[wayland/ozone] Chromium based programs don't start | gpu errors

PixelHamster · 2024-12-28 05:21:45

I've been having these issues since around 2024-12-07. Same setup ect was working fine for over a year now.

HW:
- GPU 1: NVIDIA GeForce GTX 1650 Ti Mobile [Discrete]
- GPU 2: Intel UHD Graphics @ 1.15 GHz [Integrated]

uname -a:
- Linux statice 6.12.7-arch1-1 #1 SMP PREEMPT_DYNAMIC Fri, 27 Dec 2024 14:24:37 +0000 x86_64 GNU/Linux

drivers loaded with modesetting (I boot with EFISTUB, nvidia modules are configured via mkinitcpio.conf)
- Dec 28 03:21:19 statice kernel: nvidia-modeset: Loading NVIDIA UNIX Open Kernel Mode Setting Driver for x86_64 565.77 Release Build (archlinux-builder@)

merlijn@statice ~ % vesktop --enable-features=UseOzonePlatform --ozone-platform-hint=auto
APPIMAGE env is not defined, current application is not an AppImage
checkForUpdatesAndNotify called, downloadPromise is null
[6578:1228/033745.961820:ERROR:gpu_process_host.cc(982)] GPU process exited unexpectedly: exit_code=512
[6578:1228/033816.314907:ERROR:command_buffer_proxy_impl.cc(324)] GPU state invalid after WaitForGetOffsetInRange.
[6578:1228/033816.320695:ERROR:gpu_process_host.cc(982)] GPU process exited unexpectedly: exit_code=512
[6578:1228/033846.657590:ERROR:command_buffer_proxy_impl.cc(324)] GPU state invalid after WaitForGetOffsetInRange.
[6578:1228/033846.662239:ERROR:gpu_process_host.cc(982)] GPU process exited unexpectedly: exit_code=512
vesktop 4.48s user 0.79s system 5% cpu 1:31.79 total

The chromium browser behaves similar and eventually launches without gpu acceleration:

merlijn@statice ~ % chromium --enable-features=UseOzonePlatform --ozone-platform-hint=auto
[244781:244781:1228/055936.801039:ERROR:object_proxy.cc(576)] Failed to call method: org.freedesktop.portal.Secret.RetrieveSecret: object_path= /org/freedesktop/portal/desktop: org.freedesktop.DBus.Error.UnknownMethod: No such interface “org.freedesktop.portal.Secret” on object at path /org/freedesktop/portal/desktop
[244781:244781:1228/055936.801052:ERROR:secret_portal_key_provider.cc(150)] Failed to retrieve secret: No response from portal.
[244949:1:1228/060006.774960:ERROR:command_buffer_proxy_impl.cc(131)] ContextResult::kTransientFailure: Failed to send GpuControl.CreateCommandBuffer.
[244844:1:1228/060006.775012:ERROR:command_buffer_proxy_impl.cc(131)] ContextResult::kTransientFailure: Failed to send GpuControl.CreateCommandBuffer.
[244875:10:1228/060006.775011:ERROR:command_buffer_proxy_impl.cc(131)] ContextResult::kTransientFailure: Failed to send GpuControl.CreateCommandBuffer.
[244781:244781:1228/060006.779993:ERROR:gpu_process_host.cc(982)] GPU process exited unexpectedly: exit_code=512
[244949:1:1228/060037.069708:ERROR:command_buffer_proxy_impl.cc(331)] GPU state invalid after WaitForGetOffsetInRange.
[244875:10:1228/060037.069744:ERROR:command_buffer_proxy_impl.cc(331)] GPU state invalid after WaitForGetOffsetInRange.
[244781:244781:1228/060037.075743:ERROR:gpu_process_host.cc(982)] GPU process exited unexpectedly: exit_code=512
[244949:1:1228/060107.363752:ERROR:command_buffer_proxy_impl.cc(331)] GPU state invalid after WaitForGetOffsetInRange.
[244781:244781:1228/060107.369884:ERROR:gpu_process_host.cc(982)] GPU process exited unexpectedly: exit_code=512
[244844:7:1228/060107.598224:ERROR:command_buffer_proxy_impl.cc(131)] ContextResult::kTransientFailure: Failed to send GpuControl.CreateCommandBuffer.
[244845:7:1228/060107.599188:ERROR:command_buffer_proxy_impl.cc(131)] ContextResult::kTransientFailure: Failed to send GpuControl.CreateCommandBuffer

