Summary:
The NVIDIA implementation of glXSwapBuffers will, by default, queue up
to two frames for presentation before blocking. KWin's compositor,
however, assumes that calls to glXSwapBuffers will always block until
the next vblank when rendering double buffered. This assumption isn't
valid, as glXSwapBuffers is specified as being an implicit glFlush,
not an implicit glFinish, and so it isn't required to block. When this
assumption is violated, KWin's frame timing logic will
break. Specifically, there will be extraneous calls to
setCompositeTimer with a waitTime of 0 after the non-blocking buffer
swaps, dramatically reducing desktop responsiveness. To remedy this,
a call to glXWaitGL was added by Thomas Luebking after glXSwapBuffers
in 2015 (see bug 346275, commit
8bea96d701). That glXWaitGL call is
equivalent to a glFinish call in direct rendering, so it was a good
way to make glXSwapBuffers behave as though it implied a glFinish
call.
However, the NVIDIA driver will by default do a busy wait in glFinish,
for reduced latency. Therefore that change dramatically increased CPU
usage. GL_YIELD can be set to USLEEP (case insensitive) to change
the behavior and use usleep instead. When using the NVIDIA driver,
KWin will disable vsync entirely if GL_YIELD isn't set to USLEEP
(case sensitive, a bug in KWin).
However, the NVIDIA driver supports another environment variable,
__GL_MaxFramesAllowed, which can be used to control how many frames
may be queued by glXSwapBuffers. If this is set to 1 the function
will always block until retrace, in line with KWin's expectations.
This allows the now-unnecessary call to glXWaitGL to be removed along
with the logic to conditionally disable vsync, providing a better
experience on NVIDIA hardware.
Reviewers: #kwin, davidedmundson, zzag
Reviewed By: #kwin, davidedmundson, zzag
Subscribers: kwin, davidedmundson, zzag
Tags: #kwin
Differential Revision: https://phabricator.kde.org/D19867
Summary:
Otherwise it thinks outputs couldn't be created and it leaves.
Problem is that after this fix it crashes when it actually does things with:
kwin_wayland: /home/apol/devel/frameworks/kwin/composite.cpp:646: void KWin::Compositor::aboutToSwapBuffers(): Assertion `!m_bufferSwapPending' failed.
Test Plan: As said, it proceeds correctly and crashes elsewhere
Reviewers: #kwin, davidedmundson
Reviewed By: #kwin, davidedmundson
Subscribers: davidedmundson, kwin
Tags: #kwin
Differential Revision: https://phabricator.kde.org/D19921
Summary:
The NVIDIA implementation of glXSwapBuffers will, by default, queue up
to two frames for presentation before blocking. KWin's compositor,
however, assumes that calls to glXSwapBuffers will always block until
the next vblank when rendering double buffered. This assumption isn't
valid, as glXSwapBuffers is specified as being an implicit glFlush,
not an implicit glFinish, and so it isn't required to block. When this
assumption is violated, KWin's frame timing logic will
break. Specifically, there will be extraneous calls to
setCompositeTimer with a waitTime of 0 after the non-blocking buffer
swaps, dramatically reducing desktop responsiveness. To remedy this,
a call to glXWaitGL was added by Thomas Luebking after glXSwapBuffers
in 2015 (see bug 346275, commit
8bea96d701). That glXWaitGL call is
equivalent to a glFinish call in direct rendering, so it was a good
way to make glXSwapBuffers behave as though it implied a glFinish
call.
However, the NVIDIA driver will by default do a busy wait in glFinish,
for reduced latency. Therefore that change dramatically increased CPU
usage. GL_YIELD can be set to USLEEP (case insensitive) to change
the behavior and use usleep instead. When using the NVIDIA driver,
KWin will disable vsync entirely if GL_YIELD isn't set to USLEEP
(case sensitive, a bug in KWin).
