Skip to content
Snippets Groups Projects
New look: Off

Implement Anbox Wayland backend

    • Closed Imported Issue created by Administrator @root

    to force the Wayland backend, we need to export EGL_PLATFORM=wayland SDL_VIDEODRIVER=wayland. Unfortunately Anbox doesn't work with it.

    [ 2020-03-08 11:46:03] [Renderer.cpp:131@initialize] Failed to select EGL configuration
Stack trace (most recent call last) in thread 27758:
#2    Object "/lib/ld-musl-x86_64.so.1", at 0x7fc49a1e227c, in
#1    Object "/usr/bin/anbox", at 0x5637247e875e, in backward::SignalHandling::sig_handler(int, siginfo_t*, void*)
#0    Object "/usr/bin/anbox", at 0x5637247e85af, in backward::SignalHandling::handleSignal(int, siginfo_t*, void*)
zsh: segmentation fault  anbox session-manager

    I could reproduce this error with both Intel driver on x86_64 with Sway and Pinephone with Plasma Mobile.

    This is apparently caused by the requirement of Anbox on pbuffers. Since every available EGL configuration doesn't support it, because wayland EGL doesn't implement EGL, there is no EGL configuration that can be selected. I've tried working around it a bit, but I always had segfaults either in the session-manager or inside Android's surfaceflinger.

    Support for it should be added upstream. A good issue to bump is this one, https://github.com/anbox/anbox/issues/1270.

    Edited
    To upload designs, you'll need to enable LFS and have an admin enable hashed storage. More information

    Child items 0

    • No child items are currently assigned. Use child items to break down this issue into smaller parts.

    Activity

    • Administrator
      Administrator @root · Imported
      Author Owner

      The relevant part of src/anbox-3ed2e6d5c360d57b6aa61386e279adf3ff155ded/src/anbox/graphics/emugl/Renderer.cpp:

      117   s_egl.eglBindAPI(EGL_OPENGL_ES_API);
118
119   // Create EGL context for framebuffer post rendering.
120   GLint surfaceType = EGL_WINDOW_BIT | EGL_PBUFFER_BIT;
121   const GLint configAttribs[] = {EGL_RED_SIZE, 8,
122                                  EGL_GREEN_SIZE, 8,
123                                  EGL_BLUE_SIZE, 8,
124                                  EGL_SURFACE_TYPE, surfaceType,
125                                  EGL_RENDERABLE_TYPE, EGL_OPENGL_ES2_BIT,
126                                  EGL_NONE};
127
128   int n;
129   if ((s_egl.eglChooseConfig(m_eglDisplay, configAttribs, &m_eglConfig,
130                              1, &n) == EGL_FALSE) || n == 0) {
131     ERROR("Failed to select EGL configuration");
132     return false;
133   }

      By Antoine Fontaine on 2020-03-08T11:49:48

    • Administrator added type::bug + 1 deleted label · Imported

      added type::bug + 1 deleted label

      • Administrator
        Administrator @root · Imported
        Author Owner

        Curious, will it pass this check if EGL_PBUFFER_BIT is removed?

        By Alexey Min on 2020-03-08T21:52:43

        Edited by Ghost User
      • Administrator
        Administrator @root · Imported
        Author Owner

        The anbox session manager still shows Failed to select EGL configuration and segfaults seconds later. Worth trying.

        By Antoine Fontaine on 2020-03-08T21:52:43

      • Administrator
        Administrator @root · Imported
        Author Owner

        I tried on my computer, but I didn't see the same result:

        [ 2020-03-08 23:08:46] [Renderer.cpp:170@initialize] Using a surfaceless EGL context
[ 2020-03-08 23:08:46] [Renderer.cpp:203@initialize] Failed: Initialize set of configs

        And the session manager doesn't segfault. Instead, it SIGABRT inside the image: Abort message: 'no suitable EGLConfig found, giving up'

