Occasional crash on Wayland (Linux)

jakobkmar commented 2 years ago

When starting a compose application on Linux under Wayland, it crashes with the following error:

# Problematic frame:
# C  [libskiko-linux-x64.so+0x3b6230]  skgpu::v1::AtlasTextOp::createDrawForGeneratedGlyphs(GrMeshDrawTarget*, skgpu::v1::AtlasTextOp::FlushInfo*) const+0x280

Core Dump

System: OS: Fedora Linux 35 (64bit) Windowing System: Wayland on Gnome 41.2 GPU: NVIDIA GeForce RTX 2060/PCIe/SSE2 CPU: Intel® Core™ i9-9900K

The version of Compose I used: 1.0.1-rc2

jakobkmar commented 2 years ago

This also affects JetBrains Toolbox.

jakobkmar commented 2 years ago

Edit: this only happens occasionally, not every time. Could also be an issue with the compositor itself, but not sure. I am still leaving this issue open as there might be something related to Compose anyways?

What does the core dump indicate, is it an issue with Compose or the compositor?

olonho commented 2 years ago

Interesting, hard to say from the crashdump, as workaround please use direct software rendering.

olonho commented 2 years ago

I.e. with env variable SKIKO_RENDER_API=DIRECT_SOFTWARE.

jakobkmar commented 2 years ago

Just wanted to add:

this is an issue with nvidia cards only
although I wrote that this crash is "occasional", it happens 95% of the time

Is it planned to fix this? Would be great as software rendering is very slow :(

jakobkmar commented 2 years ago

Could be related to this issue: https://gitlab.gnome.org/GNOME/mutter/-/issues/2045 If it is, this is a bug in the proprietary Nvidia Drivers.

jakobkmar commented 2 years ago

The crash seems to be fixed now by a newer version of the Nvidia Drivers, so the original issue is resolved.

However it would still be great to have actual native Wayland support and not run Compose Applications through XWayland. Should I convert this issue to that?

Skaldebane commented 2 years ago

I have had the same issue with an Intel iGPU. It happens only when using Wayland with specific types of apps (which use a LazyList with large lists, for example). It doesn't only crash the app, but it crashes XWayland as a whole, causing the entire desktop environment (in my case GNOME) to crash.

esp-er commented 2 years ago

I've had the same issue as @Skaldebane on Linux with Mesa AMDVLK drivers using XWayland. My app was using LazyLists and it would occasionally cause the whole desktop environment to restart. Yesterday I switched from compose 1.1.1 to 1.2.0-alpha-dev(kotlin 1.6.21) and I haven't had a single crash yet, so the situation might be improving. Hopefully proper Wayland support in OpenJDK gets added soon, I know it's being worked on with project Wakefield.

Side note: On Wayland and my machine, my application feels much smoother(e.g animations, scrolling etc) with less input lag when adding -Dskiko.vsync.enabled=false to the launch command. I assume that this is because Wayland already does vsync globally, and with vsync enabled on the XWayland-end we end up syncing twice, adding input lag and frame output gets janky?

Skaldebane commented 2 years ago

@esp-er That's interesting! I already use 1.2.0 development versions on some other testing projects, but not yet on those with LazyList. I'll definitely give it a shot and see!

Thanks for that side note though! Animations always felt super laggy for me compared to Android, so that might as well be the reason.

p0058781 commented 5 months ago

I have had the same issue with an Intel iGPU. It happens only when using Wayland with specific types of apps (which use a LazyList with large lists, for example). It doesn't only crash the app, but it crashes XWayland as a whole, causing the entire desktop environment (in my case GNOME) to crash.

I have similar issue that i can reproduce consistently. For me when i use Window with DpSize.Unspecified and a composable inside that has Modifier.fillMaxSize(), XWayland and/or iGPU crashes. In my case Gnome doesn't crash but one interesting thing is that then all XWayland apps wont render, just black screen, until i reboot

[158532.952508] i915 0000:00:02.0: [drm] Resetting rcs0 for preemption time out
[158532.952590] i915 0000:00:02.0: [drm] AWT-EventQueue-[662018] context reset due to GPU hang
[158532.981889] i915 0000:00:02.0: [drm] GPU HANG: ecode 12:1:85dffffb, in AWT-EventQueue- [662018]
[158532.981892] GPU hangs can indicate a bug anywhere in the entire gfx stack, including userspace.
[158532.981892] Please file a _new_ bug report at https://gitlab.freedesktop.org/drm/intel/issues/new.
[158532.981893] Please see https://drm.pages.freedesktop.org/intel-docs/how-to-file-i915-bugs.html for details.
[158532.981893] drm/i915 developers can then reassign to the right component if it's not a kernel issue.
[158532.981893] The GPU crash dump is required to analyze GPU hangs, so please always attach it.
[158532.981893] GPU crash dump saved to /sys/class/drm/card2/error
[158540.632444] i915 0000:00:02.0: [drm] Resetting rcs0 for preemption time out
[158540.632555] i915 0000:00:02.0: [drm] Xwayland[3930] context reset due to GPU hang
[158540.666161] i915 0000:00:02.0: [drm] GPU HANG: ecode 12:1:85dffffb, in Xwayland [3930]
[158622.340381] i915 0000:00:02.0: [drm] Resetting rcs0 for preemption time out
[158622.340465] i915 0000:00:02.0: [drm] AWT-EventQueue-[664085] context reset due to GPU hang
[158622.371511] i915 0000:00:02.0: [drm] GPU HANG: ecode 12:1:85dffffb, in AWT-EventQueue- [664085]
[158631.083491] i915 0000:00:02.0: [drm] Resetting rcs0 for preemption time out
[158631.083592] i915 0000:00:02.0: [drm] AWT-EventQueue-[664085] context reset due to GPU hang
[158631.108887] i915 0000:00:02.0: [drm] GPU HANG: ecode 12:1:85dffffb, in AWT-EventQueue- [664085]
[158638.763368] i915 0000:00:02.0: [drm] Resetting rcs0 for preemption time out
[158638.763471] i915 0000:00:02.0: [drm] Xwayland[662576] context reset due to GPU hang
[158638.796316] i915 0000:00:02.0: [drm] GPU HANG: ecode 12:1:85dffffb, in Xwayland [662576]
[158646.446461] i915 0000:00:02.0: [drm] Resetting rcs0 for preemption time out
[158646.446554] i915 0000:00:02.0: [drm] AWT-EventQueue-[664085] context reset due to GPU hang
[158646.477526] i915 0000:00:02.0: [drm] GPU HANG: ecode 12:1:85dffffb, in AWT-EventQueue- [664085]
[158654.124038] i915 0000:00:02.0: [drm] Resetting rcs0 for preemption time out
[158654.124148] i915 0000:00:02.0: [drm] Xwayland[662576] context reset due to GPU hang
[158654.158585] i915 0000:00:02.0: [drm] GPU HANG: ecode 12:1:85dffffb, in Xwayland [662576]
[158662.656075] i915 0000:00:02.0: [drm] Resetting rcs0 for preemption time out
[158662.656193] i915 0000:00:02.0: [drm] Xwayland[662576] context reset due to GPU hang
[158662.687435] i915 0000:00:02.0: [drm] GPU HANG: ecode 12:1:85dffffb, in Xwayland [662576]

okushnikov commented 1 month ago

Please check the following ticket on YouTrack for follow-ups to this issue. GitHub issues will be closed in the coming weeks.

JetBrains / compose-multiplatform

Occasional crash on Wayland (Linux) #1631