However, the NVIDIA driver supports another environment variable,
__GL_MaxFramesAllowed, which can be used to control how many frames
may be queued by glXSwapBuffers. If this is set to 1 the function
will always block until retrace, in line with KWin's expectations.
This allows the now-unnecessary call to glXWaitGL to be removed along
with the logic to conditionally disable vsync, providing a better
experience on NVIDIA hardware.
Reviewers: #kwin, davidedmundson, zzag
Reviewed By: #kwin, davidedmundson, zzag
Subscribers: kwin, davidedmundson, zzag
Tags: #kwin
Differential Revision: https://phabricator.kde.org/D19867
Summary:
This patch rewrites large parts of the Wayland platform plugin, in order to
facilitate the testing of multi output behavior in nested KWin sessions.
For that a new class WaylandOutput is introduced, which is based on
AbstractOutput and by that shares functionality with our virtual and DRM
platform plugins.
The EGL/GBM and QPainter backends have been remodelled after the DRM one,
sharing similiarities there as well now.
Pointer grabbing has been rewritten to support multiple outputs, now using
pointer locking instead of confining and drawing in this case onto a sub-
surface, which get dynamically recreated in between the different output
surfaces while the cursor is being moved.
Window resizing is possible if host supports xdg-shell, but currently the
mode size does not yet fill the new window size.
The number of outputs can be set by command line argument `--output-count`,
scaling is also supported by setting the argument `--scale`.
Further steps could be:
* Enabling automatic fill of resized windows via Wayland mode change
* Multiple diverging initial sizes and scale factors for mulitple outputs
**Watch it in action:** https://youtu.be/FYItn1jvkbI
Test Plan: Tested it in live session.
Reviewers: #kwin, davidedmundson
Reviewed By: #kwin, davidedmundson
Subscribers: davidedmundson, zzag, kwin
Tags: #kwin
Differential Revision: https://phabricator.kde.org/D18465
Summary:
If hardware cursor support is not available when using the drm backend for
Wayland compositing, the software cursor texture will not be updated when the
cursor image changes, and it will still be drawn when no cursor image is set
(such as when running a full-screen game). Furthermore, the drmModeSetCursor
and drmModeMoveCursor functions will still be unnecessarily called when the
cursor is moved or hidden.
To correct this, SceneOpenGL should connect Platform::cursorChanged as opposed
to Cursor::cursorChanged to its texture update function, as only the former
will be emitted when the cursor is updated and the compositor should check if
the cursor is hidden and the software cursor image is not null before rendering
it. DrmBackend::moveCursor and DrmBackend::hideCursor should also return
immediately if using a software cursor.
Reviewers: #kwin, davidedmundson
Reviewed By: #kwin, davidedmundson
Subscribers: kwin
Tags: #kwin
Differential Revision: https://phabricator.kde.org/D18376
Summary: Like we do on inverted mode
Test Plan: Not much, because we don't support portrait mode on intel hardware just yet. I can see the cursor properly rotated before drm tells us to fall back to the previous configuration.
Reviewers: #kwin, #plasma, graesslin
Reviewed By: #kwin, #plasma, graesslin
Subscribers: zzag, kwin
Tags: #kwin
Differential Revision: https://phabricator.kde.org/D19710
Summary:
My change in 304528e80b has been reported to
cause issues for PostMarketOS https://gitlab.com/postmarketOS/pmaports/issues/204
I originally submitted the change to activate Framebuffer devices
as secondary framebuffer devices are not on by default, (such as a
secondary DisplayLink device). Changing attributes usually forced
some framebuffer devices to turn on, however I have found a more
accurate way to turn on framebuffer devices, without messing with
the color layout.
Reviewers: #kwin, davidedmundson, graesslin, zzag
Reviewed By: #kwin, zzag
Subscribers: kwin
Tags: #kwin
Differential Revision: https://phabricator.kde.org/D19663
Summary:
Package structure plugins are not that special so move them to
the corresponding directory.
Test Plan: Compiles.
