1849237 - Crash in [@ libnvidia-egl-wayland.so.1@0x4b20] playing an Youtube video in Wayland on proprietary NVIDIA driver. Fine with Nvidia driver 545.

Status: RESOLVED WORKSFORME

(Core :: Graphics, defect)

Description

[develop7]

Repro: play a youtube video for a few seconds

Expected: FF doesn't crash
Instead: it crashes

Crash report: https://crash-stats.mozilla.org/report/index/ff03037c-e2eb-4486-95f7-d9af70230817

Reason: SIGSEGV / SI_KERNEL

Top 10 frames of crashing thread:

0  libnvidia-egl-wayland.so.1  libnvidia-egl-wayland.so.1@0x4b20  
1  libnvidia-egl-wayland.so.1  libnvidia-egl-wayland.so.1@0x4c04  
2  libffi.so.8  ffi_call_unix64  /usr/src/debug/libffi-3.4.4/src/x86/unix64.S:104
3  libffi.so.8  ffi_call_int  /usr/src/debug/libffi-3.4.4/src/x86/ffi64.c:673
4  libnvidia-egl-wayland.so.1  libnvidia-egl-wayland.so.1@0x4bbf  
5  libffi.so.8  examine_argument  /usr/src/debug/libffi-3.4.4/src/x86/ffi64.c:359
6  ?  @0x00007fe7a3ab299f  
7  libffi.so.8  ffi_call  /usr/src/debug/libffi-3.4.4/src/x86/ffi64.c:710
8  ?  @0x00007fe700000000  
9  firefox  arena_dalloc  

Comment 1

[BugBot [:suhaib / :marco/ :calixte]]

The bug has a crash signature, thus the bug will be considered confirmed.

Comment 2

[trevnels]

I have also been seeing this crash frequently for the past few days on NVIDIA driver 535.98 on Wayland. For me, it seems like entering/exiting fullscreen on YouTube will often trigger the crash. I've also seen the same crash on other sites using WebGL, such as Google Maps. I can't reliably reproduce the crash every time I use YouTube, etc. - it seems to happen at random.

Here's a small sample of my recent crash reports:
https://crash-stats.mozilla.org/report/index/7db233e8-41ce-43fe-afa5-708cd0230818
https://crash-stats.mozilla.org/report/index/688fe74f-291d-408a-b7fa-a1d930230818
https://crash-stats.mozilla.org/report/index/17c36f04-c3c0-43dc-8a5a-660600230817

Comment 3

[trevnels]

I'm still crashing with webgl.disabled=true, so it seems like the problem extends further than just WebGL. It now seems like resizing the browser window often causes a crash - might be related to the crashing I saw with entering/exiting fullscreen on YouTube.

Comment 4

[Alexandru Trif, Desktop Test Engineering [:atrif]]

Hello and thank you for the report! I tried installing NVIDIA driver 535.98 on Ubuntu 22.04 to try to replicate the issue but unfortunately this led to a system crash. I will try again after I repair my machine. Until then I'm moving this to Core: Widget: GTK component to receive engineering input.

L.E: It seems that the version I get after updating the Nvidia drivers is 535.86.05 with GTX1080. I tried reproducing the issue with Firefox 116.03, 117, and the latest nightly (2023-08-23) by loading Google Maps, opening youtube videos, entering and exiting fullscreen, loading webgl content, and resizing browser but without any luck.

Comment 5

[Martin Stránský [:stransky] (ni? me)]

Please attach your about:support page. AFAIK we don't use dmabuf/va-api by default on NVIDIA.

Comment 6

[Martin Stránský [:stransky] (ni? me)]

Bug 1788573 may be related.

Comment 7

[Erik Kurzinger]

https://bugzilla.mozilla.org/show_bug.cgi?id=1788573 was a segfault in libnvidia-eglcore.so whereas this is happening in libnvidia-egl-wayland.so, so I don't think it's the same issue. Also that bug was pretty specific to our suspend / resume path which doesn't seem to be the case here.

On the other hand, https://bugzilla.mozilla.org/show_bug.cgi?id=1840360 could be related. Especially if this is triggered by going full-screen which would cause the compositor to send a new set of formats / modifiers.

Comment 8

[trevnels]

Attachment: about:support

(In reply to Martin Stránský [:stransky] (ni? me) from comment #5)
> Please attach your about:support page. AFAIK we don't use dmabuf/va-api by default on NVIDIA.

Comment 9

[trevnels]

Whoops, above is under Xorg. Will grab about:support under Wayland in a second. Most of the config overrides are from https://github.com/elFarto/nvidia-vaapi-driver/, which attempts to provide compatibility for va-api on NVIDIA devices - this may be related to the crashes, though other users of nvidia-vaapi-driver on wayland I've talked to are not seeing crashes.

Comment 10

[trevnels]

Attachment: about:support under Wayland

Comment 11

[trevnels]

Based on my (admittedly limited) understanding of the situation, my guess is that a regression took place sometime after dmabuf was blocklisted on NVIDIA in Bug 1788573, which recently started affecting those who have dmabuf force-enabled due to the vaapi compatibility layer linked above. This would explain the lack of people I've spoken to being able to reproduce the issue - people without the override would be unaffected since they would not have dmabuf enabled currently. I'll try disabling dmabuf to see if that works as a temporary fix.

Looking at the calendar and driver release dates, I can also recall that this behavior started right after upgrading to NVIDIA driver 535.98. Not sure if this is relevant for the Firefox Bugzilla, but all Electron applications on my system also began consistently segfaulting on startup around that time unless the --disable-gpu flag is passed. These are not using nvidia-vaapi-driver, but do seem to use dmabuf based on their logs. If these crashes are related it may be indicative of an upstream driver regression surrounding dmabuf instead of a problem within Firefox specifically.

Comment 12

[develop7]

Attachment: about:support on Wayland in FF 118 (affected)

about:support in affected FF 118 coming right up

Comment 13

[develop7]

Whoopsie, please disregard (and delete maybe?) the above attachment and related comment, I actually can't repro the crash in my FF 118.3 (aurora).

Comment 14

[trevnels]

Got the NVIDIA driver 535.104.05 update this morning and I have not been able to repro the crash since then with FF 117 either. Aforementioned issues with Electron also seem to be gone.

Comment 15

[trevnels]

(In reply to trevnels from comment #14)

Got the NVIDIA driver 535.104.05 update this morning and I have not been able to repro the crash since then with FF 117 either. Aforementioned issues with Electron also seem to be gone.

Looks like I spoke too soon, just saw another crash today.

Comment 16

[Martin Stránský [:stransky] (ni? me)]

Force enabled dmabuf/va-api. It's somehow expected to see issues here.

Comment 17

[develop7]

Attachment: about:support on Wayland in FF 118 (affected)

Just have reproduced issue, here's about:support

Comment 18

[develop7]

(In reply to develop7 from comment #17)
related crash https://crash-stats.mozilla.org/report/index/745b1ba3-cf9d-4fd5-ba1e-eb44e0230907#tab-details

Comment 19

[develop7]

Couldn't reproduce it with NVIDIA video driver upgraded to v545.29.02.

Comment 20

[Erik Kurzinger]

This may be fixed by the following egl-wayland commit https://github.com/NVIDIA/egl-wayland/commit/3f9889c35e4a159da3b6008303957ab109863f0b

Is the issue reproducible with version 1.1.13 of that library?

Comment 21

[BugBot [:suhaib / :marco/ :calixte]]

Closing because no crashes reported for 12 weeks.