[SOLVED] SIGKILL to an OpenGL program results in a full system hang

weirddan455 · 2025-09-22 16:26:06

I was going to post this on the Gilab issue tracker but account creation is disabled so I'm waiting on an email back.

This is a regression in 6.16.8.arch2-1 in a patch to the amdgpu driver. Rolling back to 6.16.8.arch1-1 resolves this issue.

I'm a developer for DOSBox Staging so that is the first application I noticed it in. I can also reproduce it in Firefox and I assume most/all OpenGL (and possibly Vulkan) applications.

To reproduce, simply

kill -KILL PID

. The window will linger on screen. Sometimes the system will remain usable for a minute or so and then it will lock up. Mouse stops working, unable to switch to console with Ctrl+Alt+Fkey. I had to hard power off to recover.

This is the logs from journalctl from when I SIGKILL'd Firefox:

Sep 22 10:55:35 arch firefox[1711]: Exiting due to channel error.
Sep 22 10:55:35 arch firefox[1642]: Exiting due to channel error.
Sep 22 10:55:35 arch firefox[1595]: Exiting due to channel error.
Sep 22 10:55:35 arch firefox[1457]: Exiting due to channel error.
Sep 22 10:55:35 arch firefox[1619]: Exiting due to channel error.
Sep 22 10:55:35 arch firefox[1524]: Exiting due to channel error.
Sep 22 10:55:35 arch kernel: [drm:amdgpu_job_submit [amdgpu]] *ERROR* Trying to push to a killed entity
Sep 22 10:55:39 arch systemd[1]: systemd-timedated.service: Deactivated successfully.
Sep 22 10:55:43 arch systemd[1]: systemd-hostnamed.service: Deactivated successfully.
Sep 22 10:55:44 arch kwin_wayland[788]: kwin_wayland_drm: Pageflip timed out! This is a bug in the amdgpu kernel driver
Sep 22 10:55:44 arch kwin_wayland[788]: kwin_wayland_drm: Please report this at https://gitlab.freedesktop.org/drm/amd/-/issues
Sep 22 10:55:44 arch kwin_wayland[788]: kwin_wayland_drm: With the output of 'sudo dmesg' and 'journalctl --user-unit plasma-kwin_wayland --boot 0'
Sep 22 10:55:45 arch kwin_wayland[788]: kwin_wayland_drm: Pageflip timed out! This is a bug in the amdgpu kernel driver
Sep 22 10:55:45 arch kwin_wayland[788]: kwin_wayland_drm: Please report this at https://gitlab.freedesktop.org/drm/amd/-/issues
Sep 22 10:55:45 arch kwin_wayland[788]: kwin_wayland_drm: With the output of 'sudo dmesg' and 'journalctl --user-unit plasma-kwin_wayland --boot 0'
Sep 22 10:55:46 arch kwin_wayland[788]: kwin_wayland_drm: Pageflip timed out! This is a bug in the amdgpu kernel driver
Sep 22 10:55:46 arch kwin_wayland[788]: kwin_wayland_drm: Please report this at https://gitlab.freedesktop.org/drm/amd/-/issues
Sep 22 10:55:46 arch kwin_wayland[788]: kwin_wayland_drm: With the output of 'sudo dmesg' and 'journalctl --user-unit plasma-kwin_wayland --boot 0'
Sep 22 10:55:47 arch kwin_wayland[788]: kwin_wayland_drm: Pageflip timed out! This is a bug in the amdgpu kernel driver
Sep 22 10:55:47 arch kwin_wayland[788]: kwin_wayland_drm: Please report this at https://gitlab.freedesktop.org/drm/amd/-/issues
Sep 22 10:55:47 arch kwin_wayland[788]: kwin_wayland_drm: With the output of 'sudo dmesg' and 'journalctl --user-unit plasma-kwin_wayland --boot 0'
Sep 22 10:55:48 arch kwin_wayland[788]: kwin_wayland_drm: Pageflip timed out! This is a bug in the amdgpu kernel driver
Sep 22 10:55:48 arch kwin_wayland[788]: kwin_wayland_drm: Please report this at https://gitlab.freedesktop.org/drm/amd/-/issues
Sep 22 10:55:48 arch kwin_wayland[788]: kwin_wayland_drm: With the output of 'sudo dmesg' and 'journalctl --user-unit plasma-kwin_wayland --boot 0'
Sep 22 10:55:49 arch kwin_wayland[788]: kwin_wayland_drm: Pageflip timed out! This is a bug in the amdgpu kernel driver
Sep 22 10:55:49 arch kwin_wayland[788]: kwin_wayland_drm: Please report this at https://gitlab.freedesktop.org/drm/amd/-/issues
Sep 22 10:55:49 arch kwin_wayland[788]: kwin_wayland_drm: With the output of 'sudo dmesg' and 'journalctl --user-unit plasma-kwin_wayland --boot 0'
Sep 22 10:55:50 arch kwin_wayland[788]: kwin_wayland_drm: Pageflip timed out! This is a bug in the amdgpu kernel driver
Sep 22 10:55:50 arch kwin_wayland[788]: kwin_wayland_drm: Please report this at https://gitlab.freedesktop.org/drm/amd/-/issues
Sep 22 10:55:50 arch kwin_wayland[788]: kwin_wayland_drm: With the output of 'sudo dmesg' and 'journalctl --user-unit plasma-kwin_wayland --boot 0'
Sep 22 10:55:51 arch kwin_wayland[788]: kwin_wayland_drm: Pageflip timed out! This is a bug in the amdgpu kernel driver
Sep 22 10:55:51 arch kwin_wayland[788]: kwin_wayland_drm: Please report this at https://gitlab.freedesktop.org/drm/amd/-/issues
Sep 22 10:55:51 arch kwin_wayland[788]: kwin_wayland_drm: With the output of 'sudo dmesg' and 'journalctl --user-unit plasma-kwin_wayland --boot 0'

