[regression] Starting in Mesa 12.0.2, Qt/QML apps don't render any more (logs say "QEGLPlatformContext: eglMakeCurrent failed: 3001")

[dinamic]

nothing works, all apps, dash are just black windows - Ubuntu 16.10 + proposed (x86, and using Nvidia 8600GT with nouveau drivers)

and they fail with this error "QEGLPlatformContext: eglMakeCurrent failed: 3001"

[Daniel van Vugt (vanvugt)]

Thank you for taking the time to report this bug and helping to make Ubuntu better. Please execute the following command only once, as it will automatically gather debugging information, in a terminal:
   apport-collect 1620934

When reporting bugs in the future please use apport by using 'ubuntu-bug' and the name of the package affected. You can learn more about this functionality at https://wiki.ubuntu.com/ReportingBugs.

[Daniel van Vugt (vanvugt)]

Or more simply, please detail what graphics hardware (or VM) you are running on.

[dinamic (dinamic6661)]

it's a desktop install x86, and using Nvidia 8600GT with nouveau drivers

[dinamic (dinamic6661)]

it worked fine until the last update

[Daniel van Vugt (vanvugt)]

Thanks. This is probably the same issue as bug 1553328. Mesa and many other bits have changed since that was reported. So the symptoms could easily have changed.

We (mir-team) should re-test nouveau for bug 1553328 and possibly change the description to match this bug if appropriate now.

[Daniel van Vugt (vanvugt)]

Just tried nouveau for the first time in a long time on a pure 16.10 machine.

Mir and Unity8 seems to work fine for me. Although not for that long, before I encountered a complete system freeze. And the usual nouveau errors in the kernel log.

[Daniel van Vugt (vanvugt)]

Error 3001 is EGL_NOT_INITIALIZED. No idea how you got there though.

Please try attaching ~/.cache/upstart/unity8.log* and /var/log/lightdm/unity-system-compositor.log* so we can look for any errors that might have come earlier.

[dinamic (dinamic6661)]

i also have proposed enabled

[dinamic (dinamic6661)]

hm.. i think the bug is Qt? gtk mir apps seems to work fine.

the problem is with Qt/KDE apps and ubuntu apps. ubuntu terminal is black, the dash is black, in music app i can see the background but it doesn't have any content, same with the settings

[dinamic (dinamic6661)]

segfault libQt5Gui.so.5.6.1
segfault libEGL.so.1.0.0

[Daniel van Vugt (vanvugt)]

unity8.log is a little bit flooded and even unity8.log.1 isn't old enough to show us start-up. Can you please provide unity8.log.[2-9] from when the bug is occurring?

[Larry Price (larryprice)]

I've started experiencing the same issues on yakkety, even after reinstalling unity8-desktop-session. My machine is a 64-bit with integrated Intel graphics - unity8-dash.log attached.

[Daniel van Vugt (vanvugt)]

And after updating now me too (Intel graphics).

unity8-dash.log:
[2016-09-09:10:15:04.738] QEGLPlatformContext: eglMakeCurrent failed: 3001
[2016-09-09:10:15:04.738] QEGLPlatformContext: eglMakeCurrent failed: 3001
[2016-09-09:10:15:04.754] QEGLPlatformContext: eglMakeCurrent failed: 3001
[2016-09-09:10:15:04.754] QEGLPlatformContext: eglMakeCurrent failed: 3001
[2016-09-09:10:15:04.770] QEGLPlatformContext: eglMakeCurrent failed: 3001
[2016-09-09:10:15:04.770] QEGLPlatformContext: eglMakeCurrent failed: 3001
[2016-09-09:10:15:04.786] QEGLPlatformContext: eglMakeCurrent failed: 3001
[2016-09-09:10:15:04.786] QEGLPlatformContext: eglMakeCurrent failed: 3001
[2016-09-09:10:15:04.802] QEGLPlatformContext: eglMakeCurrent failed: 3001
[2016-09-09:10:15:04.802] QEGLPlatformContext: eglMakeCurrent failed: 3001
[2016-09-09:10:15:04.819] QEGLPlatformContext: eglMakeCurrent failed: 3001

[Daniel van Vugt (vanvugt)]

Native Mir demos still render correctly. So the problem is specific to the Ubuntu/Qt toolkit side.

[Daniel van Vugt (vanvugt)]

Sorry for the noise. I don't know which project to target but it's not Mir, Mesa or Unity8...

[Michał Sawicz (saviq)]

@dinamic, can you please look at /var/log/apt/history.log and try to estimate which upgrade broke it?

