1972290 - [KDE][HDR] MOZ_RELEASE_ASSERT(!mAttachedToSurface && !mIsAttachedToCompositor)

Status: RESOLVED FIXED

(Core :: Widget: Gtk, defect, P3)

Description

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

Looks like the same buffer is reused too early:

#10 MOZ_CrashSequence (aAddress=0x0, aLine=104) at /raid/src/objdir/dist/include/mozilla/Assertions.h:248
#11 0x00007f8f968b083b in mozilla::widget::WaylandBuffer::BorrowBuffer (this=0x7f8f394545b0, aSurfaceLock=...) at /raid/src/widget/gtk/WaylandBuffer.cpp:103
#12 0x00007f8f968bb26d in mozilla::widget::WaylandSurface::AttachLocked (this=0x7f8f2d3c2100, aSurfaceLock=..., aWaylandBuffer=[(mozilla::widget::WaylandBufferDMABUF *)] = {...})
    at /raid/src/widget/gtk/WaylandSurface.cpp:1129
#13 0x00007f8f901ca30b in mozilla::layers::NativeLayerWaylandRender::CommitSurfaceToScreenLocked (this=0x7f8f5379b450, aProofOfLock=...) at /raid/src/gfx/layers/NativeLayerWayland.cpp:1000
#14 0x00007f8f901c594a in mozilla::layers::NativeLayerWayland::UpdateLayer (this=0x7f8f5379b450, aScale=2) at /raid/src/gfx/layers/NativeLayerWayland.cpp:725
#15 0x00007f8f901c4e56 in mozilla::layers::NativeLayerRootWayland::UpdateLayersLocked (this=0x7f8f504d51c0, aProofOfLock=...) at /raid/src/gfx/layers/NativeLayerWayland.cpp:393
#16 0x00007f8f901c629c in mozilla::layers::NativeLayerRootWayland::CommitToScreen (this=0x7f8f504d51c0) at /raid/src/gfx/layers/NativeLayerWayland.cpp:473
#17 0x00007f8f90a08d6e in mozilla::wr::RenderCompositorNative::CompositorEndFrame (this=0x7f8f4ea434b0) at /raid/src/gfx/webrender_bindings/RenderCompositorNative.cpp:278

Comment 1

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

Looks like a specific KDE scenario and our layered code bug. KDE fails to release wl_buffer and we're committing/updating all layers after any change, not only changed ones. So we attach mFront buffer again event if it's not changed and potentially already attached.

Comment 2

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

Attachment: WIP: Bug 1972290 [Wayland/HDR] Update only changed properties

Comment 3

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

The patch here reduces rendering overhead but root cause is somewhere else - we can't reuse the same wl_buffer for root surface (tmp one) if it's already attached. It happens on popup map/unmap.

We may consider to create regular SW-WR for popups instead of composited ones.

Comment 4

[Phabricator Automation]

Comment on attachment 9495398 [details]
WIP: Bug 1972290 [Wayland/HDR] Update only changed properties

Revision D254140 was moved to bug 1973888. Setting attachment 9495398 [details] to obsolete.

Comment 5

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

Attachment: Bug 1972290 [Wayland/HDR] Allow to reuse already attached wl_buffer r?emilio

Comment 6

[Pulsebot]

Pushed by [e-mail]:
https://github.com/mozilla-firefox/firefox/commit/a62f1787fc3e
https://hg.mozilla.org/integration/autoland/rev/19e8f6786a2f
[Wayland/HDR] Allow to reuse already attached wl_buffer r=emilio

Comment 7

[Sandor Molnar[:smolnar]]

https://hg.mozilla.org/mozilla-central/rev/19e8f6786a2f