Reviewers: #kwin
Subscribers: kwin
Tags: #kwin
Differential Revision: https://phabricator.kde.org/D18685
Summary:
When Compositor finishes compositing, it destroys EffectsHandlerImpl,
which in its turn tries to unload all effects. But there is a problem...
EffectsHandlerImpl has platform-specific hooks to ungrab keyboard and
also stop mouse interception. Given that any call made to a virtual function
in the destructor of a base class(EffectsHandlerImpl) won't go to a derived
class(EffectsHandlerImplX11), keyboard won't be ungrabbed even if effect
that grabbed it is already gone.
BUG: 399572
Reviewers: #kwin, davidedmundson
Reviewed By: #kwin, davidedmundson
Subscribers: kwin
Tags: #kwin
Differential Revision: https://phabricator.kde.org/D19178
Summary:
Deleted has lifetime bounded to effects and scene.
We can't discard Deleted before EffectsHandler is destroyed because
effects that referenced it may call unrefWindow.
On the other hand, the fact that Deleted may outlive scene doesn't
make sense because Deleted exist purely for animation purposes and
nothing more.
This change arranges lifetime of Deleted so it's more reasonable.
Reviewers: #kwin, davidedmundson
Reviewed By: #kwin, davidedmundson
Subscribers: graesslin, kwin
Tags: #kwin
Differential Revision: https://phabricator.kde.org/D18914
Summary:
On Wayland we cannot switch from OpenGL to QPainter compositor as this
would break any running OpenGL application. KWin registers it's
EGLDisplay to Wayland and without OpenGL this doesn't make sense any
more. We are not able to render OpenGL buffers in the QPainter
compositor.
While it's theoretically possible to switch from QPainter to OpenGL it
doesn't make any sense for the same reason. Any running OpenGL
application would be using llvmpipe and could not be switched to proper
OpenGL.
This change stores the selected compositing type in Platform and the
implementations can use it to restrict the supported compositors. On X11
we don't need this, all other Platforms implement the restriction. Thus
it's no longer possible to switch the backends at runtime.
Test Plan:
Adjusted tests run, no runtime test as gui doesn't support
switching to QPainter anyway.
Reviewers: kwin
Tags: #kwin
Differential Revision: https://phabricator.kde.org/D19084
Summary:
Using the Wayland protocol for OpenGL is no longer used or useful. KWin
internal windows only use OpenGL through QtQuick. We either have the
sharing context (KWin uses OpenGL for compositing) or we have the
QPainter compositor which also turns QtQuick to use software renderer.
Thus a situation where the Wayland platform context is useful doesn't
exist any more. Removing it helps getting the QPA plugin Wayland free.
Test Plan:
Run nested KWin triggering Outline once with OpenGL and once
with QPainter compositor.
Reviewers: #kwin
Subscribers: kwin
Tags: #kwin
Differential Revision: https://phabricator.kde.org/D19012
Summary:
Despite plasma frameworks doing it's own scaling with fonts, it's been
requested to use kwin/wayland scaling.
Like DRM, when kscreen is not used, scale value is loaded from a config
file.
Config format is
[HWComposerOutputs][0]
Scale=N
The 0 is to map similarly to DRM and support multi-screen, but with a
screen index
rather than a UUID based on EDID.
Because we don't support multi screen this is always 0 for now.
Test Plan: Ran with the config value unset and with the config value at
Scale=3.
Reviewers: #kwin, romangg
Reviewed By: #kwin, romangg
Subscribers: bshah, romangg, nicolasfella, zzag, kwin
Tags: #kwin
Differential Revision: https://phabricator.kde.org/D18810
Summary:
The virtual platform did not actually present content and spin the
compositor. Let's do this from now on , first for the EGL GBM backend.
Test Plan: All tests pass according to Vlad
Reviewers: #kwin, zzag
Reviewed By: #kwin, zzag
Subscribers: graesslin, zzag, kwin
Tags: #kwin
Differential Revision: https://phabricator.kde.org/D18436
Summary:
This gives us XdgOutput support a requirement now that wayland server
reports that we support it and Qt5.12 then expects it to be sent.