        03-08 22:52:02.931  4088  4088 F DEBUG   : *** *** *** *** *** *** *** *** *** *** *** *** *** *** *** ***
03-08 22:52:02.931  4088  4088 F DEBUG   : Build fingerprint: 'Android/anbox_x86_64/x86_64:7.1.2/N2G47J/lambda02221206:userdebug/test-keys'
03-08 22:52:02.931  4088  4088 F DEBUG   : Revision: '0'
03-08 22:52:02.931  4088  4088 F DEBUG   : ABI: 'x86_64'
03-08 22:52:02.932  4088  4088 F DEBUG   : pid: 4077, tid: 4077, name: surfaceflinger  >>> /system/bin/surfaceflinger <<<
03-08 22:52:02.932  4088  4088 F DEBUG   : signal 6 (SIGABRT), code -6 (SI_TKILL), fault addr --------
03-08 22:52:02.933  4088  4088 F DEBUG   : Abort message: 'no suitable EGLConfig found, giving up'
03-08 22:52:02.933  4088  4088 F DEBUG   :     rax 0000000000000000  rbx 00007f613abb6be8  rcx 00007f613a6a4987  rdx 0000000000000006
03-08 22:52:02.933  4088  4088 F DEBUG   :     rsi 0000000000000fed  rdi 0000000000000fed
03-08 22:52:02.933  4088  4088 F DEBUG   :     r8  00007ffce49b1080  r9  00007ffce498d758  r10 0000000000000008  r11 0000000000000206
03-08 22:52:02.933  4088  4088 F DEBUG   :     r12 0000000000000fed  r13 0000000000000006  r14 00007f613a441a6a  r15 00007ffce498ded0
03-08 22:52:02.933  4088  4088 F DEBUG   :     cs  0000000000000033  ss  000000000000002b
03-08 22:52:02.933  4088  4088 F DEBUG   :     rip 00007f613a6a4987  rbp 0000000000000016  rsp 00007ffce498d798  eflags 0000000000000206
03-08 22:52:02.935  4088  4088 F DEBUG   :
03-08 22:52:02.935  4088  4088 F DEBUG   : backtrace:
03-08 22:52:02.935  4088  4088 F DEBUG   :     #00 pc 000000000008d987  /system/lib64/libc.so (tgkill+7)
03-08 22:52:02.935  4088  4088 F DEBUG   :     #01 pc 000000000008a421  /system/lib64/libc.so (pthread_kill+65)
03-08 22:52:02.935  4088  4088 F DEBUG   :     #02 pc 00000000000302c1  /system/lib64/libc.so (raise+17)
03-08 22:52:02.935  4088  4088 F DEBUG   :     #03 pc 00000000000287dd  /system/lib64/libc.so (abort+77)
03-08 22:52:02.935  4088  4088 F DEBUG   :     #04 pc 000000000001064f  /system/lib64/libcutils.so (__android_log_assert+287)
03-08 22:52:02.935  4088  4088 F DEBUG   :     #05 pc 000000000003d92d  /system/lib64/libsurfaceflinger.so
03-08 22:52:02.935  4088  4088 F DEBUG   :     #06 pc 000000000003d345  /system/lib64/libsurfaceflinger.so
03-08 22:52:02.936  4088  4088 F DEBUG   :     #07 pc 0000000000043278  /system/lib64/libsurfaceflinger.so (_ZN7android14SurfaceFlinger4initEv+936)
03-08 22:52:02.936  4088  4088 F DEBUG   :     #08 pc 000000000000130d  /system/bin/surfaceflinger
03-08 22:52:02.936  4088  4088 F DEBUG   :     #09 pc 000000000001c934  /system/lib64/libc.so (__libc_init+84)
03-08 22:52:02.936  4088  4088 F DEBUG   :     #10 pc 00000000000011c4  /system/bin/surfaceflinger
03-08 22:52:02.936  4088  4088 F DEBUG   :     #11 pc 0000000000000000  <unknown>
03-08 22:52:02.954    13    13 W         : debuggerd: resuming target 4077
03-08 22:52:02.959    32    32 I ServiceManager: service 'media.audio_flinger' died
03-08 22:52:02.959    32    32 I ServiceManager: service 'media.audio_policy' died
03-08 22:52:02.959    32    32 I ServiceManager: service 'media.sound_trigger_hw' died
03-08 22:52:02.959    32    32 I ServiceManager: service 'media.radio' died
03-08 22:52:02.960    32    32 I ServiceManager: service 'media.player' died
03-08 22:52:02.960    32    32 I ServiceManager: service 'media.resource_manager' died
03-08 22:52:02.960    32    32 I ServiceManager: service 'netd' died
03-08 22:52:02.970    40    40 E         : eof
03-08 22:52:02.970    40    40 E         : failed to read size
03-08 22:52:02.970    40    40 I         : closing connection
03-08 22:52:02.970    32    32 I ServiceManager: service 'batterystats' died
03-08 22:52:02.970    32    32 I ServiceManager: service 'appops' died
03-08 22:52:02.970    32    32 I ServiceManager: service 'power' died
03-08 22:52:02.970    32    32 I ServiceManager: service 'display' died

        By Antoine Fontaine on 2020-03-08T23:15:50

        Edited by Administrator
      • Administrator
        Administrator @root · Imported
        Author Owner

        If I short-circuit that part, and replace if (!surfaceless_supported) by a if (1), I have this error instead. On x86_64/intel drivers.