[Daniel van Vugt (vanvugt)]

The regression appears to have happened for me in this upgrade:

Start-Date: 2016-09-09 09:40:23
Commandline: apt full-upgrade
Requested-By: dan (1000)
Upgrade: ubuntu-settings:amd64 (16.10.2, 16.10.3), suru-icon-theme:amd64 (16.10+16.10.20160829.3-0ubuntu1, 16.10+16.10.20160908-0ubuntu1), libaudit-common:amd64 (1:2.6.5-1ubuntu1, 1:2.6.6-1ubuntu1), qtdeclarative5-qtmir-plugin:amd64 (0.4.8+16.10.20160831-0ubuntu1, 0.4.8+16.10.20160906-0ubuntu1), qtmir-desktop:amd64 (0.4.8+16.10.20160831-0ubuntu1, 0.4.8+16.10.20160906-0ubuntu1), docbook-xml:amd64 (4.5-7.3, 4.5-8), gnome-software:amd64 (3.20.1+git20160617.1.0440874.ubuntu-xenial-0ubuntu1, 3.20.1+git20160617.1.0440874.ubuntu-xenial-0ubuntu2), ubuntu-software:amd64 (3.20.1+git20160617.1.0440874.ubuntu-xenial-0ubuntu1, 3.20.1+git20160617.1.0440874.ubuntu-xenial-0ubuntu2), ubuntu-mobile-icons:amd64 (16.10+16.10.20160829.3-0ubuntu1, 16.10+16.10.20160908-0ubuntu1), libsodium18:amd64 (1.0.10-1, 1.0.11-1), light-themes:amd64 (16.10+16.10.20160829.3-0ubuntu1, 16.10+16.10.20160908-0ubuntu1), ubuntu-artwork:amd64 (1:16.10+16.10.20160829.3-0ubuntu1, 1:16.10+16.10.20160908-0ubuntu1), libaudit-dev:amd64 (1:2.6.5-1ubuntu1, 1:2.6.6-1ubuntu1), pciutils:amd64 (1:3.3.1-1.1ubuntu2, 1:3.3.1-1.1ubuntu3), mythes-en-us:amd64 (1:5.2.0-1, 1:5.2.1-1), mtools:amd64 (4.0.18-2, 4.0.18-2ubuntu1), libpci3:amd64 (1:3.3.1-1.1ubuntu2, 1:3.3.1-1.1ubuntu3), gnome-software-common:amd64 (3.20.1+git20160617.1.0440874.ubuntu-xenial-0ubuntu1, 3.20.1+git20160617.1.0440874.ubuntu-xenial-0ubuntu2), ubuntu-mono:amd64 (16.10+16.10.20160829.3-0ubuntu1, 16.10+16.10.20160908-0ubuntu1), libaudit1:amd64 (1:2.6.5-1ubuntu1, 1:2.6.6-1ubuntu1)
End-Date: 2016-09-09 09:40:50

[Daniel van Vugt (vanvugt)]

Although it could also have happened a day or two ago in a different upgrade. I might not have noticed.

[Michał Sawicz (saviq)]

The only relevant piece there seems to be qtmir, but looking at its changelog [1] there's no related changes there...

[1] https://launchpad.net/ubuntu/+source/qtmir/0.4.8+16.10.20160831-0ubuntu1

[Daniel van Vugt (vanvugt)]

Yeah it could have happened a day or two ago. Probably been a couple of days since I tested Unity8 apps.

[Daniel van Vugt (vanvugt)]

What we need is someone with a slightly out of date yakkety that doesn't have the bug. Then upgrade it, check the regression now occurs and work from that.

[Daniel d'Andrada (dandrader)]

This only happens in yakkety. The same qtmir, qtubuntu and unity8 runs fine in xenial. Lost a couple of hours on this issue yesterday and my "solution" was to reboot my test laptop into a xenial partition so I could unblock myself.

[Daniel d'Andrada (dandrader)]

Maybe the culprit is different Qt versions between xenial and yakkety?

[Daniel d'Andrada (dandrader)]

qtbase-opensource-src did get a new release on september 3rd. But xenial stable-phone-overlay and yakkety seem to be carrying the very same qt version, so scrap my theory from comment #30...

[Larry Price (larryprice)]

In my case, the Dash shows up as just a black screen with the Dash icon and "Scopes" and never finishes loading. When opening up System Settings, it sits on a white screen with just the titlebar. When opening up the web browser, I got a crash and a weird update notifier log (attached) in addition to this message in the regular application-legacy-webbrowser-app-log:

    Loading module: 'libubuntu_application_api_desktop_mirclient.so.3.0.0'
    could not open containers config file "/home/lrp/.local/share/libertine/ContainersConfig.json"
    [0912/145919:ERROR:gl_surface_egl.cc(253)] No suitable EGL configs found.