It also provides an easy starting point for supporting wayland scaling
and a slight step towards multi-output.
Test Plan:
Ran on phone. Got output. Turning screen on and off works.
The double tap doesn't work great, but that was true before
(via bshah's hack to build on the current state)
Reviewers: #kwin, romangg
Reviewed By: #kwin, romangg
Subscribers: romangg, nicolasfella, zzag, kwin
Tags: #kwin
Differential Revision: https://phabricator.kde.org/D18808
Summary: Required for kwin_wayland to find it's own plugin.
Test Plan: My kwin_wayland was crashing and gdb showed wrong plugin being loaded
Reviewers: #kwin
Subscribers: kwin
Tags: #kwin
Differential Revision: https://phabricator.kde.org/D19010
Summary:
We have a mix of different doxygen comment styles, e.g.
/*!
Foo bar.
*/
/**
* Foo bar.
*/
/** Foo bar.
*/
/**
* Foo bar.
*/
/**
* Foo bar.
**/
To make the code more consistent, this change updates the style of all
doxygen comments to the last one.
Test Plan: Compiles.
Reviewers: #kwin, davidedmundson
Reviewed By: #kwin, davidedmundson
Subscribers: kwin
Tags: #kwin
Differential Revision: https://phabricator.kde.org/D18683
Summary:
Lanczos filter shouldn't be in KWin core because it's quite specific about
rendering backend.
Test Plan: Windows in the desktop grid look normal.
Reviewers: #kwin, davidedmundson
Reviewed By: #kwin, davidedmundson
Subscribers: mart, kwin
Tags: #kwin
Differential Revision: https://phabricator.kde.org/D18774
Summary:
KWindowSystem provides a plugin interface to have platform specific
implementations. So far KWin relied on the implementation in
KWayland-integration repository.
This is something I find unsuited, for the following reasons:
* any test in KWin for functionality set through the plugin would fail
* it's not clear what's going on where
* in worst case some code could deadlock
* KWin shouldn't use KWindowSystem and only a small subset is allowed
to be used
The last point needs some further explanation. KWin internally does not
and cannot use KWindowSystem. KWindowSystem (especially KWindowInfo) is
exposing information which KWin sets. It's more than weird if KWin asks
KWindowSystem for the state of a window it set itself. On X11 it's just
slow, on Wayland it can result in roundtrips to KWin itself which is
dangerous.
But due to using Plasma components we have a few areas where we use
KWindowSystem. E.g. a Plasma::Dialog sets a window type, the slide in
direction, blur and background contrast. This we want to support and
need to support. Other API elements we do not want, like for examples
the available windows. KWin internal windows either have direct access
to KWin or a scripting interface exposed providing (limited) access -
there is just no need to have this in KWindowSystem.
To make it more clear what KWin supports as API of KWindowSystem for
internal windows this change implements a stripped down version of the
kwayland-integration plugin. The main difference is that it does not use
KWayland at all, but a QWindow internal side channel.
To support this EffectWindow provides an accessor for internalWindow and
the three already mentioned effects are adjusted to read from the
internal QWindow and it's dynamic properties.
This change is a first step for a further refactoring. I plan to split
the internal window out of ShellClient into a dedicated class. I think
there are nowadays too many special cases. If it moves out there is the
question whether we really want to use Wayland for the internal windows
or whether this is just historic ballast (after all we used to use
qwayland for that in the beginning).
As the change could introduce regressions I'm targetting 5.16.
Test Plan:
new test case for window type, manual testing using Alt+Tab
for the effects integration. Sliding popups, blur and contrast worked fine.
