[Issue]: AMDGPU Display Core Timeout Causes GDM Greeter Crash on Idle

[SidShetye]

Problem Description

Apologies if this isn't the best repo for this amdgpu display core issue. Originally posted at https://gitlab.freedesktop.org/drm/amd/-/issues?show=eyJpaWQiOiI0OTIxIiwiZnVsbF9wYXRoIjoiZHJtL2FtZCIsImlkIjoxNDY1MDJ9 but it's been crickets there. Please let me know what's the best way to get this in front of the relevant AMD engineer(s).

Summary

Display fails to wake from idle/blank on AMD Strix Halo GPU. Keyboard/mouse input does not restore display; VT switch (Ctrl+Alt+F1/F2) required to recover. GDM greeter Wayland session crashes with "Child process was already dead" error. Root cause appears to be AMDGPU Display Core timeout in optc35_disable_crtc.

System Information

Hardware:

• GPU: AMD Strix Halo [Radeon Graphics / Radeon 8050S Graphics / Radeon 8060S Graphics] (rev c1)
  • PCI ID: c6:00.0 Display controller: Advanced Micro Devices, Inc. [AMD/ATI]
  • Subsystem: Device 2014:801d
• Display: Single HDMI-A-1 connected monitor (4K capable: 3840x2160)

Software:

• Distribution: Fedora 43
• Kernel: 6.18.6-200.fc43.x86_64 (and also now on 6.18.7-200.fc43.x86_64)
• Desktop: GNOME Shell 49.2 / Mutter 49.2 on Wayland
• GDM: 49.0 (or similar - Fedora 43 default)
• linux-firmware: linux-firmware-20260110-1.fc43.noarch
  Current Kernel Parameters:

BOOT_IMAGE=(hd1,gpt2)/vmlinuz-6.18.6-200.fc43.x86_64 root=UUID=9255c567-ae5b-44e7-b514-062bec516a4b ro rootflags=subvol=root rhgb quiet amdgpu.gttsize=126976 ttm.pages_limit=32505856

GNOME Settings:

• org.gnome.desktop.session idle-delay = 900 (15 minutes)
• org.gnome.desktop.screensaver lock-enabled = true
• org.gnome.desktop.screensaver lock-delay = 0

Symptom

When the display goes idle/blank (after 15 minutes of inactivity), keyboard and mouse input fail to wake the display or show the lock screen. The screen remains completely black with no backlight. The only recovery method is a VT switch (Ctrl+Alt+F1 or F2), which restores the display and allows login.

Steps to Reproduce

1. Boot Fedora 43 with GNOME Wayland session on AMD Strix Halo GPU
2. Wait for display to idle/blank (15 minutes idle-delay, or manually lock screen)
3. Attempt to wake display with keyboard or mouse input
4. Expected: Lock screen appears
5. Actual: Display remains black, no response to input
6. Press Ctrl+Alt+F1 or F2 (VT switch)
7. Actual: Display recovers, login prompt appears

Reproducibility: 100% - occurs every time display idles

Log Evidence

Boot-time AMDGPU Display Core Timeouts

From previous boot (2026-01-27 18:39):

Jan 27 18:39:37 vega kernel: amdgpu 0000:c6:00.0: [drm] REG_WAIT timeout 1us * 100000 tries - optc35_disable_crtc line:162
Jan 27 18:39:38 vega kernel: amdgpu 0000:c6:00.0: [drm] REG_WAIT timeout 1us * 100000 tries - optc35_disable_crtc line:165
Jan 27 18:40:02 vega gdm[1611]: Gdm: Child process -2003 was already dead.

