1768260 - Nvidia/Wayland startup crash in [@ wl_log]: "error 7: failed to import supplied dmabufs: Arguments are inconsistent (for example, a valid context requires buffers not supplied by a"

Status: RESOLVED FIXED

(Core :: Graphics, defect)

Description

[Daniel Holbert [:dholbert]]

Today's Nightly is startup-crashing on my Lenovo ThinkPad when I use Ubuntu 22.04 in "Ubuntu-on-Wayland" mode (choosing that at the Ubuntu login screen using the little gear at the bottom right that appears at the password prompt).

Crash reports:
bp-72f6bbf9-4c85-4d12-a98e-e8af90220506
bp-4dc4a4a3-111b-44fc-917d-383190220506
bp-7929e9dc-a043-4728-aaf9-ccb6f0220506
bp-5208a2af-04c0-4a90-8cbd-414be0220506

mozregression range:
https://hg.mozilla.org/integration/autoland/pushloghtml?fromchange=188e42a20c32d29291defd8c6372f15283143c76&tochange=b2bea8e0b400e5e9a994cf148733505a92c0c624

--> Seems to be caused by bug 1767172

Crash report: https://crash-stats.mozilla.org/report/index/72f6bbf9-4c85-4d12-a98e-e8af90220506

MOZ_CRASH Reason: [destroyed object]: error 7: failed to import supplied dmabufs: Arguments are inconsistent (for example, a valid context requires buffers not supplied by a

Top 10 frames of crashing thread:

0 libxul.so mozilla::widget::WlCrashHandler widget/gtk/nsWaylandDisplay.cpp:336
1 libwayland-client.so.0 wl_log src/wayland-util.c:409
2 libwayland-client.so.0 display_handle_error src/wayland-client.c:1017
3 libffi.so.8 ffi_call_unix64 
4 libffi.so.8 ffi_call_int.lto_priv.0 
5 libwayland-client.so.0 wl_closure_invoke.constprop.0 src/connection.c:1025
6 libwayland-client.so.0 dispatch_event.isra.0 src/wayland-client.c:1583
7 libwayland-client.so.0 wl_display_dispatch_queue_pending src/wayland-client.c:1971
8 libnvidia-egl-wayland.so.1 libnvidia-egl-wayland.so.1@0x0000000000004bf9 
9 libnvidia-egl-wayland.so.1 libnvidia-egl-wayland.so.1@0x0000000000004baf 

Comment 1

[BugBot [:suhaib / :marco]]

Set release status flags based on info from the regressing bug 1767172

Comment 2

[Daniel Holbert [:dholbert]]

On a different machine -- a Microsoft Surface 3 -- latest Nightly runs just fine (also using Wayland on Ubuntu 22.04). so there's some hardware-specific nature here, i.e. this might not be reproducible for everyone.

Comment 3

[Darkspirit]

Also occured in bug 1739780 a few months ago: Which 'egl-wayland' version (Nvidia) do you have?

Comment 4

[Daniel Holbert [:dholbert]]

Also FWIW: on my affected system, my STR are just "start Firefox Nightly".'

An outline of a browser window briefly appears, and then I get a crash reporter dialog. This happens regardless of Firefox profile -- I can repro with a fresh profile, or my regular profile, or from mozregression with its default ephemeral profile.

On my affected system, I've got Ubuntu configured "Using NVIDIA driver metapackage from nvidia-driver-510 (proprietary, tested)" in Ubuntu's Software and Updates dialog.

And for some reason, at the login screen, my wayland-vs-xorg options are

Ubuntu
Ubuntu with Wayland

vs. on my unaffected system, the Surface 3, my login-screen wayland-vs-xorg options are

Ubuntu
Ubuntu with X.org

(i.e. the default seems to be different between these two systems; and the system that has wayland as the non-default option is the one that's affected by this issue.)

Comment 5

[Daniel Holbert [:dholbert]]

Attachment: output of `eglinfo` on the affected system (logged in w/ wayland, with bug reproducing)

Comment 6

[Daniel Holbert [:dholbert]]

(In reply to Darkspirit from comment #3)

Also occured in bug 1739780 a few months ago: Which 'egl-wayland' version (Nvidia) do you have?

Looks like I have:

Package: libnvidia-egl-wayland1
Architecture: amd64
Version: 1:1.1.9-1.1

Comment 7

[Darkspirit]

Login screen explanation:
Xorg is default ("Ubuntu") on Nvidia,
Wayland is default ("Ubuntu") on Intel&AMD.