        [ 2020-03-08 23:14:01] [Renderer.cpp:166@initialize] Failed to create pbuffer surface: error=0x3009
Stack trace (most recent call last) in thread 4831:
#2    Object "/lib/ld-musl-x86_64.so.1", at 0x7fad1098c27c, in
#1    Object "/usr/bin/anbox", at 0x55ae6009875e, in backward::SignalHandling::sig_handler(int, siginfo_t*, void*)
#0    Object "/usr/bin/anbox", at 0x55ae600985af, in backward::SignalHandling::handleSignal(int, siginfo_t*, void*)
zsh: segmentation fault  anbox session-manager

        By Antoine Fontaine on 2020-03-08T23:18:17

        Edited by Administrator
      • Please register or sign in to reply
    • Administrator mentioned in issue #487 (closed) · Imported

      mentioned in issue #487 (closed)

      By Antoine Fontaine on 2020-03-10T20:31:45

      • Administrator
        Administrator @root · Imported
        Author Owner

        I'll just leave it here (copied from matrix)

        https://hg.mozilla.org/integration/autoland/rev/9250f12dd57c

        > Maybe you can use surfaceless EGL with an FBO.
>
> When the KHR_surfaceless_gles2 extension is present you should be able
> to do eglMakeCurrent with only an EGLContext and no EGLSurface.

If you're using the wayland egl platform you can just create a dummy
wl_egl_window and then create an EGL window surface for that.

        By Alexey Min on 2020-03-10T22:00:39

        Edited by Ghost User
      • Administrator
        Administrator @root · Imported
        Author Owner

        This is saddly all magic to me at the moment. I don't think I'd be able to implement that yet.

        By Antoine Fontaine on 2020-03-10T22:00:38

      • Administrator
        Administrator @root · Imported
        Author Owner

        After reading that mozilla commit the solution looks to be to not try to select EGL config with pbuffers, and do not use pbuffers later at all (use wayland surfaces instead, that are created differently)

        That may or may not be easy :(

        By Alexey Min on 2020-03-10T22:10:20

      • Administrator
        Administrator @root · Imported
        Author Owner

        I managed to go past the EGL configuration selection several times, but then it segfaults inside the image. I hope that doesn't mean the Android image should be fixed. Between the sheer requirements in size, the compilation time and the glibc dependency, that wouldn't seem fun to debug.

        By Antoine Fontaine on 2020-03-10T22:23:03

      • Administrator
        Administrator @root · Imported
        Author Owner

        a patch with things I tried, in case it could be useful: fix-wayland.patch

        By Antoine Fontaine on 2020-03-10T22:28:22

      • Administrator
        Administrator @root · Imported
        Author Owner

        I hope that doesn't mean the Android image should be fixed.

        It doesn't, android should be almost generic inside, it's the anbox part that needs fixing (IMO)

        By Alexey Min on 2020-03-11T06:39:08

      • Please register or sign in to reply
    • Administrator changed title from Force Anbox to use the Wayland backend to Implement Anbox Wayland backend · Imported

      changed title from Force Anbox to use the Wayland backend to Implement Anbox Wayland backend

      By Antoine Fontaine on 2020-03-10T22:13:47

    • Administrator changed the description · Imported

      changed the description

      By Antoine Fontaine on 2020-03-10T22:13:47

    • Administrator mentioned in issue #495 (closed) · Imported

      mentioned in issue #495 (closed)

      By Antoine Fontaine on 2020-03-18T08:47:07

    • Administrator changed milestone to %Anbox · Imported

      changed milestone to %Anbox

      By Oliver Smith on 2020-03-22T02:10:29

    • Administrator mentioned in issue #556 (closed) · Imported

      mentioned in issue #556 (closed)

      By Antoine Fontaine on 2020-04-21T08:56:14

    • Administrator added help wanted label · Imported

      added help wanted label

    • Administrator
      Administrator @root · Imported
      Author Owner

      Anbox has been dropped from Alpine completely, so I'm closing this.

      See https://gitlab.alpinelinux.org/alpine/aports/-/merge_requests/33087

      By Newbyte on 2022-04-19T09:47:29

    • Administrator closed · Imported

      closed

    Please register or sign in to reply