Reviewers: #kwin
Subscribers: kwin
Tags: #kwin
Differential Revision: https://phabricator.kde.org/D18228
Summary:
Mesa requires XESetWireToEvent xlib callbacks to be called
when DRI2 is used. This is done by the GLX integration in
the Qt's xcb plugin, but Qt 5.12 initializes the GLX integration
only when required, e.g. when a window with OpenGL support is
created or when availability of OpenGL is checked.
So force initialization of the GLX integration by calling
QOpenGLContext::supportsThreadedOpenGL().
https://codereview.qt-project.org/#/c/6557/https://bugzilla.opensuse.org/show_bug.cgi?id=1120090
Reviewers: #kwin, graesslin
Reviewed By: #kwin, graesslin
Subscribers: davidedmundson, graesslin, fvogt, filipf, kwin
Tags: #kwin
Differential Revision: https://phabricator.kde.org/D18366
Summary:
Currently, when the lanczos filter attempts to release acquired resources,
the backend is already gone. To fix that we have to destroy the filter
together with SceneOpenGL2. At that moment the backend is still alive.
BUG: 403370
FIXED-IN: 5.15.0
Reviewers: #kwin, davidedmundson
Reviewed By: #kwin, davidedmundson
Subscribers: kwin
Tags: #kwin
Differential Revision: https://phabricator.kde.org/D18367
Summary:
So far we didn't try to do high DPI on kwin internal windows, such as
the user context menu and tab bars and whatever.
Due to wayland scaling they were the correct phyiscal size but upscaled.
This patch fixes our QPA to enable Qt's high-dpi support.
BUG: 402853
Note icons are still low res. This is because the global
QGuiApplication::devicePixelRatio which is the max of all connected
screens is static for the duration of the app. QIcon uses this when
determining the DPR to use. This will require a Qt change.
Test Plan:
Ran at 2x on my normal DPI screen (as that's easier to see anything)
* User action menu is high DPI
* Window deco tooltips are still fine
* Tab switcher is high DPI
* Overlay in present windows Desktop grid are still ok
Reviewers: #kwin
Subscribers: zzag, kwin
Tags: #kwin
Differential Revision: https://phabricator.kde.org/D18042
Summary:
Instead we depend on the surfaceless platform for which we recently
added support. Thus the plugin does not need to use gbm and udev
anymore. So simplifies a lot.
Test Plan: ctest (prior to breaking change) passes
Reviewers: #kwin
Subscribers: kwin
Tags: #kwin
Differential Revision: https://phabricator.kde.org/D18160
Summary:
Even though the buffer is passed through a side channel some properties
we might use are double-buffered on the surface commit.
Test Plan: Used by unit test InternalWindowTest::testScale in linked commit
Reviewers: #kwin, graesslin
Reviewed By: #kwin, graesslin
Subscribers: kwin
Tags: #kwin
Differential Revision: https://phabricator.kde.org/D18085
Summary: For unit test purposes
Test Plan: Used by unit test InternalWindowTest::testScale in linked commit
Reviewers: #kwin, graesslin
Reviewed By: #kwin, graesslin
Subscribers: kwin
Tags: #kwin
Differential Revision: https://phabricator.kde.org/D18086
Summary:
Thus we don't require vgem any more which fails with latest mesa.
Kepping the vgem variant as fallback for older systems. See T10245
Test Plan:
SceneOpenGL Test passes without init vgem, not tested with
latest mesa
Reviewers: #kwin
Subscribers: kwin
Tags: #kwin
Differential Revision: https://phabricator.kde.org/D17980
Summary:
This ensures that KWin gets the same keyboard layout config as is
configured in the session and not a default config.
BUG: 402764
FIXED-IN: 5.14.5
Test Plan:
Xephyr to verify the config is applied, otherwise completely
untested. I don't have an X session.
Reviewers: #kwin
Subscribers: kwin
Tags: #kwin
Differential Revision: https://phabricator.kde.org/D17967
Summary:
This change inits XInput extension, listens for touch events and
forwards them to our platform API. Thus touch events are forwarded on a
nested wayland session on X11.
Please note that I only tested this change on Xwayland.