Upstream bug of bug 1739780: https://github.com/NVIDIA/egl-wayland/issues/41#issuecomment-952843907

Comment 8

[Daniel Holbert [:dholbert]]

Thanks for clarifying the login screen scenario.

Ah, it looks like other apps are indeed affected; epiphany startup-crashes as well.

(Though, to reiterate, Firefox Nightly seems to have been fine before bug 1767172's patch, i.e. before today's Nightly. If bug 1767172 isn't super critical, I wonder if we should back it out and see if we can re-land it with adjustment to avoid tripping over this issue?)

Comment 9

[Darkspirit]

IIUC, the problem seems to occur when having Intel+Nvidia hybrid graphics and using Nvidia PRIME (=preferring Nvidia over Intel).
Firefox might have used the wrong but working GPU/driver before, maybe Mesa/Intel or Mesa/Nvidia.

Does Firefox start fine with these two possible workarounds?

• $ __EGL_VENDOR_LIBRARY_FILENAMES=/usr/share/glvnd/egl_vendor.d/50_mesa.json path/to/firefox
• Switching to Intel-only in nvidia settings.

Comment 10

[Daniel Holbert [:dholbert]]

(In reply to Darkspirit from comment #9)

Does Firefox start fine with these two possible workarounds?

• $ __EGL_VENDOR_LIBRARY_FILENAMES=/usr/share/glvnd/egl_vendor.d/50_mesa.json path/to/firefox

That works, yeah.

• Switching to Intel-only in nvidia settings.

For some reason, the "Intel (Power Saving Mode)" option is grayed out for me in NVIDIA Settings "PRIME Profiles" section. My current selection (default AFAIK) is NVIDIA On-Demand, and my only available alternative is "NVIDIA (Performance Mode)"

Comment 11

[Daniel Holbert [:dholbert]]

Attachment: screenshot of dholbert's nvidia-settings "PRIME Profiles" options

Comment 12

[Darkspirit]

Is Nvidia-only ("Performance Mode") also broken? (possibly a reboot required)
Or is just on-demand broken?

Comment 13

[Daniel Holbert [:dholbert]]

(In reply to Daniel Holbert [:dholbert] from comment #10)

For some reason, the "Intel (Power Saving Mode)" option is grayed out for me in NVIDIA Settings "PRIME Profiles" section

(ah, maybe that was just because the graphical app doesn't automatically have root privileges. I did see a reference to prime-select in the terminal output, and I tried running that, and it told me it needed root; and I was able to do sudo /usr/bin/prime-select intel.

Firefox still startup-crashes, but I assume I need a reboot; trying that now...)

(In reply to Darkspirit from comment #12)

Is Nvidia-only ("Performance Mode") also broken? (possibly a reboot required)
Or is just on-demand broken?

I'll test this too.

Comment 14

[Daniel Holbert [:dholbert]]

The Intel "PRIME Profile" is fine; it successfully avoids the crash. (chosen via sudo /usr/bin/prime-select intel + reboot)

The NVIDIA/performance-mode "PRIME Profile" does not avoid the crash. (chosen via sudo /usr/bin/prime-select nvidia + reboot)

Comment 15

[Daniel Holbert [:dholbert]]

(In reply to Daniel Holbert [:dholbert] from comment #8)

(Though, to reiterate, Firefox Nightly seems to have been fine before bug 1767172's patch, i.e. before today's Nightly. If bug 1767172 isn't super critical, I wonder if we should back it out and see if we can re-land it with adjustment to avoid tripping over this issue?)

Update: chatted with emilio and CosminS|sheriffduty, and we're going to go ahead and back out bug 1767172 and respin Nightly, to minimize the chance of this affecting any other Nightly users on configurations that are susceptible to this. We can follow up in bug 1767172 to see why it cause this on Monday.

(In the meantime, I'm happy to continue testing stuff if it's useful, though!)

Comment 16

[Daniel Holbert [:dholbert]]

FIXED-by-backout (bug 1767172 comment 4).

Emilio figured this out in matrix; essentially, before he landed bug 1767172's patch yesterday, we were crashing early in glxtest, and we'd fall back to software-webrender as a result.

bug 1767172 removed the code that was crashing in glxtest, so we think the driver is OK and we launch with full webrender and then crash.

He's re-landing bug 1767172 with a minimal tweak to let us resume crashing in glxtest on this configuration, so that we fall back to software-webrender.