[Larry Price (larryprice)]

The System Settings log could actually contain some interesting information. It seems to imply that the launcher is out of memory.

[Michał Sawicz (saviq)]

@larryprice the black Scopes and blank Settings are splash screens, basically means the app never rendered.

[Daniel van Vugt (vanvugt)]

And the app never rendered because its call to eglMakeCurrent failed with error 3001 (EGL_NOT_INITIALIZED).

[Daniel van Vugt (vanvugt)]

Another prime candidate that possibly triggered this bug:
   https://launchpad.net/ubuntu/+source/mesa/12.0.2-1ubuntu1
Released on 7 September (which would have been 8 Sept here).

It could easily be that our Qt apps are (and always have been) requesting some EGL attributes that Mesa can not or should not provide. That could be a Mesa bug or a Qt bug that's been around for much longer and Mesa only just became pedantic enough to say no. Remember EGL clients from the 'mir-demos' package still render OK so this is something specific to Q* clients.

[Daniel van Vugt (vanvugt)]

CONFIRMED. Downgrading Mesa to 12.0.1 fixes the problem:
https://launchpad.net/ubuntu/+source/mesa/12.0.1-3ubuntu2/+build/10506751

So the regression was triggered by:
https://launchpad.net/ubuntu/+source/mesa/12.0.2-1ubuntu1

However that doesn't mean Mesa is buggy necessarily. Our Qt EGL setup code might be requesting something invalid and it only just became a fatal error. In that case our Qt code would need fixing rather than Mesa. Indeed non-Qt clients don't have the bug at all.

[Daniel van Vugt (vanvugt)]

Enabling debugging I can't see anything EGL-related failing before the errors start in unity8-dash.log. Although rendering appears to fail:

[2016-09-13:12:45:18.825] ubuntumirclient: Created surface with geometry: QRect(
0,24 510x712) title: "Scopes" role: 1
[2016-09-13:12:45:18.825] ubuntumirclient.graphics: Requested format: QSurfaceFormat(version 2.0, options QFlags(), depthBufferSize 24, redBufferSize 8, greenBufferSize 8, blueBufferSize 8, alphaBufferSize -1, stencilBufferSize 8, samples -1, swapBehavior 2, swapInterval 1, profile 0)
Actual format: QSurfaceFormat(version 2.0, options QFlags(), depthBufferSize 24, redBufferSize 8, greenBufferSize 8, blueBufferSize 8, alphaBufferSize 8, stencilBufferSize 8, samples 0, swapBehavior 0, swapInterval 1, profile 0) with associated Mir pixel format: XRGB8888
[2016-09-13:12:45:18.825] ubuntumirclient: UbuntuWindow(window=0x55e51c635990, s
creen=0x55e51c53cd80, input=0x55e51c53cbe0, surf=0x55e51ca3eae0) with title 'Sco
pes', role: '1'
[2016-09-13:12:45:18.825] ubuntumirclient: setVisible (window=0x55e51c635990, vi
sible=true)
[2016-09-13:12:45:18.825] ubuntumirclient: updateSurfaceState (window=0x55e51c63
5990, surfaceState=restored)
[2016-09-13:12:45:18.826] file:///usr/share/unity8//Dash/Dash.qml:39: ReferenceE
rror: window is not defined
[2016-09-13:12:45:18.826] file:///usr/share/unity8//Dash/GenericScopeView.qml:64
4: TypeError: Cannot read property 'activeFiltersCount' of null
[2016-09-13:12:45:18.826] file:///usr/share/unity8//Dash/GenericScopeView.qml:64
3: TypeError: Cannot read property 'filters' of null
[2016-09-13:12:45:18.826] file:///usr/share/unity8//Dash/Dash.qml:265: ReferenceError: scopeStyle is not defined
[2016-09-13:12:45:18.826] ubuntumirclient.input: customEvent(type=mir_event_type_surface_output)
[2016-09-13:12:45:18.826] User agent string: "release=16.10&scopes-api=1.0.6+16.10.20160617-0ubuntu1&plugin=0.5.7+16.10.20160624.2-0ubuntu2&unity8=8.14+16.10.20160831.3-0ubuntu1"
[2016-09-13:12:45:18.827] qt.scenegraph.renderloop: exposureChanged() QQuickWindowQmlImpl(0x55e51c635990)
[2016-09-13:12:45:18.827] qt.scenegraph.renderloop: handleExposure() QQuickWindowQmlImpl(0x55e51c635990)
[2016-09-13:12:45:18.827] qt.scenegraph.renderloop: - adding window to list
[2016-09-13:12:45:18.827] qt.scenegraph.renderloop: - starting render thread
[2016-09-13:12:45:18.829] ubuntumirclient.graphics: EGL vendor: Mesa Project
[2016-09-13:12:45:18.829] ubuntumirclient.graphics: EGL version: 1.4 (DRI2)
[2016-09-13:12:45:18.829] ubuntumirclient.graphics: EGL extensions: EGL_EXT_buffer_age EGL_EXT_image_dma_buf_import EGL_KHR_create_context EGL_KHR_get_all_proc_addresses EGL_KHR_gl_renderbuffer_image EGL_KHR_gl_texture_2D_image EGL_KHR_gl_texture_cubemap_image EGL_KHR_image EGL_KHR_image_base EGL_KHR_image_pixmap EGL_KHR_reusable_sync EGL_KHR_surfaceless_context EGL_MESA_configless_context EGL_MESA_drm_image EGL_MESA_image_dma_buf_export
[2016-09-13:12:45:18.829] ubuntumirclient.graphics: EGL configuration attributes:
[2016-09-13:12:45:18.829] EGL_BUFFER_SIZE: 32
[2016-09-13:12:45:18.829] EGL_ALPHA_SIZE: 8
[2016-09-13:12:45:18.829] EGL_BLUE_SIZE: 8
[2016-09-13:12:45:18.829] ...

