[kwin] [Bug 524279] New: Night Light temperature transitions cause transient scanout corruption

https://bugs.kde.org/show_bug.cgi?id=524279

            Bug ID: 524279
           Summary: Night Light temperature transitions cause transient
                    scanout corruption
    Classification: Plasma
           Product: kwin
      Version First 6.6.4
       Reported In:
          Platform: Bazzite
                OS: Linux
            Status: REPORTED
          Severity: normal
          Priority: NOR
         Component: wayland-generic
          Assignee: [email protected]
          Reporter: [email protected]
  Target Milestone: ---

**Environment:**

* **KWin / Plasma:** 6.6.4 (Wayland)
* **GPU:** AMD RX 6800 (Navi 21)
* **Driver / Mesa:** amdgpu, Mesa 26.0.4
* **Kernel:** 6.17.7-ba29.fc43.x86_64
* **OS:** Fedora 43 (bazzite-deck)
* **Monitors:** Three DisplayPort outputs available; reproduces with any two
connected (including two native DP links).

**Description**

While KWin's Night Light color temperature is *changing*, the displays show
brief graphical corruption (flickering blocks of incorrect pixels). The
corruption stops immediately as soon as the temperature settles.

The corruption is **never present in a screen recording**, indicating it is
introduced during scanout rather than in the composited image.

It is far more frequent with two or more outputs enabled, but does still occur
- rarely and faintly - with a single output.

**Steps to reproduce**

1. Connect two monitors. Any two outputs reproduce this; 1920x1080@60 at scale
1.0, SDR, sRGB, no ICC profile is simply the most minimal configuration I
confirmed it on.
2. Ensure the cursor remains perfectly still to avoid introducing plane
changes.
3. Run the following DBus script to sweep color temperatures:

```bash
for t in 6500 6000 5500 5000 4500 4000 3500 3000 2500 2000 1500 1000 \
         1500 2000 2500 3000 3500 4000 4500 5000 5500 6000 6500; do
  qdbus6 org.kde.KWin /org/kde/KWin/NightLight org.kde.KWin.NightLight.preview
$t
  sleep 0.4
done
qdbus6 org.kde.KWin /org/kde/KWin/NightLight
org.kde.KWin.NightLight.stopPreview
```

**Expected Results:**
The color temperature changes smoothly with no visual artifacts.

**Actual Results:**
Transient corruption appears on the monitors throughout the transition,
generally manifesting within ~2 seconds of the script starting.

**Technical Investigation & CRC Verification**

The corruption is strictly tied to the color temperature transition, not the
application of a static transform. I verified this via CRTC CRC capture
(`/sys/kernel/debug/dri/*/crtc-*/crc`, source `auto`):

| Condition | Frames | Distinct CRCs |
| --- | --- | --- |
| Idle, no transform (6500K) | 3600 | 1 |
| Static transform held at 3000K | 600 | 1 |
| Static transform held at 3000K (repeat) | 480 | 1 |

There were zero dropped frames and the counters were fully consecutive. A
settled color transform produces a bit-identical image indefinitely.

**Variables Ruled Out**

* **Display bandwidth / prefetch headroom:** Reproduces with `min_ttu_vblank`
at 414.9 µs - clear of the ~414 µs `dram_clk_change` requirement, and roughly
7.5× the margin of the configuration where I first noticed the problem.
* **Refresh-rate mismatch:** Reproduces with both outputs at exactly 60 Hz.
* **Resolution / Scaling:** Reproduces at 1920x1080, scale 1.0.
* **GPU load:** `gpu_busy_percent` reads only 5–9% during the corruption.
* **Logs:** No underflows, no `flip_done` timeouts, and no DC errors are logged
during the event.

**Additional Context (Main Loop Stalls)**

`NightLight.preview()` is synchronous. With one output, the DBus round trip is
0.4 ms median / 0.6 ms max. With two outputs, I observed occasional excursions
to 13–31 ms. While testing a 3-output configuration, `Libinput: … event
processing lagging behind by 42ms` was repeatedly logged. Note: corruption
still occurs on setups with a 0.6 ms max round trip, so the main loop stall is
not a strict requirement for the visual artifact, but may be related.

I have `amdgpu_dm_dtn_log` captures and CRTC CRC traces from these
reproductions if they would be useful.

-- 
You are receiving this mail because:
You are watching all bug changes.