Wake Failure Event (2026-01-28 09:07)

Timeline:

2026-01-28T09:07:23 - User attempts authentication (display still black, auth fails)
2026-01-28T09:07:26 - User performs VT switch (Ctrl+Alt+F1/F2)
2026-01-28T09:07:26 - Kernel: "fbcon: Taking over console"
2026-01-28T09:07:26 - Kernel: "Console: switching to colour frame buffer device 480x135"
2026-01-28T09:07:33 - Successful authentication after VT switch
2026-01-28T09:07:34 - GDM error: "Gdm: Child process -2104 was already dead."
2026-01-28T09:07:34 - New user Wayland session starts successfully
2026-01-28T09:07:34 - Old greeter session stopped (systemd[2086]: org.gnome.Shell@wayland.service)

GDM Logs:

2026-01-28T09:07:23-08:00 vega gdm-password][156601]: pam_unix(gdm-password:auth): authentication failure; logname= uid=0 euid=0 tty=/dev/tty1 ruser= rhost=  user=sid
2026-01-28T09:07:34-08:00 vega gdm[1615]: Gdm: Child process -2104 was already dead.

Kernel Logs (VT Switch Recovery):

2026-01-28T09:07:26-08:00 vega kernel: fbcon: Taking over console
2026-01-28T09:07:26-08:00 vega kernel: Console: switching to colour frame buffer device 480x135

Systemd Logs (Greeter Session Termination):

2026-01-28T09:07:34-08:00 vega systemd[2086]: Stopped target gnome-session-wayland@gnome-login.target
2026-01-28T09:07:34-08:00 vega systemd[2086]: Stopping org.gnome.Shell@wayland.service - GNOME Shell on Wayland...
2026-01-28T09:07:34-08:00 vega systemd[2086]: Stopped org.gnome.Shell@wayland.service - GNOME Shell on Wayland.
2026-01-28T09:07:34-08:00 vega systemd[2086]: org.gnome.Shell@wayland.service: Consumed 1min 49.375s CPU time, 403M memory peak.

Analysis

1. AMDGPU Display Core timeout occurs in optc35_disable_crtc (DCN 3.5 CRTC disable function)
2. This timeout appears to cause the GDM greeter Wayland compositor to crash
3. The crashed greeter leaves the display in a non-responsive state
4. VT switch forces a console framebuffer takeover, which breaks the deadlock
5. Pattern is consistent across multiple boots and wake events

Additional Context

• Strix Halo is very new hardware (2025 launch) - driver support may be incomplete in kernel 6.18
• No suspend/resume involved - this is purely an idle display blank/wake issue
• No coredumps generated (greeter may be timing out rather than segfaulting)
• Issue does NOT occur:
  • During normal boot
  • When manually switching VTs
  • In console-only mode (multi-user.target)
• Issue DOES occur:
  • Every time GNOME Wayland session idles/blanks display
  • Both automatic idle and manual lock screen trigger it

Workarounds Tested

1. amdgpu.runpm=0 - Previously added to cmdline, but removed in current boot (not present)
2. Increased GTT size and TTM limits (currently in use, no effect on this issue)

Potential Solutions to Test

1. Kernel/firmware updates - Strix Halo support may require kernel 6.19+ or linux-firmware updates
2. Enable DRM debugging - Add drm.debug=0x1e log_buf_len=4M to capture detailed DC debug logs
3. Switch to X11 temporarily - Bypass Wayland compositor issue (workaround only)
4. amdgpu.dc=0 - NOT viable: disabling Display Core results in no video output on modern GPUs

Questions for AMD Engineers

1. Is DCN 3.5 (optc35_disable_crtc) fully supported in kernel 6.18.6?
2. What is the expected timeout behavior when disabling CRTC on Strix Halo?
3. Are there known issues with Strix Halo display power management in this kernel version?
4. Should specific kernel parameters be used for Strix Halo to avoid display core timeouts?
5. Is there a minimum kernel/firmware version recommended for Strix Halo GPUs?

Files Available

• Full debug notes: DEBUG-display-doesnt-resume.md
• System journals available upon request
• Can provide full drm.debug=0x1e logs if needed

Contact

Ready to provide additional logs, test patches, or try debugging steps as directed.

Operating System

Fedora 43 Workstation

CPU

AMD RYZEN AI MAX+ 395 w/ Radeon 8060S

GPU

AMD RYZEN AI MAX+ 395 w/ Radeon 8060S

ROCm Version

none

ROCm Component

No response

Steps to Reproduce

No response

(Optional for Linux users) Output of /opt/rocm/bin/rocminfo --support

No response

Additional Information

No response