Last edited by weirddan455 (2025-09-24 22:42:27)

seth · 2025-09-22 20:07:28

So basically https://github.com/archlinux/linux/comm … dbcc2ca2f1 *triggers* this for you??
But this is only on SIGKILL, not SIGTERM?
See the upstream bugs, the revert is likely a mitigation of the more frequent problem (that does not involve sigkill), not a proper fix.

I had to hard power off to recover.

https://wiki.archlinux.org/title/Keyboa … el_(SysRq)

weirddan455 · 2025-09-22 20:37:25

Yes, the revert triggers the problem. The commit that was reverted is titled "fix task hang from failed job submission during process kill" and now I'm having problems with hanging during a process kill. SIGTERM works fine, it's only SIGKILL that's problematic.

seth wrote:

See the upstream bugs, the revert is likely a mitigation of the more frequent problem (that does not involve sigkill), not a proper fix.

Yes, I read the https://gitlab.freedesktop.org/drm/amd/-/issues/4568 but I cannot reproduce the Firefox Webrender issue on my system. From what the kernel devs are saying on that thread, it sounds like a Firefox bug and the fix would be to patch Firefox. Maybe the kernel devs will write a proper fix in "don't break userspace" fashion but this revert is causing more problems than it solves IMO.

IckyVicky · 2025-09-22 21:09:29

Same exact issue for me, it started to happen after a system reboot today, and it is being a whole nightmare right now, cannot kill any app that isn't the terminal or the system decides to die. Gonna have to rollback aswell.

Last edited by IckyVicky (2025-09-22 21:10:14)

seth · 2025-09-22 21:14:16

@weirddan455, bump this if you don't get a gitlab account in reasonable time

@IckyVicky, do you habitually SIGKILL processes? Why?

billli11 · 2025-09-22 22:39:07

Looks like linux-6.16.8.arch3-1 and linux-zen-6.16.8.zen3-1 is coming with the revert reverted.

https://github.com/archlinux/linux/rele … 16.8-arch3
https://github.com/zen-kernel/zen-kernel/releases

https://gitlab.archlinux.org/archlinux/ … 5fbb21a037
https://gitlab.archlinux.org/archlinux/ … 7128662aad

Edit: Now in testing
https://archlinux.org/packages/core-tes … _64/linux/
https://archlinux.org/packages/extra-te … linux-zen/

Last edited by billli11 (2025-09-22 22:57:00)

seth · 2025-09-23 07:06:41

Sure?
https://github.com/archlinux/linux/comp … 16.8-arch3

billli11 · 2025-09-23 11:25:48

@seth

Yes
arch2 tag
https://github.com/archlinux/linux/comm … 16.8-arch2
arch3 tag
https://github.com/archlinux/linux/comm … 16.8-arch3

Revert "drm/amdgpu: fix task hang from failed job submission during process kill" no longer in arch3

weirddan455 · 2025-09-24 22:43:02

I can confirm 6.16.8-arch3-1 fixes the problem so marking as solved.

Arch Linux

#1 2025-09-22 16:26:06

[SOLVED] SIGKILL to an OpenGL program results in a full system hang

#2 2025-09-22 20:07:28

Re: [SOLVED] SIGKILL to an OpenGL program results in a full system hang

#3 2025-09-22 20:37:25

Re: [SOLVED] SIGKILL to an OpenGL program results in a full system hang

#4 2025-09-22 21:09:29

Re: [SOLVED] SIGKILL to an OpenGL program results in a full system hang

#5 2025-09-22 21:14:16

Re: [SOLVED] SIGKILL to an OpenGL program results in a full system hang

#6 2025-09-22 22:39:07

Re: [SOLVED] SIGKILL to an OpenGL program results in a full system hang

#7 2025-09-23 07:06:41

Re: [SOLVED] SIGKILL to an OpenGL program results in a full system hang

#8 2025-09-23 11:25:48

Re: [SOLVED] SIGKILL to an OpenGL program results in a full system hang

#9 2025-09-24 22:43:02

Re: [SOLVED] SIGKILL to an OpenGL program results in a full system hang

Board footer