chrome://gpu says

* Canvas: Software only, hardware acceleration unavailable
* Canvas out-of-process rasterization: Disabled
* Direct Rendering Display Compositor: Disabled
* Compositing: Software only. Hardware acceleration disabled
* Multiple Raster Threads: Enabled
* OpenGL: Disabled
* Rasterization: Software only. Hardware acceleration disabled
* Raw Draw: Disabled
* Skia Graphite: Disabled
* Video Decode: Software only. Hardware acceleration disabled
* Video Encode: Software only. Hardware acceleration disabled
* Vulkan: Disabled
* WebGL: Software only, hardware acceleration unavailable
* WebGL2: Software only, hardware acceleration unavailable
* WebGPU: Disabled
* WebNN: Software only, hardware acceleration unavailable

What does work now:
Chromium based programs launch without the ozone flags fine under xwayland or xwayland-satellite
Firefox video playback and acceleration things
MPV video playback

Sidenote:
Launching sway (broke at the same time that these gpu errors manifested)
At the time of the breakage sway could not start due to segfaults in libgallium from mesa, however downgrading mesa did not fix the chromium issue so it might be unrelated.

seth · 2024-12-28 20:34:45

Please use [code][/code] tags, not "quote" tags. Edit your post in this regard.

Are you trying to run chromium on vulkan?
https://wiki.archlinux.org/title/Chromium#Vulkan

Which GPU does sway run on, intel or nvidia?

eglinfo -B

Can you disable either of them to make everyting run on the other?
Does chromium leave you w/ any https://wiki.archlinux.org/title/Core_d … _core_dump ?

dext · 2025-01-01 13:32:51

I think I have a very similar issue. On Wayland (KDE), all Chrome-based apps take 1-2 minutes to start. There's no problem on xorg session. Same thing with Flatpack and Pacman.

Laptop:

Intel(R) Core(TM) i7-10750H CPU
Intel Cometlake (Gen9)

GL_VERSION: 4.6 (Compatibility Profile) Mesa 24.3.2-arch1.1
GL_RENDERER: Mesa Intel(R) UHD Graphics (CML GT2)
GL_VENDOR: Intel

Apps start after this error happens 3 times:

[3:0101/142121.378937:ERROR:gpu_process_host.cc(1007)] GPU process exited unexpectedly: exit_code=512
[3:0101/142151.491719:ERROR:gpu_process_host.cc(1007)] GPU process exited unexpectedly: exit_code=512
[3:0101/142221.608135:ERROR:gpu_process_host.cc(1007)] GPU process exited unexpectedly: exit_code=512

seth · 2025-01-04 21:23:55

Do you also have a hybrid system?
Did you maybe enable vulkan support for chromium?

dext · 2025-01-05 12:27:36

seth wrote:

Do you also have a hybrid system?

Yes, Intel + Nvidia and it works well (nvidia-prime).

seth wrote:

Did you maybe enable vulkan support for chromium?

I don't have this file

~/.config/chromium-flags.conf

and I never even thought about doing anything to enable Vulcan, I'm even surprised this is a topic.

Both Brave and Google Chrome in

*://gpu

have "Vulkan: Disabled"

One extra mention: I've moved to KDE Plasma Wayland in ~September and had zero problems. These issue showed up around November, I can't recall if I've done any config changes, just system updates.

seth · 2025-01-05 15:32:56

Meh.

strace -f -tt -o /tmp/humongouschromiumstrace.txt chromium

You can

gzip -c /tmp/humongouschromiumstrace.txt | curl -F 'file=@-' 0x0.st

to stream it gzipped into 0x0.st (or look at it yourself, notably whether any nvidia libraries/files/objects show up) - be aware that it's likely gonna be a big file.

dext · 2025-01-05 16:45:16

It's a gigantic flood of:

1374420 17:20:05.854085 ioctl(20, DRM_IOCTL_I915_GEM_EXECBUFFER2, 0x7fff695e2940) = -1 ENOMEM (Cannot allocate memory)

I can't find any related and open issues about this.

I've found a workaround, for me this is only related to Electron apps, all I have to do is to run them with:

--disable-gpu

But, I can't find any global places to apply this.

~/.config/electron-flags.conf

Doesn't work.

