsing the EDID blob failed

[mexisme]

Actually, perhaps I am wrong.

Have just noticed that clipcatd seems like it might be the root cause of Niri having high CPU.

This seems to be happening to me, as well.

Niri 0.1.9 Distro: NixOS GPU: AMD Radeon 780M CPU: AMD Ryzen 9 7940HS

Just tried it, using the Mod+P option to trigger the power-off-monitors action.

One additional weird thing is Niri can't seem to parse the EDID blob, as I get the following in the journal:

Sep 25 16:47:06 xxx niri[2579]: 2024-09-25T04:47:06.844301Z  WARN niri::backend::tty: error getting EDID info for eDP-1: error parsing EDID
Sep 25 16:47:06 xxx niri[2579]: Caused by:
Sep 25 16:47:06 xxx niri[2579]:     Parsing the EDID blob failed

[FlyingWombat]

I used to get this issue, with v25.01. But as of commit 4c98b874862c2e6df7f71bdf36df0ba527690fbb it doesn't seem to happen anymore. That or some other update fixed it. I never pinned down exactly what the cause was.

Update: I also updated niri to v25.02 (as packaged by my distro, Arch, as opposed to building it myself from git), and don't encounter this issue.

[YaLTeR]

Huh, interesting. Guess this can be closed then?

[xbwwj]

can confirm same experience as FlyingWombat. The IME CPU problem no longer happens after I switched to git main build.

[YaLTeR]

There were some fixes to IME in Smithay which are included in niri v25.02. So maybe they also fixed this issue. Well in any case good to know