https://bugs.kde.org/show_bug.cgi?id=520084
Bug ID: 520084
Summary: Excessive DRM_IOCTL_MODE_ATOMIC affecting input
latency
Classification: Plasma
Product: kwin
Version First 6.6.4
Reported In:
Platform: NixOS
OS: Linux
Status: REPORTED
Severity: normal
Priority: NOR
Component: general
Assignee: [email protected]
Reporter: [email protected]
Target Milestone: ---
DESCRIPTION
I've been measuring desktop latency on KWin Wayland, and noticed it degrading
in two scenarios:
1. Bursty input from a mouse (amplitude of the spike seems correlated with the
polling rate of the device)
2. A specific app being composited (not-minimized and on the current virtual
desktop). The app is the Zed editor (zed.dev)
I don't know what the root cause is, but I've found a pretty good proxy metric
for the degradation, hence naming a specific IOCTL in the title:
```console
$ kwincg="/sys/fs/cgroup$(systemctl --user show -p ControlGroup --value
plasma-kwin_wayland.service)"
$ sudo bpftrace -e '
tracepoint:syscalls:sys_enter_ioctl
/cgroup == cgroupid("'$kwincg'") && args->cmd == 0xc03864bc/
{
$flags = *uptr((uint32 *)args->arg);
@mode_atomic[$flags & 0x100 ? "test" : "commit"] = count();
}
interval:s:1
{
print(@mode_atomic);
clear(@mode_atomic);
}
'
```
Specifically: on an idle desktop, at 120Hz refresh rate (VRR disabled), it
bottoms out at ~4 of these ioctls with the ATOMIC_TEST_ONLY flag and ~4
without, per second. With the Zed window shown, it jumps to its max sustained
rate of 120/s with the flag and 120/s without it.
When this happens, the click-to-photon latency floor jumps higher, and the
distribution of measurements gets wider as well.
A similar impact is felt whenever I start to move my mouse (the bursty behavior
I described earlier):
```
# idle desktop baseline:
@mode_atomic[commit]: 4
@mode_atomic[test]: 4
@mode_atomic[commit]: 4
@mode_atomic[test]: 4
@mode_atomic[commit]: 4
# start of continuous movement, ioctl rate overshoots:
@mode_atomic[commit]: 127
@mode_atomic[test]: 1839
@mode_atomic[commit]: 128
@mode_atomic[test]: 947
@mode_atomic[test]: 120
# movement continues, but now the rate of ioctls stabilizes
@mode_atomic[commit]: 120
@mode_atomic[test]: 120
@mode_atomic[commit]: 120
# movement stops, going back to idle:
@mode_atomic[test]: 64
@mode_atomic[commit]: 65
@mode_atomic[test]: 5
@mode_atomic[commit]: 5
```
STEPS TO REPRODUCE
1. Start monitoring ioctls with bpftrace
2. Open a Zed editor window
3. Observe the sustained rate maxing out
OBSERVED RESULT
While the zed window is being composited:
```
@mode_atomic[commit]: 120
@mode_atomic[test]: 120
```
EXPECTED RESULT
Should stay at a low level while idle, same as with every other app's window
not causing this behavior:
```
@mode_atomic[commit]: 4
@mode_atomic[test]: 4
```
SOFTWARE/OS VERSIONS
Operating System: NixOS 26.05
KDE Plasma Version: 6.6.4
KDE Frameworks Version: 6.25.0
Qt Version: 6.11.0
ADDITIONAL INFORMATION
No such effect, as far as I can tell, is present in full-screen apps.
--
You are receiving this mail because:
You are watching all bug changes.