Also, for Flatpak, modifying Desktop files doesn't work, for example:

/var/lib/flatpak/exports/share/applications/org.standardnotes.standardnotes.desktop

What does work is:

standardnotes --disable-gpu
flatpak run org.standardnotes.standardnotes --disable-gpu

I kind of fill like it's time for a new laptop. Probably close to no one uses Iris Xe anymore.

seth · 2025-01-05 21:45:56

https://gitlab.freedesktop.org/mesa/mes … ests/20449 - old, suggesting RAM pressure. But that doesn't fit "works on X11/xwayland"

--disable-gpu skips the part where electron figures that this won't work and moves straight to the software renderer, exporting LIBGL_ALWAYS_SOFTWARE=1 to the electron client will likely also work

mesa moved to 24.3.0 on 11/22, try to downgrade to 24.2.7 - there're plenty of amd/vega users w/ amdgpu crashes on 24.3.x …

dext · 2025-01-05 22:27:35

seth wrote:

--disable-gpu skips the part where electron figures that this won't work and moves straight to the software renderer, exporting LIBGL_ALWAYS_SOFTWARE=1 to the electron client will likely also work

Thank you, I have failed to figure out how to pass "--disable-gpu" to all electron-based flatpaks, but I'm able to pass LIBGL_ALWAYS_SOFTWARE=1 individually on every electron flatpack I use. This kind of solves it for me.

It's not memory pressure, I have 32GB machine, 12GB free.

I don't want to downgrade Mesa, besides Electron, everything else works just fine with GPU acceleration.

I would not mark this as solved:
- If someone needs GPU acceleration in some electron app, it's still broken.
- I think LIBGL_ALWAYS_SOFTWARE affects everything, not just Electron. I have like 5 electron apps, I'm fine doing them individually. But if someone has more apps, or multiple users, it'd be a problem.

seth · 2025-01-05 22:50:26

I don't want to downgrade Mesa, besides Electron, everything else works just fine with GPU acceleration.

The point is that it would probably be nice to know whether the issue is coming from there (or elsewhere)

dext · 2025-01-06 09:46:50

I've tried the downgrade. I removed my changes, to be able to reproduce the problem again. I installed:
https://archive.archlinux.org/packages/ … kg.tar.zst

After reboot, there's no difference, strace shows exactly the same. In my case this is not it.

seth · 2025-01-06 11:47:05

Do you get the same w/ the LTS kernel?
If it's not mesa, the kernel seems more likely than every single electrog client being affected at once.

Unfortunately comet lake (gen 10) isn't covered by https://wiki.archlinux.org/title/Intel_ … _Xe_driver

dext · 2025-01-06 11:56:18

seth wrote:

Do you get the same w/ the LTS kernel?

Unfortunately I can't test LTS. Long story short I blew up my system years ago when root fs run out of space during upgrade. I managed to recover everything but the Linux LTS install. I'm happy to do extra tests to figure this out, but I'm afraid to touch that part.

seth · 2025-01-06 12:03:40

What?
We'll just fix that

lsblk -f
cat /proc/cmdline

dext · 2025-01-06 12:44:36

I had linux-lts on ignore for years. I just run pacman -Syu linux-lts and it worked flawlessly, all the file conflicts I had are gone. Pacman is amazing.

I can confirm, there is no issue on LTS 6.6.69.

One more thing, I just realized Signal is Electron based, no issue there. So for me, it's not all Electron apps, but it's only Electron apps.

jeff_bh_221 · 2025-01-06 18:43:01

Same problems, tried lots envvars, downgrades. Long stroy, but, finally, I figured out this, and simply replace nvidia(nvidia-lts) with nvidia-dkms, solved!
Tested with lts,stable,zen kernel. maybe not need to edit cmdline, Possibly? I don't have any nvidia paraments in kernel cmdline.
Try it, might be help.

seth · 2025-01-06 18:45:53

That is extremely unlikely since nvidia and nvidia-lts are built from nvidia-dkms, chances are you're currently not using any nvidia module?

dkms status
lsmod | grep nvidia

dext · 2025-01-06 19:29:11

OK, my fault, nvidia-lts was missing. I tried rebooting with and without to LTS and I can confirm: the issues show up when nvidia modules are loaded.

But I'm very confused now. I have everything running on Intel, don't I? Looks like it. Do you know how to make sure? How just presence of the nvidia module could create this problem?

glxinfo | grep -E "OpenGL vendor|OpenGL renderer"
OpenGL vendor string: Intel
OpenGL renderer string: Mesa Intel(R) UHD Graphics (CML GT2)

nvidia-smi shows no processes are running.

jeff_bh_221 · 2025-01-07 06:31:11

also, but solved, digged a little deeper. Guess some paraments inside of chrome, chromium, Electron, etc would like use high-performance gpu?
here I found some from https://bbs.archlinux.org/viewtopic.php?id=284426 , may be help:

touch a file in ~/.config/environment.d/ if setup envvars by systemd.

