Not all combinations of connectors, crtcs and planes will work
on all hardware, so we need to test the pipelines before using
them.
BUG: 433107
CCBUG: 435265
There's no need to guard the code, gbm and drm format definitions
are the same. Using GBM_BO_FORMAT_* even caused bugs, as that is
an enum and not a proper format identifier.
BUG: 441253
IN_FORMATS contains information about which buffer formats can be
scanned out with a given drm plane. Using these plane-specific
format+modifier combinations as well as explicit modifiers in
general can yield bandwidth and performance improvements,
especially in multi-gpu systems.
When the last output gets disconnected, create a virtual output as a
placeholder until we have access to a physical output again. While
this placecholder never gets rendered to, with virtual outputs in
general that is possible (with gbm and qpainter atm) and can be done
for future use cases like wireless displays.
CCBUG: 420160
CCBUG: 438839
DrmPipeline is now what contains all the thing related to drm calls,
instead of DrmOutput. This allows for some more flexibility and tidies
the code up a bit. Additionally instead of rolling back changes if
presentation fails, changes are directly tested with atomic test only
commits.
This reverts commit 5a22deda3b.
We still need more work to finish the DrmPipeline. At the moment, there
are a few major issues, e.g. some outputs not turning on, output
transforms not working correctly, a crash when changing dpms mode.
Let's merge this change back once all major issues are fixed and after
more testing.
DrmPipeline is what now contains all the drm bits related to
modesetting and presentation, instead of that being in DrmOutput.
This gives a lot more freedom for managing drm resources and
enables far better usage of the atomic API with guaranteed
immutability for failed tests.
The value that the DrmCrtc::resIndex() function returns is better known
as "pipe index." This change renames the method to match the terminology
used by kernel developers and other compositor developers.
At the moment, the session code is far from being extensible. If we
decide to add support for libseatd, it will be a challenging task with
the current design of session management code. The goal of this
refactoring is to fix that.
Another motivation behind this change is to prepare session related code
for upstreaming to kwayland-server where it belongs.
If the file descriptor of the DRM device is greater than FD_SETSIZE, the
stack will be corrupted. However, it is highly unlikely that we ever hit
this case because DRM devices are opened at startup of kwin, so the file
descriptors should small.
In order to prevent the potential stack corruption, this change replaces
the usage of select() with poll().
Unlike select(), the api of poll() is much more sensible. Back 20 or so
years ago the main argument against poll() was that it's not implemented
by all platforms. But, nowadays, it's supported on all major platforms.
There are a couple of reasons not to use the lambda:
* It is unnecessary. The DrmGpu has the DRM file descriptor
* If a crash occurs somewhere in the lambda, the backtrace will be hard
to read
* Instead of processing events in the destructor of the DrmBackend
class, we should keep dispatching events without involving
QCoreApplication::processEvents() until all page flips are completed.
Once in a while, we receive complaints from other fellow KDE developers
about the file organization of kwin. This change addresses some of those
complaints by moving all of source code in a separate directory, src/,
thus making the project structure more traditional. Things such as tests
are kept in their own toplevel directories.
This change may wreak havoc on merge requests that add new files to kwin,
but if a patch modifies an already existing file, git should be smart
enough to figure out that the file has been relocated.
We may potentially split the src/ directory further to make navigating
the source code easier, but hopefully this is good enough already.
At the moment, our frame scheduling infrastructure is still heavily
based on Xinerama-style rendering. Specifically, we assume that painting
is driven by a single timer, etc.
This change introduces a new type - RenderLoop. Its main purpose is to
drive compositing on a specific output, or in case of X11, on the
overlay window.
With RenderLoop, compositing is synchronized to vblank events. It
exposes the last and the next estimated presentation timestamp. The
expected presentation timestamp can be used by effects to ensure that
animations are synchronized with the upcoming vblank event.
On Wayland, every outputs has its own render loop. On X11, per screen
rendering is not possible, therefore the platform exposes the render
loop for the overlay window. Ideally, the Scene has to expose the
RenderLoop, but as the first step towards better compositing scheduling
it's good as is for the time being.
The RenderLoop tries to minimize the latency by delaying compositing as
close as possible to the next vblank event. One tricky thing about it is
that if compositing is too close to the next vblank event, animations
may become a little bit choppy. However, increasing the latency reduces
the choppiness.
Given that, there is no any "silver bullet" solution for the choppiness
issue, a new option has been added in the Compositing KCM to specify the
amount of latency. By default, it's "Medium," but if a user is not
satisfied with the upstream default, they can tweak it.
At the moment, the gbm_device for the primary device is destroyed before
the EGLDisplay is destroyed. This results in a crash in Mesa.
In order to fix the crash, this change ensures that the EGLDisplay is
destroyed before the gbm device.
Xaver Hugl
Méven Car
Ömer Fadıl Usta
Vlad Zahorodnii
Aleix Pol
Adriaan de Groot
Bhushan Shah