Test Plan: Run nested kwin_wayland with two outputs and looked into debug console
Reviewers: #kwin
Subscribers: kwin
Tags: #kwin
Differential Revision: https://phabricator.kde.org/D17369
Summary:
This is a preparation step to move this to ECM (or possibly CMake),
which makes sense now that qtbase is using this too (for Qt6).
Reviewers: #build_system, cgiboudeaux
Reviewed By: cgiboudeaux
Subscribers: cgiboudeaux, kwin
Tags: #kwin
Differential Revision: https://phabricator.kde.org/D17547
Summary:
This brings KWin a step closer to be run from build dir without having
to install at all. The integration tests are adjusted so that the
virtual platform is still found which makes the code be closer to what
is used in normal kwin_wayland.
Test Plan: ctest passes, manually verified correct plugin is loaded
Reviewers: #kwin
Subscribers: kwin
Tags: #kwin
Differential Revision: https://phabricator.kde.org/D17388
Summary:
When suspending compositing, SceneOpenGLShadow cannot cleanup cached
decoration shadow textures because the effects handler is already gone.
This sometimes can result in a crash when running kwin_x11 --replace
(we're hitting an assert statement).
To fix that, let's use the scene instead of the effects handler for
making the OpenGL context current.
Test Plan:
No longer hit the assert statement:
ASSERT: "m_cache.isEmpty()" in file /home/vlad/Workspace/KDE/src/kde/workspace/kwin/plugins/scenes/opengl/scene_opengl.cpp, line 2025
Application::crashHandler() called with signal 6; recent crashes: 1
QCoreApplication::applicationFilePath: Please instantiate the QApplication object first
KCrash: crashing... crashRecursionCounter = 2
KCrash: Application Name = kwin_x11 path = /home/vlad/Workspace/KDE/usr/bin pid = 5407
KCrash: Arguments: /home/vlad/Workspace/KDE/usr/bin/kwin_x11 --replace
Reviewers: #kwin, graesslin
Reviewed By: #kwin, graesslin
Subscribers: kwin
Tags: #kwin
Differential Revision: https://phabricator.kde.org/D17306
Summary:
The idea of opening the compositing kcm was to show a warning. But that
broke quite some time ago without noticing. We had two ways:
* pass through --args command line argument
* use dbus call to already open kcm
Neither of the two ways is working. The kwincompositing doesn't parse
the arguments and the dbus interface doesn't exist any more.
Following the advice to remove functionality nobody noticed that it is
broken, this is removed with this change. This probably broke with
introducing the new KCM which happened IIRC for Plasma 5.0.
BUG: 393845
FIXED-IN: 5.14.0
Reviewers: #kwin, #plasma
Subscribers: kwin
Tags: #kwin
Differential Revision: https://phabricator.kde.org/D13235
Summary:
The current refresh rate is stored in OutputInterface. Move the getter
therefore in the AbstractOutput class.
Reviewers: #kwin, davidedmundson
Reviewed By: #kwin, davidedmundson
Subscribers: kwin
Tags: #kwin
Differential Revision: https://phabricator.kde.org/D16793
Summary: Removes unneeded includes and reorders the code in a sensible way.
Reviewers: #kwin
Subscribers: davidedmundson, kwin
Tags: #kwin
Differential Revision: https://phabricator.kde.org/D16792
Summary:
Set the mode in AbstractOutput and call into the plugin
for final change on hardware.
Reviewers: #kwin, davidedmundson
Reviewed By: #kwin, davidedmundson
Subscribers: davidedmundson, kwin
Tags: #kwin
Maniphest Tasks: T10016
Differential Revision: https://phabricator.kde.org/D16786
Jimi Huotari
Aleix Pol
Erik Kurzinger
Volker Krause
Roman Gilg
Vlad Zagorodniy
Nerdopolis Turfwalker
Laurent Montel
Martin Flöser
David Edmundson
Alexander Volkov
l10n daemon script