DXVK_FILTER_DEVICE_NAME="Intel"
VKD3D_FILTER_DEVICE_NAME="Intel"
__GLX_VENDOR_LIBRARY_NAME="mesa"
__EGL_VENDOR_LIBRARY_FILENAMES="/usr/share/glvnd/egl_vendor.d/50_mesa.json"
__NV_PRIME_RENDER_OFFLOAD="0"
__VK_LAYER_NV_optimus="non_NVIDIA_only"
WLR_RENDER_DRM_DEVICE="/dev/dri/renderD128"

check your device with command

lspci | grep VGA
ls -al /dev/dri/by-path/

dext · 2025-01-07 09:15:54

No change. I've set everything as instructed:

% cat ~/.config/environment.d/electron.conf
DXVK_FILTER_DEVICE_NAME="Intel"
VKD3D_FILTER_DEVICE_NAME="Intel"
__GLX_VENDOR_LIBRARY_NAME="mesa"
__EGL_VENDOR_LIBRARY_FILENAMES="/usr/share/glvnd/egl_vendor.d/50_mesa.json"
__NV_PRIME_RENDER_OFFLOAD="0"
__VK_LAYER_NV_optimus="non_NVIDIA_only"
WLR_RENDER_DRM_DEVICE="/dev/dri/renderD129"

% lsgpu
card0                    10de:1f95                         drm:/dev/dri/card0
└─renderD128                                               drm:/dev/dri/renderD128
card1                    Intel Cometlake (Gen9)            drm:/dev/dri/card1
└─renderD129                                               drm:/dev/dri/renderD129

I did run:

systemctl --user daemon-reload
systemctl --user show-environment

Everything is set. The issue is still exactly the same (I'm confirming with strace every time I test)

For someone coming late to the party, TLDR:
Some electron apps take minutes to start if nvidia module is present. Intel Comet Lake + Nvidia 1650 hybrid. Errors on stdout:

[3:0107/100904.107832:ERROR:gpu_process_host.cc(1002)] GPU process exited unexpectedly: exit_code=512

Flood of errors in strace:

ioctl(20, DRM_IOCTL_I915_GEM_EXECBUFFER2, 0x7ffd63248e10) = -1 ENOMEM (Cannot allocate memory)

Not all Electron apps, but only Electron apps.

So far we tried to downgrade mesa to 24.2.7, and tried booting with LTS kernel. No change.

Working workarounds so far:
- pass --disable-gpu param, or LIBGL_ALWAYS_SOFTWARE=1 env to the electron app
- don't load Nvidia module.

seth · 2025-01-07 09:53:03

Do you have https://archlinux.org/packages/extra/x8 … kan-intel/
Try to also export "VK_DRIVER_FILES=/usr/share/vulkan/icd.d/intel_icd.i686.json" and "DRI_PRIME=pci-0000_00_02_0" but make sure your IGP is actually 00:02.0
nb. that having all those exprts in /etc/environment will probably kill your efforts to actually run stuff on the nvidia GPU, you probably want to localize them around electron™

And finally (please, god please - i want to shit on that thing SOOOOOO much. Please)
The order of your cards is reversed, intel is 1 and nvidia is 0…
Try to add "nvidia_drm.modeset=1" to the https://wiki.archlinux.org/title/Kernel_parameters (it *has* to be the kernel parameter, using modprobe.conf won't do)

dext · 2025-01-07 10:20:20

Sorry, but VK_DRIVER_FILES and DRI_PRIME made no difference (I used the values as you provided)

lspci | grep -i vga                               
00:02.0 VGA compatible controller: Intel Corporation CometLake-H GT2 [UHD Graphics] (rev 05)

I had nvidia_drm.modeset=1 set already:

cat /sys/module/nvidia_drm/parameters/modeset
Y

Your hypothesis is, that for some reason Electron is trying to use Nvidia instead of Intel? Could it be as stupid as trying to call DRM_IOCTL_I915_GEM_EXECBUFFER2 to Nvidia driver?

EDIT, PS: I had vulkan-intel as well.

