1743833 - Wayland/Nvidia: Tab crash in [@ NvGlEglGetFunctions ] on bilibili's video page

Status: RESOLVED DUPLICATE of bug 1737834

(Core :: Graphics: CanvasWebGL, defect)

Description

[Alynx Zhou]

User Agent: Mozilla/5.0 (X11; Linux x86_64; rv:94.0) Gecko/20100101 Firefox/94.0

Steps to reproduce:

First set MOZ_ENABLE_WAYLAND=1, and open firefox for sometime (won't happen when firefox just started, typically after open for 1 hour or more), then open www.bilibili.com and open random video.

Actual results:

Sometimes the tab will crash.

Expected results:

Just open webpage as usual.

Comment 1

[Alynx Zhou]

Some other info:

NVIDIA driver has been 495.44 for a long time and not updated.

Can be reproduced on both stable and nightly.

X11 seems not have this problem, and if I disable wayland for firefox, then restart it in wayland, it could works fine for some time.

Not every time bilibili's video tab crash, but it happens often.

I have some crash report here:

https://crash-stats.mozilla.org/report/index/babc9254-3e0c-4dd1-ab47-642d10211201

https://crash-stats.mozilla.org/report/index/36306019-2ef3-4086-9316-b8d5d0211201

Crash report says that NvGlEglGetFunctions in libnvidia-eglcore.so.495.44 crashed, and it's from https://hg.mozilla.org/mozilla-central/file/15c1262d6e8f4524fa63dcc9cc2572038ffb3a92/gfx/gl/GLContext.cpp#l907 for nightly, which seems to be a workaround for NVIDIA driver in bug https://bugzilla.mozilla.org/show_bug.cgi?id=1367570.

Comment 2

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

The Bugbug bot thinks this bug should belong to the 'Core::Widget: Gtk' component, and is moving the bug to that component. Please revert this change in case you think the bot is wrong.

Comment 3

[Alynx Zhou]

Refer to the crash report I think this is more a WebGL/Graphics bug.

Comment 4

[Darkspirit]

Thanks for the report!

Comment 5

[Alynx Zhou]

I've checked my pacman.log:

NVIDIA has no version for nearly a month.

Linux kernel updated from 5.15.4 to 5.15.5 in 2021-11-29.

firefox stable updated from 94.0.1 to 94.0.2 in 2021-11-19.

This bug only happens rapidly today, I think it happens less then 2 times yesterday and never happens before, which really confused me because I nearly touched nothing.

I've already refreshed my firefox, tried private mode, disabled all extensions, it still happens.

Hoping those can help for addressing the problem.

Comment 6

[Robert Mader [:rmader]]

Can you paste your about:support here? (copy as text and paste it here - bugzilla will offer to make an attachment out of it) Thanks!

Comment 7

[Darkspirit]

Which egl-wayland version do you have?

Comment 8

[Alynx Zhou]

(In reply to Darkspirit from comment #7)

Which egl-wayland version do you have?

[2021-11-25T08:00:25+0800] [ALPM] upgraded egl-wayland (1:1.1.9+2+gdaab854-1 -> 2:1.1.9+r3+g582b2d3-1)

Comment 9

[Alynx Zhou]

Attachment: about:support

Comment 10

[Alynx Zhou]

(In reply to Robert Mader [:rmader] from comment #6)

Can you paste your about:support here? (copy as text and paste it here - bugzilla will offer to make an attachment out of it) Thanks!

https://bugzilla.mozilla.org/show_bug.cgi?id=1743833#c9

PS I am not sure why my comments continue to remove Crash Signature, I am sorry for the inconvenience.

Comment 11

[Alynx Zhou]

Also, I am using GNOME, but I am not sure mutter is related to EGL problem...

Comment 12

[Andrew Osmond [:aosmond] (he/him)]

Do you think the workaround for bug 1367570 is not necessary when we are using EGL?

https://searchfox.org/mozilla-central/rev/a3bddeeb9436fc7d0efba7adbd5c8ab31c2750cc/gfx/gl/GLContext.cpp#907

Comment 13

[Alynx Zhou]

Not sure if this is related, but recently I've got GL_FRAMEBUFFER_UNSUPPORTED for some opengl applications.

Minecraft reports GL_FRAMEBUFFER_UNSUPPORTED.

glxgears runs but the window is just black.

gtk4-demo reports Gsk:ERROR:../gtk/gsk/ngl/gsknglcommandqueue.c:1261:gsk_ngl_command_queue_create_render_target: assertion failed (glCheckFramebufferStatus (GL_FRAMEBUFFER) == GL_FRAMEBUFFER_COMPLETE): (0x00008cdd == 0x00008cd5)

I am not sure if this is related and where the problem is.

Comment 14

[Alynx Zhou]

A re-login temporary fixes GL_FRAMEBUFFER_UNSUPPORTED, but after using for hours it occurs again.

Comment 15

[Darkspirit]

EGL might intermittently or permanently be broken on experimental Xwayland and experimental Wayland: https://github.com/NVIDIA/egl-wayland/issues
Please try Nightly on Gnome X11 instead, it should be more reliable.

Comment 16

[Alynx Zhou]

(In reply to Darkspirit from comment #15)

EGL might intermittently or permanently be broken on experimental Xwayland and experimental Wayland: https://github.com/NVIDIA/egl-wayland/issues
Please try Nightly on Gnome X11 instead, it should be more reliable.

I can only reproduce this bug with Wayland, if I use MOZ_ENABLE_WAYLAND=0, it never occurs, should I report a bug for egl-wayland?

Comment 17

[Alynx Zhou]

I am trying to downgrade mutter to 41.1 to see if things changed.

Comment 18

[Alynx Zhou]

It works fine with mutter 41.1 for hours, I'll open an issue for mutter for help.

Comment 19

[Alynx Zhou]

https://gitlab.gnome.org/GNOME/mutter/-/issues/2045
Issue for mutter.

Comment 20

[Alynx Zhou]

It turns out that there are memory leaks when allocating buffers for hardware cursor in NVIDIA driver, which makes other GL program fails to create framebuffer after leaking too much. I'll close this bug and thanks for your help here.