[Daniel van Vugt (vanvugt)]

Maybe... Mesa only just started to care about mixing and matching GL with GLES in invalid ways. I know we have done it in the past and possibly still do. Maybe today is the day Mesa finally started checking the validity of our GL calls...?

[Gerry Boland (gerboland)]

Note it is only Unity8 which is mixing GL and GLES calls, clients are not doing that. Qt clients are only doing GL calls on desktop.

Nothing obvious comes to mind here.

It may be Qt's EGL configuration code is failing - code that QtUbuntu is using. Another consumer of that same code is the EGL backend for the XCB QPA plugin - the non-default - as usually the GLX backend is used.

Could someone with a problematic system log into Unity7, and for a simple qml file, run:
QT_XCB_GL_INTEGRATION=xcb_egl qmlscene ~/test.qml
and see does it fail?

[Daniel d'Andrada (dandrader)]

On 13/09/2016 08:49, Gerry Boland wrote:
> Could someone with a problematic system log into Unity7, and for a simple qml file, run:
> QT_XCB_GL_INTEGRATION=xcb_egl qmlscene ~/test.qml
> and see does it fail?

It works fine.

[Timo Aaltonen (tjaalton)]

Probably caused by

commit de695014eb6c819b050a4f7f2f952c4466047292
Author: Nicolas Boichat <email address hidden>
Date: Fri Jul 15 16:24:20 2016 +0800

    egl/dri2: dri2_make_current: Set EGL error if bindContext fails

[Timo Aaltonen (tjaalton)]

    Without this, if a configuration is, say, available only on GLES2/3, but
    not on GLES1, and is rejected by the dri module's bindContext call,
    eglMakeCurrent fails with error "EGL_SUCCESS".

    In this patch, we set error to EGL_BAD_MATCH, which is what CTS/dEQP
    dEQP-EGL.functional.surfaceless_context expect.

[Daniel van Vugt (vanvugt)]

See also bug 1620994 which happened in the upgrade to 12.0.2

[Timo Aaltonen (tjaalton)]

so, will you fix mir soon or should I revert that mesa commit for now?

[Daniel van Vugt (vanvugt)]

tjaalton: Yes please revert the Mesa commit if it solves the problem.

Either way, we don't yet know what/where to fix in Mir/Unity8/Qt*

[Timo Aaltonen (tjaalton)]

that wasn't the commit to revert, and Chris fixed the mir egl patch. I'll upload it in a bit

[Launchpad Janitor (janitor)]

This bug was fixed in the package mesa - 12.0.3-1ubuntu1

---------------
mesa (12.0.3-1ubuntu1) yakkety; urgency=medium

  * Merge from Debian.

  [ Christopher James Halse Rogers ]
  * egl-platform-mir.patch: Updated for upstream changes to dri2_initialize
    in egl_dri2.c. (LP: #1620934, #1620994)

mesa (12.0.3-1) unstable; urgency=medium

  * New upstream release.
  * Enable vc4 on arm64.

 -- Timo Aaltonen <email address hidden> Fri, 16 Sep 2016 10:07:05 +0300