Last edited by dext (2025-01-07 10:22:13)

seth · 2025-01-07 10:29:20

cat /sys/module/nvidia_drm/parameters/modeset

cat /proc/cmdline

It is *not* about the parameter itself, it's about killing the simplydumb device

Iff that's really not it, post your mkinicpio.conf; try to remove the kms hook, add the i915 module and regenerate the initramfs.

dext · 2025-01-07 10:50:48

# cat /sys/module/nvidia_drm/parameters/modeset
Y

# cat /proc/cmdline
initrd=\intel-ucode.img initrd=\initramfs-linux.img rd.luks.name=443807fa-6b8e-427f-906e-13757e99de8b=cryptlvm root=/dev/volgroup/root resume=/dev/mapper/volgroup-swap nvidia_drm.modeset=1 rw

# cat /etc/mkinitcpio.conf | grep -v "#" | grep -vE "^$" 
MODULES=()
BINARIES=()
FILES=()
HOOKS=(base systemd autodetect modconf keyboard sd-vconsole block sd-encrypt lvm2 filesystems fsck)

Seth, thank you for all you help. I just want to be open instead of ghosting: I need to start using my computer, and reboots are very costly for me. I'm personally fine with setting LIBGL_ALWAYS_SOFTWARE=1 for the apps I have the problem with. Furthermore, I have my Intel+Nvidia setup working as expected, and it's starting to look like a proper solution for Election problem could change this.

seth · 2025-01-07 11:07:44

Add i915 to the MODULES array, regenerate the initramfs and see what happens after the next reboot (whenever that is - there's no rush)
Idk why the card order is reversed if it's not the simpledrm device but rn that's my last best guess as to why electron is -apparently- trying to run the mesa stack on the nvidia GPU (on wayland)…

I have my Intel+Nvidia setup working as expected, and it's starting to look like a proper solution for Election problem could change this.

Nah, you're running on the intel chip anyway - fixing the card order won't change that.

Arch Linux

#1 2024-12-28 05:21:45

[wayland/ozone] Chromium based programs don't start | gpu errors

#2 2024-12-28 20:34:45

Re: [wayland/ozone] Chromium based programs don't start | gpu errors

#3 2025-01-01 13:32:51

Re: [wayland/ozone] Chromium based programs don't start | gpu errors

#4 2025-01-04 21:23:55

Re: [wayland/ozone] Chromium based programs don't start | gpu errors

#5 2025-01-05 12:27:36

Re: [wayland/ozone] Chromium based programs don't start | gpu errors

#6 2025-01-05 15:32:56

Re: [wayland/ozone] Chromium based programs don't start | gpu errors

#7 2025-01-05 16:45:16

Re: [wayland/ozone] Chromium based programs don't start | gpu errors

#8 2025-01-05 21:45:56

Re: [wayland/ozone] Chromium based programs don't start | gpu errors

#9 2025-01-05 22:27:35

Re: [wayland/ozone] Chromium based programs don't start | gpu errors

#10 2025-01-05 22:50:26

Re: [wayland/ozone] Chromium based programs don't start | gpu errors

#11 2025-01-06 09:46:50

Re: [wayland/ozone] Chromium based programs don't start | gpu errors

#12 2025-01-06 11:47:05

Re: [wayland/ozone] Chromium based programs don't start | gpu errors

#13 2025-01-06 11:56:18

Re: [wayland/ozone] Chromium based programs don't start | gpu errors

#14 2025-01-06 12:03:40

Re: [wayland/ozone] Chromium based programs don't start | gpu errors

#15 2025-01-06 12:44:36

Re: [wayland/ozone] Chromium based programs don't start | gpu errors

#16 2025-01-06 18:43:01

Re: [wayland/ozone] Chromium based programs don't start | gpu errors

#17 2025-01-06 18:45:53

Re: [wayland/ozone] Chromium based programs don't start | gpu errors

#18 2025-01-06 19:29:11

Re: [wayland/ozone] Chromium based programs don't start | gpu errors

#19 2025-01-07 06:31:11

Re: [wayland/ozone] Chromium based programs don't start | gpu errors

#20 2025-01-07 09:15:54

Re: [wayland/ozone] Chromium based programs don't start | gpu errors

#21 2025-01-07 09:53:03

Re: [wayland/ozone] Chromium based programs don't start | gpu errors

#22 2025-01-07 10:20:20

Re: [wayland/ozone] Chromium based programs don't start | gpu errors

#23 2025-01-07 10:29:20

Re: [wayland/ozone] Chromium based programs don't start | gpu errors

#24 2025-01-07 10:50:48

Re: [wayland/ozone] Chromium based programs don't start | gpu errors

#25 2025-01-07 11:07:44

Re: [wayland/ozone] Chromium based programs don't start | gpu errors

Board footer