[REGRESSION] i915: s2idle lid-open resume leaves eDP panel black until an input event (Raptor Lake; works on 6.18 LTS)

[徐庆]

To: regressions@xxxxxxxxxxxxxxx

Cc: intel-gfx@xxxxxxxxxxxxxxxxxxxxx, dri-devel@xxxxxxxxxxxxxxxxxxxxx, linux-kernel@xxxxxxxxxxxxxxx

Subject: [REGRESSION] i915: s2idle lid-open resume leaves eDP panel black until an input event (Raptor Lake; works on 6.18 LTS)

Hi,

a regression report, following Documentation/admin-guide/reporting-regressions.rst:

Regression range: v6.18..v7.2 (last known good: linux-lts 6.18.49-2-lts;

first tested broken: 7.2.3-arch1-2; intermediate 7.0/7.1/7.2 untested)

Symptom:

After closing the lid (systemd-logind suspends to s2idle) and reopening it,

the machine resumes fine (kernel resume log clean, ~1 s, Wi-Fi and input

work), but the internal eDP panel stays on a static black frame. The screen

only lights up after any input event (key press or mouse move), which then

triggers the compositor to repaint. Reproducible on every lid cycle with

kernel 7.2.3; the identical userspace on linux-lts 6.18.49-2-lts does NOT

show the problem.

Reproduce:

1. Kernel 7.2.3, GNOME 50.4 Wayland.

2. Close lid -> "PM: suspend entry (s2idle)".

3. Wait ~5 s, open lid.

4. Keep hands off keyboard/mouse: screen stays black indefinitely.

5. Press any key: screen lights up instantly.

Environment:

- Laptop: HONOR GLO-GXXX (MagicBook), Intel Raptor Lake-P Iris Xe

  [8086:a7a0] (i915) driving the internal eDP panel; NVIDIA RTX 3050

  [10de:25ab] offload-only, not connected to the panel.

- Arch Linux, kernel 7.2.3-arch1-2 (Arch default config), GNOME Shell/mutter

  50.4 Wayland, gdm autologin.

- /sys/power/mem_sleep: "[s2idle] deep" (s2idle in use).

- uname -a: Linux archlinux 7.2.3-arch1-2 #1 SMP PREEMPT_DYNAMIC Thu, 03 Sep

  2026 17:55:06 +0000 x86_64 GNU/Linux

Measurements during the black period (so the panel is NOT off):

- intel_backlight brightness=96000/96000, bl_power=0 (backlight on);

  connector connected; lid open.

- i915 debugfs /sys/kernel/debug/dri/1/state: CRTC-0 active=1, planes

  scanning out continuously; no modeset when the screen finally lights up.

=> panel powered/backlit/scanned out all along; the compositor receives no

   event that would make it redraw after s2idle resume until user input.

Ruled out by testing: lid ACPI events (clean), session-freeze setting,

GNOME lock-on-suspend, backlight/panel power loss, NVIDIA involvement,

compositor restart, EC/firmware lid handling (6.18 LTS on same machine is

fine). "deep" (S3) wakes the screen but the machine hard power-cycles ~15 s

after S3 resume (separate firmware issue), so s2idle is the required mode.

Suspect area: i915 s2idle-resume / frame-presentation changes between v6.18

and v7.2 affecting Raptor Lake (PSR/self-refresh, async flip, or vblank

handling after resume). Willing to run drm.debug traces, bisect, or test

patches.

Full journal excerpts and sampling data are attached.

#regzbot introduced: v6.18..v7.2

Attachment: s2idle-regression-evidence.txt
Description: Binary data