chromium 111+ builds crashy under sway with output scale != 1

[edrex]

Describe the bug

Chromium (under sway wayland) crashes either at startup or when windows are moved between my two displays a few times. Reproducible with an empty profile. First appears in de134a1

I initially reported this upstream, but I've been unable to reproduce running the official builds with steam-run (chrome://gpu looks about the same as with nixpkgs build) so I'm now convinced this is something particular to the nixpkgs builds.

TODO

• Try to reproduce on another wayland env like plasma
  • couldn't reproduce on startplasma-wayland, Hyprland, or river, trying to replicate conditions/procedure (laptop hidpi output scale 2.25, using keybind to rapidly cycle maximize or fullscreen window between outputs). Text is a little blurry with the fractional scaling on hidpi output on all those compositors, but seems crisp always on sway. maybe related? Anyway, this seems to be a sway-only bug.
• I haven't heard from anyone else with this issue. I'm on an i915 igpu laptop (gen7) update: it seems like @yashlala has the same crash with Arch Linux builds, while primeos can't repro with similar hardware/software.
• I'm 95% sure this started before I enabled the xanmod kernel, but I should check that reproducible with default kernel
• Test with both outputs scale=1
  • eDP-1 scale=1: no crash 🤔
  • scale=2: crash
• Build chromiumDev and see if it's present there.
  • solve tmpfs out of space first. update: added swap, and set tmpfs size to 100% yield 16GB space
  • Especially curious whether 1356014 - [Linux][Wayland] Non-integer scaling support - chromium makes any difference. That commit seems to be in the current dev + beta. It has since been reverted but the revert isn't in the channels yet afaict. Reland "[Linux/Ozone/Wayland] Support fractional-scale-v1" (4370091) · Gerrit Code Review is the one to watch going forward.
• Also, with above build, test with sway@HEAD which has changes related to the fractional scaling protocol

Notify maintainers

@primeos

Metadata

[user@system:~]$ nix-shell -p nix-info --run "nix-info -m"
 - system: `"x86_64-linux"`
 - host os: `Linux 6.1.20-xanmod1, NixOS, 23.05 (Stoat)`
 - multi-user?: `yes`
 - sandbox: `yes`
 - version: `nix-env (Nix) 2.13.3`
 - nixpkgs: `/nix/var/nix/profiles/per-user/root/channels/nixos`

[primeos]

We've recently updated Chromium to M112 (#224726) - is that version affected as well?

I'm also running Chromium under Sway via Ozone/Wayland (--enable-features=UseOzonePlatform --ozone-platform=wayland) but don't remember any crashes lately (there were definitely some issues in the past though (Ozone/Wayland is still(?) experimental) - e.g. I remember the crashes when repeatedly moving Chromium windows between displays but that was probably a year ago or so and got fixed with the next major release).

I have a gen7 Intel CPU+GPU as well.

[yashlala]

I found the issue you filed upstream. I'm having the same issue on Arch Linux, with the same chromium versions, on an Intel gen7 CPU+GPU. I also couldn't reproduce this with the latest official chromium build.

I recognize this is the NixOS bug tracker; just thought I'd provide the backtraces and about:gpu on my end, in case the issues are related.

[edrex]

M112?

112.0.5615.49 (Official Build) (64-bit) from 0e19daa is still affected. Unrelated but what's the story with chrome://version/ saying (Official Build)?

Test command: nix run nixpkgs/nixos-unstable#chromium -- --user-data-dir=$(mktemp -d)

Forgot to post the log here (it's in the upstream issue tho):

[139105:139105:0404/203928.426871:ERROR:wayland_event_watcher.cc(38)] libwayland: xdg_wm_base@15: error 4: wrong configure serial: 69136
[0404/203928.475957:ERROR:elf_dynamic_array_reader.h(64)] tag not found
[0404/203928.477108:ERROR:process_memory_range.cc(75)] read out of range
fish: Job 1, 'chromium --user-data-dir=(mktem…' terminated by signal SIGTRAP (Trace or breakpoint trap)

@yashlala does your log look like that? I don't see those call sites in your backtrace, although I'm not great at reading them.

[yashlala]

@edrex It's similar.

Test command: chromium --user-data-dir=$(mktemp -d)

Output:

[15233:15233:0408/122327.921960:ERROR:wayland_event_watcher.cc(38)] libwayland: xdg_wm_base@15: error 4: wrong configure serial: 13347

[0408/122327.930374:ERROR:elf_dynamic_array_reader.h(64)] tag not found
[1]    15233 trace trap (core dumped)  chromium --user-data-dir=$(mktemp -d)

I've added package version numbers and log output to the Gist now.

[edrex]

I just tested with startplasma-wayland, river, and Hyprland, trying to reproduce equivalent conditions as much as possible (esp fractional scaling of my hidpi laptop display to 2.25) and I can only repro on sway. I do notice that text looks a little blurry on those compositors on the fractional scaling output, while on sway it looks crisp all the time. 🤔

[edrex]

@yashlala can you confirm no crash with all outputs scale set to 1? (wlr-randr --output eDP-1 --scale 1 etc, or similar with way-displays)

[edrex]

@primeos could you possibly try two outputs with one set to a fractional scale eg 1.5? Usually it either crashes immediately or after swapping the window between outputs a few times (<< 10). Waiting on chromiumDev build.

[yashlala]

@edrex I can confirm that. I can also confirm that there are no crashes with all outputs scaled to 3; there need to be different scaling factors. Here's a link to the Arch issue I opened, where I record my findings.

[yashlala]

Do note that I don't have an issue when moving windows, only when creating new ones.

I half-remember problems with moving windows yesterday, but couldn't reproduce it today. I wouldn't rule out the "two separate issues" case yet.

[edrex]

Should be fixed by #229759

(see my comment in upstream bug for details)